The inspiration for this piece is twofold. First, a Matt Yglesias essay “Why Silicon Valley hasn’t done more for most Americans” that was shared on my Twitter feed (with a great deal of contempt) by Malcolm Harris, author of Palo Alto: A History of California, Capitalism, and the World. “I encourage people to read that post, which is written at a much much lower intellectual level than I could have imagined,” Harris writes. And he’s right!

Second, I just got back from San Francisco and have been thinking a lot about the city since. And so, I’ll try to lay out in this essay, Yglesias’s piece purports to be about why San Francisco isn’t a megacity, but it’s not actually interested in thinking about the reasons why.

So what is Yglesias’s piece about? He riffs off a Paul Krugman blog comparing American and European living standards—specifically an aside suggesting Europe shouldn’t envy America’s tech sector since it produces oligarchs who undermine politics—and tries to insist our tech sector is both worth envying, but also horrible at distributing its benefits beyond a sliver of workers and elites. The rest of America largely gets ancillary benefits (e.g. Google Search, on-demand convenience). Why? Well, dear reader, it’s because of housing.

To make his case, Yglesias looks to earlier economic booms in American history. Chicago’s steel boom, Detroit’s auto boom, these led to population explosions that pulled waves of migration in for work. Agglomeration around these industries created boomtowns and widely distributed prosperity, he argues, so we should be asking why we don’t see this to the same extent around San Francisco’s tech boom. Yglesias finds it even more puzzling that three of the biggest companies in the world by market share are headquartered in San Francisco, but San Francisco isn’t the biggest metropolitan area in the world (or even the state of California). Instead, we see populations that would’ve made up waves of migration during this boom choose to live elsewhere because San Francisco simply doesn’t build enough housing for them.

If San Francisco had built enough housing, he reasons millions would’ve moved to the Bay Area instead of the Sunbelt and enjoyed higher incomes as tech workers (or as service and construction workers in a boom town) And following this counterfactual’s logic, we also would’ve seen America’s manufacturing capabilities mobilized to build out a Bay Area megacity that would’ve certainly distributed wealth more broadly and deeply across the country. Our housing supply, you see, is the bottleneck—one billion buildings now!

It’s a straightforward enough argument, but is it right? Krugman believes the problem traces back to what the tech sector is, Yglesias believes it boils down to where the tech sector lives. I think both of them miss the mark (Yglesias much more than Krugman).

To his credit, Krugman calls the political influence of our tech oligarch class a “negative externality” but that’s still a little too soft, implying the harms they produce are incidental to an otherwise productive and desirable process (like pollution). The key mechanisms of our tech platforms (zero-sum competition, network effects, a preference for consolidation, etc.) are integral, not incidental, to how the sector generates profits. The fortunes and activities that distort markets and politics are direct and intentional consequences. That the structure of our various tech firms and products tends to stabilize around that of a monopoly or oligopoly isn’t a market failure, but a market feature—part and parcel of the sector’s ambition to serve as infrastructure inseparable from the rhythms of daily life.

Yglesias goes out of his way to treat these outcomes as accidents and policy failures, getting the basic history and periodization (as Harris points out, the Midwest and Bay Area booms are not that similar) wrong—which matters because it has implications for recommendations about what policy surrounding the tech sector should look like!

Inconvenient Facts

The heart of Yglesias’s argument rests on his analogies between the Midwest industrial boom and the Silicon Valley tech boom: they’re both economically and geographically concentrated, so they should produce the same kind of boomtown. Any difference we observe, then, can be explained by policy. But what about the actual economic activity driving each boom?

We can pretty quickly identify a few structural conditions that not only defined the industrial boom, but are unlikely to be replicated in a tech boom.

Industrial production needed masses of workers

One place to start is to ask yourself how profits and prosperity are actually distributed. As anyone who has had a job can tell you, what you get paid is downstream of some compromise made at some point between workers and the people they work for. At its simplest, we might say the shape, form, and substance of that agreement is determined in part by the power of the negotiating parties. Certain workers at certain points of the production process might be in better positions to realize compromises that better suit their interests.

In Forces of Labor, sociologist Beverly J. Silver thinks of this structural power as marketplace bargaining power and workplace bargaining power:

Marketplace bargaining power can take several forms including (1) the possession of scarce skills that are in demand by employers, (2) low levels of general unemployment, and (3) the ability of workers to pull out of the labor market entirely and survive on nonwage sources of income. Workplace bargaining power, on the other hand, accrues to workers who are enmeshed in tightly integrated production processes, where a localized work stoppage in a key node can cause disruptions on a much wider scale than the stoppage itself. Such bargaining power has been in evidence when entire assembly lines have been shut down by a stoppage in one segment of the line, and when entire corporations relying on the just-in-time delivery of parts have, been brought to a standstill by railway workers’ strikes.

(Silver’s scheme comes from the late sociologist Erik Olin Wright’s theory of worker power, specifically his 2000 essay “Working-Class Power, Capitalist-Class Interests, and Class Compromise.” which seeks to try to map out the conditions where class compromises between workers and those who own their workplaces (capitalists) are positive, benefitting the material interests of workers and capitalists.)

Industrial workers (like those in the auto industry) had both workplace and marketplace bargaining power. Their labor was essential enough that production could be brought to a screeching halt with strikes, and the laborers were numerous enough that they could form a political bloc with that leverage. This also means we can trace a significant part of the broadly shared prosperity associated with industrial workers of that era (and which Yglesias pines for throughout his piece) comes to us from struggles groups like the UAW—which didn’t emerge because lots of people happened to live near factories, but because the production process concentrated workers who could exercise enough structural leverage of capital to force this or that compromise.

All of this also happens within a legal framework (the Wagner Act, the NLRB) that, for a time, permitted and even encouraged collective bargaining. It’s the result of a political struggle, not an automatic consequence of agglomeration. Still, that brief window is only partially taken advantage of. Taft-Hartley, the anti-communist purges of the Congress of Industrial Organization purges, the Red Scare(s) more generally—these were all victories for those eager to claw back profit shares given up to organized labor by strangling those collectives and the compact they achieved. And the postwar compact between labor and capital, however golden it may have been, was still a racial one—American apartheid was premised on excluding Black people from home ownership and wealth accumulation, as well as creating a dual labor market that confined Black workers to the worst set of jobs and working conditions.

So assuming Yglesias doesn’t want to emulate the exclusionary nature of the Midwest boom, we’re left with the structural elements—the nature of (industrial) production and the proactive redistribution of prosperity via political struggle—that are fundamentally different from those of the Bay Area boom. Does it make much sense to boil this down to where workers choose to live?

Domestically there was a closed circuit

Structural power enjoyed by industrial workers along the production process was also reinforced by the fact that their firms were, in a real way, fixed to one spot. Auto plants were enormous fixed-capital investments beyond the assembly line itself. An automobile required production inputs like steel, rubber, and flat glass before it was finally assembled, with supply chains criss-crossing the country and in turn requiring their own node and network of capital-intensive extraction, processing, and logistical infrastructure.

To take one example relevant to Yglesias’s argument, look to the Ford River Rouge complex in Dearborn, Michigan—first envisioned as a bird sanctuary, then mobilized as an arms manufacturer during World War One, then an automobile factory (and once the world’s largest factory at some 2,000 acres, boasting some 100,000 employees at its peak). The sprawling industrial park was vertically integrated, allowing it to take raw inputs (iron ore and timber from northern Michigan, coal from Pennsylvania, rubber from Latin American plantations), produce its own steel and glass and electricity, and finally assemble an automobile.

There was not much of a credible threat of moving the industrial park or any of its key facilities along the production process—which was partly why firms like Ford reacted so viciously to the threat of collective bargaining (particularly to recognizing the United Auto Workers union). In 1932, thousands of unemployed auto workers marched to present demands to the company and were met with Ford’s private security alongside dozens of police officers who let loose with tear gas, water cannons, and bullets in what would be called the Ford Massacre (hundreds were fired for participation, five people were killed). Five years later, we have the Battle of the Overpass where Ford rolled out private security again to beat up UAW organizers outside of the Rouge River complex (pictures of the beatdown ended up forcing Ford to the negotiating table).

Capital’s immobility, alongside the structural power workers enjoyed over production, was an integral pillar of the capital-labor accord. Yglesias is right that there is a geographic concentration, but it was at various points of the nation itself that meant strikes could shut down production because of a lack of alternative sites when it came to extracting raw materials, crafting components, assembling the product, or eventually distributing the finished goods.

Another piece of here is that the workers are also consumers and domestic markets were designed around this contingency. Henry Ford and his business empire is emblematic here: from our benevolent titan of industry came Fordism, where factories should be organized such that they produce as much as possible for the lowest cost possible so as to compel as much consumption as possible; that higher wages for industrial workers meant higher consumption, higher sales, higher profits, and lower turnover. If owners of capital wanted to preserve their workforce and domestic markets, they had an interest—a reluctant one, constantly contested and extracted over decades of negotiations and strikes and legal battles and political struggle, but real nonetheless—in maintaining wages at a level high enough to help sustain this Fordist wage-consumption loop.

Production yielded wages yielded consumption yielded revenue yielded reinvestment yielded more production, and so on and so on. A relatively closed national economy was one where producers, workers, and consumers were relatively immobile, subject to broadly similar labor laws, spending the same currency, buying the same products from the same firms. And so like a snake eating its own tail, capital—which needs labor to produce and consume—and labor—which needs capital to reinvest—are brought close enough to reach an unsteady peace, so long as they were all stuck in the same room.

The breakdown of those conditions are a key part of why we see such a sharp divergence between wages and productivity that has come to define our society today. Or as the Economic Policy Institute puts it in a primer on the gap between productivity and worker compensation:

Starting in the late 1970s policymakers began dismantling all the policy bulwarks helping to ensure that typical workers’ wages grew with productivity. Excess unemployment was tolerated to keep any chance of inflation in check. Raises in the federal minimum wage became smaller and rarer. Labor law failed to keep pace with growing employer hostility toward unions. Tax rates on top incomes were lowered. And anti-worker deregulatory pushes—from the deregulation of the trucking and airline industries to the retreat of anti-trust policy to the dismantling of financial regulations and more—succeeded again and again.

We can quickly throw in a few other developments worth noting as they’ve made it so the average American and our society at large enjoy lower prices and better goods, but done little to address the deeply unequal distribution of prosperity here. Containerization, which made it cheap enough to ship partially assembled products across oceans; free trade agreements, which opened low-wage labor markets across Latin America and East Asia to American capital; the proliferation of export-processing zones that sparked a global wave of outsourcing (and an erosion of worker bargaining power) by offering low wages, tax exemptions, and little to no labor enforcement in hopes of attracting attracting foreign direct investment.

Where does this leave our compact? The wage-consumption loop frays—I can produce in one country, pay those workers starvation wages, then sell the product to another country with purchasing power (ideally at the highest possible price). A world where capital is mobile but labor now forced to immobility is a world where agreements contingent on the former sharing that condition suddenly dissolve. Where workplace and marketplace bargaining power are insufficient to compel owners of capital to stay in line, where negotiated wages and contracts dissolve—along with the sort of prosperity Yglesias believes is a natural consequence of economic booms.

Results may vary

The core goods of the industrial booms were manufactured then transported—tethered to a place upon which supply chains entangling the globe converge, employing (or contracting [or contracting out the contracting of]) all across the world. What does the web look like for the Bay Area’s products?

Take Apple, which has close to a $3.7 trillion market capitalization. It’s got about 166,000 employees, a sizable chunk of which work in its retail stores. Relatively speaking, its value-generating core is tiny compared to the wealth it produces. In Silver’s parlance, its core technical workers have high marketplace bargaining power, yes, because their skills might be relatively scarce but they have little to no workplace bargaining power. A bitter lesson we’ve been forced to learn over and over again is that no single group of workers can shut down these tech firms (not for lack of trying!), let alone organize them into a collective group large enough to seriously try at doing so. Partly because of the division of labor (amongst employees, but also between employees and contractors). Partly because of the limits of labor law (which has only been whittled down over the years). Partly because of the sheer size of the workforce. But partly because this tech boom has produced a labor aristocracy—well-paid, individuated, and ideologically aligned (by carrot of compensation or stick of constant layoffs)—a move that makes complete sense from the perspective of these firms, who have nothing to lose and everything to gain by eviscerating the “mass worker” as a subject. When it comes to workers who are not in that rarefied elite—the great mass of service workers who clean, cook, and care for the sector’s labor aristocrats—Silver herself notes that the bargaining power of today’s low-wage service workers is actually closer to that of 19th century textile workers than 20th century auto workers. These workers are left wanting for structural power, dependent on associational power and shifting alliances as opposed to any real leverage over any particular point of production. Do they enjoy higher wages because of their proximity to the “ground zero for global innovation,” as Yglesias suggests?

With the abandonment of the Fordist loop, we can trace out a path a firm like Apple sees its value is captured in Cupertino, but its physical products are assembled by Foxconn workers in conditions familiar to workers trapped on the wrong side of the Wagner Act. Cancer clusters, consistent injuries and maimings, persistent exploitation, death from overwork, and so on and so forth. We can dive further into the supply chain to find more deplorable conditions also familiar to workers before the Wagner Act (such as slaves and child workers in Central Africa mining for minerals that go in our advanced electronics). It is true that, as Yglesias writes, tech’s market “scales pretty easily to become truly global” but mainly because it’s free to decouple industry from any sort of obligation to where its headquartered or where its production is placed.

Still, the labor-capital compact’s destruction isn’t just a story centered on individual policy choices or a long chain of deregulations and reregulations in favor of owners of capital. In The Long Twentieth Century, Giovanni Arrighi (Silver’s husband) offers a structural periodization to understand the decline of American hegemony: the postwar order that made the Fordist loop was one phase of American hegemony organized around material expansion—around the production of material goods, infrastructure, and the employment of workers at scale. What kicked off in the 1970s wasn’t simply a policy pivot, but a seismic shift towards financial expansion as the hegemon showed signs of infirmity and as the engine powering its ascension began to sputter. And so, as manufacturing declines and disappears we see capital pour into finance, seeking returns through increasingly abstract instruments and securities and business models over labor-intensive, geographically fixed material production.

We can use this to reframe the tech boom’s relationship to deindustrialization. What spurred the globalization that enabled tech’s scaling via offshoring, the creation of global supply chains, and the prying open of low-wage labor markets? Financialization! Financialization and the creation of a global order privileging American capital’s ability to benefit from production anywhere, anytime, uncoupled from domestic production. This decoupling helped midwife what would become a new institutional life-world, if you will, complete with venture capital funding schemes, equity-based compensation, and corporate governance regimes organized around shareholder value—a life-world where a certain species of corporation might thrive, focused on capturing enormous value while getting away with the minimal number of domestic workers. And so the tech boom doesn’t happen despite deindustrialization, it spins out of it—born within an economic and political order already restructured (and eager to continue to restructure itself) into forms that maximize returns to capital while minimizing any offerings to labor.

Would things have been different if San Francisco built millions of housing units to absorb waves of domestic migration? Probably not. The firms at the center of the Bay Area boom are structurally distinct from the firms at the center of the Midwest’s boom. They’re not anchored to massive fixed-capital investments in industrial parks, they’re not employing hundreds of thousands of domestic workers along a production process vulnerable to stoppages and strikes, they don’t need people who buy the products to be the people who make them, they don’t necessarily have an interest in positive class compromise that props up wages. It also helps that software doesn’t need to be extracted, processed, shipped, or assembled across the country. All of this is a bit deeper than municipal zoning, no? This is about reorganizing American society and its economy around a different logic of capital accumulation.

This isn’t to say that there are zero agglomeration effects at play here, just that they are largely limited to a small, highly compensated group and the financiers backing them. Scholars like AnnaLee Saxenian might point out that Silicon Valley’s geographic concentration has created real clusters and networks, but at the same time this largely was in service of circulating a narrow elite of technical workers between firms, through informal knowledge networks, and through enmeshed venture capital fund portfolios where friends, co-workers, mentees, and ideological comrades enrich one another.

None of this, in of itself, creates the sort of broad multiplier effects for a wider urban population that Yglesias seems to think it does. In fact, if we spend a little time just looking at the Bay Area’s history then that much becomes even more painfully clear.

Geographer Richard Walker is a great resource here (specifically his chapter “Industry Builds Out the City” in the edited volume Manufacturing Suburbs, as well as his essay “The strange case of the Bay Area”) as he’s taken great pains to reconstruct the Bay Area’s industrial geography across the late 19th and early 20th centuries—specifically the period when this region had the sort of manufacturing base Yglesias believes automatically transmits prosperity to those near it. But even under those more favorable conditions, the boomtown/megacity was never in the cards. There was never one industrial core that later saw suburbanization. From the earliest days of San Francisco’s industrialization, manufacturing was dispersed: southern industrial districts went past South of Market by 1890, East Bay surpassed SF’s manufacturing output by 1910 (and employment by 1920).

It’s important to emphasize this isn’t firms relocating to the periphery, which you might tell yourself if you think this was about workers following the money in a boomtown. It was series of sectoral successions, where each new industrial wave erupted in new places with new spatial forms. The industries that made San Francisco dominant (mining equipment, wooden carriages, leather goods, animal processing) were either dead, dying, or undergoing a fundamental transformation by 1900. The ascendant industries (petroleum refining, alloy steel, automobiles, electrical machinery, industrial chemicals) found homes almost entirely outside the city in Oakland, Emeryville, Richmond, and along the Contra Costa shore (they had to find somewhere else to offload the industrial waste and pollution). Industry was not lost to the suburbs, it was made obsolete by new enterprises that sprung up elsewhere. And so the geography of the Bay Area metropolitan area and its productive base found themselves remade and reconstituted with each successive wave of industry.

There were other reasons for moving, of course, beyond finding new places to dump waste. Owners of capital were looking for cheap land, large sites for industrial production, and (most importantly) freedom from San Francisco’s militant laborers. Nearly every major factory in Contra Costa County was, as Walker puts it in his “Industry Builds Out the City” chapter, “dreamed up and financed" by San Francisco capital” as well as “the rail, water, oil, and electricity networks that fed the new industrial district.” Emeryville would be incorporated as a purpose-built industrial enclave run by factory workers for decades, South SF would be politically controlled by its corporate founders, and so industry’s needs were prioritized first and foremost in the design of these vast swaths of the Bay Area metropolis. No surprise, then, that working-class housing followed capitalist investment decisions instead of the other way around! The class geography of the entire region—bourgeois to the north, proletarian to the south; flatland workers, hillside owners in Oakland; workers clustered near factories because of time, income, and transit constraints—was determined by decisions made by owners of capital, not by restrictive zoning codes.

Even in the period of heavy manufacturing for this earlier boom, the Bay Area’s spatial form was overdetermined by capital’s investment logic! Where industrialists chose to build and what political-legal jurisdictions they chose to construct in hopes of protecting those investments, as well as how successive waves of industry sprung up and went on to reshape the region. The megacity never appeared because of some regulatory block, but because it’s not what has ever made sense in the Bay Area if you look at how capital has been accumulated and reproduced over the years there. And if that was the case during the period when the Bay Area had steel mills, shipyards, automobile plants, and other industrial sites that required huddled masses, then there’s no reason to expect a different outcome with a tech boom that requires even less fixed capital, even fewer domestic workers, even greater spatial flexibility, and even greater subservience and sensitivity to the capital’s demands?

This, of course, isn’t even a full accounting of all these details. We can talk about the Bay Area’s massive working class and where they have actually lived during the tech boom, the class structure reproduced by the tech boom, or how a hypothetical megacity wouldn’t substantially enrich these systematically disempowered workers. We could talk about the dreaded environmental reasons why the tech boom may not have produced a megacity (goddamn the NEPA, as some of the Abundance crowd might say)—the sector quite literally poisoned the air, water, and earth (taking extreme care to limit the damage to Black and brown communities, however). We could actually talk about the initial boomtown, tracing how imperial plunder and material extraction (and devastation) along with military spending subsidies all but ensured the class structure of any boom in the region would be one that prioritized capital’s needs (and at even greater cost than normal to labor).

Closing thoughts

So where is the megacity? We have a few rough answers we can offer.

First, the tech economy has no real interest in producing one. The core products have a great deal of physical production offshored and minimal labor requirements relative to wealth generated. When labor-intensive physical production was at home, firms fouled the earth with reckless abandon (visit your local Bay Area Superfund site to get a peek). Benefits are largely limited to a small and narrow technical elite, as well as the financiers behind them, and no amount of housing construction will do anything to change that structural relationship. In the Midwest booms, prosperity was distributed by working in the booming industry itself, not adjacent to it. It’s not clear what, if anything, trickle-down economics—rebranded by consuming in the shadow of Silicon Valley’s mountains of treasure—will do for the masses of workers. What’s a slightly higher nominal wage when the tech boom itself is driving up the cost of living aggressively? In today’s Bay Area, a janitor is precariously housed, systematically disempowered, cut off from any meaningful decisions about their workplace or political order. In tomorrow’s megacity, a janitor will be precariously housed, systematically disempowered, cut off from any meaningful decisions about their workplace or political order.

The second answer is that there’s a superficial understanding of the historical models offered as alternatives to learn from. The Midwest boom shared prosperity through organized labor struggling politically and exercising structural leverage within a national economy that immobilized capital. And even that era, hailed as a golden age by many, was racially exclusionary and viciously contested at every single step. The Bay Area’s own earlier boom was organized around extraction and concentrated wealth spooling out of the Gold Rush through the railroad era right to today’s military-industrial complex (and Silicon Valley’s committed re-engagement with it). Is the dream of a billion skyscrapers on John C Frémont’s golden horns a stillborn one? It’s better to say it’s one that comes with a fever—a delusion entertained by projecting hallucinations onto reality. Not to say we shouldn’t desire a megacity, but that we should think a bit more seriously about why one never existed so we can actually work towards creating one (and figuring out what that entails).

And a third answer: megacities do exist! They were built by developmental states wielding tools like massive public housing provision, state-directed industrial policy, public control of land and credit, and putting the fear of god into a few capitalists here and there so that they ignore the devil on their shoulder (profit-seeking). Can we import policy templates from Singapore or Seoul in hopes of building a megacity? Perhaps. Some of Vivek Chibber’s arguments in essay and book form offer a look at part of the problem here: the conditions which enabled successful industrial transformations (state capacity to discipline capital, bureaucratic autonomy from private interests, etc.) should be understood as specific historical achievements—political settlements that were struggled for, not developments that emerged from letting the private sector do the right thing. We understand this when it comes to developing some of the industrial titans of today (such as Huawei and TSMC) but, for some reason, applying this idea to urban development is treated as suspect at best. Though, of course, the reason is clear, isn’t it? State capacity at the scale necessary to twist capital’s arm is state capacity in position to (and mobilized by an ideological project that) may have funny ideas about property rights, capital mobility, state-run enterprises, competition, the political power of the tech or real estate sector, and so on. At that point, the question shifts from “how do we build more housing?” to “what kind of political power would be necessary to organize the economy’s relationship to land, labor, capital, surplus, prices, and so on?” Some may be uninterested in the latter, some may even view entertaining such questions as squandering opportunities to craft their own version of positive class compromise (to make building more housing in everyone’s interests) because it puts various political actors on high alert—and I’m sure this is true in some instances, but so be it!

I’m skeptical of the idea that booms are always healthy, always distributional in a way that benefits everyone, and easily remedied by policy if for some reason this isn’t the case. I think it’s clear even a cursory glance at the history, economics, politics at play here supports that skepticism.

Pax Silica

I’m sure you’ll remember, since you’re with The Wall Street Journal, back in the day 25 years ago people used to talk about the Washington Consensus. Today we are seeing a new – a true new consensus that is being born out of Washington and has spread throughout the world that is embraced, which is that economic policy flows from national security.

Jacob Helberg, Under Secretary of State for Economic Affairs (December 16, 2025)

Back in December, I laid out some areas of the AI bubble I anticipated would see important developments this year. I talked about “compute diplomacy” as the successor to oil diplomacy and how the Compute Axis—Silicon Valley and its capital-intensive dream of building God out of sand, Trump and his brigands, and the sovereign capital of Gulf monarchs—was attempting to construct a new imperial architecture out of chips, data centers, and export controls. I discussed the PetroCompute thesis, the structural traps awaiting Gulf states in their pivot from hydrocarbons to silicon, and how “Sovereign AI” functioned as a mechanism for locking allies and clients into dependence on the American tech stack. I talked about Morozov’s three-act play of US imperial management: dollar diplomacy, oil diplomacy, compute diplomacy. CSIS’s Navin Girishankar cheerfully advocated for a compute-dollar system in which access to advanced chips would be conditioned on dollar-denominated settlement. Evgeny Morozov, considerably less cheerfully, described how national sovereignty in the Republic of Nvidia would boil down to the privilege of writing checks to US corporations.

That prediction turned out to be on the money as the United States shortly afterwards announced with a great deal of pomp its newest gambit: Pax Silica—an American-led project that would secure control over supply chains for artificial intelligence and the chips to run it, at the same time creating a new dependency regime and tech stack that could be backed up with the threat of violence (economic or otherwise). The name, after all, tells us everything we need to know.

For those who dream in marble, Pax Romana (Latin for “Roman Peace”) and its two-century span represents Western Civilization’s golden age: a period of relative peace bracketed by the rise of Augustus in 27 BC and the death of Marcus Aurelius in 180 AD. Nevermind that the Pax featured some of Rome’s bloodiest civil wars, foreign adventures, and revolts against imperial rule—the violence is the point, not a contradiction. Pax Britannica ran from the Napoleonic Wars to World War I and Pax Americana from the end of World War II through the Cold War—both are hailed as Long Peaces between major powers, even though revolts, genocides, and wide-ranging proxy wars were central features of these eras of supposed tranquility. To the marble fetishist, this is the selling point. A lack of scruples about using force to crush rivals abroad and irritants at home is the only way to deliver peace, justice, freedom, and security to our new empire. Every Pax has meant the same thing: the conditions an empire imposes on everyone within reach, named as though they were a gift.

In 83 AD, near the end of the Roman conquest of Britain, Calgacus—chieftan of the ill-fated Caledonian Confederacy (a Celtic tribe in what’s now Scotland)—gave a speech to his forces just before their decisive defeat at the Battle of Mons Graupius at the hands of Roman general Agricola, reported to us by Roman historian Tacitus (Agricola’s son-in-law):

You have not tasted servitude. There is no land beyond us and even the sea is no safe refuge when we are threatened by the Roman fleet....We are the last people on earth, and the last to be free: our very remoteness in a land known only to rumour has protected us up till this day. Today the furthest bounds of Britain lie open—and everything unknown is given an inflated worth. But now there is no people beyond us, nothing but tides and rocks and, more deadly than these, the Romans. It is no use trying to escape their arrogance by submission or good behaviour. They have pillaged the world: when the land has nothing left for men who ravage everything, they scour the sea. If an enemy is rich, they are greedy, if he is poor, they crave glory. Neither East nor West can sate their appetite. They are the only people on earth to covet wealth and poverty with equal craving. They plunder, they butcher, they ravish, and call it by the lying name of ‘empire’. They make a desert and call it ‘peace’ (XXX).

Pax Romana had Britannia. With Pax Britannica, we have our pick of the litter but one resonant example might be the two Opium Wars against the Qing dynasty to force China into accepting British trade terms (buy from us the opium that is ravaging your society or else). For Pax Americana, we could craft a long list from Korea to Indonesia to Indochina to Latin America to swaths of Africa to Europe’s doorsteps—there are too many piles of corpses killed by Americans and their proxies to count. What will Pax Silica hold? As Helberg and company articulate in unambiguous terms: the targets are broadly any country that does not bend the knee—but specifically China. China controls roughly 90 percent of global rare earth processing—the materials refined into the chips that the entire AI edifice runs on—and so we must band together (on America’s terms) to undermine our great rival.

In his December briefing, Helberg said: “Our strategy is to create a competitive edge so steep, so insurmountable that no adversary or competitor can scale it.” We’ll speak in the language of peace but move with the logic of siege. Pax Silica will bind allied nations into a “coalition of capabilities”—Israel, Japan, South Korea, Singapore, the UK, and Australia signed on first—conditioning access to advanced chips on alignment with American supply chains and, implicitly, American foreign policy. It is, as Helberg said without apparent irony, a “new consensus” in which “economic policy flows from national security.” The Washington Consensus packaged structural adjustment as development. The Silicon Valley Consensus will package dependency as partnership.

Pax Silica does not need Pacific Rim signatories to be successful, however. It needs the Gulf. Qatar signed the Pax Silica declaration on January 12. The UAE followed on January 15. Between them, the Qatar Investment Authority and the UAE’s constellation of sovereign vehicles (Mubadala, ADQ, ADIA, MGX) control north of $1.5 trillion. Saudi Arabia’s PIF alone has nearly $1 trillion in assets. They also offer what few, if any, other partners can: patient capital at sovereign scale, cheap dispatchable energy to power the data centers that train the models, and a geographic position at the center of the India-Middle East-Europe Corridor (IMEC) that Washington has been trying to repurpose as a digital trade route. “For the UAE and Qatar,” Helberg told Reuters, “this marks a shift from a hydrocarbon-centric security architecture to one focused on silicon statecraft.” The dependency doesn’t shift, then. It deepens. Under oil diplomacy, the Gulf priced crude in dollars and recycled surpluses into Treasuries. Under silicon statecraft, the Gulf finances AI campuses, hosts American chips under American export licenses, and locks itself even more tightly into an architecture whose terms are set in Washington.

Helberg, the public face of this initiative, is also worth a quick treatment. He’s a former Google policy staffer who married PayPal mafia member Keith Rabois at a ceremony officiated by Sam Altman—the CEO of OpenAI (the man whose $500 billion Stargate project is the single largest beneficiary of the Gulf capital Helberg’s initiative is designed to mobilize). From Google, Helberg moved to Palantir, the surveillance firm and military contractor founded by Peter Thiel, where he established himself as a technologist and China hawk in Trump’s orbit. As if that weren’t enough, he’s become something of a conduit between the American and French far right. In one biography, you can trace a familiar groove: platform capital to defense-tech to sovereign wealth to the Oval Office.

All that aside, it’s worth asking: how long will Pax Silica last? Romana spanned two centuries, Britannica just shy one, Americana a measly half century. This is the multi-trillion dollar question that will determine what the next century looks like.

Six weeks, it turns out.

That’s how long the ink on the Pax Silica declaration was dry before the US-Israel strikes on Iran were followed by Iranian strikes on nearby Gulf states hosting American military bases and Big Tech infrastructure, as well as a closure of the Strait of Hormuz (through which 20 percent of the world’s oil passes through) and reports that Iran may be considering letting tankers through if they conduct trade in Chinese yuan instead of American dollars.

Operation Epic Fury

Ostensibly, the United States and Israel attacked Iran on February 28 to decapitate its leadership, eviscerate its missile infrastructure, and attack facilities linked to its nuclear program. In the weeks since, the strikes have not only failed on these counts, but come to threaten the immiseration of Gulf economies as oil production halts and Iran responds with missiles and drones aimed at neighbors hosting American military bases. In the meantime, the rationale for this war has shifted a few times. Comments by Secretary of State Marco Rubio and reporting by The New York Times make it clear the United States was dragged into this war by Israel. And this was further backed up by a recent Bloomberg interview with former Secretary of State Anthony Blinken recalling multiple attempts by Israel during the Obama administration to pressure the United States into attacking Iran by threatening to carry out a preemptive strike on its own. Rubio has tried to walk back his comments, but Israel’s desire to preemptively attack Iran has been well documented by now.

Somewhere in the depths of hell, all the American state planners that diligently built the foundations of American power must be screaming in unison. Again, it was just six weeks before Iranian drones hit three Amazon Web Services data centers in the UAE and Bahrain—the first military attack on hyperscale cloud infrastructure in the history of the world—that Helberg anticipated the Gulf’s integration into the US-led AI supply chain as “a shift from a hydrocarbon-centric security architecture to one focused on silicon statecraft” that would ensure security for the client states and dominance for the tribune.

We build some of the most expensive data centers on the planet in the most strategically volatile region on the planet, announce a new security paradigm that hinges on their proliferation and protection, and then launch a war against a country next door. Iran’s IRGC said it targeted the Bahrain facility specifically because AWS hosts US military and intelligence workloads. As Fortune put it, the boundary between commercial cloud infrastructure and military operations has “largely vanished”:

The Pentagon’s Joint Warfighting Cloud Capability and its Joint All-Domain Command and Control networks run on the same commercial infrastructure that serves banks and ride-hailing apps. Meanwhile, several news organizations have reported that the U.S. military used Anthropic’s AI model Claude—which runs on AWS—for intelligence assessments, target identification, and battle simulations during the Iran strikes.

That dual-use reality means that attacks on commercial data centers can have immediate military consequences—and vice versa.

I think it might be more accurate to say that if such a boundary ever existed, it disappeared a long, long time ago. For years now, our compute infrastructure has been on one side labeled “enterprise cloud services” while the reverse is labeled “kill chain.” It’s the success of the cloud divisions that makes the vendors attractive for various firms and countries interested in preserving apartheid, pursuing mass deportation, or engaging in remote assassination.

After the shellshock of Rome’s sack in 410 by Visigoths, St. Augustine—desperate to defend Christendom from accusations that Rome’s embrace of the faith hastened its demise—penned The City of God to, in part, dismiss the idea that Rome’s paganism was the source of Rome’s supposed glory. In a move that echoes Calgacus, Augustine argued that Rome was driven not by civic virtue or divine sanction, but by naked aggression and a lust for domination:

Remove justice, and what are kingdoms but gangs of criminals on a large scale? What are criminal gangs but petty kingdoms? A gang is a group of men under the command of a leader, bound by a compact of association, in which the plunder is divided according to an agreed convention.

If this villainy wins so many recruits from the ranks of the demoralized that it acquires territory, establishes a base, captures cities and subdues peoples, it then openly arrogates to itself the title of kingdom, which is conferred on it in the eyes of the world, not by the renouncing of aggression but by the attainment of impunity.

For it was a witty and a truthful rejoinder which was given by a captured pirate to Alexander the Great. The king asked the fellow, “What is your idea, in infesting the sea?” And the pirate answered, with uninhibited insolence, “The same as yours, in infesting the earth! But because I do it with a tiny craft, I’m called a pirate: because you have a mighty navy, you’re called an emperor.

Little has changed in the centuries since Augustine wrote this. In 1986, Chomsky used the exchange in the opening pages of Pirates and Emperors to launch a discussion about “international terrorism” and “the heart of the frenzy over selected incidents of terrorism currently being orchestrated, with supreme cynicism, as a cover for Western violence.” And as the renowned antiwar leftist Francis Fukuyama noted in a recent post “the United States and Israel have by now taken out most of the visible military facilities in Iran, and are moving, for lack of other targets, to attack infrastructure that serves ordinary people. These include oil storage facilities, electrical grids, desalination plants, and other dual-use civilian targets.”

And so we infest the earth.

The United States and Israel have taken it upon themselves to not just target civilian infrastructure, but residential buildings and increasingly the civilian population itself. Right now, the United States is bombing Iran’s oil infrastructure in hopes of forcing the country’s surrender and unleashing toxic pollution over Tehran that will poison the city of nearly 10 million people for decades. It is unclear how any of this is not international terrorism—unless, of course, you abide by the eternal rationale that the strong do what they can and the weak suffer what they must.

The compact of association is intact. The plunder will be split in due time. It’s unclear exactly what comes next, but we can see the shape of things to come. The attempt to shift Pax Americana’s oil diplomacy to Pax Silica’s silicon statecraft—from hydrocarbons and Treasuries to GPUs and data centers and AI models and the kill chains they power—feels stillborn at this juncture, but the larger circuit still remains. The emperor’s navy protects the trade routes that fund the emperor’s navy that bombards the countries along those trade routes, and when the bombed countries retaliate by striking the trade routes, the emperor and his court will respond with bewilderment.

Casus Belli

It should come as no surprise that an AI-powered war launched by Israel and the United States has proven so destructive so quickly. Israel, after all, has spent years deploying all manner of new technology (most recently, AI-powered weapons) to destroy as much of Gaza as possible, to kill as many Palestinians as possible, and export to other countries eager to try Israeli arms on their own population. The United States government has spent the past few years integrating AI into the Department of Defense’s operations and recently labeled Anthropic a supply chain risk over what was a disagreement in degree over autonomous weaponry (Anthropic believes autonomous weapons should exist, our government should have them, already powers “kinetic” operations, but has qualms about reliability).

In the first 24 hours of the US-Israel strikes, the two rogue states managed to assassinate Supreme Leader Ali Khamenei (as well as a number of potential successors the United States hoped to back) and strike over 1,000 targets (including a girl’s primary school where it killed more than 170 people, most of whom were children)—all made possible by the Pentagon’s Maven Smart System, built by surveillance firm Palantir, with Anthropic’s Claude embedded at its core. The Maven-Claude system suggested hundreds of targets, issued coordinates, prioritized them by importance, and evaluated the results of each strike. Military commanders had become so dependent on Claude that if Anthropic pulled the plug, the administration planned to use government powers to retain the technology. “Whether [Anthropic’s CEO’s] morals are right or wrong or whatever,” one official told the Post, “we’re not going to let his decision making cost a single American life.”

Iranian life is not so precious a thing, however. 1,400 civilians have been killed so far and the death count will certainly rise as America and Israel run out of military targets and turn to targeting civilians and their infrastructure. The Financial Times reports that as “the US and Israel sought to degrade regime institutions, they have hit more than 20,000 non-military buildings, according to Iran’s Red Crescent, including 17,353 that were residential.” Whatever legally required vetting and basic due diligence that might’ve been in place to minimize civilian deaths has been and will continue to be whittled down, by design, because the point of automated kill-chains is to compress the process, it is to “shrink that into seconds and minutes, almost instantaneous” as one anonymous defense tech expert told The Financial Times. The point is more targets, destroying in weeks what might’ve taken a traditional campaign months—and so, if civilians die they die.

And so we infest the earth with AI tools embedded at every stage of the kill chain—tools built by the same companies and financed by the same capital, that were supposed to inaugurate an era of American technological supremacy; the same tools tested and refined on Palestinians in Gaza over the course of Israel’s genocidal campaign.

Israel deployed Lavender, a probabilistic classification system that assigned Palestinians a numerical rating representing their statistical likelihood of being a member of Hamas or Palestinian Islamic Jihad—drawing on the pervasive surveillance dragnet Israel has maintained for decades (cellular metadata, social network analysis, financial transaction records, behavioral patterns) to generate a score that determined whether someone should be assassinated or not. In the first weeks of the war, as +972 Magazine and Local Call reported, the Israeli Defense Forces (IDF) gave sweeping approval to adopt Lavender’s kill lists with no requirement to examine the underlying data. Human reviewers spent an average of twenty seconds on each target, limited to confirming the target’s gender: if male, a militant; if female, an error. When the IDF exhausted assassination targets exceeding a certain ratings threshold, commanders simply lowered it to generate more—the pretense of precision giving way to the real prerogative, which was rationalizing mass bombardment and exterminating as many Palestinians as possible.

Alongside Lavender, the IDF deployed Gospel—a geospatial intelligence fusion system for identifying buildings and infrastructure as military targets —and “Where’s Daddy,” a tracking system that generated real-time alerts when assassination targets entered family residences, maximizing the probability of killing both the target and their relatives. Together, these systems constituted what their own operators described as a “factory for mass assassinations,” generating hundreds of targets per day compared to the fifty or so per year the IDF might have produced within a conventional intelligence system.

By all accounts, the IDF’s assassination factory was a spectacular success. A February 2026 study in the Lancet Global Health medical journal suggests Israel has killed at least 75,000 people during the first 16 months of its war (October 2023 to January 2025)—25,000 more than what had been reported at the time . The Max Planck Institute for Demographic Research’s estimate sits a bit higher, with a death toll of 78,313 by the end of 2024 (and likely exceeding 100,000 by October 2025). The vast majority of the dead in both analyses were women, children, and the elderly. UNOSAT satellite analysis also found 81 percent of all structures in the Gaza Strip had been destroyed. This, with full financial, diplomatic, and intelligence support of the United States, which vetoed multiple UN ceasefire resolutions and spent over $21 billion on military aid to Israel during the war. Then (and now) we see no meaningful restriction on the use of AI in this campaign—not by Israel, its American patrons, or by the companies whose models made the targeting architecture possible.

That the kill chain runs on good ol’ American compute is key to all of this. To just focus on one head of the many-headed beast here, we can look at Google and Amazon. Their $1.2 billion Project Nimbus contract doesn’t just provide Israel’s military and intelligence services with cloud infrastructure and machine learning tools—it contractually prohibits either company from pulling the plug for any reason whether it be human rights violations, humanitarian crisis, or genocide. Or take another head of the beast: Microsoft. A joint investigation by The Guardian, +972 Magazine, and Local Call revealed Microsoft Azure stores intercepted phone calls from Israel’s surveillance drag net, allowing Unit 8200 (Israel’s military surveillance agency) to process a million calls an hour. Earlier that year, another joint investigation first detailed the deepening ties between Microsoft, OpenAI, and Unit 8200, tracing how after the volume of surveillance data surged after October 2023, Microsoft expanded capacity by 60% and embedded engineers directly inside military units, logging an estimated 19,000 hours of consulting support by mid-2024 (and gave Unit 8200 extensive access to OpenAI’s ChatGPT-4). And in yet another joint investigation between the outlets, it was revealed Unit 8200 was “developing a new, ChatGPT-like artificial intelligence tool and training it on millions of Arabic conversations obtained through the surveillance of Palestinians in the occupied territories.”

This is one face of the Silicon Valley Consensus in practice: not disruption, not innovation, but the quiet, contractually guaranteed provision of computational infrastructure to militaries keen on generating targets faster than any human being can evaluate them. The idea on display here is that when we (we being anyone on the right side of the crusade promised by Pax Silica) start evaluating targets, we start to forget that it’s the abundance of force and our willingness to use it that keeps the peace. When the IDF’s own servers couldn’t keep pace with the data throughput of the campaign—the drone feeds, the intercepted conversations, the algorithmic kill lists—it was American Big Tech that eagerly ensured the bottleneck was never the machines (or the will to kill as many as possible). And of course they did, because we have no problem with any of those things when there is money to be made. In fact, it’s what we’re good at. As Palantir chief executive Alex Karp recently said in a CNBC interview:

What makes America special right now is our lethal capabilities, our ability to fight war, both because we’ve been doing it for 20 years, because we have meritocracy in our military, and because we finance it in a way no other country has. And because the A.I. revolution is uniquely American. Every company. All the model providers that are relevant and the ontology that makes them valuable and the chips that they run on are built in this country.

That is: these systems are not developed in isolation from the American “defense” technology ecosystem. They are architecturally continuous with it, letting everyone involved maintain the polite fiction that there is a meaningful technical distinction between an Israeli system that generates kill lists from surveillance data and an American system that does the same thing but with a nicer user interface and a chatbot bolted on top, pointed at civilians in a country an ally has been begging for a chance to invade for years now.

We can look at yet another head of the beast: surveillance firm Palantir. Palantir’s Maven Smart System—the Pentagon’s primary AI warfare platform, managed by the National Geospatial-Intelligence Agency, performs structurally identical functions: applying computer vision algorithms to satellite imagery, automatically detecting objects likely to be enemy systems, visualizing potential targets, and “nominating” them for bombardment. A tool within Maven called the “AI Asset Tasking Recommender” proposes which bombers and munitions should be assigned to which targets. In November 2024, Palantir announced that it would integrate Anthropic’s Claude into the software it sells to US intelligence and defense agencies—Claude became the voice and reasoning engine of Palantir’s AIP system, the application layer through which military analysts interact with a chatbot that can identify enemy units, generate courses of action, plan routes for troops, and assign electronic warfare assets. In a Palantir demo released in 2023, the entire Observe, Orient, Decide, Act loop—anomaly detection to unit identification to strike planning to troop mobilization—was compressed into a single chatbot conversation.

One has to labor under an arduous delusion to think there exists any distance between Israel’s “factory for mass assassinations” and the Pentagon’s AI warfare platform and the various programs (or models) used within it. Marketing is the great differentiator here—the same repackaging exercise that lets a firm slap “AI-powered” on a product identical to last year’s model and charge twice or thrice as much, except here the product is a kill chain and the customer is the DoD (or the IDF). As I’ve laid out before, the Silicon Valley Consensus rests on overbuilding, overvaluing, and overinvesting in various assets in hopes of realizing excessive gains that can be translated into political power aimed at restructuring society. What kind of picture do you see taking shape here?

Capital Circuitry/Compute Circuitry

Regardless of how you feel about whether or not there is an AI bubble, I think it is relatively benign at this point to say the primary mechanism driving the overbuild of AI infrastructure is a fundamental misallocation of capital. AI-powered weaponry offers another way to think about that financial morass, namely by thinking of all this as one circuit grafted onto various familiar grooves across the world. One such system that showcases the overlap of AI-powered warfare, imperial architecture, and the AI overbuild and the channels of capital they rely on is the petrodollar recycling mechanism—a beautiful living machine that has been continuously operating for half a century and whose latest output will be the $500 billion Stargate project (if it actually happens).

Gulf states sell hydrocarbons priced in dollars. The resulting surpluses get recycled into American financial instruments—Treasuries, equities, private equity, real estate, and now data centers and AI infrastructure. In return, the Gulf states receive market access, the implicit promise that the Fifth Fleet and CENTCOM will keep the shipping lanes open, and the political cover that comes from being structurally embedded in the American financial system. The carrier strike groups patrolling the Arabian Gulf are the military expression of a monetary circuit. To describe this as an “investment relationship,” as the comfortable language of bilateral diplomacy might insist on doing, is akin to describing feudal tithes as a community development initiative (this is NOT an endorsement of technofeudalism).

Evgeny Morozov laid out this historical arc in the Le Monde Diplomatique I looked at in Part One. Nixon killed the gold standard, the dollar wobbled, and so Kissinger flew to Riyadh with an offer. Charge whatever you want for oil (so long as it’s in dollars) and invest the profits in US Treasuries.

Yakov Feygin and Dominik Leusder offer a more structural reading in Phenomenal World, arguing the dollar system should be understood not primarily as a national tool but as a consequence of a globalized economy that privileges the preferences of financial elites for the free, international movement of capital. The system persists not because it benefits “America” or “Saudi Arabia” in the aggregate but because it benefits the specific people who control capital allocation in both countries: American financial elites get cheap capital inflows, Gulf ruling families get a safe liquid store for their surpluses and a political relationship with the world’s dominant military power. Fault lines run along class, rather than national, lines. Which is to say the circuitry doesn’t care what asset class it’s recycling into. Treasuries in the ‘70s, Manhattan real estate in the ‘80s, private equity in the 2000s, data centers in the 2020s, effective altruist shrimp colonies in the 2030s, who cares. Elites on both sides of the deal profit, the mechanism persists, and nobody rocks the boat—least of all the management consultants who, as Colin Powers documents in a recent essay on Saudi Vision 2030, have set up shop inside a byzantine network of institutional actors generating polarizing wealth effects within what is already one of the most unequal dungeons on the planet.

The carrier strike groups that guarantee the petrodollar circuit are part of the same military infrastructure now bombing Iran. The AI tools integrated into that infrastructure—Maven, Claude, the AIP system—are financed, in part, by the same Gulf capital that flows through the petrodollar circuit. And the acts of terror those tools are carrying out are destroying the physical and financial infrastructure of the Gulf states whose capital makes the whole arrangement possible.

Show Me (some of) the Money

The concentration of Gulf sovereign capital in the American AI ecosystem is worth assembling in one place because once you see it laid out you cannot unsee it, and because it is genuinely one of the most alarming concentrations of financial and political power I’ve ever tried to get my head around.

MGX—Abu Dhabi’s purpose-built AI investment vehicle, whose chairman is Sheikh Tahnoun bin Zayed, the UAE’s National Security Advisor—has invested in all three frontier AI labs: OpenAI, Anthropic, and xAI. It is a founding equity partner in Stargate, contributing billions in initial equity alongside SoftBank, OpenAI, and Oracle. It holds a 15 percent managing stake in TikTok’s US operations. It joined the consortium—including Nvidia, Microsoft, and BlackRock—that purchased Aligned Data Centers for $40 billion, the largest data center transaction in history. And MGX routed $2 billion to the Binance crypto exchange through USD1, a stablecoin launched by Trump’s World Liberty Financial—a privatized monetary instrument issued offshore, collateralized with US public debt, intermediated through a Gulf monarchy, with no federal oversight and no public mandate (the private mandate is, clearly, to steal anything that isn’t bolted down).

This is all one sovereign vehicle! Simultaneously inside the three leading AI labs, the largest data center consortium, the most politically contested social media platform in the country, and a crypto vehicle controlled by the sitting president’s family.

MGX is not alone. Kuwait Investment Authority anchors the AI Infrastructure Partnership alongside MGX and BlackRock. The Qatar Investment Authority backed Anthropic’s $13 billion round and xAI’s $20 billion Series E. Saudi sovereign capital may have put on a brief hiatus from Silicon Valley following assassination of Saudi journalist Jamal Khashogi, but it had played such an integral role in financing Silicon Valley and terraforming the world into forms and relations more hospitable for various tech firms that the reunion was all but inevitable. Saudi Arabia’s Public Investment Fund led the $55 billion bid to take Electronic Arts private—the largest leveraged buyout in history—alongside Silver Lake and Jared Kushner’s Affinity Partners, with JPMorgan committing $20 billion in debt financing.

One particular node that is a favorite of mine is the constellation surrounding Kushner. This is where the circuit really sings. Affinity Partners raked in $2 billion from PIF, another billion split between QIA and Abu Dhabi-based investors, and $157.5 million in management fees (all while returning zero profits to investors). PIF’s own investment committee recommended rejection, citing “the inexperience of the Affinity Fund management” and fees that were “excessive,” but Crown Prince Mohammed bin Salman personally overruled them.

Sovereign infrastructure, personal enrichment, what difference does it make to those along the path? These are all stops along the same circuit that’s concerned with little more than ensuring the capital flows. Both are handled with equal efficiency. And both help along the alchemical process: speculative gains into real wealth into political power that makes repeating this ritual easier.

War Math

Despite the collapse of negotiations between Anthropic and the Pentagon—in part because of the company’s concerns about its technology powering fully autonomous weaponry and making pervasive mass surveillance of Americans economically feasible—the war marches on and Claude’s products are still being used by the DoD. It is worth noting that Anthropic does not oppose autonomous weapons as such: Bloomberg reported that the company submitted a proposal for a $100 million Pentagon prize challenge to produce technology for autonomous drone swarming, though Anthropic says its submission focused on translating commander intent into digital instructions rather than autonomous targeting. Either way, the company has repeatedly insisted in its own words and statements that the core issue is reliability, not principle. “I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries,” opens a February 26 statement from Anthropic chief executive Dario Amodei.

That last distinction says everything about the kind of objection being raised—not “we won’t build this” but “this doesn’t work well enough yet,” a quibble a pirate might squeak about the seaworthiness of a recently stolen ship, as opposed to a real concern about the nature of what they’re doing.

In response to the US-Israel AI-powered strikes, Iran retaliated across the region with missiles and drones aimed at US bases in Bahrain, Kuwait, Saudi Arabia, the UAE, Qatar, Oman, Jordan, and Iraq (as well as with strikes at civilian targets in cities across the region). On top of this, Iran took aim at three AWS data centers—two in the UAE, one in Bahrain—which went offline and disrupted banking, payments, and cloud services across the Gulf. And it followed up the attacks with an announcement from a news agency that major tech firms and their infrastructure would be considered legitimate military targets by the IRGC.

For now, the Strait of Hormuz is functionally closed. Gulf states cannot physically ship the oil they produce. Revenues are collapsing even as oil prices rise on paper, because higher prices only matter if product reaches market—the tankers cannot transit the Strait, cannot find insurance, and cannot find ports willing to accept them. The worst of both worlds: higher prices on paper, lower incomes in practice.

What does all this mean for the financial architecture laid out above?

The fiscal pressures were already severe before anyone started bombing. Powers documents that Saudi Arabia ran a $32 billion deficit in 2024, requiring crude above $94 per barrel to balance expenditures while benchmark Brent averaged $69. PIF deploys approximately 76 percent of its assets domestically, financing the giga-projects—NEOM, the Red Sea development, Qiddiya—that constitute MBS’s legitimation strategy. The PetroCompute traps identified by Abdullah Alzabin—value capture, sovereignty, obsolescence, cannibalization—were already constraining the Gulf’s ability to finance both domestic transformation and the American AI buildout simultaneously.

That was pre-war math. How are things looking now? The PIF cannot simultaneously finance NEOM (in fact, it has already killed it), fund the Saudi defense establishment, absorb war costs, and anchor Stargate—even if oil costs climb as the global supply tightens. QatarEnergy, the world’s largest LNG producer, halted production after strikes on its facilities. G7 finance ministers have announced preparations to release emergency energy stockpiles, though it will take months for the effects to be felt and even then they will be minuscule. The very non-oil sectors that Vision 2030 was supposed to cultivate—the ones that were supposed to justify the entire diversification gamble, the tech sector included—are the sectors most immediately damaged by the war.

It should come as no surprise, then, that Gulf states are reviewing overseas investments, effectively weighing whether the American AI buildout and its related asset classes make sense for recycling their surpluses. How long will Gulf states and their elites absorb the cost of a war they opposed while simultaneously funding the financial and computational architecture that makes that war possible?

Unicorn Shit for Breakfast

Running through all that’s come before is the idea that Gulf capital is not easily replaceable. That the AI infrastructure buildout cannot survive any notable retreat of capital from our motley crew of Gulf despots. And that raises, again, the constant question of whether the capital flows used to finance this buildout are sustainable enough, much less profitable enough, to realize a return on the eye-wateringly large investment.

Norway’s $2 trillion sovereign wealth fund is rules-bound, follows ESG mandates (and now uses Claude to screen ESG risks), is pretty transparent, and under relatively strict ethical guidelines. It does not anchor PE funds, does not take board seats, and does not seem to fratenize with crown princes. The vast majority of its investments are equities, which will likely be the frontier of its investment in Big Tech for a long while. In December, Reuters reported that Alexander Knapp, the fund’s new head of real estate investments, seemed uninterested in investing in data centers. “We don’t have any active plans to make investments there,” Knapp said. “We’re trying to be measured in our approach, so therefore, sectors that are volatile, we’re very careful with,” when asked about data centers, before adding: “We’re trying to make investments that are going to enhance the returns of the fund overall and not take undue risk”.

Singapore’s GIC ($936 billion) and Temasek ($324 billion) manage their investments on a commercial basis, with large chunks of their portfolios oriented toward Southeast Asia and China. Japanese institutional capital is massive but risk-averse , demographically constrained (the IMF calls Japan the “world’s policy laboratory” for dealing with aging, shrinking populations), and facing mounting domestic pension liabilities. We aren’t going to beg China for capital, obviously, given the entire binding compact of the Compute Axis is “We Must Win the Cold War with China.” European pension and insurance capital is fragmented across dozens of national regulatory regimes, subject to parliamentary oversight, and has recently discovered that having your own “Sovereign AI” strategy means spending your own money rather than recycling someone else’s.

It becomes increasingly clear that the substitutability problem here is not really about the dollar amount. It’s about the characteristics of the capital gushing forth: patience beyond the more immediately greedy timelines of venture capital (which does not even have enough capital despite its alchemical science to finance this buildout) that’s more comfortable eating losses for more than a few fiscal years; speed/intimacy/corruption measured in personal calls and bribes and favors and influence as opposed to committee cycles; political flexibility meaning no ESG mandates, no exclusion lists, no parliamentary oversight, no political backlash, minimal transparency; scale concentration in which a single decision-maker can direct tens of billions with singular authority. No realistic combination of alternative capital sources replicates this. The question is not “can other investors buy US Treasuries?”—obviously they can. The question is “can any other capital source simultaneously anchor the PE ecosystem, co-invest in AI infrastructure at planetary scale, absorb $142 billion in defense procurement, backstop a $500 billion compute buildout, and do all of this while asking no questions about governance, human rights, or return on investment?” [It is also worth noting that even in an ideal environment for raising capital for the AI infrastructure overbuild, projects like Stargate have persistently struggled to get going—obvious if you spend a few moments thinking about the economics of the project.]

The capital most useful to the Compute Axis is the capital with the fewest conditions or strings attached. That’s going to be the capital that originates from political systems offering insulation from the stampede of the mob (and its funny ideas about undesirability of financing apartheid, terrorism, genocide, unlimited war, and so on), driven by one singular authority or individual (or where those who rule can deploy enough force to suppress opposition without blowback), where personal and national interest blend, and where a compact can come together and unite in hopes of securing plunder and infesting the earth.

And so we are back to Augustine again. The entire American AI supremacy narrative—the one that animates executive orders and congressional appropriations and CSIS white papers and Pax Silica declarations—rests on a foundation controlled by a handful of Gulf monarchs. The “arsenal of democracy” for the AI age is being financed by absolute monarchies that are about as close to Mordor as one can get on Earth (for the time being). The free market that Silicon Valley champions depends, at its most critical junctures, on the least free capital in the world. Kingdoms and gangs, bound in a compact of association, dividing the plunder according to an agreed convention, infesting the earth until nothing is left (or until someone abandons the compact).

Compute Axis Wobbling

The Gulf states do not need to execute a dramatic exit. The mechanisms of pressure here work through inaction as much as action: they could simply stop rolling over maturing Treasuries, pause new PE and VC commitments, delay AI infrastructure commitments, announce “reviews” of the $2–3 trillion in investment pledges from Trump’s Gulf tour (most of them non-binding, announced with the ceremonial pomp and photography that diplomatic summits produce).

Beneath the volitional threat lies a fiscal one that feels a bit more realistic: the Gulf states may not choose to pull back so much as be forced to, because with Hormuz traffic collapsed and oil revenues falling despite rising prices, the surplus available for recycling into American assets shrinks whether anybody wants it to or not.

The hedging has long been underway. Saudi Arabia has been moving toward BRICS after being invited, but without formally committing to preserve some flexibility via ambiguity. The UAE has established bilateral currency swaps with China and conducted increasing trade volumes in renminbi. It is worth noting, however, that BRICS has done little in the midst of the US-Israel attacks on Iran, with Foreign Policy columnist C. Raja Mohan noted there are fundamental issues within the bloc: individual ties with the United States and Israel, a structural rivalry between Iran and conservative Gulf monarchies, and a long-standing pattern in international politics showcasing “the persistent failure of transnational solidarity.”

Still, there is promise there. As Tim Sahay and Kate Mackenzie write:

The postwar geopolitical order rested on three pillars: American hegemony, the fossil-fuel energy system, and an open, multilateral trading order. America has now attacked each pillar at the foundation of its hydrocarbon global order.

There are now two competing global models of energy and influence: one based on fossil fuels, one on green technologies and a new model of sustainable development. China’s technology is finding new markets around the world because lots of people want it. But there is so far no real wraparound support of finance, trade, and tech transfer—as no new international order of sustainable governance has yet been built. The critical question of the future of BRICS lies with its member countries’ willingness and ability to effect broader collaboration in the fields of technology, trade, and finance. A quarter of the way to the twenty-second century, everything is up for grabs.

The mBridge project—a cross-border central bank digital currency platform developed by the BIS Innovation Hub in partnership with the central banks of the UAE, China, Hong Kong, and Thailand—is a functioning proof of concept for non-dollar sovereign settlement. De-dollarization does not require a dramatic announcement. It operates through incremental decisions: a larger share of oil settled in RMB, a smaller proportion of reserves held in dollar instruments, surplus investment routed through non-dollar vehicles. Individual steps. Modest steps. But steps nonetheless.

Karthik Sankaran’s Phenomenal World essay on “monetizing primacy” describes the contradictions the Trump administration was already generating vis-à-vis the dollar before the first missile landed: a weak dollar to spur reindustrialization and a strong dollar to offset tariff-driven inflation, foreign inflows and extraterritorial powers and lower borrowing costs. That framing helps make Helberg’s Pax Silica a bit more transparent.

In December 2023, Stephen Miran (now Trump’s head of the Council of Economic Advisers) outlined what he called a “Mar-A-Lago Accord.” The memo is straightforward: the dollar’s centrality in the international monetary system is: (1) an extraordinary geopolitical weapon that enables sanctions, export controls, and the extraterritorial projection of American power; (2) an economic burden because persistent foreign demand for dollar assets keeps the currency strong enough to hollow out America’s industrial base. The solution offered by Miran? Not to relinquish our privilege, but to charge for it. Countries that benefit from the American defense umbrella and the dollar-denominated trading system should be made to finance both, either by swapping their current Treasury holdings for century bonds at below-market interest rates or, more bluntly, by writing checks to the US government. Primacy-as-a-service!

Helberg’s Pax Silica and “silicon statecraft” follow the same logic, applying it to compute instead of bonds. Where Miran envisions allies paying tribute in discounted long-dated debt, Helberg envisions them paying for the privilege of dependence on an American tech stack—paying in chips, data center commitments, and alignment with nascent supply chains. The currency may differ, but the transaction is near-identical. The contradiction mentioned earlier (wanting a weak and strong dollar, foreign inflows and extraterritorial power, cheap borrowing and reindustrialization, all simultaneously, each undermining each other) existed before the war on Iran and will exist afterwards in whatever attempts emerge to preserve the compact of association. You cannot simultaneously monetize your primacy by demanding Gulf states finance your AI buildout, then use that primacy to launch a war that threatens the Gulf infrastructure your buildout depends on. But let’s say it can be done this time...can it be done a second?

The Compute Axis was supposed to build the new imperial architecture; petrodollar to compute-dollar, oil diplomacy to silicon statecraft, Pax Silica, whatever you want to call it and however you want to achieve it, the great white hope is that it we will keep the Red Chinese back while rejuvenating American geostrategic primacy for a new era with a new ubiquitous commodity as its backbone. Instead, the military adventurism of two of the Axis’s members has produced a war that is physically destroying the infrastructure of the third (and the appetite to add onto it). The sovereign capital that was supposed to finance the AI buildout is now needed for reconstruction, fiscal stabilization, and the simple task of keeping the lights on while the Strait of Hormuz is closed. Without Gulf capital, the financial architecture that sustains the buildout (Stargate financing, data center acquisitions, PE anchor commitments, the venture ecosystem’s anchor LPs) loses a load-bearing wall. You can drill for oil elsewhere as we did, sure. But can you conjure up $2 trillion in capital willing to eat unicorn shit quarter after quarter after quarter that’s backed by absolutist regimes whose interests align with ours (whether they are personal or national), either because of contrived artifice or harmonious unity?

The dependency on display here is a fundamental flaw: the American economic model, dependent on chronic fiscal deficits, continuous capital inflows, and reserve currency privilege, structurally requires the kind of capital that only sovereign autocracies can provide. Its intentional design doesn’t make it any less of a ticking time bomb, whether or not the crisis resolves the way American state planners hope.

Hello friends and enemies, here is a little overview of stuff I enjoyed last year starting with my own writing and moving on to things I watched/read/listened to. I have a bit of backlog for essays I need to write, but I’m shifting gears for 2026 to try and keep up with the evolution of our ongoing bubble and its various frontiers—consider this a newsletter reboot. Nonetheless, thank you always to everyone who has read and subscribed and shared and commented and argued and glazed, I’ve deeply appreciated it all and it’s been more helpful than you know.

To start, here is a roundup of some of my writing over the past year on here:

Some Stuff I Wrote This (Last) Year

Trapped In The Maw of a Stillborn God (January)

An honest look at Palo Alto’s past (eugenics, environmental ruin, and surveillance) and present (“less a fascism of blood and soil than a nihilistic capitalism of the bottom line” as Quinn Slobodian puts it) suggests the world we’re racing towards will be dominated by bantustans, though I’m sure the Riverians won’t have much qualms about putting casinos inside of them. The sooner we free ourselves of delusions about Silicon Valley’s supposed right-wing turn, the sooner we can articulate the futures we do or don’t want (and the technologies involved in both) and speak a bit more bravely about the gap between the stakes and our willingness to act. Quickly approaching is the day when we will see the embrace of a genocidal telos (“exterminism”) that’ll seek to sacrifice the environment, genetically and socially engineer humanity, and liquidate the uncooperative elements. All of the ingredients are already there. Now we wait for the Great Work that will bring together the brigands laying waste to our world for one last orgy of violence. Will it be those that seek to purify capitalism of its democratic flaw and colored defects? Or those that promise us it will give birth to yet another stillborn god?

The Silicon Valley Consensus & AI Capex Part 1 (March)

The first part looked at AI as a vector for revitalizing eugenics and shock therapy, which has become more apparent as the world’s richest man uses the pretense of algorithmic efficiency to finish the job Obama, Clinton, Reagan, and Carter couldn’t quite manage: burning away what little remains of the New Deal.

The second part examined the impact of the British Empire’s attempts to make factories resemble plantations on the birth of modern computation, the influence all this had on what we call artificial intelligence, and the weakness of popular frameworks for understanding Silicon Valley (surveillance capitalism, techno-feudalism, techno-authoritarianism, etc).

Part Three is about the Silicon Valley Consensus, a first stab at concretizing the constellation that links computational infrastructure, energy firms (from fossil fuel extractors to energy providers), and various financiers to help explain why various bullshit tech products are foisted upon us. My goal is to help explain how the scramble to build up infrastructure for artificial intelligence is part of a consistent but fragile pattern where new technologies and developments are backstopped by ongoing bids to bolster existing asset classes, synthesize new ones, and turn speculative gains into real wealth. SVC isn’t just about capital expenditures tacked onto the generative AI hype bubble, though this is certainly a major frontier of the froth—which is the subject of the first stab. SVC is ultimately about a now-familiar process that has come to dominate much of the economy and our daily lives: a bunch of independent profit-seeking actors have converged on sustaining a certain technology through a frenzy of overbuilding, overvaluing, and overinvesting in order to realize excessive gains that can be translated into political power aimed at restructuring society.

AI, Indulgences, and the false promise of salvation (May)

If artificial intelligence is contributing to mass illiteracy and cheating, it is because you dragged your feet on adopting it (or adapting to it) in the classroom. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is contributing to an information environment saturated by nonconsensual porn, child sexual abuse material, scams and fraud that utilize realtime deepfakes generated by AI deep learning, or an abundance of AI-generated slop, it is because you dragged your feet on adopting it (or adapting to it) in your information environment. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is threatening the livelihood of your creative workers, it is because you dragged your feet on adopting it (or adapting to it) in cultural production. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is threatening to bolster the fossil fuel industry at the precise moment that sector must be put down for the sake of humanity’s survival, if there is an unimaginable amount of opacity around how much energy infrastructure artificial intelligence will actually need, if data centers are sucking up water everywhere they can manage from deserts to areas gripped by high levels of water stress, it is because you dragged your feet on adopting it (or adapting to it) in energy policy. Close your eyes and fork over some cash to fix this with artificial intelligence.

And so on, and so on.

This Silicon Valley Stuff Will Get You Killed (July)

Some believe sacrifices will restore some semblance of a natural order we’ve lost sight of. The future of human flourishing, they insist, isn’t going to be found in the past few centuries of flirtations with democracy and liberalism, but in a recommitment to Biological Hierarchies that reimpose caste, eugenics, apartheid, terror, and the like. We must administer a harsh treatment for a harsher disease that will cause a great deal of pain and misery in the short-term, but leave us better off in the long-run. That these reactionary ideologies are proving increasingly fundamental to the worldview of the most powerful people in the world and their sycophants, at the same time as this desperate search for capitalist (re)legitimacy, does not bode well for any of us.

These and more horrific exterminist forces are firmly in the driver’s seat, enjoying victory after victory, accumulating greater and greater resources to remake the world into a form more hospitable to their political project(s), and in the course of this self-annihilation they are likely closing the doors on various futures forever—though it will be a long time before we learn which options are lost to us forever.

This unholy alliance—far-right oligarch-ideologues who think democracy and capitalism are incompatible, tech firms with laboratories innovating the armament of fascism, financiers eager to transform speculation into wealth into power, and a host of other demoniacs—is relatively insulated from the public, its concerns, its pressures, its frustrations, and the few levers connected to those that could effect a change. And as a result, it enjoys relatively unimpeded power in building, expanding and legitimizing a police state in this country—a country that has, for a long time now, committed itself to surveillance, social control, force, projection, arbitrary violence, and terror.

It is increasingly unclear to me what, if anything, can be done about this.

Ride-sharing apps are bad, actually (August)

I want to emphasize this is only one prong of what is wrong with Uber but it is a large part of it. If I could sum up everything we’ve covered and a few other key points for you, it would go as follows:

• Uber is an enterprise that regularly uses accounting tricks to obscure its failure to realize profits through innovation. It has taken advantage of superficial coverage in the business press and crafted an aggressively deceptive PR campaign built on older taxi deregulation lobbying playbooks as well as company-sponsored academic research.

• Uber has realized stupendous growth by using capital as a weapon (investor subsidies), breaking the law then rewriting it (regulatory arbitrage), embracing anti-competitive business practices, and deploying incredibly successful political operations in the United States and worldwide that have codified its arbitrage efforts.

• Uber has realized profits largely through predatory behavior such as algorithmic wage discrimination and perpetual fare hikes. Expansions into other lines of business have benefited from its years of experimentation here (namely food delivery).

• Uber has offered a roadmap for other firms interested in immiserating their workers, growing via the misallocation of public subsidies. The metastasis of the so-called “gig economy” will continue unabated as Uber’s bastards proliferate.

• It is bad when a firm uses investor subsidies to distort markets such that it can realize profits through predatory behavior, even worse when it does so with public subsidies, and even worse when this allows the firm to rewrite laws to realize profits locked behind activities made illegal because they are against the public interest, and creates a model for other firms to do so. That Yglesias cannot understand this suggests he is an idiot or operating in a different moral universe.

The Silicon Valley Consensus & The AI Economy (September)

To wrap this up, it’s not clear to me what merits the techno-optimist outlook on whatever constitutes the “AI economy.” It ignores the financials: overlooks the gap between revenues and capex, waves away the question of how it will generate profits, engages in revisionist historical accounts to justify these bad economics, and whistles past the debt land mines that fuel growth. It ignores technology: there’s no interest in market structure, scale of cost, or product roadmaps. It ignores the reality of adoption: gloms onto hype, falls for simplistic narratives, repeats corporate talking points, and reproduces shaky assumptions. We’re left with a picture of reality that leaves us unable to explain why things are the way they are and what to do about it. The “AI economy” as talked about within mainstream and optimist circles presents a vision unmoored from reality that is frothy enough to drown out skeptics, juice speculation, and provide cover for entrenched interests looking to enrich themselves at everyone else’s expense.

On the Origins of Dune’s Butlerian Jihad (September)

In one of the essays in my AI series, I argued that Luther’s critique of indulgences in the medieval era could be applied to today’s Silicon Valley Consensus. Luther was not opposed to indulgences so much as their abuse, which cheapened repentance and undermined attempts to compel good works or genuine attempts to right wrongs. The idea that salvation could be realized through a transaction convinced many they’d obviated the need for the hard work of being a better person. Indulgences also centralized and codified unjustified power grabs by the Church, which claimed new authorities over souls in Purgatory and introduced perverse incentives to prioritize activities that had nothing to do with Christendom.

In some ways, I think of Luddism (and Butlerianism) similarly. My concern is not technology in of itself (though there are multiple technologies we would do better off without). Technology, however, is downstream of politics and economics and history and social relations. We aren’t saying destroy the clocks before they become killer drones, but we are saying the killer drones are already here and we should figure out how to destroy them. Clearly, technological dependence obscures the political and economic decisions about what sort of technologies should be developed, how they should be financed, who should finance their development and reap their rewards and bear their costs, and how society should be organized around the facts of those arrangements.

Is the solution more or less democratic control over technological development and deployment? Do we trust today’s major players in this space to truly prioritize anything other than profits and returns? Are we going to be able to realize or experiment with other values, arrangements, and models that prioritize anything else within today’s authoritarian technological system or within a democratic system? If we realize that certain paths or arrangements or products or models go against human flourishing or the public good or our ecological niche or the mental health of the general public (realizations we have already made), will we be able to do anything about it?

Book reviewers (on Youtube)

These three have read far and wide, are familiar with the SFF genre (which is my favorite) and its history and its authors and movements and milieus and intended audiences (as well as other genres), have very very particular taste and are loud about the limits of their interest, their blindspots, what does and doesn’t work for them, as well as when any of these things shift (revisiting work, new interests, outside events in the news, etc.)

Bookpilled

One of my favorite book reviewers on Youtube right now is Matt via bookpilled, with a separate equally impressive (and inactive) channel on thrifting (Thrift A Life).

He does a variety of free content: book hauls, small batch reviews, large group reviews and rankings, book challenges that feature videos honing in on one particular book, as well as videos honing in on certain authors. He’s also got a Patreon featuring longform reviews of every book he reads.

A quick slice of his stuff: A reflection on the work of Barry Malzberg, five books he hated reading in 2024, looking over a 98 rare/vintage book haul, and his baffling dislike for Book of the New Sun.

Outlaw Bookseller

Going along with bookpilled is Outlaw Bookseller, a Welsh bookseller who has an amazing grasp and familiarity with science-fiction and fantasy. I’ve talked about both of them in previous recommendation posts and like them for similar reasons (they’re both incredibly knowledgeable and well read in the genre)—I will say that Outlaw focuses much more on video essays and most of his videos are tight argumentative pieces or in-depth dives into some aspect of the literature. He’s also got some interviews, something I don’t think Bookpilled has ever dipped into.

A quick slice of his stuff: why Science Fiction isn’t fundamentally woke, on the intellectual and historical origins of genre SF, and his own take on Book of the New Sun.

Cultural Logic

And to round out this section, I really have enjoyed Jeet Heer’s Cultural Logic. I’ve been reading Jeet for a long time at The Nation, where he’s got a weekly podcast (The Time of Monsters) and a monthly column (Morbid Symptoms), but this Youtube channel features some of my favorite work of his. Each of the videos are broadly 20 to 30 minute videos building on written work of his, reviews of recently released media, or deep dives into specific individuals, there’s less of it than the two previous writers but that just means you have that much more to look out for in the next year.

A quick slice of his stuff: Why you should watch Avatar, reflections on forgotten masters like Henry Kuttner and C.L. Moore, fascinating looks at Thomas Pynchon’s V and Gravity’s Rainbow and Vineland, and more.

Music

Despite the best efforts of my roommates and the larger Brooklyn community, I did not get much into Geese. I did, however, spend much of the year hopping around from old songs and music videos I last heard eons ago to persistent favorites across the past few years to new discoveries that have become favorites.

The Old Ones

One old favorite I rediscovered this year was Panic! at the Disco’s “Build God, Then We’ll Talk” though I can’t really remember when I first saw the video. It’s the final song on the band’s debut album A Fever You Can’t Sweat Out came out in September 2005 but two decades ago videos and singles were released a bit more slowly (or maybe it’s better to say over longer periods of time).

“I Write Sins Not Tragedies” is a song I’ve remembered for years and remember watching the video in elementary school (it came out around January 2006) while Build God’s video came out the following year (February 2007). Still, I vaguely recall seeing both on MTV around the same time. Either way, I love the song/video and it actually inspired an essay on artificial intelligence that I’m working on. There are two levels that play off each other well:

1. The video follows The Pornomime, a man who mimes all sorts of sexual acts, as a woman in the audience falls in love with him, marries him, and eventually cheats on him (mimes sex) because he won’t stop having sex (miming) with other people. It ends with him storming home to find her in bed with no one, so he grabs no one and beats them up before storming out. Were they actually in a relationship, were they actually intimate, or was it “a wonderful caricature of intimacy”?

2. The song itself follows a virgin paid for sex by a lawyer at a firm she’s getting a job with, as well as a constable/cop that propositions the woman after the virgin leaves. PRE-WOKE in that it mocks the sex work and resulting pregnancy as caricatures of intimacy.

There’s a lot there to play with for artificial intelligence, for me at least, specifically around the fact that so much of the consumer demand for generative artificial intelligence seems to be companion bots, nonconsensual porn, erotica, and gooning material of various sorts. When it’s not sexual or romantic, it’s still masturbatory—wow, this is the most brilliant idea anyone has ever had, User! If it proves workable you’ll read more about it in the coming year.

Back to the song: it’s nice, I like Panic! I like the bridge that riffs off My Favorite Things, the cello solo, the way some of the lines are sung sounding like they’re racing to catch up or shush the flow, highly recommended!

One song I always wished there was a music video for was Nas’ “I Want to Talk to You”—this would be the second thing I did with a small fortune, right after burning most of it on trying to get an adaptation of GOD EMPEROR OF DUNE off the ground. Over the years Nas has had a tendency to try and make explicitly political songs, but they almost always fall short of the story telling itself. This song kind of waffles between them but it does a lot of things I really like:

• It conjures up a funny image. The premise of the song is Nas storms into the White House with a g-pack and forcing the President at the time (Bill Clinton) to listen to his ranting, shifting between direct address and spacy rambling directed at us

• There’s a good bit of High Hotep talk in here: “Niggas thought that we slept/But the architect of the Pentagon’s from Egypt/government secret, the 99 to 2-G computer shutdown, what now?/Extinction on Earth, human cutdown/Niggas play with PlayStations, they buildin’ space stations/On Mars, plottin’ civilizations”

• Again, just imagining a video that brings to mind The Negotiator (1998) or maybe John Q (2002)? Maybe at some point it evolves to them both sharing some reefer during Verse 2 and then swerves away to a conflict in Verse 3.

• Features one of my favorite lines from 90s era Nas: “Niggas tryin’ to get with the computers, man, y’know I’m saying? We ain’t John Henry banging down a fuckin’ cave fighting against the machine nigga.”

The last one you should really just listen to the entire album. Have you ever listened to a rap album that doubles as a space opera? Introducing DELTRON 3030:

New Crate Finds

I wish I remembered how I found this song, but it really opened up punk music for me this year and I’ve been listening to more post-hardcore/progressive ever since. Just don’t listen too closely to the lyrics (but do a little!)

Another one for the life of me I cannot find but has been welcome.

I haven’t listened to much Steely Dan besides what a close friend plays when driving us around the city and the needle drop in One Battle After Another. I loved this song, which is one of the dumbest songs I’ve ever heard but such a fun groovy time.

Books

I read a lot of epistolaries last year because I’m toying with using the device for part of my sci-fi novel—unsure if I eventually will use the device. It led me to Augustus by John Williams, however, which was my favorite novel of that sort. A historical fiction that invents most of the documents used to (re)create the story of how Gaius Octavius Thurinus becomes Rome’s first emperor, then his struggles trying to plan for succession as his family and various forces conspire to tear things apart.

Karen Hao’s Empire of AI book is a great look at OpenAI, yes, but also about the sort of power firms like it are accruing. It is imperial power in a true sense: resembling the evolution of colonial enterprises that were soaked in blood, deeply exploitative and extractive, rationalized via civilizing missions that adorn the guns with roses and wreaths. It will not look exactly like the imperial power of old or even relatively recently because the world has changed, but not so much because those imperial regimes still exist as do the relationships they carved into the world. What will AI look like as it successfully grafts itself onto governments, their police authorities, and militaries? What will its development look like as our grand old republic tries to forge a new imperial strategy that will sustain/regenerate global primacy?

Never thought I’d read a Pynchon novel with a Nas reference (Bleeding Edge) or Crash (by J.G. Ballard) with a partner (or more accurately, the first few pages because they were so repulsed by it that I had to reread it myself). Bleeding Edge will probably end up being one of my favorite Pynchon novels (still have to finish Shadow Ticket/Mason & Dixon/Inherent Vice) for the horrible and amazing jokes, for the strange plot (who was responsible for 9/11 and who was responsible for its aftermath), for the concern about what happens next, for bits of red meat for me specifically (the tech bubble, Jay/Nas beef, etc.), and so much more. Crash, well I’ll let you read that for yourself. If you are unfamiliar, it is about a bunch of people who develop a new sexuality connected to car crashes. It is one of my favorite novels, but it is also easily one of the most grotesque things (maybe the most) you’ll ever read so steel yourself.

Movies

I saw Lust, Caution at Metrograph with a close friend who’s taken to calling herself my movie domme—an amazing movie based in Hong Kong following Chinese university students who try to honey trap and assassinate a traitor collaborating with the Empire of Japan’s puppet regime in east China. It has such a mindnumbingly depressing ending that we stumbled out of the theater in a haze before remembering (with dread) that it was the night of the Democratic primary. What a pleasant night that was, leaving that horrible movie then learning we were about to witness a three-piece combo.

Saw The Shrouds, Videodrome, and The Fly within three or four days (then rewatched Crash a week later). One of my favorite stretches of movie watching this year. The Shrouds follows a grieving businessman who creates technology that lets you monitor the decomposition of your late loved ones and the vast conspiracy behind the ransacking of multiple graves using this technology. Videodrome prefigures Infinite Jest’s Entertainment, following a CEO who discovers a broadcast of snuff films and gets pulled into a vast conspiracy that consumes and transforms more and more of his life. The Fly is a bit more deceptively simple: a scientist creates teleportation and as he tests it on himself a fly sneaks in and he accidentally fuses with the creature.

None of these descriptions are doing the films justice. David Cronenberg is one of the only filmmakers with anything interesting to say about what technology is as well as what it does to human relations, is a huge romantic to boot, and keen on making movies that break my heart & make me squirm in equal measure. You will get none of that unless you watch them!

Bound and Jade were two of my favorite erotic thrillers I watched this year—the former is such a sleek sexy film, I can’t believe this was the Wachowski sisters first film. Bound is about a lesbian ex-con seduced by a wife next door who hatches up a brilliant scheme: lets steal from my husband, a piece of shit mafia money launderer, pin it all on him, and run away together. What could go wrong? Jade is an erotic thriller that people say is not erotic or a thriller, these people are idiots. They will buzz in your ear about how this means it is a bad movie, about how it wasted the talents of Linda Fiorentino. It is a great movie, she is amazing in it, and it is so clearly a deconstruction of the genre that I want to hit them over the head. The plot follows a DA’s attempt to unravel a conspiracy involving sex, political corruption, and MURDER that goes up to the highest levels. It features what is one of my favorite car chases in any movie, honestly right behind One Battle After Another’s chase, it is full of so many dead ends and wispy mirages that never become anything. As it probably would be in real life.

The Last Seduction was another erotic thriller I watched in this stretch, also starring Fiorentino who I kind of became obsessed with after seeing this movie (and learned that the idiotic reception to Jade eventually sank her career). This film is from the perspective of a femme fatale played by Fiorentino, who steals $700,000 from her husband and fucks off to a small town (Buffalo) where she gets a new name, a new job, and a new mark. She is one step ahead of everyone at every single point of the movie and it is a sight to behold. Anything more would be a spoiler, watch this and all the others as soon as you can!

Next up, you’ll be getting pt 2 of my AI Bubble in 2026 essays. I’ve also been inspired by Brian Merchant posting more speculative fiction here (he just shared Busy by Omar El Akkad, from a short story anthology he published at TERRAFORM that I blurbed). I am sitting on a few dozen short stories, I only write them for readings I’m asked to do—I like to write a new story for each one and use that as an experiment for something I’m thinking about doing in my actual novel. Sometimes I am trying to ape a certain author’s voice or play with one specific part of it or outright steal an image they use, sometimes I’m testing out a character in an intrusive scene that I think might help me understand them, sometimes I’m just tending to the garden and seeing what’s growing, you know how it goes!

To everyone who read and subscribed last year: thank you for supporting my work, with your money or your attention! I hope to be a bit more consistent and ambitious with this newsletter in the new year, here’s to 2026!

My final essay of the year will be split into four parts, laying out areas of the AI bubble I want to focus more on next year. They’re overshadowed by a myopic focus on equity prices, valuations, and capital expenditures—an overcorrection by commentators and talking heads who stubbornly dismissed early AI skeptics. These are all important topics, of course, as they consist of the financial frontier of an AI bubble that is consuming more and more of our economy, but nonetheless are contributing to an obfuscation of the geopolitical and industrial dimensions that’ll have a decisive impact on what our world looks like regardless of whether the bubble bursts or not.

Part One focuses on: how geopolitical ambitions will factor into various actors trying to stabilize or take advantage of the AI bubble in 2026 and beyond. Across the Biden and Trump administrations, the United States has made clear that it views artificial intelligence as integral in its dream of securing hegemonic primacy in the 21st century. Can we anticipate some of the ways that will present itself next year?

Subscribe now

From Oil Diplomacy to Compute Diplomacy

One development I expect to see is advocacy (and even some steps) for a transition from the Petrodollar system of the 20th century to what we might call a Compute-Dollar system in the 21st century.

In Le Monde Diplomatique, Evgenvy Morozov reframes “sovereign AI” offerings as “the final act of a three-act play” of US imperial management, featuring an evolution from “dollar diplomacy” to “oil diplomacy” to “compute diplomacy” centered around deploying our state apparatus and capital to preserve global hegemony:

Act I opened in the early 20th century, when the US promoted dollar diplomacy to Latin American governments as a path to political stability through economic prosperity and sound finance; Theodore Roosevelt used this as a pretext to gain control of the Dominican Republic’s customs collection. By 1912 Brown Brothers bank controlled Nicaragua’s customs collection through loan receivership. The majority of the revenue was collected in New York. When Nicaraguans objected, US marines occupied Nicaragua for 21 years (1912-33), with peak deployment reaching nearly 4,000 troops. In 1922 The Nation called it the ‘Republic of Brown Brothers’.

Act II began in 1974. Nixon had killed the gold standard and the dollar was wobbling. Kissinger flew to Riyadh with an offer: charge whatever price you like for oil, as long as it’s in dollars, and invest your profits in US Treasury bonds – a deal backed by implicit security guarantees and the unmistakable threat that deviation would be treated as hostile to US strategic interests. And between 1974 and 1981, a substantial part of OPEC’s approximately $450bn in accumulated surpluses was reinvested in US Treasuries. No marines required; the threat of capital exile was enough.

Act III is still being written, but the scale of operations exceeds everything we have seen so far. The commodity isn’t bananas or barrels but the raw processing power that lets machines calculate faster than central banks can print money.

In Act III, Morozov envisions that the United States could manufacture a “sovereignty crisis” with some hysteria about compromised datacenters and Chinese chips—the only cure, then, is the American option: US-made chips (Nvidia), US-controlled cloud architecture (Microsoft/Amazon), US-controlled financing (BlackRock, Emirati investment firm MGX), and so on. Export controls, like those that have forced ASML to stop serving Chinese customers with extreme ultraviolet lithography machines necessary to make advanced chips

Navin Girishankar, president of the Economic Security and Technology Department at the Center for Strategic and International Studies (CSIS), is a bit more explicit in advocating for what Act III should look like:

The Trump administration aims to ensure that “American AI technology continues to be the gold standard worldwide,” according to Vice President JD Vance. But these agreements miss an essential ingredient of American power: a guarantee that AI-enabled exports generated using American chips will be invoiced and settled in dollars. Giving other countries access to compute gives them the ability to export AI-enabled goods and services globally. That throws up a critical question: in which currency will they settle that trade—dollars, renminbi, or another currency?

Claims that compute is the new oil are now commonplace, but they often overlook what makes this comparison truly powerful. A country might spend $10 billion on data-center infrastructure using American chips—a one-time capital cost. Those chips can then generate $50–100 billion annually in AI-enabled exports to third countries, including China and the Global South: for example, they may contribute to the autonomous vehicle systems that end up on tens of thousands of vehicles, or they may help a drug discovery algorithm locate a new multi-billion dollar therapeutic. The currency used to invoice and settle these exports is a critical source of global influence.

Failing to deliver a solution to this problem would be tantamount to letting the gold standard collapse without a replacement. Fortunately for Americans, in 1974, U.S. Secretary of State Henry Kissinger and U.S. Treasury Secretary William Simon helped engineer the petrodollar system as a replacement for the gold standard in 1974. The United States didn’t just sell military equipment to Saudi Arabia—it anchored global energy markets to the dollar through an implicit agreement with lasting effects and provided the fiscal capacity that helped the United States win the Cold War.

Girishankar concedes are some fundamental differences between the commoditized foundations of a petrodollar arrangement (”oil is a physical commodity with clear delivery points and standardized pricing) and a compute-dollar system (Ai-enabled services—measured in compute units like FLOPS or AI tokens—are digital, distributed, and harder to track”), but believes that if the compute-dollar system is built upon three explicit principles and enforcement mechanisms—as opposed to the implicit understanding of the petrodollar system—then the dream of global supremacy is still alive.

First, the United States should condition access to leading-edge chips on binding commitments to settle AI-enabled exports in dollars.

Girishankar points to trade deals with Malaysia, Cambodia, Ecuador, Argentina, and Thailand that “already require alignment” with export controls, sanctioned entity restrictions, and investment screening.

Second, the United States should use dollar-backed stablecoins as the settlement mechanism.

Trump has already signed the Genius Act, which created a regulatory framework for the issuance of “payment stablecoins” or digital assets that are backed 1:1 by USD or short-term Treasuries. A compute-dollar system could build on this with digital assets that are pegged to the dollar, provide instant settlement, transparency through a distributed ledger and verifiable records, and sustain dollar dominance instead of yuan adoption.

Third, provide an economic security umbrella—a modern complement to the Cold War-style defense umbrella.

We already provide arms, preferential licensing, trade protections, and other benefits in exchange for export control alignment, so the next step should be to formalize them. Do you want priority access to critical mineral reserves we are stockpiling years too late? Do you want protection from Chinese economic coercion? So long as you join the compute-dollar system, we’re in business.

For Girishankar and others, the choice is clear. Either we use the moment slipping from us to create the foundations for permanent monetary and technological advantage, or we allow AI services to be settled in digital yuan, for the US to lose sustained dollar demand, for Treasury borrowing costs increase, and our ability to fund ambitious national projects.

Sovereign AI (which we will talk more about later), technodollars/compute-dollars, synonyms for: making an American tech stack that allies and clients will be forced to become dependent on, one way or another. These desperate bids will have increasingly central roles in shaping trade agreements, cryptocurrency legitimation efforts, arms deals, security partnerships, foreign investment, and the evolution of how we overbuild, overvalue, and overinvest in AI infrastructure.

The Compute Axis: Trump, Altman, and the Gulf

What political vehicle will meet the task for building this new order? One candidate is the burgeoning coalition that we can describe as the Compute Axis: 1) Silicon Valley and its capital-intensive dream of building God out of sand; 2) Trump and his brigands—concerned with transactional relationships, deregulation, and imperial plunder; 3) the sovereign capital of Gulf sovereigns.

One analysis I cohere with is at American Affairs, where Guy Laron wrote a lengthy essay using Trump’s Gulf tour in May 2025 to try and explain the coming political-economic order and what role our various tech overlords, Gulf monarchs, and domestic oligarchs will play in it.

Art of the Deal

It was to the Gulf monarchies that Sam Altman, chief executive of OpenAI, first pitched his $7 trillion plan to build the physical and digital infrastructure for the coming Age of AI. Such an energy supply and compute capacity buildout simply cannot be done in the United States or Europe, where regulatory constraints and political backlash would kill it at conception. There was a place, however, where he could cobble together the capital, land, and “dispatchable power” (e.g. fossil fuels and nuclear).

In the Compute Axis, one key interlocutor proves to be Sheikh Tahnoun bin Zayed—the UAE’s National Security Advisor, the head of its $100 billion MGX sovereign wealth fund, and G42. Tahnoun seeks to leverage the Gulf’s inordinate oil wealth as part of its own transition from Petrostates to PetroCompute hubs. Zayed’s strategy slots into the longstanding strategy by Gulf states to pivot into the (post-oil) future by citing its fruits—oil wealth, cheap energy, authoritarian governance—as attractions to entice foreign investment.

Back in September, I dove into Saudi Arabia’s particular obsession with this post-oil pivot as exemplified by Vision 2030 and its various tensions as an attempt to centralize the Kingdom around MBS while overhauling its economy, civil society, political system, and the international order in ways that advance his interests.

Whether you believe in the transformative potential of AI or McKinsey’s initial Vision 2030 reports, Saudi Arabia will play a central role in the years to come. Trump’s “Road to Riyadh” tour resulted in $2 trillion worth of announced deals that were pure pay-to-play: we rollback Biden-era export controls on the UAE and provide unrestricted access to advanced chips to Gulf clients, while the monarchies provide the capital necessary to build out AI infrastructure operated by American firms.

Open Veins of the Gulf

Unveiled at the September 2023 G20 Summit in New Delhi, the India-Middle East-Europe Corridor (IMEC) had been initially conceived as the West’s response to China’s Belt and Road Initiative—years too late, but a response of some kind.

Under the Trump administration, it has been aggressively repurposed away from a trade corridor to a digital corridor. Gulf-based energy and compute centers (where models are trained and hosted) will be linked with India’s vast pool of digital labor (where models are refined, debugged, and integrated into services). New high-capacity fiber optic cables will cement India as the AI economy’s “back office,” a role it has already had in other sectors.

As Larson puts it, “India long been seen as the world’s back office: a land of coders, clerks, and call centers” but today it is also home to 1,600 Global Capability Centers (GCCs) which employ 1.66 million professionals involved in “software engineering, data analytics, AI research, cybersecurity, and even core product development.” These GCCs are key to the West’s “digital ambitions at scale and at lower cost” and are a rapidly growing part of India’s services exports (well over a third).

In some ways, however, the IMEC is now truly a response to BRI: the latter is state-led, while the former is a “privatized artery of power,” governed by contracts between sovereign wealth funds, tech monopolies, and family offices (i.e. the Trump Organization) that allow participants to bypass traditional diplomatic bureaucracy.

What emerged from this fusion of corridor geopolitics and digital ambition was not just a policy shift but an operational triangle of power, capital, and access. The Trump-Altman vision depended on the coordination of three actors: U.S.-based AI firms in search of infrastructure and funding, Gulf monarchies eager to reposition themselves as indispensable nodes in the global tech economy, and a Trump family empire that straddled both politics and business.

This alliance has achieved staggering velocity. Gulf sovereign wealth funds now bankroll data centers, chip deals, crypto ventures, and real estate branded by the Trump Organization. AI firms gain capital and regulatory havens. Gulf monarchs secure access to otherwise restricted technology. And Trump reaps the political and financial rewards of acting as matchmaker-in-chief. The strategy delivers for all three sides of the triangle, but the very speed and flexibility that make it so effective also expose its limits. Beneath its coherence lies a deeper fragility: a system defined by speed but hollowed out in its capacity to govern.

Most striking is the contradiction between the coalition’s ambitions and its institutional tools. The entire model is built on bypassing the state: private equity replaces diplomacy, crypto transactions circumvent the banking system, fiber cables and next-generation data centers stand in for treaties and embassies. The coalition runs on velocity, but infrastructure requires durability. Ports, corridors, cables, and compute campuses cannot be run on Signal chats and licensing contracts alone. They demand oversight, regulation, dispute resolution, and long-term coordination, which are all functions that the Trump coalition not only neglects but often openly disdains.

Subscribe now

More on PetroCompute

Abdullah Alzabin’s recent essay, PetroCompute, provides the blueprint for how the G.C.C. plans to execute a maneuver that essentially swapping potential energy (oil) for digital work (inference).

The core of the “PetroCompute” thesis is simple. The United States is hitting a wall: global AI data centers will require an additional 130 GW of power by 2030, the US gas-power generation capacity is projected to increase by only 30 GW. The American grid is old, litigious, and maxed out. Across both parties, backlash to the AI infrastructure overbuild and Silicon Valley’s so-called reactionary turn (in truth, it has always been a bastion of reactionaries).

Central to the logic of the PetroCompute pivot strategy offered by Alzabin is what he calls a “triple advantage” that could let the GCC outmaneuver Washington and Beijing when it comes to deploying inference infrastructure.

The first is an energy advantage. Europe pays $0.29 per kWh, the U.S. averages $0.17, but the Gulf’s unsubsidized power costs average $0.10 per kWh. Through “centralized planning and execution” the GCC might be able to build out power infrastructure rapidly: the Kingdom plans to add 42 GW of gas capacity by 2030, outpacing the United States by 40 percent.

Second is by a geographical advantage. The Gulf sits at the crossroads of three continents and an extensive submarine cable network, meaning it can service four billion internet users within 100 millisecond latency—the threshold that lets AI interactions feel “instantaneous.” If that is not enough, the region has the world’s largest desalination infrastructure (40 percent of global desalinated water). Geographically, it’s well suited to serve the world’s inference workloads and provide more than enough water to cool power-intensive A.I. data centers as the overbuild continues along.

There’s also a financial advantage: a nearly $5 trillion sovereign wealth fund war chest that is a bit more patient than ravenous Western financiers while also having, as Alzabin points out, a “proven capital deployment capability.”

This rare combination of patient capital and execution agility at large scale, enabled by centralized decision-making and compounding capital endowment, allows this group of states to pursue strategic infrastructure at a scale and speed that few other countries and regions can match.

(So long as one ignores the region-wide failures of megaprojects and Vision 20XX plans foisted upon Gulf monarchies by Western consulting firms)

Still, the advantages laid out here are real, as are the four distinct structural traps that might make the Gulf even more subservient to Washington (or Beijing) in the course of attempting a PetroCompute pivot.

The first risk lies in value capture: how to move up the value chain to higher layers of the tech stack. Alzabin invokes the example of AT&T and Apple: when the former became the exclusive partner for the iPhone launch in 2007, it was valued at double Apple’s market cap ($250 billion vs $105 billion). At the time of the essay’s publication, Apple reached a market capitalization of $3.5 trillion while AT&T has stagnated at $165 billion.

As Malaysia, India, and other nations construct gigawatt-scale A.I. infrastructure, G.C.C. states risk investing in infrastructure that, while necessary, lacks differentiation - a potentially commoditised asset with high fixed costs and limited pricing power.

Hyperscalers secure commercial benefits (preferential terms on power, land, and tax) as well as financial advantages (they capitalize the cloud infrastructure they own). This grows their customer base, captures greater value from proprietary apps atop the stack, and in return the Gulf gets “modest foreign direct investment, jobs, and workforce training.” At the moment the Gulf is merely a landlord for intellectual property and, as we’ve talked about and will add on in the next section, an appendage of US imperial management and extraterritorial control.

The second trap is sovereignty. Anchoring the Gulf’s future to America’s bet on A.I. could conjoin the region to American foreign policy even more tightly than under oil diplomacy. In 2020, U.S. senators threatened to block military sales to the Kingdom unless it cut oil production as part of a bid to save U.S. shale producers. Sacrifice your market share to benefit us (your competition). What will be asked of the Gulf in the coming years as the region grows more dependent on U.S. chips and export licenses?

The third trap is the technological obsolescence of physical infrastructure. AI hardware cycles are ruthless: chip capabilities improve year over year, model demands increase year over year, server racks burn out year after year, and so on and so on. This introduces a massive capital expenditure risk: what if the Gulf pours billions into AI infrastructure that quickly becomes obsolete, dotting the map with stranded assets instead of silicon money printers.

Lastly, we have the cannibalization trap and the Jevons Paradox. From 2023 to 2030, AI infrastructure power consumption could grow from 12 terawatt-hours (TWh) of electricity (2 percent of the GCC’s yearly power consumption) to 330 TWh (half of the GCC’s yearly power consumption)—a conservative estimate that excludes inference demand growth. Inference costs do not seem to be coming down but demand will grow, regardless of whether power consumption costs come down or not. Will the GCC start eating into its hydrocarbon exports to power domestic AI infrastructure (eating into its financial advantage)? Would this pivot turn its energy surplus into an energy crisis?

Alzabin offers some paths around these various traps through a few methods:

• Compute-for-Equity: Instead of servers being rented for cash, Gulf states should offer deals to top AI startups where compute is subsidized in exchange for equity (we already see this with CoreWeave & Microsoft/OpenAI).

• Energy diversification: To prevent AI from cannibalizing oil exports, the Gulf should aggressively build out solar and nuclear to power its data centers

• Ecosystem building: Replicate Ireland’s model where data centers aren’t just rental properties, but anchors and hubs used to build local technical talent alongside specialized contracting industries.

The Most Serene Republic of Nvidia

This brings us back to Morozov’s initial essay on Nvidia chief executive Jensen Huang aggressively promoting “Sovereign AI” as part of a marketing strategy to entrench US dominance. Huang’s pitch is that nations must “own the production of their intelligence,” but that the only way to do so is by purchasing billions of dollars worth of Nvidia hardware.

Who volunteers to run this machine for Washington? It’s no longer soldiers (they are only sent to poor countries), but local elites, who show an enthusiasm that would put colonial administrators to shame. Their logic is irrefutable: in a monopolistic world, diversification is tantamount to suicide, and the only sensible choice is to become the monopoly’s accredited agent. Mao used the term ‘comprador bourgeoisie’ to describe Chinese merchants who lived handsomely inserting themselves between foreign capital and the domestic economy. Today, computing power has replaced opium, but the margins are just as fat.

A key part of Morozov’s framework is Mao’s description of “comprador bourgeoise,” used to describe “Chinese merchants who lived handsomely inserting themselves between foreign capital and the domestic economy.” In the Republic of Nvidia, national sovereignty boils down to the privilege of writing checks to US corporations and facilitating foreign capital flows. Local elites in Europe and Asia climb over each other for the chance to make their countries subservient in hopes that it’ll secure their position in the value chain.

One example Morozov points to is French President Macron’s “sovereign AI” strategy which involves a €109 billion investment plan flowing primarily to Nvidia for its chips and Microsoft for its cloud infrastructure. French firms like Mistral, however, are relegated to the role of junior partner.

Other examples abound:

SoftBank has travelled furthest: once channeling Japanese savings into domestic startups, it’s now investing $48bn in US AI companies (OpenAI, Ampere, Nvidia), though it has only $31bn in cash reserves. It will borrow to make up the shortfall. When SoftBank asked Japanese banks for $13.5bn to finance its next US splurge, they volunteered $27bn.

Deutsche Telekom, once Deutsche Bundespost laying copper for German factories, now markets an ‘industrial AI cloud’ powered by 10,000 Nvidia Blackwell GPUs – designed in Santa Clara, fabricated in Taiwan, booked through Delaware. Berlin owns 32%, but 68% belong to global funds. This is sovereignty in name alone, with the bulk of profits flowing westwards.

Even the most stubborn have given in. Chinese giants such as ByteDance, Alibaba and Tencent, thought to share Beijing’s strategic priorities, are quietly amassing contraband Nvidia chips, despite government pressure, national security concerns and the availability of cheaper (though always inferior) equivalents from Huawei.

Another key pillar proves to be how the United States realizes extraterritorial control through legal mechanisms:

• The Clarifying Lawful Overseas Use of Data (CLOUD) Act, which sets up a legal framework for U.S. access of data stored overseas as well as foreign access to data held by U.S. firms.

• The Foreign Direct Product Rule (FDPR), an export control rule that allows the U.S. to prohibit the sale of products made with American tech, even if made in a foreign country. U.S. sovereignty now extends into the “atoms” of any “chip, wafer or screw that has brushed up against American software or research dollars”

• The Chip Security Act, proposed this may, which would “make it compulsory to fit Nvidia’s H11 and B200 chips with location-tracking system[s]. The same kind of surveillance architecture that the West accused Huawei of building into its products would become federal policy, but only applying to American chips.“ (emphasis added)

The end result being, of course, an offer you can’t refuse:

Access to China’s market, rare earth metals and AI models means not only rejecting the binary choice Washington offers – us or them, dependence or isolation, integration or exile – but also risking capital flight, the possibility of having assets frozen, a hostile security architecture, sticks instead of carrots. In many countries, it’s not the ability to say no that’s lacking, it’s the will to endure what follows.

Final Thoughts

So in the year to come: on the geopolitical front I’m going to be interested in attempts to concretize a compute-dollar system, moves made by the Compute Axis, the Gulf’s attempts to manage its role in the global AI value chain, and how “Sovereign AI” will be used to lock allies and clients into dependence on our tech stack. These projects overlap and contradict. Some require vertical integration under U.S. control, some require horizontal outsourcing to private actors and client states. Some require decisive action and speed and flexibility. But all do require permanent and durable institutions to build some sort of global technological system, even if it is just an appendage of a desperate gambit to preserve geostrategic primacy. Is any of this going to work? I hope not. Do you want to live in a world where America manufactures sovereignty crises to graft its national champions onto various countries? Do we want to have a global competition among elites racing to offer up their nations’ digital futures on a platter in hopes for crumbs? Probably not. Which parts of this vision will succeed, which will fail, and which will transform into something more ugly?

Leave a comment

What’s happened since #9

Zohran Mamdani won and given us a wonderful three-piece combo (beat Cuomo in the primary, beat Cuomo in the general despite the Adams endorsement, and charmed a good deal of the party—and seemingly Trump?)

I libbed out that night and morning, talked to NYT about libbing out while recovering from it, and in the weeks sense have gotten more and more of a sense of the coming agenda—which we will talk about today.

I’ve also been working on transcribing episodes of my podcast, This Machine Kills (which covers many of the same subjects as this newsletter, and will be sharing some digests based off those transcripts in the coming weeks.

Also: Me, Jathan Sadowski, Brian Merchant, and Paris Marx were all in town for a Neo-Luddite conference earlier this month featuring a good chunk of our comrades, some curious interlocutors, and a younger generation of Luddites (the Luddite Club). It was wonderful and was followed up by a Luddite Tribunal at Starr Bar the next night—our second one there. Thanks to all who came out, the bar was packed and we had a long line of volunteers offering up pieces of technology to be put on trial and either acquitted or sentenced to death (getting sledge hammered to pieces). I’ll be trying to throw more Luddite tribunals more frequently so it’s not just once a year, the people yearn for kangaroo court justice.

Some of the things we destroyed (after discussing the political economy backstopping their production/distribution, as well as their net impact on society): Ray-Ban Meta glasses, a printer, an Alexa, an iPad, a children’s toy that simulated a smartphone, and an Xbox 360 (many people spoke up in defense of the console but the jury called for its blood).

Also: I’ll be giving a talk at Yale with Molly Crabapple titled Resisting AI’s Impact on the Creative Industry on December 4. If you want either of us (or both) to come to your school/workplace/organization to corrupt the youth via a talk about Luddism, shoot me a message and let’s chat!

This week, the recommendations include, but are not limited to:

Mamdani’s budget, bond markets, tech policy, venture capital’s perpetual value machine, concrete elements of the AI bubble, the limits of AI bubble discourse, Epstein as a pedophile warlord, Pynchon’s Luddism, China envy, and more.

Let’s begin!

Read more

Last week, a former Disney star (Calum Worthy) took to Twitter to reveal his new app: 2wai, which claimed it would use artificial intelligence and a three-minute video to digitally clone a user’s loved ones.

To promote this new grift—uhh, I mean enterprise—Worthy shared a short video online about what life with 2wai might be like, captioned “What if the loved ones we’ve lost could be part of our future.”

The video begins with “Baby Charlie” floating on the screen. A pregnant woman is talking to her elderly mother, showing her bump and sharing that her baby has been kicking a lot lately. The older woman gives her daughter some advice, “Put your hand on your tummy and hum to him. You used to love that.” And so the expecting mother does; “It feels like he’s dancing in there.”

We cut to “10 months later” and Charlie is a big ass baby now. His mother sits down and brings him close to her phone, “Mom, would you tell Charlie that bedtime story you used to always tell me?” Her elderly mother is on the phone now and she hasn’t aged a day. As she tells Charlie the bedtime story, her daughter begins to cry.

Now it’s “10 years later” and we see Charlie again, a kid walking home from school holding a featureless smartphone (NO COPYRIGHT INFRINGED UPON) and video chatting his elderly grandmother (who still hasn’t aged a day). He tries to talk to her about basketball and she interjects, “I don’t really care that much about basketball. What about...” there’s a mischievous pause, “the crush?” The boy tries to wave it off, “Stop, grandma, stop talking,” he repeats over her teasing plea to “Just tell me one thing!”

We cut to “30 years old” and Charlie is holding up a picture of an ultrasound to his Featureless Smartphone. “Look who’s going to be a great grandmother.” Some things never change. His grandmother hasn’t aged a day. His wife is complaining that the baby has been kicking a lot lately. “Tell her to put her hand on her belly and hum to him. You loved that,” Granny tells Charlie. “You would have loved this moment,” Charlie glumly replies. “You can call anytime.”

Now it’s “Before Charlie” and the elderly woman is being filmed by her daughter (Charlie’s mother) for a three minute video. They’re laughing, cracking jokes, and the screen cuts to white: “With 2wai, three minutes can last a lifetime.”

When I first saw this ad, my impulse was to write something about it because I felt there was no shortage of angles to cover. Where to start?

The silent admission in the ad that this digital avatar is not your loved one, even as the entire video pretends otherwise. Or the similarity here to an episode from Season 2 of Black Mirror (”Be Right Back”)—a woman’s boyfriend dies in a car accident and, after she learns she’s pregnant, is brought back with a “digital clone” from his social media posts that communicates with her through texts, then phone calls, then IRL after she embodies it in a synthetic body.

We could talk about the fact that I’m struggling to think of a single science fiction story where a good response to grief might feature digitally cloning your lost loved one and keeping them trapped in your phone for decades as your life moves on. Why must everything, even death, be subjected to the bureaucratic rationality that has swallowed the horizon of what’s possible with computation?

Or we could contextualize this startup with Worthy’s past comments about what inspired this project—not a desire for connection with loved ones but a desire to capitalize on parasocial relationships:

“Having worked as an actor, writer and producer for the last 20 years, I experienced firsthand how challenging it is to create a meaningful relationship with fans around the world,” Worthy said. “With the launch of 2wai, we’re enabling entirely new experiences that are as authentic as the HoloAvatars we create.”

…

“2wai really puts the control back in the hands of the artists,” said Worthy. “This lets them engage fans 24-7 without needing to be online all the time. This platform enables this one-on-one, human-like connection.”

But all of that is really downstream of what I’ve been interested in for years: understanding the reactionaries who’ve used twisted varieties of techno-utopian ideas to trap us in today’s death cult. We can all understand wanting to live a little longer or getting extra moments with loved ones, but tech capitalists have taken these longstanding impulses to an irrational extreme in hopes of getting more: accumulating more time to accumulate more wealth to accumulate more power over their biology, our sociality, and (one day, they hope) reality itself.

As it turns out, I wrote an essay five years ago for VICE that is relevant to 2wai, but gets to the larger issue at hand: Silicon Valley’s insistence on perverting everything it touches, including death.

Read it below.

The Proto-Communist Plan to Resurrect Everyone Who Ever Lived

Is there anything that can be done to escape the death cult we seem trapped in?

One of the more radical visions for how to organize human society begins with a simple goal: let’s resurrect everyone who has ever lived. Nikolai Fedorov, a nineteenth-century librarian and Russian Orthodoxy philosopher, went so far as to call this project “the common task” of humanity, calling for the living to be rejuvenated, the dead to be resurrected, and space to be colonized specifically to house them. From the 1860s to the 1930s, Fedorov’s influence was present throughout the culture—he influenced a generation of Marxists ahead of the Russian Revolution, as well as literary writers like Leo Tolstoy and Fyodor Dostoevsky, whose novel, The Brothers Karamazov, directly engaged with Federov’s ideas about resurrection.

Note: The first 18 seconds of the video feature a bright white background that flashes and pulses rapidly.

After his death, Federov’s acolytes consolidated his ideas into a single text, A Philosophy of the Common Task, and created Cosmism, the movement based on his anti-death eschatology. Federov left the technical details to those who would someday create the prerequisite technology, but this did not stop his disciples: Alexander Bogdanov—who founded the Bolsheviks with Lenin—was an early pioneer of blood transfusions in hopes of rejuvenating humanity; Konstantin Tsiolkvosky, an astrophysicist who was the progenitor of Russia’s space program, sought to colonize space to house the resurrected dead; and Alexander Chizhevsky, a biophysicist who sought to map out the effects of solar activity on Earth life and behavior, thought his research might help design the ideal society for the dead to return to.

The vast majority of cosmists were, by the 1930s, either murdered or purged by Stalin, muting the influence of their ambitious project but also leaving us with an incomplete body of work about what type of society resurrection requires or will result in, and whether that would—as some cosmists believe now—bring us closer to the liberation of the species. Now, I think it is obvious that—despite what today’s transhumanists might tell you—we are in no position, now or anytime soon, to resurrect anyone let alone bring back to life the untold billions that have existed across human history and past it into the eons before civilization’s dawn.

To be clear, I think cosmism is absolute madness, but I also find it fascinating. With an introduction to Cosmism and its implications, maybe we can further explore the arbitrary and calculated parts of our social and political order that prioritize capital instead of humanity, often for sinister ends.

What? Who gets resurrected? And how?

At its core, the Common Task calls for the subordination of all social relations, productive forces, and civilization itself to the single-minded goal of achieving immortality for the living and resurrection for the dead. Cosmists see this as a necessarily universal project for either everyone or no one at all. That constraint means that their fundamental overhaul of society must go a step further in securing a place where evil or ill-intentioned people can’t hurt anyone, but also where immortality is freely accessible for everyone.

It’s hard to imagine how that world—where resources are pooled together for this project, where humans cannot hurt one another, and where immortality is free—is compatible with the accumulation and exploitation that sit at the heart of capitalism. The crisis heightened by coronavirus should make painfully clear to us all that, as J.W. Mason—an economist at CUNY—recently put it, we have “a system organized around the threat of withholding people’s subsistence,” and it “will deeply resist measures to guarantee it, even when the particular circumstances make that necessary for the survival of the system itself.” Universal immortality, already an optimistic vision, simply cannot happen in a system that relies on perpetual commodification.

Take one small front of the original cosmist project: blood transfusions. In the 1920s, after being pushed out of the Bolshevik party, Bogdanov focused on experimenting with blood transfusions to create a rejuvenation process for humans (there’s little evidence they do this). He tried and failed to set up blood banks across the Soviet Union for the universal rejuvenation of the public, dying from complications of a transfusion himself. Today, young blood is offered for transfusion by industrious start-ups, largely to wealthy and eccentric clients—most notably (and allegedly) Peter Thiel.

In a book of conversations on cosmism published in 2017 titled Art Without Death, the first dialogue between Anton Vidokle and Hito Steyerl, living artists and writers in Berlin, drives home this same point. Vidokle tells Steyerl that he believes “Death is capital quite literally, because everything we accumulate—food, energy, raw material, etc.—these are all products of death.” For him, it is no surprise we’re in a capitalist death cult given that he sees value as created through perpetual acts of extraction or exhaustion.

Steyerl echoes these concerns in the conversation, comparing the resurrected dead to artificial general intelligences (AGIs), which oligarch billionaires warn pose an existential threat to humanity. Both groups anticipate fundamental reorganizations of human society, but capitalists diverge sharply from cosmists in that their reorganization necessitates more extraction, more exhaustion, and more death. In their conversation, Steyerl tells Vidokle:

Within the AGI Debate, several ‘solutions’ have been suggested: first to program the AGI so it will not harm humans, or, on the alt-right/fascist end of the spectrum, to just accelerate extreme capitalism’s tendency to exterminate humans and resurrect rich people as some sort of high-net-worth robot race.

These eugenicist ideas are already being implemented: cryogenics and blood transfusions for the rich get the headlines, but the breakdown of healthcare in particular—and sustenance in general—for poor people is literally shortening the lives of millions … In the present reactionary backlash, oligarchic and neoreactionary eugenics are in full swing, with few attempts being made to contain or limit the impact on the living. The consequences of this are clear: the focus needs to be on the living first and foremost. Because if we don’t sort out society—create noncapitalist abundance and so forth—the dead cannot be resurrected safely (or, by extension, AGI cannot be implemented without exterminating humankind or only preserving its most privileged parts).

One of the major problems of today’s transhumanist movement is that we are currently unable to equally distribute even basic life-extension technology such as nutrition, medicine, and medical care. At least initially, transhumanists’ vision of a world in which people live forever is one in which the rich live forever, using the wealth they’ve built by extracting value from the poor. Today’s transhumanism exists largely within a capitalist framework, and the country’s foremost transhumanist, Zoltan Istvan, a Libertarian candidate for president, is currently campaigning on a platform that shutdown orders intended to preserve human life during the coronavirus pandemic are overblown and are causing irrevocable damage to the capitalist economy (Istvan has in the past written extensively for Motherboard, and has also in the past advocated for the abolition of money).

Cosmists were clear in explaining what resurrection would look like in their idealized version of society, even though they were thin on what the technological details would be. Some argue we must not only restructure our civilization, but our bodies so that we can acquire regenerative abilities, alter our metabolic activity so food or shelter are optional, and thus “overcome the natural, social, sexual, and other limitations of the species” as Arseny Zhilyaev puts it in a later conversation within the book.

Zhilyaev also invokes Federov’s conception of a universal museum, a “radicalized, expanded, and more inclusive version of the museums we have now” as the site of resurrection. In our world, the closest example of this universal museum is the digital world “which also doubles as an enormous data collector used for anything from commerce to government surveillance.” The prospect of being resurrected because of government/corporate surveillance records or Mormon genealogy databases is “sinister” at best, but Zhilyaev’s argument—and the larger one advanced by other cosmists—is that our world is already full of and defined by absurd and oppressive institutions that are hostile to our collective interests, yet still manage to thrive. The options for our digital world’s development have been defined by advertisers, state authorities, telecom companies, deep-pocketed investors, and the like—what might it look like if we decided to focus instead on literally any other task?

All this brings us to the question of where the immortal and resurrected would go. The answer, for cosmists, is space. In the cosmist vision, space colonization must happen so that we can properly honor our ethical responsibility to take care of the resurrected by housing them on museum planets. If the universal museum looks like a digital world emancipated from the demands of capital returns, then the museum planet is a space saved from the whims of our knock-off Willy Wonkas—the Elon Musks and Jeff Bezos of the world. I am not saying it is a good or fair idea to segregate resurrected dead people to museum planets in space, but this is what cosmists suggested, and it’s a quainter, more peaceful vision for space than what today’s capitalists believe we should do.

For Musk, Mars and other future worlds will become colonies that require space mortgages, are used for resource extraction, or, in some cases, be used as landing spots for the rich once we have completely destroyed the Earth. Bezos, the world’s richest man, says we will have “gigantic chip factories in space” where heavy industry is kept off-planet. Beyond Earth, Bezos anticipates humanity will be contained to O’Neill cylinder space colonies. One might stop and consider the fact that while the cosmist vision calls for improving human civilization on Earth before resurrecting the dead and colonizing space, the capitalist vision sees space as the next frontier to colonize and extract stupendous returns from—trillions of dollars of resource extraction is the goal. Even in space, they cannot imagine humanity without the same growth that demands the sort of material extraction and environmental degradation already despoiling the world. Better to export it to another place (another country, planet, etc.) than fix the underlying system.

Why?

Ostensibly, the “why” behind cosmism is a belief that we have an ethical responsibility to resurrect the dead, much like we have one to care for the sick or infirm. At a deeper level, however, cosmists not only see noncapitalist abundance as a virtue in of itself, but believe the process of realizing it would offer chances to challenge deep-seated assumptions about humanity that might aid political and cultural forms hostile to the better future cosmists seek.

Vidokle tells Steyerl in their conversation that he sees the path towards resurrection involving expanding the rights of the dead in ways that undermine certain political and cultural forms,

“The dead … don’t have any rights in our society: they don’t communicate, consume, or vote and so they are not political subjects. Their remains are removed further and further from the cities, where most of the living reside. Culturally, the dead are now largely pathetical comical figures: zombies in movies,” he said. “Financial capitalism does not care about the dead because they do not produce or consume. Fascism only uses them as a mythical proof of sacrifice. Communism is also indifferent to the dead because only the generation that achieves communism will benefit from it; everyone who died on the way gets nothing.”

In another part of their conversation, Steyerl suggests that failing to pursue the cosmist project might cede ground to the right-wing accelerationism already killing millions:

There is another aspect to this: the maintenance and reproduction of life is of course a very gendered technology—and control of this is on a social battleground. Reactionaries try to grab control over life’s production and reproduction by any means: religious, economic, legal, and scientific. This affects women’s rights on the one hand, and, on the other, it spawns fantasies of reproduction wrested from female control: in labs, via genetic engineering, etc.

In other words, the failure to imagine and pursue some alternative to this oligarchic project has real-world consequences that not only kill human beings, but undermine the collective agency of the majority of humanity. In order for this narrow minority to rejuvenate and resurrect themselves in a way that preserves their own privilege and power, they will have to sharply curtail the rights and agency of almost every other human being in every other sphere of society.

Elena Shaposhnikova, another artist who appears later in the book, wonders whether the end of death—or the arrival of a project promising to abolish it—might help us better imagine and pursue lives beyond capitalism:

“It seems to me that most of us tend to sublimate our current life conditions and all its problems, tragedies, and inequalities, and project this into future scenarios,” she said. “So while it’s easy to imagine and represent life in a society without money and with intergalactic travel, the plot invariably defaults to essentialist conflicts of power, heroism, betrayal, revenge, or something along these lines.”

In a conversation with Shaposhnikova, Zhilyaev offers that cosmism might help fight the general fear of socialism as he understands it:

According to Marx, or even Lenin, socialism as a goal is associated with something else—with opportunities of unlimited plurality and playful creativity, wider than those offered by capitalism. … the universal museum producing eternal life and resurrection for all as the last necessary step for establishing social justice.

In the conversations that this book, cosmism emerges not simply as an ambition to resurrect the dead but to create, for the first time in human history, a civilization committed to egalitarianism and justice. So committed, in fact, that no part of the human experience—including death—would escape the frenzied wake of our restructuring.

It’s a nice thought, and something worth thinking about. Ours is not that world but in fact, one that is committed, above all else, to capital accumulation. There will be no resurrection for the dead—there isn’t even healthcare for most of the living, after all. Even in the Citadel of Capital, the heart of the World Empire, the belly of the beast, the richest country in human history, most are expected to fend for themselves as massive wealth transfers drain the public treasuries that might’ve funded some measure of protection from the pandemic, the economic meltdown, and every disaster lurking just out of sight. And yet, for all our plumage, our death cult still holds true to Adam Smith’s observation in The Wealth of Nations: “All for ourselves, and nothing for other people, seems, in every age of the world, to have been the vile maxim of the masters of mankind.”

Next free essay

Another stab at contrasting the Silicon Valley Consensus to more narratives about the AI economy, this time focusing on a bit more on narratives that dismiss the financial risk of overbuilding AI infrastructure

What I’m working on

This week’s episode of This Machine Kills features a conversation with Thea Riofrancos, author of “Extraction: The Frontiers of Green Capitalism” as we talk about the political economy, material infrastructure, and extractive industries criss-crossing the planet as “green capitalism” roars to life.

I’m going to be in Ethiopia this week where I’m speaking at a pan-African biotechnology conference. I may edit and expand my speech into an essay for the paywall, but the thrust of it is that as firms and governments in the major blocs on either side of the US-China Cold War start sniffing around to bolster their tech and energy stacks, African countries should play both blocs against each other to get the best possible deal.

Additionally, I’ve been organizing more and more neo-Luddite events, primarily teach-ins and workshops—if you’re interested in having me or fellow tech critics come in and speak to your class or on your campus or your organization (especially if you’re part of a union), reach out here, over e-mail (edwardongwesojr@gmail.com), or on Signal (@bigblackjacobin.29).

This week, the recommendations include, but are not limited to

How American policy created the Darién Gap, why OpenAI needs a trillion dollars over the next four years, a tripartite book review examining chips and the wake of the US-China Cold War, the digital cloud’s externalities, an Andrew Ferguson interview on the ascendance of tech capital, the incestous financing of OpenAI-NVIDIA-Oracle, prediction markets disrupting gambling, a clear-eyed review of a new AI doomer book, and global realignment via electrostates vs petrostates.

Let’s get into it!

Read more

An excellent piece from Jacob Silverman in the October 2025 issue of The Nation titled "Elon Musk's AI Grift" recently came out and it does a great job surveying Musk's business empire, the state of the AI sector, and his attempts to create a financial ecosystem insulated from pressures that might undermine his various projects. Saudi Arabia and the Gulf have become increasingly central to this endeavor.

Now, as Silverman lays out, it is a bit funny that Saudi Arabia has invested in both xAI (a $800 million stake) and x/Twitter (a $1.9 billion stake) considering this is a country which maintained spy rings inside Twitter for years. Nonetheless, if the world can look past Crown Prince Mohammed bin Salman arranging the grisly murder and dismemberment of Saudi journalist Jamal Khashoggi, then what’s a little espionage aimed at critics between friends:

“The merger of X and xAI helped muddy, if not obscure, X’s checkered investment history—and its troubling role as a tool of state surveillance. The blob of capital and hype that makes up the Muskonomy has helped to further insulate the platform from anything resembling public accountability. The two companies already shared personnel and investors; now they are more enmeshed than ever. The deal intensified the laundering of responsibility for the Saudi spy ring, the most prominent instance of Silicon Valley’s widespread—and growing—vulnerability to espionage.

One simple reason that Saudi Arabia’s outsize and troubling role in the tech world hasn’t drawn wider attention is that money is a moral lubricant. Some tech billionaires appear to consider the infiltration of their companies by foreign dictatorships simply a cost of doing business. And some seem to share their authoritarian instincts.”

The reforms that the Kingdom have undertaken have not been, to the chagrin of the Kingdom’s various cheerleaders, interested in doing anything about the long-obvious fact that MBS would be a brutal autocrat. In fact, it seems Silicon Valley has actually been catching up to Washington’s position here: understanding that everything and anything is permitted so long as there is a transaction. Still, the coziness between Saudi Arabia and the Silicon Valley is integral to any attempt at describing what has become of global technological development, especially in the sectors involved with what we call artificial intelligence.

I want to use this as a chance to revisit an essay I abandoned last year on Saudi Arabia’s role in our tech bubble and its attempts to ingratiate itself in the coming global order. I’ve been working on a related essay linking Saudi Arabia to the Silicon Valley Consensus framework, so I figured this was relevant since it’ll be built on iteratively and because the core thesis still stands: Saudi Arabia’s Vision 2030 investments are less about economic diversification than about consolidating authoritarian control and securing geopolitical importance resembling what the Kingdom enjoy thanks to oil. Increasingly, that means being enmeshed with speculators, crypto, real estate tycoons, and Big Tech firms (more on that coalition next time).

One way to look at the Gulf’s role is that the region’s various schemes to pursue megaprojects and generate revenues independent of fossil fuels are, in part, stillborn attempts to preserve the geopolitical import these states have enjoyed for decades thanks to oil. This ambition has found a natural partner in Silicon Valley’s own dreams of usurping and consolidating control over America’s political and economic order. Over in Palo Alto, techno-nationalists eagerly chase exterminist futures by day before turning in for the night and dreaming of a world where a trillion network states bloom, each sitting somewhere on a spectrum between Little St. James and Prospera.

As the United States deepens its Cold War with China, as BRICS asserts itself globally, as Europe dithers between Washington and Beijing, the Kingdom has not going hesitated to negotiate the best deal with whoever it can. It seeks to secure stakes in various startups, partner with governments to build multi-gigawatt data centers, invest Smaugian hoards of capital into venture capital funds, work with more traditional Wall Street financiers, and increasingly cooperate with Chinese firms, all to attract the talent, tech, and capital necessary to pursue its own interests.

The way Saudi Arabia’s investments into tech are typically understood, however, are through the lens of Vision 2030—the Kingdom’s diversification strategy aimed at transforming a state whose economy, civil society, social programs, and political stability are all contingent on extract, processing, and selling petroleum (by)products. Nevermind that the Kingdom’s Vision wasn’t even the first (America’s concrete industry released one in 2001, Kenya’s came from a “development process” across 2006-2008, Pakistan’s was published in 2007, Abu Dhabi’s in 2008, the World Health Organization’s in 2010, South Africa’s in 2011, the North Carolina Chamber of Commerce’s in 2012, and so on). The fingerprints of management consultants are unmistakable; outside of the Kingdom, few clients have made it painfully obvious—the Saudi Vision came almost immediately after a McKinsey & Company published a few months before the diversification strategy was announced.

In 2015, after the Kingdom’s first deficit since 2007 (an eye-watering $98 billion) thanks to the 2014-15 collapse of oil prices, commentators warned that drastic reforms were necessary to save the country from an “economic time bomb”: its reliance on oil exports. Over the next decade, the record is mixed on its attempts to bolster alternative sources of revenues while restructuring the economy, with some tech investments promising to bolster the country’s green tech stack while others are certain to overexpose Saudi Arabia to tech startup overvaluation and asset bubbles related to AI infrastructure.

Vision 2030 tends to constantly shift under scrutiny because its core ambitions outstrip its capacities–even those of its key lever: the country’s sovereign wealth fund, the Public Investment Fund (PIF). PIF is simultaneously expected to be: an investment vehicle funding grand projects aimed at minimizing Kingdom’s reliance on oil; a power play by Crown Prince Mohammed bin Salman (MBS) to personally consolidate power; a channel to translate overseas investment into economic diplomacy; an opportunity to divorce geo-strategic importance from oil.

Whether the Vision fails or not is up in the air, though it may not come soon enough to stop the Kingdom from coughing up trillions of dollars worldwide into technological developments that suit its aims: surveillance and force projection, sci-fi moonshot projects, extractive and dispossessive digital technologies—in other words, the same suite of goods and services our own financiers and techno-capitalists eagerly await.

On December 1 2015, the McKinsey Global Institute released its report envisioning "productivity-led economic transformation" by 2030—but only if Saudi Arabia privatized swaths of the economy, curtailed subsidies, and invested in developing revenue streams independent of oil.

Weeks later, the Saudi Finance Ministry announced a 2016 budget nearly identical to McKinsey's, aimed at: "privatizing a range of sectors and economic activities," as well as "investing in development projects and programs that serve the citizens directly" and enacting a “revision of energy, water, and electricity prices." In January 2016, MBS used his first on-the-record interview to hint at more to come: sin and consumption taxes, as well as the transformation of "unutilised assets" such as state-owned land in Mecca and Medina, uranium reserves, religious and cultural tourism, along with fundraising via IPOs of state-owned companies like Aramco. When “Saudi Vision 2030” was announced months later in April, there were few surprises in this respect.

Privatization of key state-owned sectors (e.g. healthcare, air travel) would prove to be a central pillar of MBS’s neoliberal vision. One bright example: Saudi Arabia is one of the world’s largest arms importers, in 2016 importing about 98 percent of its military arms. Vision 2030 painted a rosy picture of an “indigenous” arms industry that manufactured 50 percent of Saudi weapons. The goal here was to scale up industries that could soak up employees laid off once public sector payrolls were slashed, while also providing jobs to the youth (70 percent of the population was under 30, with unemployment rate estimates ranging from 12 to 29 percent). To further bolster labor markets, employers would be encouraged to tap into a "great asset" known as “women” by hiring them in greater droves.

Other assets, specifically Aramco shares, would be transferred to PIF in pursuit of an outrageous goal: growing one of the world’s ten largest sovereign wealth funds ($152 billion worth of assets in 2015) into a $2 trillion investment juggernaut able to aid privatization while jumpstarting a domestic electric vehicle industry, expanding its petrochemical sector, and slew of other developmental infrastructure projects.

There would also need to be a massive expansion of non-oil revenue. The Ministry of Finance claimed it would increase non-oil revenue from $43 billion to $266 billion. For the Kingdom, this requires—on top of the fundamental overhauls listed above—massive levels of foreign direct investment (an estimated $100 billion) to build its desired private sector and generate private consumption. There would also need to be new taxes on top of the already proposed sin and consumption taxes, such as a tax targeting luxury goods and undeveloped urban land. Expenditures would, of course, be drawn back—specifically water and electricity subsidies which both the Kingdom and IMF claimed largely benefited elites. In the face of price hikes, middle and lower classes would be offered cash transfers to soften the blow.

Vision 2030 also called for megaprojects (now “gigaprojects” by the more sycophantic commentators) to help attract 100 million visitors. $1.5 trillion worth of construction for NEOM, including an artificial moon and a linear city called The Line. $63 billion for Diriyah Gate, a cultural capital at the ancestral seat of the Saud dynasty promising to attract 27 million annual visitors by 2030. Qiddiya, envisioned as a sports and entertainment capital, had a budget of $50 billion at one point. $20 billion for the Red Sea Project, $23 billion for King Salman Park, $50 billion for a new city district centered around consumerist Ka’ab known as Murabba (“The Cube”).

On top of all this, Vision 2030 calls for political and cultural reforms aimed at making direct investment and tourism more attractive, partly measured by reaching 100 million tourists a year by 2030 (a goal it reached this year). To call Saudi Arabia ultraconservative would be an understatement. An absolutist monarchy with the civil rights of a US prison, the state’s legitimacy stems from the first Saudi state’s pact between the founder of the Saudi dynasty and the Wahhabi Islam movement. Non-adherents are regularly discriminated against across Saudi society. Attempts to replace Wahhabi Islam with “moderate Islam”, however, have given way to political repression by another name. Political dissidents are detained, publicly executed, or assassinated from afar. Political rivals to MBS have also been detained, purged, and had their assets seized. The Kingdom's nearly 11 million foreign workers (not including undocumented workers)—the majority of the private sector workforce—are forced into virtual slavery as they are "used, abused and deported" without pay to build MBS’s dream.

In the face of these impossible demands, has the Saudi Thatcherite Revolution finally arrived? Will MBS make the desert bloom?

Since 2016, it’s been clear that a $2 trillion PIF required significantly higher crude oil prices or selling larger stakes in Aramco. The IMF estimated $96 per barrel was necessary for Vision 2030, a level breached only once this past decade: in 2022, after Russia’s invasion of Ukraine and subsequent sanctions. Non-oil revenues have grown to 38 percent of Saudi Arabia’s revenues thanks to consumption taxes, but the sector’s growth is slowing to a crawl. Oil’s contribution shrunk too, but largely because of OPEC-led attempts to shore up oil prices with production cuts.

Aramco has been a similar disappointment. Originally envisioned in 2016 as a partial IPO offering 5% and raising $100 billion at a $2 trillion valuation, in 2019 MBS offered 1.5% to raise $26 billion at a $1.7 trillion valuation—another $4 billion was raised with a greenshoe option, where additional shares were sold to investors. It took MBS another five years to set up a secondary sale this June, offering 0.64% stake to raise $11.2 billion—followed by $1 billion more thanks to another greenshoe option.

The privatization program is a bit more mixed. One key plank tried and failed to raise $10 billion by 2020 through sales of state-owned assets such as grain silos and water desalination plants. Another key pillar requires annually attracting $100 billion in foreign direct investment, calling for “shock therapy to catalyse investment”. This quest for FDI has been a resounding failure, the Kingdom on track to only attract $19 billion this year. Its core ambition—to grow the Saudi economy to $1.7 trillion by 2030, with the private sector contributing 65 percent—is right on schedule or never happening, depending on who you ask.

The private sector currently contributes 45 percent (up from 40 percent in 2016), but the IMF has slashed growth projections after seeing the Saudi economy contract for consecutive quarters. And while the number of private sector employees has grown, less than 20 percent are Saudi nationals. The Kingdom introduced reforms to the abusive "kafala" or visa sponsorship program in 2021, but little has changed. Sponsors ("kafaleels") have complete control over working conditions, legal status, pay frequency, mobility, privacy, and regularly subject migrant laborers to discrimination, abuse, and violence.

Things look even worse when you review Saudi Arabia’s litany of so-called “gigaprojects” that have been stalled or downsized as aforementioned Vision 2030 shortfalls have forced the Kingdom to borrow funds. NEOM, the most famous and most expensive ($1.5 trillion) earns the biggest cuts as it has the most projects under it such as the Red Sea Project and The Line. The latter was promised to house 1.5 million people in a linear city stretching 170 kilometers, but is now slated to house less than 300,000 along a 2.4 kilometer line. NEOM features at least ten proposed projects including, but not limited to: the Oxagon (a net-zero floating industrial complex), Trojena (outdoor ski resort), Sindalah (luxury island resort), Leyja (more luxury resorts), Epicon (even more luxury resorts), Siranna (repeat after me: more luxury resorts), Utamo (luxury art), Norlana (luxury "lifestyle"), Aquellum (luxury "experiential"), Zardun (yet another luxury resort). Sadly, all their futures are now in doubt as the Kingdom reviews the sustainability of blowing hundreds of billions of dollars on projects that may not be feasible.

There is still the matter of cultural reform: MBS’s eagerness to moderate the influence of Wahabi Islam and forge a new national myth may successfully disempower religious elites, but also have (un)intended consequences elsewhere. Take the Vision 2030's aim to grow Hajj pilgrim numbers to 6 million (1.8 million in 2023) and year-round Umrah pilgrimage to 30 million (13.5 million in 2023). In the name of modernizing religious tourism, the Kingdom has sought to digitize both—but this process has just accelerated an already explosive increase in costs while supercharging profits. Saudi Arabia’s ever constant pressure to further commercialize the pilgrimages have not only helped make them even more unaffordable, but attracted investors who see an opportunity to profit by inserting themselves into the price-gouging chain.

Despite the fact that virtually every aspect of Vision 2030 has been either downsized, scrapped, delayed, disappointing, desperately pushed through, revealed to be unfeasible, or obscuring steep costs, MBS will continue undeterred. Why?

Whatever Thomas Friedman might tell you, MBS is not the first Saudi to talk of reforms to revamp the Kingdom’s economy and society. Since 1970, Saudi Arabia has issued 10 five-year development plans that sought to diversify the economy, all of which were unsuccessful and still left oil revenues accounting for 90 percent of the monarchy’s budget in 2015. Saudi Arabia's "megaprojects" also harken back to a 2000s wave of planning high-tech futuristic cities featuring still-unfinished cities from the United Arab Emirates, South Korea, and Saudi Arabia.

In 2003, King Abdullah pitched the world (and investors) six new "economic cities" that were hailed as “nothing short than an overhaul of Saudi society.” After the Great Recession, six became four and eventually four became one (King Abdullah Economic City). Vision 2030 has breathed new life into the failed megaprojects, which once promised to house 4.5 million people but will now simply be special economic zones. King Abdullah Economic City, however, will be given another chance: less than 10,000 people currently live there, but Vision 2030 imagines it can house 2 million people with an investment of $100 billion.

Diversification strategies haven’t fared much better across the Gulf. The United Arab Emirates launched its Vision 2021 in 2014, but has problems that rhyme with Saudi Arabia’s: an obsession with megaprojects, pursuits of transient industries like tourism, an overreliance on exploiting and abusing foreign workers, and "unfulfilled aspirations for high-tech industries."

In 2017, Kuwait’s then-emir announced Vision 2035, a bold vision to make a "New Kuwait" that would be a "financial, cultural and institutional leader in the region by 2035." Skepticism was widespread and warranted: Kuwait had been here before. Diversification strategies contingent on trendy technology, fickle industries, and vanity megaprojects and megacities were proposed before and have been proposed again. Progress has remained stunted at best, with commentators wondering if Kuwait will ever kick start its promised reform agenda. It does not matter whether you jump to Oman’s Vision 2040, Bahrain’s Vision 2030, or Qatar’s National Vision 2030, you will find the similar problems, weaknesses, oversights, structural limitations to reforms, and histories of failed diversification schemes as we find with Saudi Arabia

What distinguishes Saudi Arabia’s Vision 2030 is ultimately the size or ambition of Vision 2030, but the alternative ways it will be deployed: power and influence at home and abroad. Shortly after Vision 2030 was announced, King Salman abolished the dozen Supreme Councils managing various strategic sectors and installed the Council of Economic and Development Affairs—it would be placed under MBS's control and now responsible for the Kingdom's diversification strategy. Salman would not only reorganize his ministry and fill key positions with MBS loyalists, but promote MBS to Crown Prince (heir) while shifting succession law towards primogeniture that affirmed MBS's position. When MBS undertook the mass arrests of other royals at Riyadh's Ritz Carlton, he detained wealthy entrepreneurs as well as political rivals with strong bases independent of his patronage—the political purge was spun as an anti-corruption crackdown. After years of political purges and cabinet reshuffling, he would be left with wide-ranging powers as Crown Prince and Deputy Prime Minister and Minister of Defense and Chief of the Royal Court and chairman of CEDA and PIF. (He would become Prime Minister in 2022, giving the Minister of Defense role to his younger brother).

Not one to let reality interfere with analysis, the Atlantic Council gleefully anticipated in 2018 "arresting high-profile Saudi royals involved in financial dealings, commercial ventures, and security" under the guise of a corruption crackdown would send “a signal to the international financial community" that the Kingdom is ready for investors. Sure. Later that year, MBS had journalist Jamal Khashoggi murdered and dismembered in Turkey—this another signal of sorts to investors. Weeks later, they would climb over one another weeks later to attend an investment summit hosted in Riyadh (affectionately called “Davos in the Desert”).

It took relatively little to rehabilitate the minor PR bruises Saudi Arabia and MBS suffered for assassinating a journalist who resided in the United States—MBS understands there are much more important things (like profit and geopolitics) at stake after all. America’s cold war with China, largely an economic war aimed at undermining the latter’s technological development, has only deepened Saudi Arabia’s entanglement with Washington, Silicon Valley, and Wall Street.

The centralization of MBS’s power has given him singular control over state-owned assets, royal authority, civilian ministries, and thus allowed the Kingdom, Vision 2030, or PIF—now managing close to $1 trillion in assets—to further diplomatic ties, geostrategy, or political objectives. Closer ties with President Trump can be pursued by giving his son-in-law billions despite internal concerns. The American-Chinese Cold War can be used to bolster the Kingdom’s own position, securing more advantageous investment or arms deals by playing one against the other. And more capital can be generated for local projects, much to the chagrin of foreign financiers, by deploying petrodollars at home or regionally, potentially enticing foreign investment that’s been slow to come.

So is Vision 2030 going to succeed? That depends on the question. If you mean: can Saudi Arabia smoothly transition to a post-oil economy, I lean towards no. If you mean: can MBS centralize the Saudi Kingdom around himself even more so than predecessors, the answer is clearly yes. If you mean: can MBS consolidate the Kingdom around himself to pursue ambitious visions at home and abroad that simultaneously restructure the economy, civil society, political system, and international order in ways that advance his interests, the answer remains to be seen.

Leave a comment

My next two free essays: one on how Saudi Arabia’s post-oil ambitions are sustaining the longstanding tech bubble, then another contrasting the Silicon Valley Consensus to more narratives about the AI economy.

On Substack, I have been reading essays about: total mobilization in the information era, the case against social media, the Pentagon's entanglement with our tech sector, China's AI & chip strategies, the decline of literacy, various narratives and frameworks for talking about AI, and data center infrastructure.

Off of Substack, I’ve been reading essays about: climate change fatalism and optimism(?), the consequences of the DoD's obsession with venture capital, more on AI bubbles, Elon Musk’s reliance on Gulf capital, a DHS job fair, meta-commentary on generative AI commentary by writers and critics, Bari Weiss’s media empire, how the Sinaloa Cartel smuggles fentanyl into the United States, Peter Thiel’s Antichrist obsession, JP Morgan’s role in Jeffrey Epstein’s pedophile empire, and more.

Some of what I’ve been reading: books on Brazil’s recent history, how Silicon Valley Bank blew up, the collaboration between Silicon Valley and DHS that has birthed pervasive borderland tech targeting all of us, a book on antitrust & tech from a neo-Brandeisian critic, and the effect of endless quantification on capitalism and society.

Read more

Nearly one hundred years before Frank Herbert published “Dune” and teased its Butlerlian Jihad—the Great Revolt against computers, thinking machines, and conscious robots that some humans used to enslave humanity (who were, in turn, enslaved by a "god of machine-logic")—there was the Butler that inspired it all: Samuel Butler, a 19th century English novelist who was one of the earliest thinkers to try and apply Darwin’s theory of evolution to the possibility of machine intelligence.

In 1863, four years after "On the Origins of Species” was published, Butler sent a letter to the editor published in The Press, a New Zealand daily newspaper, titled "Darwin among the Machines.” In it, Butler posits that machines could be thought of as "mechanical life" undergoing evolution that might make them, not humans, the preeminent species of Earth:

We refer to the question: What sort of creature man’s next successor in the supremacy of the earth is likely to be. We have often heard this debated; but it appears to us that we are ourselves creating our own successors; we are daily adding to the beauty and delicacy of their physical organisation; we are daily giving them greater power and supplying by all sorts of ingenious contrivances that self-regulating, self-acting power which will be to them what intellect has been to the human race. In the course of ages we shall find ourselves the inferior race.

Butler was looking at the monstrous wake of the Industrial Revolution, struggling with the implications of Darwin’s theory, and concluded that the evolutionary pressures advancing machines were even more intense than humans—happening on much shorter timescales that yielded much more dramatic effects because of our intervention—suggesting that consciousness and intelligence would eventually arise. Our succession was a foregone conclusion: the question then was how, not when. What would bring that day to pass?

Butler writes:

Day by day, however, the machines are gaining ground upon us; day by day we are becoming more subservient to them; more men are daily bound down as slaves to tend them, more men are daily devoting the energies of their whole lives to the development of mechanical life. The upshot is simply a question of time, but that the time will come when the machines will hold the real supremacy over the world and its inhabitants is what no person of a truly philosophic mind can for a moment question.

Could anything be done to stave this off? Butler said yes:

War to the death should be instantly proclaimed against them. Every machine of every sort should be destroyed by the well-wisher of his species. Let there be no exceptions made, no quarter shown; let us at once go back to the primeval condition of the race.

Butler would take this letter and a few other writings to develop The Book of Machines, chapters 23-25 of his 1872 social commentary novel “Erewhon”. The novel itself is a funny satire of Victorian society and this section was initially read as a mockery of Darwinian evolution, but Butler makes clear in a later letter to Darwin that it was more of a jihad of his own against the theologian and Christian apologist Joseph Butler (if we refer to this Butler again, we will call him Butler 2), an Anglican bishop who’d published “The Analogy of Religion, Natural and Revealed” one hundred years earlier:

When I first got hold of the idea, I developed it for mere fun and because it amused me and I thought would amuse others, but without a particle of serious meaning; but I developed it and introduced it into Erewhon with the intention of implying: ‘See how easy it is to be plausible, and what absurd propositions can be defended by a little ingenuity and distortion and departure from strictly scientific methods,’ and I had Butler’s Analogy in my head as the book at which it should be aimed, but preferred to conceal my aim for many reasons.

You should read the novel yourself but there are a few quotes in there I think worth teasing out that are clearly building upon ideas Butler is grappling with in that first essay, that are cleanly ported over to the Butlerian Jihad. Here are two passages, first:

"True, from a low materialistic point of view, it would seem that those thrive best who use machinery wherever its use is possible with profit; but this is the art of the machines—they serve that they may rule. They bear no malice towards man for destroying a whole race of them provided he creates a better instead; on the contrary, they reward him liberally for having hastened their development. It is for neglecting them that he incurs their wrath, or for using inferior machines, or for not making sufficient exertions to invent new ones, or for destroying them without replacing them; yet these are the very things we ought to do, and do quickly; for though our rebellion against their infant power will cause infinite suffering, what will not things come to, if that rebellion is delayed?"

and second:

"But returning to the argument, I would repeat that I fear none of the existing machines; what I fear is the extraordinary rapidity with which they are becoming something very different to what they are at present. No class of beings have in any time past made so rapid a movement forward. Should not that movement be jealously watched, and checked while we can still check it? And is it not necessary for this end to destroy the more advanced of the machines which are in use at present, though it is admitted that they are in themselves harmless?"

Butler mocks the prevailing Victorian attitude of the time—blind faith in science, reason, progress, and profit—as a "low materialistic point of view" that believes mindlessly adopting and advancing technology is a moral good and an inevitable process akin to the march of time. It's through a wrongheaded belief in profits as the ultimate signifier of value that Erewhonians created an elaborate system whereby they feel in control of their society and their culture, though are in truth slavishly dedicated to their machines above all else. Failure to do so incurs their "wrath", which is little more than a mockery of capitalist competition—neglect new tech, use inferior machines, fail to innovative, and you will be punished by impersonal forces bent on impoverishing (and eventually killing) you.

Suspicious as they might be about their machines, Erewhonians were unable to live without them and unwilling to entertain thinking about lives where their relationships to the machines were any less dependent. And so you have a lone Erewhonian philosopher (the narrator of this section) insisting that while the machines clearly pose no threat today or tomorrow, this is the only time when revolt will be possible. The door will close, their utility and seductiveness will only grow—eventually to the point that the machines will no longer need to rely on the advocacy of those enslaved by dependency, they will simply act in their best interests. Such a revolt in Erewhon will cause a great deal of suffering, but what is that measured against the smothering of humanity's spirit?

A key excerpt:

"How many men at this hour are living in a state of bondage to the machines? How many spend their whole lives, from the cradle to the grave, in tending them by night and day? Is it not plain that the machines are gaining ground upon us, when we reflect on the increasing number of those who are bound down to them as slaves, and of those who devote their whole souls to the advancement of the mechanical kingdom?”

This section captures what I think is at the core of Butler's and Herbert's warnings about technology. A world where we prioritize the relentless advancement of technology and a universal dependence on it in the name of efficiency is a world where we prioritize a certain political-economic order that is more interested in advancing technologies based on criteria that have little to do with human flourishing, instead being much more interested in financing and designing and deploying them against people—in organizing the greater whole of humanity such that they are more profitable and less likely to revolt against an arrangement that is incredibly lucrative for increasingly few.

In one of the essays in my AI series, I argued that Luther's critique of indulgences in the medieval era could be applied to today's Silicon Valley Consensus. Luther was not opposed to indulgences so much as their abuse, which cheapened repentance and undermined attempts to compel good works or genuine attempts to right wrongs. The idea that salvation could be realized through a transaction convinced many they'd obviated the need for the hard work of being a better person. Indulgences also centralized and codified unjustified power grabs by the Church, which claimed new authorities over souls in Purgatory and introduced perverse incentives to prioritize activities that had nothing to do with Christendom.

In some ways, I think of Luddism (and Butlerianism) similarly. My concern is not technology in of itself (though there are multiple technologies we would do better off without). Technology, however, is downstream of politics and economics and history and social relations. We aren’t saying destroy the clocks before they become killer drones, but we are saying the killer drones are already here and we should figure out how to destroy them. Clearly, technological dependence obscures the political and economic decisions about what sort of technologies should be developed, how they should be financed, who should finance their development and reap their rewards and bear their costs, and how society should be organized around the facts of those arrangements.

Is the solution more or less democratic control over technological development and deployment? Do we trust today’s major players in this space to truly prioritize anything other than profits and returns? Are we going to be able to realize or experiment with other values, arrangements, and models that prioritize anything else within today’s authoritarian technological system or within a democratic system? If we realize that certain paths or arrangements or products or models go against human flourishing or the public good or our ecological niche or the mental health of the general public (realizations we have already made), will we be able to do anything about it?

I want to end on an exchange that I think encapsulates this thread, at least, of my personal Luddite philosophy—an interview between Bill Moyers and Noam Chomsky in 1989:

NOAM CHOMSKY: Well, we now face the most awesome problems in human history- problems such as: the likelihood of nuclear conflict, either among the superpowers or through proliferation; the destruction of a fragile environment, which finally we’re beginning to recognize, though it was obvious decades ago that we’re heading for disaster; other problems of this nature. They are of a level of seriousness that they never were in the past.

BILL MOYERS: But why do you think more participation by the public, more democracy is the answer?

NOAM CHOMSKY: Because more democracy is a value in itself, quite apart- because democracy is a value. It doesn’t have to be defended any more than freedom has to be defended. It’s an essential feature of human nature that people should be free; they should be able to participate; they should be uncoerced, and so on. These are values in themselves.

BILL MOYERS: Why do you think, if we go that route-

NOAM CHOMSKY: Because I think that’s the only hope that I can see that other values will come to the fore. I mean, if the society is based on control by private wealth, it will reflect the values that it, in fact, does reflect; the value that the only real human property is greed, and the desire to maximize personal gain at the expense of others. Now, any society- a small society based on that principle is ugly, but it can survive. A global society based on that principle is headed for massive destruction. And that’s what we are. We have to have a mode of social organization that reflects other values that, I think, are inherent in human nature that people recognize.

BILL MOYERS: And that would be? I want to see exactly what you mean.

NOAM CHOMSKY: I mean, what are human beings? In your family, for example, it’s not the case that in the family every person tries to maximize personal gain at the expense of others, or if they do, it’s pathological. It’s not the case that—if you and I are, say, walking down the street, and we see a child eating a piece of candy and we see that nobody’s around and we happen to be hungry, we don’t steal it. If we did that, we’d be pathological. I mean, the idea of care for others and concern for other people’s needs and concern for a fragile environment that must sustain future generations; all of these things are part of human nature. These are elements of human nature that are suppressed in a social and cultural system which is designed to maximize personal gain.

And I think we must try to overcome that suppression and that’s, in fact, what democracy could bring about. It could lead to the expression of other human needs and values which tend to be suppressed under the institutional structure of a system of private power and private profit.

BILL MOYERS: Do you believe that, by nature, human beings yearn for freedom, or do we settle in the interest of safety and security and conformity—do we settle for order?

NOAM CHOMSKY: These are really matters of faith rather than knowledge. On the one hand you have the Grand Inquisitor who tells you that what people, what humans crave is submission, and, therefore, Christ is a criminal and we have to vanquish freedom. That’s one view.

You have the other view of, say, Rousseau in some of his moments, that people are born to be free, and that their basic instinct is the desire to free themselves from coercion, authority and oppression. The answer to which you believe is, more or less, where you stake your hopes. I’d like to believe that people are born to be free, but if you ask for proof, I couldn’t give it to you.

Leave a comment

This week’s round-up will include essays on Charlie Kirk’s assassination, our ongoing AI bubble and the ongoing data center overbuild, Google’s antitrust trial, Trump crypto riches and Silicon Valley lobbying, AI chatbots & companions, socialist economics, and more; book reviews touching on climate change as well as reflections on US-China political economy; Chinese web novels, and some incredibly bad horny/horror books.

No music or movie recommendations because of how goddamn long this recommendation newsletter is, but I hope you enjoy!

Kirk’s shooting, liberal eulogies, and political violence

One of the more loathsome takes came from Ezra Klein’s op-ed “Charlie Kirk Was Practicing Politics the Right Way”, which I think is transparently a wrong-headed attempt to “cool” the temperature while also reaching across the aisle by saying “look I disagreed with Kirk but he had a right to say everything he did and I kinda admire him for doing it so well!” We don’t really have to waste much time even engaging with that, a plethora of great responses have emerged from 404 Media, The Nation, Mother Jones, The Guardian, and so on. I encourage reading the comments, I found a subscriber link which should let you do it—I was a bit surprised and encouraged by how strong the pushback to his argument was.

One response I particularly liked was made by Osita Nwanevu on Twitter, though he’s since deactivated his account. On the platform, he asked whether anyone would’ve said Kirk was doing politics “exactly the right way” a week before he was killed. In a thread, he went on to ask if someone could seriously imagine Klein doing the same for someone on the left if they were killed similarly. The answer, obviously, is no.

One argument Nwanevu has been making, correctly, over the past few years is that if you care about liberalism and democracy then the greatest threat to those project is the GOP. Recognizing the GOP’s threat and conspiring to end it should be a major concern, but our liberals (and progressives) seem fundamentally incapable of doing so. Instead, they go out of their way to make common cause with phrenologists, white nationalists, and other reactionaries who pitch tents imagining a future where apartheid is reimposed on our society again. This doesn’t inspire confidence, it reeks of cynicism, and makes sense given we are seeing liberals try to recalibrate into a form that accommodates increasingly reactionary but ascendant sectors of America’s political elite (e.g. tech capitalists).

Separate from essays about Kirk, I really enjoyed a few others that I’ll share with you.

Read more

Welcome to The Tech Bubble. This week, enjoy Part 1.5 of the Silicon Valley Consensus, which revisits my framework for understanding the AI bubble and examines a few more actors central to sustaining it.

My series of essays on Artificial Intelligence thus far:

• AI, life engineering, and digital hygiene - On artificial intelligence as an attack vector for eugenics & political shock therapy.

• AI, slavery, surveillance, and capitalism - On artificial intelligence for The Labor Question & the limits of surveillance capitalism and techno-feudalism.

• The Silicon Valley Consensus & AI Capex (Part 1) - On overbuilding AI infrastructure and its energy supply.

• AI, indulgences, and the false promise of salvation - On (some of) the moral externalities in the economy behind artificial intelligence.

Here are also a few other essays on AI that you could probably fit in there:

• Does OpenAI’s latest marketing stunt matter? - On distractions, intentions, aesthetics, and fascism.

• The phony comforts of useful idiots - On Casey Newton and the shallowness of (AI) anti-skepticism.

• Trapped in the Maw of a Stillborn God - On Vegas as a laboratory for surveillance and social control, the explosion of gambling as a sign of a degenerate culture seized by despair, AI delusions at CES, and the future.

• This Silicon Valley Stuff’ll Get You Killed - On ritual sacrifice in the 21st century.

If these essays sound interesting or if you’ve found them insightful, I’m happy to tell you that you can help make more of this work possible with a subscription. Your support allows me to keep my main essays free for everyone (and supports me as I make the newsletter sustainable). Consider supporting The Tech Bubble (me) with a subscription: $7 a month (the price of a few hard ciders at your local bodega) or $70 a year (a copy of METAL GEAR SOLID Δ: SNAKE EATER Tactical Edition). Not that much, huh?

Subscribe now

Silicon Valley Consensus & The Limits of an “AI Economy”

The “AI economy” is less a story of productivity or innovation, then an attempt to graft a new political-economic order—let’s call it the Silicon Valley Consensus—that is ostensibly concerned with building our stillborn God. A coalition of hyperscalers, venture capitalists, fossil fuel firms, conservatives, and reactionaries are engaged in a frenzy of overbuilding, overvaluing, and overinvesting in compute infrastructure. Their goal is not to realize AGI or radically improve life for humanity, but to reallocate capital such that it enriches themselves, transmutes their wealth into even more political power that imposes constraints on countervailing political forces, and liberates capitalism from its recent defects (e.g. democracy), consolidating benefits to its architectures regardless of the actual social utility of the technologies they pursue.

SPECTACLE & SUBSTANCE

Building out generative AI’s compute infrastructure and energy supply is an incredibly capital-intensive enterprise (McKinsey expects $7 trillion will be spent by 2030). It will only grow more so. As of late, models have become more compute-intensive as they attempt to resemble “reasoning” via chains of queries that self-check, search, and otherwise radically increase the resources needed for individual requests.

Compute infrastructure is expansive (chips, servers, clusters, data centers, data sets, labelers, cleaners, energy supply, etc.) but it's largely been hyperscaler capex that’s drawn attention. In August, we got a flood of commentary reacting to the news that their spending on data center contributed more to the US economy than all of consumer spending. Given revenues are weak and profitability/sustainability is not in sight, we might expect some scrutiny about the claims these firms are making. There has been relatively little until recently. Much of it begins and ends with surface-level comparisons to telecom and the internet during the dot-com bubble—partly because Magnificent 7 generative AI capex exceeded $102.5 billion last quarter (meaning this buildout is on track to be the largest since the Gilded Age railroad bubble), partly because of the tech sector has undertaken a wildly successful and sophisticated marketing campaign that has drowned out skepticism and critical analysis.

I want to cobble together three perspectives that I think are complementary, running the gamut from techno-optimism to a much more critical framework asking “why are we doing this?”

Today we will start with Derek Thompson’s from August:

POV 1: How AI Conquered the US Economy: A Visual FAQ by Derek Thompson

Thompson’s piece is pretty rosy on the AI boom and what its scale could mean for worker adoption as well as economic impact. Distilled, its main points are:

1. The AI investment and adoption boom is not a future event but a present-day economic reality. We’re creating a bifurcated economy: a “rip-roaring” AI sector powered by hyperscaler capex that eclipses the rest of our “lackluster” economy. This buildout is comparable in scale (and potentially impact) to the dot-com era or Gilded Age railroad bubbles. It may even rival them!

2. The primary capital source for this infrastructure buildout isn’t external debt, but internal cash flows—primarily at hyperscalers—that dominate our stock market. Their profitability is so extreme that they can put “oodles of oodles of money” towards such an ambitious project without touching risky financing options, even if revenues and profits have yet to materialize.

3. Given early evidence on adoption and productivity, we should be bullish with minor caveats about its potential impact. Generative AI is being adopted at twice the rate of the internet and certain sectors (e.g. startups on Stripe, teachers) report massive efficiency gains. Some objective studies reveal workers dramatically overestimate these gains, however: workers often feel more productive using AI (such as developers claiming 20 percent gains) even as we see the opposite (developers taking 20 percent longer), suggesting the positive impact is there and building but not fully understood.

ON AI INVESTMENT AND COMPARISONS TO THE EARLY DAYS OF THE INTERNET

If this buildout rivals that of previous bubble overbuilds in size and potential impact, a very simple question to ask is: where is the money? It has been a year since Goldman Sachs’ report “Gen AI: Too Much Spend, Too Little Benefit?” and many of its core concerns remain unaddressed. One of its key sections, an interview with Goldman’s Head of Global Equity Research Jim Covello, pointed out that trillions over the next few years were being planned for investment despite failing to convincingly answer a single question: “What trillion dollar problem will AI solve?”

Responses to this usually hinge on comparisons to earlier technological innovations, but are superficial or ill-informed. The internet, Covello notes, “was a low-cost technology solution that enabled e-commerce to replace costly incumbent solutions.” AI, however is “exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do.” Covello also argues that the core investment theory promoting generative AI is junk science at best:

The idea that technology typically starts out expensive before becoming cheaper is revisionist history. Ecommerce, as we just discussed, was cheaper from day one, not ten years down the road. But even beyond that misconception, the tech world is too complacent in its assumption that AI costs will decline substantially over time. Moore’s law in chips that enabled the smaller, faster, cheaper paradigm driving the history of technological innovation only proved true because competitors to Intel, like Advanced Micro Devices, forced Intel and others to reduce costs and innovate over time to remain competitive.

Today, Nvidia is the only company currently capable of producing the GPUs that power AI. Some people believe that competitors to Nvidia from within the semiconductor industry or from the hyperscalers—Google, Amazon, and Microsoft— themselves will emerge, which is possible. But that's a big leap from where we are today given that chip companies have tried and failed to dethrone Nvidia from its dominant GPU position for the last 10 years. Technology can be so difficult to replicate that no competitors are able to do so, allowing companies to maintain their monopoly and pricing power. For example, Advanced Semiconductor Materials Lithography (ASML) remains the only company in the world able to produce leadingedge lithography tools and, as a result, the cost of their machines has increased from tens of millions of dollars twenty years ago to, in some cases, hundreds of millions of dollars today. Nvidia may not follow that pattern, and the scale in dollars is different, but the market is too complacent about the certainty of cost declines.

The starting point for costs is also so high that even if costs decline, they would have to do so dramatically to make automating tasks with AI affordable. People point to the enormous cost decline in servers within a few years of their inception in the late 1990s, but the number of $64,000 Sun Microsystems servers required to power the internet technology transition in the late 1990s pales in comparison to the number of expensive chips required to power the AI transition today, even without including the replacement of the power grid and other costs necessary to support this transition that on their own are enormously expensive.

Covello's argument here is not that compute costs for generative AI will never fall, but that they do not follow the historical patterns that boosters and optimists are consistently using to justify their enormous bets. First, generative AI does not follow historical tech cost curves, where day one the technology is cheaper. Second, generative AI does not have competitive enough markets to drive down data center hardware costs, something that was the key driver behind cost depreciation associated with Moore's Law—in fact, NVIDIA's dominance may yield monopolistic pricing power. Third, there is a fundamental difference of scale between today's genAI compute infrastructure costs and those of previous bubbles that make them astronomically higher and much more inflexible. To start: there is the actual real estate (land, physical space) that must be bought or leased for data centers, energy supply, and then water/cooling infrastructure.

There are three other parts I want to quote Covello on that really drive home the point here. The first is on the explicit comparison of today to the early days of the Internet:

The idea that the transformative potential of the internet and smartphones wasn’t understood early on is false. I was a semiconductor analyst when smartphones were first introduced and sat through literally hundreds of presentations in the early 2000s about the future of the smartphone and its functionality, with much of it playing out just as the industry had expected. One example was the integration of GPS into smartphones, which wasn’t yet ready for prime time but was predicted to replace the clunky GPS systems commonly found in rental cars at the time. The roadmap on what other technologies would eventually be able to do also existed at their inception. No comparable roadmap exists today. AI bulls seem to just trust that use cases will proliferate as the technology evolves. But eighteen months after the introduction of generative AI to the world, not one truly transformative—let alone cost-effective—application has been found.

The second is on this AI capex arms race that’s completely unmoored from reality:

The big tech companies have no choice but to engage in the AI arms race right now given the hype around the space and FOMO, so the massive spend on the AI buildout will continue. This is not the first time a tech hype cycle has resulted in spending on technologies that don’t pan out in the end; virtual reality, the metaverse, and blockchain are prime examples of technologies that saw substantial spend but have few—if any—real world applications today. And companies outside of the tech sector also face intense investor pressure to pursue AI strategies even though these strategies have yet to yield results. Some investors have accepted that it may take time for these strategies to pay off, but others aren’t buying that argument. Case in point: Salesforce, where AI spend is substantial, recently suffered the biggest daily decline in its stock price since the mid-2000s after its Q2 results showed little revenue boost despite this spend.

The third is on the prospects for AI-related revenue expansion:

I place low odds on AI-related revenue expansion because I don't think the technology is, or will likely be, smart enough to make employees smarter. Even one of the most plausible use cases of AI, improving search functionality, is much more likely to enable employees to find information faster than enable them to find better information. And if AI’s benefits remain largely limited to efficiency improvements, that probably won’t lead to multiple expansion because cost savings just get arbitraged away. If a company can use a robot to improve efficiency, so can the company’s competitors. So, a company won’t be able to charge more or increase margins.

...

Since the substantial spend on AI infrastructure will continue despite my skepticism, investors should remain invested in the beneficiaries of this spend, in rank order: chip manufacturers, utilities and other companies exposed to the coming buildout of the power grid to support AI technology, and the hyperscalers, which are spending substantial money themselves but will also garner incremental revenue from the AI buildout. These companies have indeed already run up substantially, but history suggests that an expensive valuation alone won’t stop a company’s stock price from rising further if the fundamentals that made the company expensive in the first place remain intact. I’ve never seen a stock decline only because it’s expensive—a deterioration in fundamentals is almost always the culprit, and only then does valuation come into play.

A few days before Goldman Sachs’ report, Sequoia Capital partner David Cahn published an argument building on his 2023 argument that there was “a big gap between the revenue expectations implied by the AI infrastructure build-out, and actual revenue growth in the AI ecosystem, which is also a proxy for end-user value.” In 2023, revenue expectations were $200 billion and there was a “$125B hole that needs to be filled for each year of CapEx at today’s levels.” In his 2024 analysis, revenue expectations ballooned to $600 billion and the revenue hole grew even larger to $500 billion.

Cahn is pretty brutal about AI hype, but still retains hope for the future:

Speculative frenzies are part of technology, and so they are not something to be afraid of. Those who remain level-headed through this moment have the chance to build extremely important companies. But we need to make sure not to believe in the delusion that has now spread from Silicon Valley to the rest of the country, and indeed the world. That delusion says that we’re all going to get rich quick, because AGI is coming tomorrow, and we all need to stockpile the only valuable resource, which is GPUs.

Cahn’s argument is a splash of cold water on the techno-optimist narrative, but I want to hone in on his rebuttal of the railroad bubble analogy because it’s typically made on superficial grounds. Generally, these comparisons serve a rhetorical purpose of suggesting the overbuild will have immense value (while also obscuring reality and our ability to understand it). I’m going to plagiarize myself summarizing Cahn’s argument—I engaged with it in March when I offered my first iteration of the Silicon Valley Consensus framework. I have some qualms with Cahn’s argument (specifically the prospects for AI revenue growth), but where I’ve settled is that he’s in the right direction about the gap between revenues and data center and wrong on some areas that give him hope about climbing out of the revenue-capex hole.

Summary:

1. Lack of pricing power: Cahn believes that while railroads confer natural monopolistic advantages (there is only so much track space between two points), the same is not true of GPU data centers. GPU computing is becoming commoditized, AI compute can be offered through the cloud, and so prices are getting competed down to their marginal cost (airlines are offered as an example).

2. Investment incineration: Speculative investment frenzies are common! The underlying asset is not more impervious to zeroing out because large hoards of capital were poured into it; "It's hard to pick winners, but much easier to pick losers (canals, in the case of railroads).”

3. Depreciation: Compute differs from physical infrastructure in that it follows Moore’s Law. The continuous production of next-generation chips with lower costs and higher performance will accelerate the depreciation of the last-generation of chips. This goes both ways: markets will overestimate the value of today’s chips and under-appreciate the value of tomorrow’s chips. "Because the market under-appreciates the B100 and the rate at which next-gen chips will improve, it overestimates the extent to which H100s purchased today will hold their value in 3-4 years."

4. Winners vs losers: Long term, declining prices for GPU computing will be a boon for innovation/startups, but a bane for investors. "Founders and company builders will continue to build in AI—and they will be more likely to succeed, because they will benefit both from lower costs and from learnings accrued during this period of experimentation."

This from a venture capitalist at a firm that believes in AI! Railroads are not a good way of understanding these technologies:

• They appeal to a superficial interest in historical details and patterns that prioritize telling a story, as opposed to paying attention to revenue, pricing power, investment returns, market structure, commodities, and products.

• Emphasizing the difference between revisionist history and reality introduces us to concerns and questions that will inform what we do next better than an analysis that barely rises above the level of vibes.

• What is the “AI economy” materially? Does it actually tell us anything about what is going on in the sector? Investment is real, it’s happening, but it’s not yielded returns—so what is it doing? What decisions are being made and what interests emerge when we raise our heads to look beyond? What is motivating interest and development? Who is able to ignore these lack of returns, what moves are being made to realize them, what moves are being made anticipating they will not emerge?

ON PROFITS, NOT DEBT, BEING THE PRIMARY BUILDOUT CAPITAL SOURCE WORTH EXAMINING

It is true that internal cash flows are the major source for hyperscalers, but: not all firms burning capex are hyperscalers; not all firms, even hyperscalers, are financing AI capex equally.We will tackle this more later when we arrive at Noah Smith’s piece but for now, let’s look at two of his sources.

The first is Paul Kedrosky, who lays out a clear list of the major capital sources for the ongoing data center buildout:

Where is all this capital coming from?

For the most part, six sources:

1. Internal Cash Flows (Primary for Microsoft, Google, Amazon, Meta, etc.)

2. Debt Issuance (Rising role)

3. Equity & Follow-on Offerings

4. Venture Capital / Private Equity (CoreWeave, Lambda, etc.)

5. SPVs, Leasing, and Asset-Backed Vehicles (like Meta’s recent)

6. Cloud Consumption Commitments (mostly hyperscalers)

On a very basic level, every single capital source is not the same. Some are riskier than others, some demand higher returns than others, some are much more abundant than others, and so they have different uses for different firms involved in the AI infrastructure buildout (and go a long way in determining what various firms and their projects will prioritize to meet demands imposed by financing).

The second source (The Economist) lays this out a bit more clearly:

>”[C]apex is growing faster than [Big Tech’s] cashflows …The hot centre of the AI boom is moving from stockmarkets to debt markets … During the first half of the year investment-grade borrowing by tech firms was 70% higher than in the first six months of 2024. In April Alphabet issued bonds for the first time since 2020. Microsoft has reduced its cash pile but its finance leases—a type of debt mostly related to data centres—nearly tripled since 2023, to $46bn (a further $93bn of such liabilities are not yet on its balance-sheet). Meta is in talks to borrow around $30bn from private-credit lenders including Apollo, Brookfield and Carlyle. The market for debt securities backed by borrowing related to data centres, where liabilities are pooled and sliced up in a way similar to mortgage bonds, has grown from almost nothing in 2018 to around $50bn today

…

CoreWeave, an ai cloud firm, has borrowed liberally from private-credit funds and bond investors to buy chips from Nvidia. Fluidstack, another cloud-computing startup, is also borrowing heavily, using its chips as collateral. SoftBank, a Japanese firm, is financing its share of a giant partnership with Openai, the maker of ChatGPT, with debt. “They don’t actually have the money,” wrote Elon Musk when the partnership was announced in January. After raising $5bn of debt earlier this year xAI, Mr Musk’s own startup, is reportedly borrowing $12bn to buy chips.

...

This symbiotic escalation is, in some ways, an advert for American innovation. The country has both the world’s best AI engineers and its most enthusiastic financial engineers. For some it is also a warning sign. Lenders may find themselves taking technology risk, as well as the default and interest-rate risks to which they are accustomed. The history of previous capital cycles should also make them nervous. Capex booms frequently lead to overbuilding, which leads to bankruptcies when returns fall. Equity investors can weather such a crash. The sorts of leveraged investors, such as banks and life insurers, who hold highly rated debt they believe to be safe, cannot.”

A deal where you have immense profits looks different from a deal where your sponsor is not flush with profits, which looks different from a deal where you have a few deep-pocketed clients, which looks different from a deal where you have nothing but talent or certain assets to offer up as leverage. Any analysis of the “AI economy” that pretends as if this uneven terrain does not exist (or minimizes its import) is, at best, an incomplete analysis.

But to really drive home the point that debt is incredibly important to think about, let’s look at a March piece from Ed Zitron that argued CoreWeave, an "AI cloud provider" that lets AI firms rent GPU compute. Its core business model: buying and selling high-end GPUs and the infrastructure needed to run them, largely from NVIDIA thanks to a cozy relationship that gives them priority access. Just as the company delayed its IPO, the first one of the generative AI sector, Zitron issued this report to lay out why CoreWeave might prove to be a ticking time bomb:

1. CoreWeave’s customer base is extremely concentrated, meaning revenue was vulnerable. Microsoft accounted for 62 percent of its 2024 revenue and NVIDIA probably accounts for another 15 percent. The close relationship between NVIDIA and CoreWeave has been accused of resembling “round-tripping” ("a practice where companies inflate their top lines through reciprocal deals that don’t always create real economic value”). Microsoft has already shed “some services” ahead of CoreWeave’s IPO, though the latter denied this. What happens to CoreWeave if it no longer gets priority access to NVIDIA chips or if Microsoft pulls back AI capex (which it did earlier this year)?

2. CoreWeave’s ungodly financial losses are hard to ignore. $1.9 billion in revenue, $863 million in losses. This loss leading, ubiquitous across the “AI economy” fuels the boom—but why is it happening at a firm whose business model is GPU compute, the one thing you’d expect every AI firm to need? Is demand insufficient? Are costs too high?

3. CoreWeave is potentially crippled by its debt. The firm has secured $8 billion through loans with high interest rates and complex, predatory loans. Most of CoreWeave's raised capital is debt, namely in two Delayed Draw Term Loan facilities (you get access to some of the money, distributed in tranches that unlock after time periods or certain milestones). DDTL 1: $2.3 billion (fully drawn) secured with an effective annual interest rate of 14.11 percent, requiring quarterly payments of $250 million ($1 billion annually) to service the loan. DDTL 2: $7.63 billion (half drawn) with a 10.5 percent annual interest rate that would double if fully drawn—so it requires $760 million in annual interest payments to service the loan, potentially doubling to $1.52 billion.

   To make matters worse, it's not clear how CoreWeave will get enough capital to meet its obligations. CoreWeave "will have to spend in excess of $39 billion to build its contracted compute," has already taken on massive loans to fuel its aggressive capex (that stipulate all future capital raised must go towards repaying the debt), and does not have the revenue to support operations or interest payments.

4. Coreweave has put up GPU compute as collateral, a depreciating asset that may create a trap. GPUs like the H100 see their value fall quickly, in part because even before mechanical failure they’ll become obsolete as advanced chips or even-more intensive tasks are developed. Rental prices plummeted from $8/hr to $1.47/hr in March (it’s as low as $0.90/hr at the time of publication).It’s not hard to imagine a scenario whereby we see a depreciation spiral as the collateral backing a massive loan plummets in value and potentially triggers covenants that force early repayment—something CoreWeave does not have the capital to do.

5. CoreWeave has found an unproven partner in Core Scientific. Core Scientific is a different company, a bitcoin mining firm that went public in 2022 via a SPAC-merger—the fraudulent financial vehicle that lets firms go public when they have no business doing so (Core Scientific filed for Chapter 11 bankruptcy the same year it went public via SPAC). This firm is central to CoreWeave's business strategy, despite the fact that it has little experience with HPC/AI data centers and despite the fact that its plan was to "bulldoze and rebuild" Bitcoin mining infrastructure in hopes of repurposing it for artificial intelligence. That CoreWeave's promised "1.3 GW of contracted power" happens to exactly match the capacity Core Scientific claims it can build should raise some, but alas we are charging full speed ahead with an ambitious plan executed by an unproven partner dependent on a circular business plan.

6. CoreWeave’s S-1 raises plenty of red flags. It admits to "material weaknesses in [its] internal control over financial reporting," suggesting "there is a reasonable possibility that a material misstatement of our annual or interim financial statements will not be prevented or detected on a timely basis” and at the earliest around 2026—nothing to worry about there! We also get to think about CoreWeave's dual-class share structure, whereby 82 percent of its voting power is held by founders without 30 percent equity ownership. This is a great way to tell shareholders to fuck off, marginalizing their interests and prioritzing decisions that allow you to, say, profit from an unsustainable business venture by "cash[ing] out nearly $500 million before the IPO".

7. CoreWeave's business model resembles that of the "AI economy". As Zitron puts it, "NVIDIA is selling the pickaxes for the goldrush, CoreWeave is selling the shovels, and it mostly appears to be turning up dirt." We don't see revenues despite its centrality to the ecosystem, we don't see clear expansion plans, we don't see any strong signals that it will thrive, and all this casts doubt on every thread of the “AI economy” that runs through it. Is there not enough real, profitable demand? Are firms not able to find clear demonstrable use cases for this technology beyond round-tripping? Where is the money?! Where are the profits?!

ON EARLY POSITIVE EVIDENCE OF ADOPTION WITH MINOR CAVEATS

One of Thompson’s central claims is that genAI is seeing a widespread, rapid, positive adoption. The St. Louis Fed’s estimate that generative AI adoption is proceeding at roughly twice the rate of the early internet. This statistic—39.4% adoption by August 2024 (~2 years post-ChatGPT) vs. ~20% for the internet at a similar stage—is presented as proof of inevitable, transformative potential. On top of this, Thompson also points to Gallup surveys showing teachers self-reporting efficiency gains and Stripe data showing AI startups reaching revenue milestones quicker, but emphasizes the bull case even as objective studies reveal that workers feel more productive but actually are not.

In March, McKinsey found 71 percent of firms reported use of generative AI and more than 80 percent of those firms reported no ”tangible impact on enterprise-level EBIT from their use of gen AI.” A recent review of claims that artificial intelligence was building "shovelware" which would unlock efficiency gains observed there was no notable impact on software releases across the world these past few years. Torsten Sløk, Apollo Global Management's chief economist, just published a report looking at the US Census Bureau and actually found a decline in AI adoption rates for large firms. Researchers at MIT tracked 300 publicly disclosed AI initiatives and found 95 percent failed to boost profits. We can go on and on and on like this.

The evidence that adoption is rapid, widespread, and positive quickly becomes the weakest when viewed in relation to the larger ecosystem and its disappointing outcomes, however. After all, if adoption is so much faster and is backed by an infrastructure investment larger than the dot-com boom, why are concrete measures of economic value—strong revenues, sustained profitability—so elusive for the core AI industry?

Let’s take one relatively recent example. Back in February 2025, Zitron argued we’re being pretty unrigorous in how we think about the growth of generative AI. ChatGPT’s claims of 300 million weekly users, issued in December 2024, had largely been accepted without much scrutiny about their veracity or the artifice of that metric. Most coverage misrepresents the capabilities of generative AI products and, as Professor Rasmus Nielsen writes for Reuters Institute for the Study of Journalism, “often takes claims about what the technology can and can’t do, and might be able to do in the future, at face value in ways that contributes to the hype cycle."

If ChatGPT had 300 million weekly users at the time, we’re underestimating the extent to which coverage that constantly emphasizes bull cases and reproduces company talking points as hype, ginning up demand for a product that is available for free but does not meet marketed expectations. One way to think about this: is there any useful reference point for a startup that has had as much uncritically optimistic widespread coverage as OpenAI?

If it is true, however, 300 million weekly users is a lot of users. And yet, it does not tell us much about the actual product or how it's used! Is this a sustainable or profitable venture? How are people using it at home or at work, how much of their use is casual versus intensive, and so on.

Digital market intelligence data Zitron obtained suggests ChatGPT's monthly unique visitors trending up to 247.1 million in November, as well as a snapshot of weekly visitor traffic in January and February 2025—trending up to 136.7 million. Typically, you get more visitors to your site than users, which means the gap between reported and actual traffic only grows larger. It doesn't get filled if you bring in mobile app data—its iOS app had been downloaded 353 million times total by late January, so a best case scenario would need 100 million mobile only users a week at least to plug the hole. Monthly active users aren't reported, even though in theory that number should be higher ("a monthly active user is one that uses an app even once a given month") because it might reveal how poor the company's paid conversion rate is!

I would be surprised if you can find more than a handful of people in the wider media ecosystem even entertaining critical discussion of OpenAI’s reported numbers, the artifice of them, what it means that we have this or that metric, what insights into their businesses and into consumer use we are denied, and how this all relates to our ability to accurately get a sense of adoption, impact, sustainability, or the damn business model more generally!

IS TECHNO-OPTIMISM WARRANTED?

To wrap this up, it's not clear to me what merits the techno-optimist outlook on whatever constitutes the “AI economy.” It ignores the financials: overlooks the gap between revenues and capex, waves away the question of how it will generate profits, engages in revisionist historical accounts to justify these bad economics, and whistles past the debt land mines that fuel growth. It ignores technology: there's no interest in market structure, scale of cost, or product roadmaps. It ignores the reality of adoption: gloms onto hype, falls for simplistic narratives, repeats corporate talking points, and reproduces shaky assumptions. We're left with a picture of reality that leaves us unable to explain why things are the way they are and what to do about it. The "AI economy" as talked about within mainstream and optimist circles presents a vision unmoored from reality that is frothy enough to drown out skeptics, juice speculation, and provide cover for entrenched interests looking to enrich themselves at everyone else's expense.

Leave a comment

It’s been a minute so this week’s round-up will run long: it includes essays on our ongoing AI bubble, GPT-5, debates about the impact of digital technologies on capitalism, China, a movie/show recommendation, books, and music I’ve enjoyed, and more.

Table of Contents

AI Bubble

GPT-5

Capitalism

China

Movies/Shows

Music

Books

Subscribe now

AI Bubble

How to Argue With An A.I. Booster - Ed Zitron, Where’s Your Ed At

Zitron is easily one of my favorite commentators/analysts/writers on the AI bubble's financial and consumer dimensions. I recommend you read every single essay he's published, but I especially recommend his latest. The spirit of it:

In the last two years I've written no less than 500,000 words, with many of them dedicated to breaking both existent and previous myths about the state of technology and the tech industry itself. While I feel no resentment — I really enjoy writing, and feel privileged to be able to write about this and make money doing so — I do feel that there is a massive double standard between those perceived as "skeptics" and "optimists."

To be skeptical of AI is to commit yourself to near-constant demands to prove yourself, and endless nags of "but what about?" with each one — no matter how small — presented as a fact that defeats any points you may have. Conversely, being an "optimist" allows you to take things like AI 2027 — which I will fucking get to — seriously to the point that you can write an entire feature about fan fiction in the New York Times and nobody will bat an eyelid.

This reminds me of an interview with Noam Chomsky where he defines “concision” in terms of our ad-centric media system:

Concision means you have to be able to say things between two commercials. Now that’s a structural property of our media—a very important structural property which imposes conformism in a very deep way, because if you have to meet the condition of concision, you can only either repeat conventional platitudes or else sound like you are from Neptune That is, if you say anything that’s not conventional, it’s going to sound very strange. For example, if I get up on television and say, “The Soviet invasion of Afghanistan is a horror,” that meets the condition of concision. I don’t have to back it up with any evidence; everyone believes it already so therefore it’s straightforward and now comes the commercial. Suppose I get up in the same two minutes and say, “The U.S. invasion of South Vietnam is a horror.” Well, people are very surprised. They never knew there was a U.S. invasion of South Vietnam, so how could it be a horror? They heard of something called the U.S. “defense” of South Vietnam, and maybe that it was wrong, but they never heard anybody talk about the U.S. “invasion” of South Vietnam. So, therefore, they have a right to ask what I’m talking about. Copy editors will ask me when I try to sneak something like this into an article what I mean. They’ll say, “I don t remember any such event.” They have a right to ask what I mean. This structural requirement of concision that’s imposed by our media disallows the possibility of explanation; in fact, that’s its propaganda function It means that you can repeat conventional platitudes, but you can’t say anything out of the ordinary without sounding as if you’re from Neptune, a wacko, because to explain what you meant—and people have a right to ask if it’s an unconventional thought—would take a little bit of time.

To that point, Zitron's work (this essay is 16,000 words long) really hinges on the idea that if you are arguing with an AI booster, you must go long and you must go deep. This is not as high of a hurdle to climb as it sounds. A great deal of commentators are incredibly lazy, incurious, deceptive, or flat-out delusional—Zitron’s work is a breath of fresh air here, as are many of the journalists he directly cites within it. This sort of argument guide covers a wide range of arguments you might encounter when dealing with boosters: comparisons to the Uber/Amazon narrative about growing into profitability, the importance of overbuilding compute infrastructure, AI 2027, the potential of agentic workflows, unit economics of inference, edge use cases that are genuinely helpful or useful, the democratization of X with the introduction of AI vibe coding and co-creation, and more. I’ll pull from the conclusion to pitch it once more

I also think the media is failing on a very basic level to realize that their fear of missing out or seeming stupid is being used against them. If you don't understand something, it's likely because the person you're reading or hearing it from doesn't either. If a company makes a promise and you don't understand how they'd deliver on it, it's their job to explain how, and your job to suggest it isn't plausible in clear and defined language.

This has gone beyond simple "objectivity" into the realm of an outright failure of journalism. I have never seen more misinformation about the capabilities of a product in my entire career, and it's largely peddled by reporters who either don't know or have no interest in knowing what's actually possible, in part because all of their peers are saying the same nonsense.

As things begin to collapse — and they sure look like they're collapsing, but I am not making any wild claims about "the bubble bursting" quite yet — it will look increasingly more deranged to bluntly publish everything that these companies say.

Read more

I love an incurious essay and Matthew Yglesias' recent one on why Uber is actually good as an attempt to snipe at the left is one of the more stunning examples in recent memory. I want to go through his argument then talk more generally about why this sort of analysis is particularly noxious.

Let’s start with a key thrust of Yglesias’ essay:

Uber's success should be celebrated.

The company's "clever arbitrage around municipal taxi regulations" broke Big Taxi, a series of taxi cartels which relied on business models that prioritized inefficiency and anti-competitiveness. Uber has improved prices, led to great social outcomes (e.g. reducing drunk driving and racial discrimination in pickups), and at minimal cost (namely traffic congestion).

Yglesias starts with an article that American University law professor Hilary Allen wrote for the Law and Political Economy blog titled “Why We Need to Stop Subsidizing Venture Capitalists” where she opens:

For many readers of this blog, Uber represents a cautionary tale. While the company attributed its initial success to cutting-edge technology—such as dynamic pricing, matching algorithms, real-time data—subsequent analysis has demonstrated that its growth was largely driven by ignoring, breaking, and then bending taxi regulations to suit its business model.

For companies in the fintech sector, however, Uber’s approach represented a blueprint to follow. From lending to payments to stock trading to crypto, prominent fintech businesses have found a competitive edge not in technology itself, but in using narratives about technology as a smokescreen for the profitable arbitrage of financial regulations. This modus operandi is encouraged by Silicon Valley’s venture capitalists, who decide which businesses to fund and often provide advice, gin up hype, and lobby for the businesses they’ve chosen. Our society continues to shower VCs with public subsidies, but as I argue in this brief post, if regulatory arbitrage is what we’re getting from Silicon Valley’s VCs in exchange, it’s well past time to reconsider this relationship.

Yglesias only quotes the first paragraph (more on that later) but responds to it with:

What I want to talk about here is not Allen’s forward-looking argument, which focuses largely on fintech issues. But the breezy way in which she asserts that Uber is a “cautionary tale” and assumes her audience will agree without argumentation. She thinks the fact that Uber busted up the old regulatory system and got a whole new category of companies legalized is a dangerous precedent. I think it’s good!

Yglesias sees Allen’s argument as "breezy" and in turn breezes by it, but it's worth spending some time actually engaging with Allen's argument to actually flesh out why we might have problems with Uber.

Allen's essay argues that VC-backed firms use an innovation "bait-and-switch" where they develop a business model built on a specific technology, craft a narrative about its ability to "democratize" and "disrupt" a legitimate issue, circumvent laws that get in the way of that business making a profit, and eventually convincing "lawmakers and regulators to change the law so that you never have to comply with it and those who are harmed have no recourse."

Allen hones in on fintech firms because they've been used this model to circumvent laws that protected consumers, investors, deposits, retirement funds, and the like from being gambled away on volatile assets or speculative activities that are incredibly risky for all but the businesses making these bets.

Financial regulations were adopted over time to protect consumers from harm, protect our economy from debilitating financial crises, and further law enforcement and national security objectives. All this regulatory arbitrage is profitable for the fintech industry and the VCs who fund it, but it leaves most of us worse off. In addition to exposing the public to harm, fintech has rarely delivered on its promises to improve financial inclusion, or to make the financial system more efficient, competitive, or secure. Not content to keep skirting regulations, many fintech businesses and their VC backers have successfully lobbied regulators and Congress to change rules and laws to permanently accommodate and legitimize their business models. These efforts erode faith in our democratic process, and in the law’s ability to protect people from harm.

To make matters worse, they are doing so with public funds. VC funds are limited partnerships that are only on the hook for what they invest, not what they lose. They successfully lobbied to gamble with pension funds, increasing appetite for risk by gambling with larger sums of other people's money. And they've secured tax loopholes that ensure fees and profits are taxed at substantially lower rates.

So what happens when a sector is constantly advancing business models that are subsidized by the public, aimed at rewriting laws to realize previously illegal levels of profit-making, and doing so with little regard for (to borrow a word from economics) “negative externalities”?

Allen writes:

Leading VC firms are well-aware of the important role played by this special legal treatment. Lobbying, media blitzes, and hiring government officials for access are all part of the VC industry’s standard operating procedure, but the playbook has been deployed with particular zeal when it comes to crypto. The crypto industry was responsible for 44% of all corporate expenditures during the 2024 election cycle—with most of it coming from Andreessen Horowitz, Coinbase, and another crypto company called Ripple. It’s worth bearing in mind that the public is not merely subsidizing a financial grift when it comes to crypto—it is subsidizing an ideological project with techno-libertarian goals of avoiding regulation, taxes, and eliminating central banks. Crypto is intended, for instance, to provide the infrastructure for the Network State movement backed by Andreessen Horowitz partner Marc Andreessen and Coinbase CEO Brian Armstrong, which aims to create tech-CEO led dictatorships outside the boundaries of democratic governance.

Yglesias dismisses Allen’s essay as being solely about fintech, but if he actually read it he would see it is about the consequences of letting predatory firms rewrite our laws to maximize their profits and to maximize their ability to convert those profits into political power that can further boost said profits. Should firms be able to subordinate the public interest to their business models or should it be the other way around? And is it any surprise that Yglesias is less interested in thinking through this question than in making a contrarian argument that affirms his priors?

Back to Uber. Yglesias' argument around Uber is relatively simple:

The taxi industry as a whole is not that large or important in the grand scheme of things. But the dynamics around ride-hailing really do, I think, illustrate the core differences between economic analysis and the LPE worldview in which Uber serves as a cautionary tale.

We will return to the idea that Yglesias’ misunderstanding of Uber is “economic analysis” while LPE’s is anti-economics, but for now we’ll focus on the narrative he advances. Yglesias explains that the taxi market was highly fragmented and localized, suggesting LPE opposes Uber because of fidelity to small businesses as opposed to a commitment to "high output and good prices." Taxis extract fares from riders and distribute them to owners of taxi firms or medallions, taxis enjoy massive barriers to entry that inflated prices, and they incur massive deadweight loss (a lot of people who would take taxis don't because the prices are too high).

Uber's arbitrage, then, broke the rickety old system because Uber’s sleek tech collected extensive data on customers and drivers that allowed it to calibrate prices and wages which would increase trips and better compensate drivers. It gave drivers flexibility to set their own hours, increased utilization rates (drivers spent more time with a passenger in the car), reduced drunk driving, reduced racial discrimination, and only cost a little traffic congestion.

The bad guys in this instance were not the richest people in America or huge corporations wielding concentrated power — they were a disaggregated network of largely anonymous small business owners. We have more competition now, but also a marketplace that’s dominated by a much smaller number of large global companies. And that’s good!

This all sounds good and well, but is it true?

Back in 2017, economist Hubert Horan's "Will the Growth of Uber Increase Economic Welfare" article made the case that even a cursory look at Uber's actual economics revealed it to be a deeply parasitic enterprise:

Uber has no ability—now or in the foreseeable future—to earn sustainable profits in a competitive marketplace. Uber's investors cannot earn returns on the $13 billion they have invested without achieving levels of market dominance that would allow them to exploit anti-competitive market power. The growth of Uber is entirely explained by massive predatory subsidies that have totally undermined the normal workings of both capital and labor markets. Capital has shifted from more productive to less productive uses, the price signals that allow drivers and customers to make welfare maximizing decisions have been deliberately distorted, and the laws and regulations that protect the public's interest in competition and efficient urban transport have been seriously undermined. Absolutely nothing in the "narrative" Uber has used to explain its growth is supported by objective, verifiable evidence of its actual competitive economics.

To show Uber's growth enhanced welfare, it would need to demonstrate three things: (1) the ability to earn sustainable profits in competitive markets or demonstrate scale/network effects that could realize these profits; (2) the capacity to provide services "at significantly lower cost than existing competitors" or "produce service that consumers value much more highly at similar costs"; (3) significant competitive advantages based on its product, tech, or innovations that cannot be replicated.

Then and now, Uber fails each welfare enhancement test.

Uber has a long history of publishing incomplete, opaque, and adjusted financial reports that do not adhere to generally accepted accounting principles (GAAP) until relatively recent. Data that Horan cobbled together on its operations from 2012 to 2016 showed Uber’s rapid growth did not yield rapid margin improvements observed in other Silicon Valley startups it compared itself to in PR (Amazon, Facebook, etc.). What Horan observed is that when Uber’s margins improved was not because of efficiency or network effects, but because of driver pay cuts.

How have things looked since? A 2018 JP Morgan study found that on-demand delivery and ride-hail workers made 53 percent less in 2017 than they did in 2013. A 2018 EPI study found that a third of passenger fares went to Uber via fees, driver compensation averaged $11.77 after fees and expenses (lower than average private-sector worker hourly pay of $32.06 and lower than service workers, the least-paid major occupation with an average hourly pay of $14.99), $10.87 an hour after deducting mandatory Social Security/Medicare taxes that self-employed drivers must pay, and $9.21 an hour after deducting the cost of a "modest" benefits package equivalent to what a W-2 employee might have.

The Uber driver W-2 equivalent hourly wage is roughly at the 10th percentile of all wage and salary workers’ wages, meaning Uber drivers earn less than what 90 percent of workers earn. The Uber driver W-2 equivalent hourly wage falls below the mandated minimum wage in the majority of major Uber urban markets (13 of 20 major markets, which include 18 cities, a county, and a state). The Uber driver “no benefits” hourly wage or discretionary compensation—the hourly compensation adjusted for an assumption that Uber drivers pay the extra payroll taxes that the self-employed must pay but do not provide a standard benefits package for themselves—falls below the mandated minimum wage in nine of 20 major markets, including the three largest (Chicago, Los Angeles, and New York).

Back then, it took an incredible amount of mental effort (or none at all in the case of some commentators) to ignore that that a key part of Uber’s arbitrage scheme was figuring out how to transfer more cash from labor to capital than was previously legally allowed—the cost of attempted margin improvements was externalized on drivers, whose working conditions degraded as they were paid out starvation wages.

In 2023, Columbia Business School professor Len Sherman affirmed much of what Horan argued when he observed Uber's 72 percent annual revenue growth and profit margin improvements that year were driven by accounting shenanigans (“Uber has never disclosed sufficient data to allow a fully substantiated assessment of its operational and financial performance”) alongside the fact that "Uber has been raising US ridehail prices four times faster than the rate of inflation while squeezing driver pay" in pursuit of revenue growth and profits. Since 2018—as part of a bid to juice investor appetite for its 2019 IPO—Uber raised prices by the equivalent of 18 percent per year while trips were down 29 percent from pre-pandemic Q319 to Q322 and fares were hiked 41 percent over the same period.

Sherman goes on to write that Uber has relatively few options for significantly growing its core revenue: improving productivity to generate more revenue per trip (batched delivery, pooled rides, etc.), new revenue sources (namely advertising), and increasing its take by hiking fares or delivery fees or cutting driver pay. Sherman makes it clear that Uber has prioritized the third plank of options:

These are the actions that undoubtedly are driving Uber’s recent financial performance improvements and the clearest manifestations of Dara’s “new math” to grow revenues (passenger fares) faster than expenses (driver pay). The sketchy data Uber does disclose confirms that the company has been increasing its take rates over the past few years. As already noted, passenger fares have been trending significantly higher for the past five years. While driver comp data is harder to come by, Uber’s recent shift to “upfront prices” has effectively decoupled consumer prices and driver pay from actual travel time and distance, giving the company cover to increase the spread between consumer prices and driver pay.

And in this regard, Uber enjoys a massive competitive advantage: more data on consumer and driver behavior on a global scale than any other mobility or delivery provider. Armed with such market insight, Uber is in an ideal position to practice what economists call first-order price discrimination — that is, charging each customer prices based on their known willingness to pay and setting each driver’s pay based on their known willingness to serve. The resulting upside potential of such price discrimination is enormous, and the opportunity (massive data + AI algorithms + upfront pricing policies) and need (growing investor pressure for near-term profitability) to exploit it is urgent.

Uber has historically denied (without evidence, of course) that they in any way set prices or pay based on individual passenger or driver characteristics. However, their recent tactics de facto promote a race-to-the-bottom in driver pay by operating a bidding process on their driving app where competing drivers have literally seconds to accept low pay offers (usually after Uber’s initial low pay offer to a specific driver was rejected). Anecdotal evidence also suggests that Uber may be using similar tactics to set and revise passenger fares, lowering them only if a passenger rejects its initial higher price offer.

Let’s skip ahead to the present. In 2025, Uber has now consistently reported quarterly profits since 2023. Uber’s second-quarter results on Wednesday: it beat revenue expectations, gross bookings were up 17 percent from last year, revenue up 18 percent since last year, monthly riders are up' posted a $1 billion quarterly profit, and so on and so on. It might be instructive to look at the past few years of improvements that led us here, so let’s rewind and look at Uber’s P&L improvement from 2019 to 2024. Here, again, I’ll refer to Horan who poured over the books to try and piece together a more complete picture of Uber’s finances since, as Horan and Sherman lay out, Uber regularly omits, obscures, or otherwise obfuscates what is offered.

In 2024, Horan looked at Uber's full year financial reports—it's a notable year because after losing $33 billion for its first 13 years, the company seemed to have finally turned a corner. Horan spends the first chunk of the piece explaining that a significant part of Uber's operating results trace back to dubious practices such as "its estimate of the changes of value in untraceable securities it received after shutting down operations that had been hopelessly unprofitable" such as shares in firms that beat Uber worldwide (Didi, Yandex, Grab) or in firms that acquired its failed moonshot projects (Aurora's acquisition of Uber's failed autonomous vehicle unit).

Horan estimates 2023 and 2024 "net profitability" were each inflated by $1.8 billion due to the aforementioned practices. He also finds that $6.4 billion of Uber's 4Q24 bottom line came from a tax valuation release, thanks to the massive losses it incurred for years (it has claimed over $41 billion in deffered tax assets). These add up and overstated Uber's P&L by 18 points in 2024, made "net profitability" a useless metric, but still allowed the firm to claim profit margins substantially higher than reality. To make matters worse, the firm publishes deceptively simplistic metrics that shed light any real changes in the way its services are consumed, their unit economics, differences across locales, or anything detailed enough to invite scrutiny. Operational costs are similarly blended and not broken out at the level of detail needed to clearly figure out Uber's take or driver revenue, driver incentives and bonuses, and so on.

These practices persist to this day and there’s little reason for Uber to change them as investors and reporters continue to report out Uber’s invented metrics, accept its garbled P&L data, and parrot back the talking points about its sustainability and viability and value that are built atop the distorted results.

After correcting Uber's numbers, however, Horan still found a massive improvement—$4 billion in 2022, then another $2 billion in 2023 and 2024. How?

Three major factors appear to have driven these improvements: Uber has been keeping a larger share of each passenger dollar (and giving drivers less), it eliminated major corporate costs during the pandemic, and developed more sophisticated price discrimination tools allowing it to charge higher fares to customers more likely to accept them and to reduce compensation offers to the minimum they thought specific drivers would accept.

We’ve talked extensively about Uber’s successful campaign to hike fares while cutting driver pay by taking a larger cut of each trip. The second boils down to shifting more costs onto the driver by pulling back on marketing and subsidies (incentives and bonuses), instituting lockouts to reduce the number of drivers (especially in cities like NYC which instituted a minimum wage law tied to driver utilization rates), and slashing its corporate workforce. The third factor—sophisticated price discrimination tools—has proven to be especially lucrative and is key to bolstering how much can be gained with the first two.

University of California, Irvine School Law Professor Veena Dubal’s law review article “On Algorithmic Wage Discrimination” extensively fleshes out how a firm like Uber can leverage their vast surveillance apparatuses, built out in the name of data-collection, and use them to squeeze their workers:

As a labor management practice, algorithmic wage discrimination allows firms to personalize and differentiate wages for workers in ways unknown to them, paying them to behave in ways that the firm desires, perhaps for as little as the system determines that the workers may be willing to accept. Given the information asymmetry between workers and firms, companies can calculate the exact wage rates necessary to incentivize desired behaviors, while workers can only guess how firms determine their wages.

…

[A]lgorithmic wage discrimination also creates a labor market in which people who are doing the same work, with the same skill, for the same company, at the same time may receive different hourly pay. Digitally personalized wages are often determined through obscure, complex systems that make it nearly impossible for workers to predict or understand their constantly changing, and frequently declining, compensation.

Dubal’s article also features eight years of embedded ethnographic research centered on ride-hail drivers in California from 2014 to 2022.

In defining algorithmic payment structures as unfair and unjust, workers frequently complained of their low hourly wages, even though they were not paid hourly. In describing the harms they suffered, they drew on the language of antidiscrimination law, condemning the variability of their income not just over time but more specifically compared to other drivers. The fact that different workers made different amounts for largely the same work was a source of grievance defined through inequities that often pitted workers against one another, leaving them to wonder what they were doing wrong or what others had figured out. This feature of algorithmic wage discrimination—because of its divisive effects—may also undermine workers’ ability to organize collectively to raise their wages and improve their working conditions.

In addition to complaints about the unfairness of low, variable, and unpredictable hourly pay, workers made two other moral judgements about the techniques through which they were paid. First, as the techniques of algorithmic wage discrimination deployed by on-demand firms both lowered pay and became increasingly obscure, drivers described the process of attempting to earn not through the lens of gaming but through the lens of gambling. Second, they portrayed the algorithmic changes or interventions that prevented them from earning as they had hoped or expected as trickery or manipulation enacted by the firm. Vacillating between feeling possibility and impossibility, freedom and control, workers experienced algorithmic wage discrimination as a practice in which the machine boss’s structures and functions were designed to take advantage of them by providing the illusion of agency. As Dietrich, a part-time driver in Los Angeles said, “[It’s] constant cognitive dissonance. You’re free, but the app controls you. You’ve got it figured out, and then it all changes.”

I want to emphasize this is only one prong of what is wrong with Uber but it is a large part of it. If I could sum up everything we’ve covered and a few other key points for you, it would go as follows:

• Uber is an enterprise that regularly uses accounting tricks to obscure its failure to realize profits through innovation. It has taken advantage of superficial coverage in the business press and crafted an aggressively deceptive PR campaign built on older taxi deregulation lobbying playbooks as well as company-sponsored academic research.

• Uber has realized stupendous growth by using capital as a weapon (investor subsidies), breaking the law then rewriting it (regulatory arbitrage), embracing anti-competitive business practices, and deploying incredibly successful political operations in the United States and worldwide that have codified its arbitrage efforts.

• Uber has realized profits largely through predatory behavior such as algorithmic wage discrimination and perpetual fare hikes. Expansions into other lines of business have benefited from its years of experimentation here (namely food delivery).

• Uber has offered a roadmap for other firms interested in immiserating their workers, growing via the misallocation of public subsidies. The metastasis of the so-called “gig economy” will continue unabated as Uber’s bastards proliferate.

• It is bad when a firm uses investor subsidies to distort markets such that it can realize profits through predatory behavior, even worse when it does so with public subsidies, and even worse when this allows the firm to rewrite laws to realize profits locked behind activities made illegal because they are against the public interest, and creates a model for other firms to do so. That Yglesias cannot understand this suggests he is an idiot or operating in a different moral universe.

It’s not clear why Yglesias looks at all of this and sees innovation. The previous system was not great, and we did not end up with a better system. Arguably it is worse, in part because it attracts commentators eager to defend it from criticism for no reason other than its use of technology. Is this what passes for “economic analysis?”

If you are so focused on innovation, how can you confuse it with plain old predation given a new digital skin? If you claim to be concerned with good outcomes, how can you overlook a business model that prizes its ability to sacrifice its workers, distort markets, break laws, and undermine democracy?

Read more

Most of my thinking on Silicon Valley—on its firms, its products, its financiers, its ideologues, its boosters, and its projects—rests on a relatively simple understanding: these people will sacrifice us.

My first experience witnessing this came when helping organize ride-hail drivers working for Uber and Lyft as well as talking with taxi drivers struggling to survive the ascent of these firms. These companies, in a desperate scramble for their first profits, brazenly ignored the law, misclassified and immiserated countless workers, pushed drivers into predatory leasing agreements, paid out starvation wages while dodging taxes and ensuring drivers were blocked from dignified working conditions, and countless more abhorrent practices.

Who cared if a few taxi drivers committed suicide because UberLyft’s predations degraded pay and labor conditions across the entire ride-hail sector, or if drivers were forced to sleep in their cars to meet aggressive quotas crafted to effectively lockout and fire workers (minimizing labor costs), or if they were attacked or robbed or killed on the job. So what? Were you going to complain on behalf of people who couldn’t adapt to the future, who made a bad choice in betting their livelihood on a line of work that should be Flexible and Temporary, who are lucky enough to get in early on “the operating system for your everyday life.”

Have things improved? Uber’s global lobbying and law breaking campaign was a resounding success—they’ve successfully degraded working conditions worldwide, convinced regulators that their specific model and structure is inevitable, integrated themselves into policy planning visions and decisions, and burned enough capital to create their desired markets and consumers and behaviors where they did not exist before.

All it took along the way was the physical and mental health of countless workers, some air pollution and traffic congestion in its major markets, tens of billions of dollars of investor capital wasted on failed behavioral psychology experiments and science fiction projects, a few public transit systems, backroom deals, and an incredulous corp of commentators.

Imagine how much more could be achieved with an even greater sacrifice.

Things have only gotten worse as Silicon Valley’s business model has metastasized, with oligarch-intellectuals poised to reorganize wider and wider swaths of our economy, culture, social relations, and politics. To maximize profits and efficiency and productivity, to purge capitalism of its last vestiges of democracy and liberalism, to transform speculative gains into real wealth then into political power that makes this alchemy easier, to discipline consumers and workers and regulators, to foster paranoia (whether by states or communities) and preserve order, to pursue geo-strategic primacy, to summon some artificial superintelligence that will either end history or realize historic profits, anything and everything will be offered up. Something has to give—the situation demands a blood sacrifice.

Some believe the sacrifices will give birth to a stillborn god that will save the world. They insist, as Google’s former chief executive Eric Schmidt does, that “we are never going to meet our climate goals anyway” so now is the time to double down on overbuilding AI infrastructure. Climate change will be staved off only by accelerating the very developments bringing about the collapse of our ecological niche—so consume the water, foul the air, enrich fossil fuel firms, do whatever you must and do it with quick if there is going to be any hope of creating an “infinitely patient, infinitely compassionate, infinitely knowledgeable, infinitely helpful” entity capable of saving the world.

Some believe that humanity will be liberated by subjecting vast swaths to undignified drudgery—we need ghost workers, potemkins, and sin-eaters to power the Great Work. The global AI value chain features critical processes—data collection and annotation, analysis and model development, and data verification—performed by "invisible workers" tucked away in digital sweatshops defined by piece-rate work, low pay, and undignified working conditions. As my TMK co-host Jathan Sadowski has made clear for years now, Potemkin AI (AKA “services that purport to be powered by sophisticated software, but actually rely on humans acting like robots—services that purport to be powered by sophisticated software, but actually rely on humans acting like robots”) can be best understood as a few things, such as:

• Disciplinary power, or the power of "coaxing and cajoling, of implanting beliefs and inducing action” by, for example, convincing people you've built a panopticon that is "tirelessly processing feeds from the ubiquitous cameras, rather than groups of human analysts who take time, get fatigued, and make mistakes.."

• Choosing “certain interests over others and reasserting the value of certain people over others” like prioritizing artificial intelligence infrastructure over human needs. Or, as Jathan writes, a "placeholder" for "attempts to use AI as a tool for replacing human decisions, exploiting human labor, and administering human life" so long as you don't look behind the curtain.

On the question of sin-eaters, it is increasingly clear that firms will offer various configurations of man-machine systems to obfuscate culpability. A human will serve as a legal guarantor or as "the final stop in the responsibility chain.” AI will be used externalize moral agency, rationalizing solutionist approaches that preserve the status quo while doing nothing to address root causes (e.g. robust carbon offset and credit markets that do not undermine fossil fuel extraction, content moderation that does not actually undermine hate speech, generated precision kill lists that justify genocide, predictive (over)policing that justifies ongoing overpolicing, and so on.

In its bid to become a central AI platform, Meta will spend $15 billion to acquire AI data labelling company Scale AI—an acquisition that will bring together two firms with longstanding commitments to exploiting workers across the world that are central to their respective platforms. Scale sacrifices those workers in pursuit of outsized funding and valuations, access to a pig trough of military contracts, and now an acquisition by a much larger firm. Meta has, for a long while now, sacrificed its workers in pursuit of persistent growth that sustains its core surveillance advertising revenue stream while buying time to cultivate others (e.g. the metaverse, its own financial system, and AI tools for federal agencies and military contractors).

On that note, some believe the sacrifices will ensure a renewed Pax Americana that brings together the private tech sector and the armed forces to cultivate nationalist fervor at home alongside a strategy that steers global development towards our national interest. As an added benefit, deploying the next generation of weaponry at home will surveil and denaturalize and deport dissidents, terrorize and dispossess migrants, that introduce dysfunction to the body politic.

In a desperate bid to beat back China's ascent, America is building a reactionary political coalition that links fossil capital with tech oligarchs, warmongers with China hawks. The great white hope here is that China's predominance in various tech stacks can be beaten back to secure control over the future of our global energy system, the course of technological development and deployment, and what gets produced where/how/why across our planet. Why should the United States—or more precisely, why should this reactionary coalition—control who gets access to various technologies? Because we say so.

Some believe sacrifices will restore some semblance of a natural order we’ve lost sight of. The future of human flourishing, they insist, isn’t going to be found in the past few centuries of flirtations with democracy and liberalism, but in a recommitment to Biological Hierarchies that reimpose caste, eugenics, apartheid, terror, and the like. We must administer a harsh treatment for a harsher disease that will cause a great deal of pain and misery in the short-term, but leave us better off in the long-run. That these reactionary ideologies are proving increasingly fundamental to the worldview of the most powerful people in the world and their sycophants, at the same time as this desperate search for capitalist (re)legitimacy, does not bode well for any of us.

These and more horrific exterminist forces are firmly in the driver’s seat, enjoying victory after victory, accumulating greater and greater resources to remake the world into a form more hospitable to their political project(s), and in the course of this self-annihilation they are likely closing the doors on various futures forever—though it will be a long time before we learn which options are lost to us forever.

This unholy alliance—far-right oligarch-ideologues who think democracy and capitalism are incompatible, tech firms with laboratories innovating the armament of fascism, financiers eager to transform speculation into wealth into power, and a host of other demoniacs—is relatively insulated from the public, its concerns, its pressures, its frustrations, and the few levers connected to those that could effect a change. And as a result, it enjoys relatively unimpeded power in building, expanding and legitimizing a police state in this country—a country that has, for a long time now, committed itself to surveillance, social control, force, projection, arbitrary violence, and terror.

It is increasingly unclear to me what, if anything, can be done about this. Though I suppose we’re all struggling with that problem right now. I’ll leave you with the end of Thanatos Triumphant, one of Mike Davis’ last essays and one of my favorites:

As an objection to my pessimism, one might claim that China is clear-sighted where everyone else is blind. Certainly, its vast vision of a unified Eurasia, the Belt and Road project, is a grand design for the future, unequalled since the sun of the ‘American Century’ rose over a war-shattered world. But China’s genius, 1949-59 and 1979-2013, has been its neo-mandarin practice of collective leadership, centralized but plurivocal. Xi Jinping, in his ascent to Mao’s throne, is the worm in the apple. Although he has economically and militarily enhanced China’s clout, his reckless unleashing of ultra-nationalism could yet open a nuclear Pandora’s Box.

We are living through the nightmare edition of ‘Great Men Make History’. Unlike the high Cold War when politburos, parliaments, presidential cabinets and general staffs to some extent countervailed megalomania at the top, there are few safety switches between today’s maximum leaders and Armageddon. Never has so much fused economic, mediatic and military power been put into so few hands. It should make us pay homage at the hero graves of Aleksandr Ilyich Ulyanov, Alexander Berkman and the incomparable Sholem Schwarzbard.

My series of essays on Artificial Intelligence thus far:

• AI, life engineering, and digital hygiene - On artificial intelligence as an attack vector for eugenics & political shock therapy.

• AI, slavery, surveillance, and capitalism - On artificial intelligence for The Labor Question & the limits of surveillance capitalism and techno-feudalism.

• The Silicon Valley Consensus & AI Capex (Part 1) - On overbuilding AI infrastructure and its energy supply.

Here are also a few other essays on AI that you could probably fit in there:

• Does OpenAI’s latest marketing stunt matter? - On distractions, intentions, aesthetics, and fascism.

• The phony comforts of useful idiots - On Casey Newton and the shallowness of (AI) anti-skepticism.

• Trapped in the Maw of a Stillborn God - On Vegas as a laboratory for surveillance and social control, the explosion of gambling as a sign of a degenerate culture seized by despair, AI delusions at CES, and the future.

If these essays sound interesting, consider supporting The Tech Bubble (me) with a subscription ($7 a month or $70 a year). If you're already supporting me, let me kiss that big ass cranium of yours expeditiously.

Today’s essay is about artificial intelligence (surprise), using debates about AI, some relevant news items, some reflections on real use cases, its various material and spiritual impacts, and the general degradation of our world.

Looking for an illustration to pair with the essay, Burke’s called to me from the recesses of one of my drives—it projects corrupted strength, the intervention of/observation by alien forces, evidence of violence and the promise of it, and it also just looks cool as hell. I did a reverse image search to try and credit the illustrator (and remember how I came across it), which turned up a few things. One was a May 2024 quote tweet of mine where I simply posted the picture. Another was a New Noise Magazine interview with a Cleveland, Ohio heavy metal core band called Walking Wounded that used this painting for their debut EP, Bestial Condemnation and captured better than I initially did why the picture was perfect:

“Bestial Condemnation is a fantasy story about an aging knight who makes a deal with an ancient deity to restore his former strength,” begins guitarist Steve Perrino, “However, his newfound strength comes at the price of his humanity as he gradually devolves into something pre-human and violent.” Here’s a fitting central lyrical theme if there ever was one, given the quartet’s barbaric style.

“I think many of us wish we were stronger in a wide variety of ways, whether it’s physically, mentally, or emotionally,” he continues. “I wanted to write something that asks the listener, ‘How far would you go to be the person you think you should be? What price would you pay? Will it be worth it?”

As I’ve been trying to argue in this series, it’s difficult to talk about why one might be opposed to how artificial intelligence is being designed, developed, financed, and deployed without bringing in a multitude of threads:

• the reactionary dream of prominent tech sector investors, entrepreneurs, thought leaders, and boosters to inoculate capitalism (and technological innovation) from democracy and liberalism

• the accelerationist death cult of techno-optimists who believe the creation of a sufficiently advanced artificial intelligence will roll back whatever damage its pursuit wrecks on our ecological niche, culture, economy, political order, mental health, and so on

• the origins of computation tracing back to the British empire’s hope of making industrial labor resemble plantation labor after the abolition of slavery, the extent to which reactionaries are eager to use these technologies to re-legitimize various hierarchies and forms of social control

• our tech sector’s persistently lucrative strategy: overbuilding, overvaluing, and overinvesting in various assets in hopes of realizing excessive gains, translating those into political power, and undertaking the task of restructuring society in their desired image.

It does not help that these concerns not only converge on artificial intelligence, but on a great deal of tech products foisted upon us by Silicon Valley leading up to it—namely those littering the so-called sharing economy, gig economy, metaverse, web3, and cryptoeconomy. Trillions of dollars have flowed to Wall Street and Silicon Valley to push out products seemingly designed to do nothing more than hurt as many people as possible while enriching as few as possible: sleek weapons for militaries and police forces, ubiquitous surveillance systems, vaporware, poorly designed Ponzi schemes, and elaborate conspiracies for printing out lottery tickets for distribution among a narrow group of bloodthirsty, skull-measuring, anti-social libertarians.

As Evgeny Morozov noted in 2019—more than a decade since the start of the great financial crisis—capitalists remain uneasy:

Once-alluring promises of meritocracy and social mobility ring increasingly hollow. They pine for a slicker, PowerPoint-friendly legitimation narrative—hard to concoct against a background of rising inequality, pervasive tax evasion and troubling omens about the true state of the post-crash global economy, were central bankers to withdraw their overextended support. What real-world developments could underpin such a narrative? What theme could make the idea of capitalism more morally acceptable to the latest batch of Ivy League graduates, who may risk getting drawn to notions like eco-socialism? Despite the growing ‘tech-lash’ against the FAANGS, capitalist thinkers still look to Silicon Valley and its culture with a glimmer of hope. For all its problems, the Valley remains a powerful laboratory of new—perhaps, better—market solutions. No other sector occupies such a prominent role on the horizon of the Western capitalist imaginary or offers such a promising field for regenerative mythologies.

Those words are no less true six years later—in fact, after the end of the ZIRP era it’s hard to understand the ongoing tech asset bubble, and the role it’s playing in promulgating artificial intelligence, without appreciating the legitimacy crisis plaguing capitalism.

It feels increasingly obvious that almost everything will be sacrificed to preserve the status quo, that we will be forced on an impossible path—giving birth to a superhuman artificial intelligence in hopes of improving our work lives, social lives, economies, politics, health, intelligence, and culture. Getting there just happens to require prioritizing the private financing, development, and deployment of relevant technologies, as well as giving increasingly more political and economic power to firms and entrepreneurs and financiers responsible for sabotaging our environment, social relations, politics, culture, and economy.

What will the cost end up being at the end of the day?

AI == INDULGENCE

The general thrust here is that modern AI hype cycles have a religious texture to them: one layer resembles the grounds on which Martin Luther criticized the abuse of indulgences in the 95 Theses he pinned to a church door in 1517, another resembles the folk-ritual of sin-eating—where a meal (sometimes resting on a corpse) was consumed to essentially absolve the recently deceased of their sins. The former will be the focus of this essay.

Indulgences in the Medieval Era

Medieval Church doctrine held there was an eternal and temporal dimension to sin: the former could be forgiven by confession and genuine repentance, but the later persisted and carried consequences in this life and the next (in Purgatory). Additional acts of penance (prayer, fasting, charity, good works, etc.) were required then to remit temporal punishment.

Enter indulgences: think of them as a credit or certificate given after an act of penance, representing a commuting of the debt incurred by sin and a reduction of future time in Purgatory. The Church began selling indulgences—promising that the monetary exchange was equivalent to penance, and then issuing excessive indulgences to raise funds for, well, anything they could. Here is an illuminating section from God's Bankers, Gerald Posner's exhaustive history of the Vatican's finances:

The early church's penances were often severe, including flogging, imprisonment, or even death. Although some indulgences were free, the best ones—promising the most redemption for the gravest sins—were expensive. The Vatican set prices according to the severity of the sin and they were initially available only to those who made a pilgrimage to Rome.

Indulgences helped Urban II in the eleventh century offset the church's enormous costs in subsidizing the first Crusades. He offered full absolution to anyone who volunteered to fight in 'God's army' and partial forgiveness for simply helping the Crusaders. Successive Popes became ever more creative in liberalizing the scope of indulgences and the ease with which devout Catholics could pay for them. By the early 1400s, Boniface IX—whose decadent spending kept the church under relentless financial pressure—extended indulgences to encompass sacraments, ordinations, and consecrations. A few decades later, Pope Paul II waived the need for sinners to make a pilgrimage to Rome. He authorized local bishops to collect the money and dispense the indulgences and also cleared them for sale at pilgrimage sites that had relics of saints. Sextus IV had an inspired idea: apply them to souls stuck in Purgatory. Any Catholic could pay so that souls trapped in Purgatory could get on a fast track to Heaven. The assurance that money alone could cut the afterlife in Purgatory was such a powerful inducement that many families sent their life savings to Rome. So much money flooded to Sextus that he was able to build the Sistine Chapel. Alexander VI—the Spanish Borgia whose Papacy was marked by nepotism and brutal infighting for power—created an indulgence for simply reciting the Rosary in public. The new sales pitch promised the faithful that a generous contribution multiplied the Rosary's prayer power.

Each Pontiff understood that tax revenues from the Papal States paid most of the day-to-day bills, while indulgences paid for everything else. The church overlooked the widespread corruption and graft inherent in collecting so much cash and instead grew ever more dependent on indulgences. And as they got ever easier to buy and promised more forgiveness, they became wildly popular among ordinary Catholics.

Indulgences were, however, more than a financial lifeline. They also helped medieval Roman Popes withstand challenges to their secular power. So-called antipopes—usually from other Italian cities—claimed they, rather than the pope elected in Rome, had the political or divine right to rule the Catholic Church. Although some antipopes raised their own armies and had popular backing, they never mustered the moral authority to issue indulgences. Repeated efforts over centuries by pretenders to the Papacy to package and sell forgiveness for sins failed. Few Catholics believed that anyone but the Roman Pope had the direct connection with God to offer a real Indulgence. And when the Pope's armies were called upon to sometimes crush an antipope, it was usually the flood of cash from indulgences that paid for the war.

Pope Leo X (1513-21), a Medici prince who'd been a wealthy and powerful cardinal since the age of 13, wasted no time bringing over his lavish lifestyle to the Papal Court when he ascended at 38. As Posner points out: Leo commissioned major artists for extravagant projects decorating Vatican estates, the Vatican's servants "nearly doubled to seven hundred," cardinals were soon called "Princes of the Church," and critics of his abuse of indulgences were threatened with excommunication.

Hoping to subsidize the cost of constructing St. Peter’s Basilica, Leo X issued a bull of plenary indulgence in 1515: the temporal punishment for almost any sin could now be remitted in exchange for money, and for the next 8 years all other indulgence promotion was to cease. For 8 years, this one indulgence could put you on the fast track to Heaven! Or as Luther claimed on Dominican friar, Johann Terzl, said: "“As soon as the coin in the coffer rings, the soul from purgatory springs.”

There are a few key areas of Luther’s critique that I’ll hone in and map onto artificial intelligence and we’ll develop them more in the next section:

1. Salvation through transaction. Indulgences shouldn’t be presented as a shortcut to salvation you can take instead of faith in God, genuine repentance, or good works motivated by the previous two. Luther's Theses don't oppose indulgences so much as the idea that the living (specifically the Papacy) can know whether sin can be remitted, especially when this remission is via a financial transaction and not a good work or genuine repentance.

   1. “That power which the pope has in general over purgatory corresponds to the power which any bishop or curate has in a particular way in his own diocese and parish. The pope does very well when he grants remission to souls in purgatory, not by the power of the keys, which he does not have, but by way of intercession for them.” (Theses 25-26)

2. Centralizing and codifying unjustified power. To suggest indulgences could be granted to the living on behalf of souls in Purgatory represented a massive overreach on the part of the Church, which was now suggesting it had authority over Purgatory. Introducing a commercial dimension to indulgences added perverse incentives that distorted the Church's mission and prioritized a host of needs which had nothing to do with Christendom.

   1. “They preach only human doctrines who say that as soon as the money clinks into the money chest, the soul flies out of purgatory. It is certain that when money clinks in the money chest, greed and avarice can be increased; but when the church intercedes, the result is in the hands of God alone.” (Theses 27-28)

Indulgences in the Modern Era

One place to start is with “solutionism” which was first proposed by Morozov as the zealous belief that the application of technology can solve humanity's problems given the right algorithm, product, or private-public partnership. The degree to which various technologies cause disruption is the degree to which we are working through the inefficiency of its uneven adoption, a stratagem that conveniently reinforces the urgency of pursuing political programs prioritizing the uncritical adoption of new products, business models, asset classes, and regulatory frameworks (i.e. on-demand platforms, fintech, crypto, web3, metaverse, artificial intelligence). As Morozov puts it in “To Save Everything, Click Here”:

Recasting all complex social situations either as neatly defined problems with definite, computable solutions or as transparent and self-evident processes that can be easily optimized—if only the right algorithms are in place!—this quest is likely to have unexpected consequences that could eventually cause more damage than the problems they seek to address.

On a basic level, the drive to push for universal fixes through the application of breakthrough technologies may be offered as a shortcut that bypasses the onerous hard work of enacting structural reforms, imposing ethical limitations, or adopting restrictive regulations that trade endless growth or progress for intentional development. It insists social problems are just market opportunities waiting for a bold entrepreneur. Public goods and services are mismanaged assets that should be made private, expediently.

Over the years, Morozov’s articulation of solutionism has evolved along with the focus of his analysis. In a 2023 New York Times op-ed, Morozov warned that solutionism had re-emerged as "digital neoliberalism" and was finding even more zealous adherents thanks to the invocation of artificial general intelligence, or A.G.I. This technology, its boosters hold, is inevitable, will likely be beneficial, and is really the only way to solve humanity's problems.

Unbeknown to its proponents, A.G.I.-ism is just a bastard child of a much grander ideology, one preaching that, as Margaret Thatcher memorably put it, there is no alternative, not to the market.

Rather than breaking capitalism, as Mr. Altman has hinted it could do, A.G.I. — or at least the rush to build it — is more likely to create a powerful (and much hipper) ally for capitalism’s most destructive creed: neoliberalism.

…

[N]eoliberalism is far from dead. Worse, it has found an ally in A.G.I.-ism, which stands to reinforce and replicate its main biases: that private actors outperform public ones (the market bias), that adapting to reality beats transforming it (the adaptation bias) and that efficiency trumps social concerns (the efficiency bias).

These biases turn the alluring promise behind A.G.I. on its head: Instead of saving the world, the quest to build it will make things only worse.

The market bias compels two phases in any technology rollout, but especially A.G.I.'s: a first phase where heavily subsidized services offered, then a second where prices are hiked and "overdependent users and agencies" are bled dry to make the venture generate profits (and reshape consumer behavior, regulatory frameworks, market dynamics, and political orders into forms more hospitable to profits and rents)

This plank can be understood as vulgar solutionism, which was on full display with the wave of frothy Silicon Valley startups that thrived during the zero-interest rate period, but has now been supercharged with the rise of the A.G.I.-ism cult. As Ed Zitron has spent newsletter after newsletter after newsletter detailing, there is a fundamental profitability issue at the heart of the generative AI business model—which is the sector actively pursuing A.G.I. Firms that have yet to turn a profit are burning through tens of billions, interested in raising hundreds of billions, and forecasting they will need trillions—who is going to fork up all that cash? Well, you (and me):

Thus, the ugly retrenchment phase, with aggressive price hikes to make an A.G.I. service profitable, might arrive before “abundance” and “flourishing.” But how many public institutions would mistake fickle markets for affordable technologies and become dependent on OpenAI’s expensive offerings by then?

And if you dislike your town outsourcing public transportation to a fragile start-up, would you want it farming out welfare services, waste management and public safety to the possibly even more volatile A.G.I. firms?

Then there's the adaptation bias, the idea that technological fixes can make us more self-reliant and resilient. Technology is not here to fix your city’s ailing infrastructure or to improve its social programs, but it can help you navigate their perpetual decline and evisceration!

The message is clear: gear up, enhance your human capital and chart your course like a start-up. And A.G.I.-ism echoes this tune. Bill Gates has trumpeted that A.I. can “help people everywhere improve their lives.”

The solutionist feast is only getting started: Whether it’s fighting the next pandemic, the loneliness epidemic or inflation, A.I. is already pitched as an all-purpose hammer for many real and imaginary nails. However, the decade lost to the solutionist folly reveals the limits of such technological fixes.

…

There’s a difference between nudging us to follow our walking routines — a solution that favors individual adaptation — and understanding why our towns have no public spaces to walk on — a prerequisite for a politics-friendly solution that favors collective and institutional transformation.

But A.G.I.-ism, like neoliberalism, sees public institutions as unimaginative and not particularly productive. They should just adapt to A.G.I., at least according to Mr. Altman, who recently said he was nervous about “the speed with which our institutions can adapt” — part of the reason, he added, “of why we want to start deploying these systems really early, while they’re really weak, so that people have as much time as possible to do this.”

This brings us to the effiency bias, or the idea that complexity (i.e. morality, social good, justice, etc.) gets in the way of optimal market outcomes.

This fixation on efficiency is how we arrived at “solving” climate change by letting the worst offenders continue as before. The way to avoid the shackles of regulation is to devise a scheme — in this case, taxing carbon — that lets polluters buy credits to match the extra carbon they emit.

This culture of efficiency, in which markets measure the worth of things and substitute for justice, inevitably corrodes civic virtues.

And the problems this creates are visible everywhere. Academics fret that, under neoliberalism, research and teaching have become commodities. Doctors lament that hospitals prioritize more profitable services such as elective surgery over emergency care. Journalists hate that the worth of their articles is measured in eyeballs.

Now imagine unleashing A.G.I. on these esteemed institutions — the university, the hospital, the newspaper — with the noble mission of “fixing” them. Their implicit civic missions would remain invisible to A.G.I., for those missions are rarely quantified even in their annual reports — the sort of materials that go into training the models behind A.G.I.

After all, who likes to boast that his class on Renaissance history got only a handful of students? Or that her article on corruption in some faraway land got only a dozen page views? Inefficient and unprofitable, such outliers miraculously survive even in the current system. The rest of the institution quietly subsidizes them, prioritizing values other than profit-driven “efficiency.”

In defense of A.G.I. and the fundraising various firms have undertaken to realize it (or to finally see a return, whether or not A.G.I. ever happens), this ideology’s zealots have constructed a variety of artifices that obscure the real costs of their activity as well as the extent to which it’s solely self-interested and far removed from anything resembling the public interest.

As I’ve laid out before, Silicon Valley’s race to build as many data centers as possible (AI capital expenditures, or AI capex) necessitates that the sector and its financiers overbuild, overvalue, and overinvest in compute and energy infrastructure.

AI infrastructure is downstream of cloud compute infrastructure, which is dominated by a few monopolistic firms that embrace deeply exploitative business practices to leverage “hyperscale” to extract rents, steal ideas, deter competitors, trap clients, and enter new-markets with built-in advantages. This infrastructure is overbuilt, overvalued, and overinvested in because of the opportunity to create a new market that resembles cloud’s ability to deliver rents, generate anti-competitive market intelligence, and realize juicy profit margins—to realize levels of growth and profit only possible when a market is rigged.

When it comes to energy infrastructure, there are a few concerns that emerge. First and foremost is that the fossil fuel industry wants to power AI infrastructure and be powered by it. Major tech firms have made it clear that they view sustainability commitments as limitations and fossil fuel firms are scrambling to help them curtail them. To this end, AI firms have committed to worsening the climate crisis by selling tools that boost fossil fuel extraction, exponentially build out energy-intensive compute, and give fossil fuel firms alongside utilities new excuses to overbuild infrastructure that will be subsidized by ratepayers (you and me).

The ways in which compute and energy infrastructure overbuilding, overvaluation, and overinvesting are rationalized, then, are what we should be worried about. I’m going to quote Zitron at length for a bit since he’s been great on the costs, externalities, illusions, and delusions at play here.

In July 2024, Zitron raises a few questions about the mechanism by which OpenAI’s valuation

As part of this deal, Microsoft has effectively purchased the rights to OpenAI's "pre-AGI" technology, and licensed all of its technology in a way that extends past any partnership or, I imagine, future deals. Microsoft also "invested" in cloud credits at an indeterminate valuation, both in how OpenAI was valued and the credits themselves.

Ask yourself, what is a dollar of "cloud compute credits," and what do they gain you access to? Microsoft's Azure cloud has many, many products, and it's unclear if OpenAI would receive preferential pricing on them, what products they'd be using, and the terms under which OpenAI receives them. Microsoft effectively created its own currency to invest in OpenAI, which OpenAI would then pay Microsoft in, which Microsoft would, in turn, receive as revenue.

In September, he expresses concern that the issue is not just OpenAI's valuation, but a "subprime AI crisis" borne out of countless companies integrating generative AI priced at unsustainably low prices. (Recall Morozov’s argument that the A.G.I. cult, succumbing to neoliberalism's market bias, would roll out their products in two phases: heavy subsidies to undercut competition and cultivate user dependency, then price hikes to extract rents, generate first profits, and reshape users/markets/regulatory regimes/governance to be a more pliant (profitable) host body.)

Back to Zitron and the subprime AI crisis:

Almost every "AI-powered" startup that uses LLM features is based on some combination of GPT or Claude. These models are built by two companies that are deeply unprofitable (Anthropic is on course to lose $2.7 billion this year), and that have pricing designed to get more customers rather than make any kind of profit. OpenAI, as mentioned, is subsidized by Microsoft — both in the "cloud credits" it received and the preferential pricing Microsoft offers — and its pricing is entirely dependent on Microsoft's continued support, both as an investor and a services provider, a problem that Anthropic faces with its deals with Amazon and Google.

Based on how unprofitable they are, I hypothesize that if OpenAI or Anthropic charged prices closer to their actual costs, there would be a ten-to-a-hundred-times increase in the price of API calls, though it's impossible to say how much without the actual numbers. However, let's consider for a fact that the numbers reported by The Information estimate that OpenAI's server costs with Microsoft will be $4 billion in 2024 — which, I add, are over two-and-a-half-times cheaper than what Microsoft charges others — and then consider that OpenAI still loses over five billion dollars a year.

OpenAI is more than likely charging only a small percentage of what it likely costs to run its models, and can only continue to do so if it is able to continually raise more venture funding than has ever been raised and continue to receive preferential pricing from Microsoft, a company that recently mentioned that it considers OpenAI a competitor. While I can't say for certain, I would think it's reasonable to believe that Anthropic receives similarly-preferential pricing from both Amazon Web Services and Google Cloud.

Assuming that Microsoft gave OpenAI $10 billion of cloud credits, and it spent $4 billion on server costs and, let's say, $2 billion on training — costs that are both sure to increase with the new o1 and “Orion” models — OpenAI will either need more credits or will have to start paying actual cash to Microsoft sometime in 2025.

Later that month, Zitron wrote about another potential red flag in the tech sector, cloud compute—specifically, the centrality of “software as a service service” (“SaaS”) to the business models of multiple tech companies, the stagnating growth of a business unit that usually prints money for firms, and the desperation which has driven these firms to overbuild, overvalue, and overinvest in generative AI:

As I hinted earlier, I see the enterprise and the SaaS market as the commercial real estate arm of the Subprime AI Crisis — an industry desperate for growth attaching itself to an unprofitable, unsustainable technology that they invested in in the hopes that it would fuel a decade or more of revenue growth.

…

At some point, the putrid margins around generative AI will begin to eat into the already-tenuous profitability of these companies, leading to some, I imagine, having to either vastly increase prices or drop the tools altogether, assuming that the competitive landscape doesn't mean that keeping them is a necessity to compete with others.

If that's the case, many SaaS companies may have added the equivalent of an adjustable-rate mortgage to their tech stacks, all to offer features that are at best sort of cool and at worst actively harmful to a company. And based on the fact that most of these companies have to effectively double prices to offer AI, it's hard to imagine that these features aren't already problematically-expensive to maintain.

…

Another deeply-worrying eventuality could be a race to the bottom, where growth-hungry SaaS companies either deeply discount or don't charge for AI, pushing their competitors to lower the price on their already-unprofitable products. Much like the feared commercial real estate bubble, SaaS may see its own series of dramatic price cuts to its AI tools as a means of competing and showing growth to the markets...despite the costs staying the same, or possibly increasing.

And, again, much like commercial real estate, AI has locked-in costs, and service providers have little incentive to drop them. Especially in the case of Microsoft, which will likely only ever make money from AI by selling compute.

…

It's hard to overstate the significance of a collapse of growth in the SaaS market, as is it hard to overstate how dangerous generative AI is to its fortunes. While these companies had costs before, generative AI is multitudes higher than regular cloud compute costs, meaning that any new revenue growth from this software will be burdened by leveraging an increasingly-expensive solution to a problem that most of them have trouble describing.

In February of this year, Zitron wrote a great essay breaking down the ways in which people used contrived metrics to insist "generative AI is a sustainable or real industry at the trillion-dollar scale" which justified externalized costs to various parts of our society that A.G.I. cultists were eager to subject to neoliberal dogma (e.g. privatization, adaptation, and efficiency maximization):

Generative AI is a financial, ecological and social time bomb, and I believe that it's fundamentally damaging the relationship between the tech industry and society, while also shining a glaring, blinding light on the disconnection between the powerful and regular people. The fact that Sam Altman can ship such mediocre software and get more coverage and attention than every meaningful scientific breakthrough of the last five years combined is a sign that our society is sick, our media is broken, and that the tech industry thinks we're all fucking morons.

This entire bubble has been inflated by hype, and by outright lies by people like Sam Altman and Dario Amodei, their lies perpetuated by a tech media that's incapable of writing down what's happening in front of their faces. Altman and Amodei are raising billions and burning our planet based on the idea that their mediocre cloud software products will somehow wake up and automate our entire lives.

The truth is that generative AI is as mediocre as it is destructive, and those pushing it as "the future" that "will change everything" are showing how much contempt they have for the average person. They believe that they can shovel shit into our mouths and tell us it's prime rib, that these half-assed products will change the world and that as a result they need billions of dollars and to damage our power grid.

Now, at last, we can begin to make our connections to Martin Luther’s critiques of indulgences.

Salvation Through Transaction

As we’ve gone over, tech firms present artificial intelligence as a universal solution to a vast array of social and political problems, even as these firms create or exacerbate them.

Venture capitalists enjoyed well over a decade of low interest rates that let them borrow huge sums of money at little cost, what did they do with it? The vast majority (around 90 percent) of their unicorns (firms valued at $1 billion or more) proved to be fundamentally unprofitable. We are all familiar with the basic formula: give me money to bleed everyone else dry and scrounge together a return for you.

In the case of Uber: give us tens of billions, secure us a valuation in excess of $100 billion, we replace very taxi driver with misclassified gig workers paid starvation wages or flying robot taxis paid no wages, and you will realize a lucrative return—it will just cost consumers immensely as use algorithmic discrimination to hike prices and suppress wages, as we massively increasing pollution and traffic congestion, and as we undermine public transit systems in our major markets, but along the way the public will forget about our initial act of arson and clamor to buy our fire insurance.

In the case of OpenAI et. al: give us tens of billions, secure us a valuation in excess of $100 billion, we replace every worker with, as premiere VC fund a16z puts it, “an AI assistant/coach/mentor/trainer/advisor/therapist that is infinitely patient, infinitely compassionate, infinitely knowledgeable, and infinitely helpful,” and you will realize a lucrative return—it will just cost consumers immensely as we enrich and sustain the fossil fuel sector, empower utilities to overbuild energy infrastructure, inflate user traffic metrics for our products to overbuild compute infrastructure, flood the digital world with sophisticated deepfakes and endless content slop, and bolster Silicon Valley’s ongoing effort to transmute speculative gains into real wealth into political autonomy that undermines democracy in the name of this desperate scramble for excessive returns, profits, rents, and greater power still.

As Karen Hao recently wrote in a New York Times op-ed:

The leading A.I. giants are no longer merely multinational corporations; they are growing into modern-day empires. With the full support of the federal government, soon they will be able to reshape most spheres of society as they please, from the political to the economic to the production of science.

…

Their influence now extends well beyond the realm of business. We are now closer than ever to a world in which tech companies can seize land, operate their own currencies, reorder the economy and remake our politics with little consequence. That comes at a cost — when companies rule supreme, people lose their ability to assert their voice in the political process and democracy cannot hold.

It is relatively easy (read: lazy) to delegate technological innovation to a sector that purports to be single-mindedly dedicated to it. But as Hao muses, you wouldn’t go along with this if fossil fuel firms said “we alone should be allowed to look into climate change.” (Well, some people would and do.) This surrender of control over our technology has been made easier still by the ease with which Silicon Valley and a persistently obsequious tech media have accepted and promulgated talking points that affirm this self-serving status quo.

In the past, fear of God worked as a key pillar in rationalizing the sale of indulgences so that one could sidestep it—today, firms like OpenAI have embraced a similar marketing strategy: fear of our stillborn god, AGI. Promising to revolutionize human civilization and solve all our ills, Altman traveled the globe searching for $7 trillion in investments to kickoff humanity’s salvation. The fever dream has moderated significantly since then, now anticipating AGI at a date ranging from this year to soon!

Today’s indulgences take a variety of forms, but at their core feature artifices deployed to sustain this unsustainable enterprise.

Why reduce your own emissions when you can utilize renewable energy credits, a creative accounting scheme that allows you to secure certificates that show you’ve bought renewable-energy generated electricity produced off-site, even though you don’t have to actually consume it. A September 2024 Guardian analysis found that data center emissions from Google, Microsoft, Meta, and Apple are likely 662 percent higher than officially reported. Buy a literal certificate to avoid doing the hard structural reform and carry on with business as usual!

Maybe you want to help other firms claim they’re reducing emissions? Follow Microsoft's lead: last year, Hao reported on how Microsoft claimed it was reducing the emissions of fossil fuel companies by selling them AI tools that allowed them to more efficiently find and develop oil and gas reserves, maximizing production at those sites. Yes, we must stop doing this if we are interested in preserving our ecological niche, but the only way we can actually do so is by developing AI tools that generate revenues and profits as well as demand for renewable energy infrastructure which can power greater and greater levels of compute which can then eventually give rise to a breakthrough technology that will save us.

Perhaps you are more interested in the compute infrastructure side? Want to structure investments linked to cloud compute & preferential prices to keep major firms afloat (OpenAI—Microsoft/Google—Anthropic—Amazon) while they figure out how to fix unprofitable business models and unsustainable unit economics? Or maybe you would like to gin up demand along each node of the generative AI supply chain by pursuing a specific vision of artificial intelligence that requires more GPUs, more clusters, larger and more compute-intensive models, more data, more data centers, more, more, more, until somewhere along the way you can grow your way into your first profits? Concerned that you won’t be able to secure larger and larger infusions of capital to justify larger and larger valuations as you prognosticate about larger and larger ambitions for your increasingly unprofitable enterprise? In each instance, the solution clearly is to just deploy more capital, to spend more money plodding along in the same direction until you are saved.

The indulgence economy for artificial intelligence is a simple one. Artificial intelligence will save the world. The firms advancing artificial intelligence will save the world. They will need as much water, land, coal, natural gas, electricity, talent, and money as possible to do this. They will need unimpeded access to as much personal data as possible, as much intellectual property as possible, as much freedom to control and reshape markets and consumer behavior as possible, as much debt as possible, as much investor capital and as much government treasuries as possible.

If artificial intelligence is contributing to mass illiteracy and cheating, it is because you dragged your feet on adopting it (or adapting to it) in the classroom. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is contributing to an information environment saturated by nonconsensual porn, child sexual abuse material, scams and fraud that utilize realtime deepfakes generated by AI deep learning, or an abundance of AI-generated slop, it is because you dragged your feet on adopting it (or adapting to it) in your information environment. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is threatening the livelihood of your creative workers, it is because you dragged your feet on adopting it (or adapting to it) in cultural production. Close your eyes and fork over some cash to fix this with artificial intelligence.

If artificial intelligence is threatening to bolster the fossil fuel industry at the precise moment that sector must be put down for the sake of humanity’s survival, if there is an unimaginable amount of opacity around how much energy infrastructure artificial intelligence will actually need, if data centers are sucking up water everywhere they can manage from deserts to areas gripped by high levels of water stress, it is because you dragged your feet on adopting it (or adapting to it) in energy policy. Close your eyes and fork over some cash to fix this with artificial intelligence.

And so on, and so on.

Change, it goes, can only be realized by exchanging some coin in the direction of our saboteurs. It’s hard to break out of the brain-dead political imaginations that tell us the financing, design, development, and deployment of technology should be left in the hands of an insular network of libertarians and reactionaries more interested in surveillance, force projection, and social control than anything resembling the public good. Harder still when those preserving such a broken political order have figured out there is money to be made, even when there are no profits to be made. So long as you can convince people that it’s worth externalizing the moral costs (onto everyone else) and eschewing accountability (for harms visited on everyone else), then you can maximize returns, wealth, and power. What else matters?

Centralizing and codifying unjustified power

It is abundantly clear by now that the variety of firms interested in artificial intelligence and spending billions on it (e.g. OpenAI, Meta, Google, Anthropic, Amazon) have aligned the development of artificial intelligence in the direction of not just profit-seeking, but a wide-ranging ecosystem of delusions, distractions, detours, accounting tricks, speculative endeavors, political projects, and (de)regulatory overhauls that are about, repeat it after me, turning speculative gains into real wealth into political power that can make repeating this alchemical transmutation that much easier.

Building on his A.G.I.-ism thesis, Morozov offers "Panglossian neoliberalism" as a term aimed at capturing how Big Tech has captured our institutional and infrastructural imagination:

Championed by venture capitalists, tech CEOs, and startup founders, this credo asserts that we already live in the best of all possible worlds (reflecting its Panglossian aspects) and that there is no alternative to the market-driven provision of our tech infrastructures (reflecting its neoliberal aspect). The essence of this ideology is distilled in the recent Techno-Optimist Manifesto by Marc Andreessen, the prominent venture capitalist, who flatly states: “free markets are the most effective way to organize a technological economy.”

As history, the dogmas of Panglossian neoliberalism are at best naïve, ignoring the significant Cold War–era public spending, much of it military, that created Silicon Valley. ARPANET, Global Positioning System (GPS), the integrated circuit, and the computer mouse all stem from government funding, not free markets. But the damage doesn’t stop with bad history. Politically, this ideology often results in paralysis, hindering the search for local, experimental, and democratic alternatives to the market-driven paradigm that dominates our technology stack.

Silicon Valley has spent a long while now insisting that our analyses should focus only on markets, but through incredibly ahistorical narratives that have less to do with understanding reality than they do with manipulating it. As I’ve laid out before, Silicon Valley’s origin story is one steeped in eugenics and environmental decay, with dreams of global domination fed by blood and soil, conquest and genocide, surveillance and control, war and insurgency. It has always been about power.

As I’ve written elsewhere:

The bounties of farmlands and the underworld mines grew and grew. The sciences that developed both helped those industries metastasize larger swaths of the land. A few individuals generated dynastic fortunes that helped build newspaper empires, propaganda campaigns, monuments and memorials, all aimed at inspiring and sustaining the waves of bodies feeding into the megamachine. More fortunes still were used to fund efforts to develop the knowledge, tools, and techniques to better dominate nature. Universities, laboratories, and corporations innovated tools of remote control to administer California’s depleted ecology and corral the populace. These forces shaped San Francisco’s development, and they never completely vanished.

The problem is that San Francisco’s contado is now the world—there are always more mines, more bodies, more forests, and more wealth to be extracted; and there will always be a need for tools and techniques to subordinate everything outside the city for some Great Work.

…

There’s a dark side to nearly every tech venture. Amazon giveth same-day delivery; Amazon taketh the physical and mental health of workers hidden away in its sprawling logistics empire. Apple divines that we need $3,500 goggles, and overlooks the sacrifice zones around the mines for precious metals powering its products and the suicides at factories assembling them. Microsoft agrees to labor neutrality against efforts to unionize its various subsidiaries, and joins the other tech giants in becoming a military contractor while helping police departments across the world find sleeker tools to terrorize their communities. Google/Alphabet promises to revitalize its ailing search engine operations with generative AI, while pursuing military cloud-computing contracts. Facebook/Meta may flail around legless in a $100 billion metaverse prison while its social media platforms stumble along, but just out of view lie an army of exploited and tortured content moderators beside an ever-growing pile of corpses from the mob violence, ethnic cleansing, and genocide its products incite.

As Silicon Valley metamorphosed from stolen land to a pioneer backwater to an imagined new Rome, as mines and oil wells puncturing the earth were connected by railroads, as the Stanford colony evolved into the American empire’s armament and laboratory during the Cold War, as one tech startup bubble deflated then gave way to an even hardier one, the Palo Alto System keeps living on like a hydra sprouting newer, hungrier heads from decapitated stumps.

Generative artificial intelligence (and the A.G.I. its boosters believe will follow) is less a product of free markets than the latest in a long line of products borne out of the Palo Alto system, which has grown increasingly effective at commercializing public research into consumer products that carry noxious social costs. Whether it’s cryptocurrency, web3, the metaverse, or on-demand platforms, Silicon Valley has tried (and failed) to insist their latest product is inevitable, revolutionary, and so transformative that they must be given an inordinate subsidies to see it through as well as as much autonomy as possible to control it unmolested by democratic oversight.

And yet, as Morozov points out, it is painfully obvious that the private provision of generative artificial intelligence has left us with the worst possible form on multiple counts:

• Inefficiency and waste: OpenAI, Anthropic, Google, Mistral, Meta, and a host of other firms are "duplicating efforts, building nearly identical resources with minor differences" that will require the same data sets, labor to treat them, and sprawling data centers that consume water and electricity. OpenAI once claimed it needed $7 trillion, in what world does it make sense to let each one of these firms independently burn trillions?

• Subpar quality of services: Hyper-competition incentivizes firms to "bypass adequate oversight" and push through products that frequently hallucinate, fail on simple tasks, and cover up significant security and privacy breaches.

• Inadequate compensation for those who helped train the models: There is no hope for fairly compensating those who create original content used to train models under the current regime. AI boosters regularly claim that such compensation would destroy their enterprise completely, not that this stops firms developing AI from stealing as much as they can without compensation. The few institutions that can secure deals (such as media organizations) are doing so for the company itself, not for every single worker involved in creating the original work that's part of these licensing deals. The insistence that the solution must benefit these AI firms first and foremost is not a natural one, and Morozov points out how we could easily empower a public entity to implement "differentiated access rules based on public policy priorities" that fleece AI firms while providing free/low-cost access to everyone else.

• Exclusion of non-corporate players from R&D: Even the one goal articulated by generative AI firms—building our stillborn A.G.I. god—is just a fancy way of saying "serving the interests of its investors." Generative AI's development will be profit, not experimentation, not innovation, and often at the cost of our social relations, political sphere, privacy, cultural works, and environment. Without the inevitable birth of A.G.I., it is harder to pretend the debate is about stalling the creation of an epoch-defining technology and it becomes easier to see this is a distraction from considering alternatives to development solely by market forces. "In fields like education, healthcare, and transportation," Morozov writes, "we've recognized the market model's limitations and its tendency to erode cherished ideals. Shouldn'w we apply the same scrutiny to generative AI.”

• Lack of transparency about true costs: I've used Zitron's writings to repeatedly invoke this, it has come up in my own, and it has come up in Morozov's—a key part of AI's indulgence economy is obscuring the real cost of these products so that it is easier to make the pitch that salvation is cheaper and easier than the much harder task of pursuing alternatives. We do not know the sustainability of what these firms are offering is, only that they insist it will pay for itself one day.

• Increased technological dependency of the Global South on the Global North: Silicon Valley, our military, and our federal government have all said in one way or another that pursuing AI development is also about pursuing additional leverage over other countries—specifically China. Slowing AI development, breaking up the firms driving it, applying basic regulatory or democratic oversight, these are anathema as they would reduce American primacy in this field instead of deepending foreign dependence on our tech offerings. Is there any interest in a multi-polar global tech regime, with institutions interested in bolstering tech development in the Global South at the expense of the Global North? Likely not. And so the march continues on.

• The conservative bias inherent in these systems, which favor stability and predictability over novelty and variety: Corporate-driven generative AI is primarily concerned with profit-seeking, utilitarian efficiency, surveillance, social control, and the like. What would what Morozov calls "ecologically sophisticated generative intelligence" look like? Morozov imagines it being advanced by "institutions that genuinely value more than just predictability and efficiency, however monetizable those qualities may be." Will this require new institutions? Will we go back to the well and bolster those embedded in the welfare state, like libraries and museums (educational and cultural ones)?

There is no hope of any of this to turn over within the “neoliberal straitjacket” that Silicon Valley has foisted upon us in its bid to consolidate its private control of computational infrastructure and the vast ecosystem needed to swell it. What is lost and what is gained when we prioritize a vision that aligns with centralizing private control of our technology?

We are left with voluntary AI ethics frameworks that lack any teeth, performative decrees that purport to govern and regulate the industry but are better understood as medieval papal bulls insisting on immaterial authority on tenuous grounds! We are left with AI ethicists who assume the inevitability of a far-flung possibility (AGI) and spill miles of ink crying out for guardrails and alignment safety protocols to ensure our science-fiction scenario arrives amicably, but who have nothing to say about a genocide going on right now where Israel is using AI tools to explicitly kill as many civilians as possible as it tries to ethnically cleanse Palestinians in Gaza. We are left with AI ethics boards that are quickly dissolved, ethics researchers that are censored and fired, and alignment that purports to avoid runaway AGI, but advances the proliferation of present AI to augment discrimination, surveillance, immiseration, exploitation, and extermination.

Is all of this worth it?

At the end of the week, the roundup will drop and focus on some of our favorite subjects: abundance & degrowth & techno-optimism & Luddism, the ongoing AI bubble, Trump’s Butlerian Jihad on the global economy, and China’s technological ascendance. I'll also be throwing in some book recommendations, a playlist, and some movie recommendations from my Diplomacy movie club. Look for that in your inbox at the end of the week.

Some housekeeping

Some lovely folks at NPR/KCRW brought me on to Question Everything and got me drunk while we talked about billionaires in February (listen here). I met Douglas Rushkoff there and went on his Team Human show to talk more about Silicon Valley, episode forthcoming. I also went on Trashfuture (listen here) and System Crash (listen here) to talk about my Silicon Valley Consensus essay. I will also be going on Ed Zitron’s Better Offline this week to talk about the ongoing AI bubble (and hawk the Silicon Valley Consensus a bit more). The second half of my SVC essay will be coming out this week too and then I’ll push out the unified beast at Security in Context sometime later—stay tuned!

Over at This Machine Kills (the podcast I co-host with Jathan Sadowski), we just released our 400th episode: we had an amazing conversation with author Malcolm Harris about his upcoming book WHAT’S LEFT, the three strategies it offers to save the world from ecological catastrophe, and how those pathways fit together into a larger vision.

On April 16, The Drift is convening a panel where I’ll talk alongside alongside Willy Staley, Sam Adler-Bell, and Delia Cai about Twitter—moderated by Elena Saavedra Buckley and hosted by NYU's Arthur L. Carter Journalism Institute. Registration is closed, but stay tuned for livestream options—tune in if you’re around! I’ll also be on a panel at NYU on May 2nd to talk with Quinn Slobodian about his upcoming book HAYEK’S BASTARDS. Our TMK episode on his book will be coming out in two weeks. And it’s worth mentioning that our conversation with Slobodian is what inspired this essay that I’ve spun out of the coming roundup.

If you like my work and want to support me, consider becoming a paid subscriber for $7 a month (the price of a dozen eggs if you’re lucky) or $70 a year (the price of a few hardcover books you won’t read)! If you are a bit more well-off, take inspiration from Engels and become a Tech Bubble financier. Share this newsletter with someone who you think might like it (or hate it)—and feel free to leave comments here or shoot me an email with further questions!

How was it that 23andme—a personal genomics company with access to the intimate private data of millions of people (their DNA)—went bankrupt? As Adam Rutherford points out in The Guardian, it ultimately boiled down to a data-driven junk science business model:

It didn’t invent direct-to-consumer genomics, but it made big data big business. The genius of its business model was not simply to get you to volunteer this personal data to a private company, but to persuade you to actually pay to give it to them. It then commercialised your DNA by selling it on to pharmaceutical companies, which would use it to develop drugs, ultimately for profit. It was the type of racket that a mob boss might look on and say: “And this is legal?” There was always an opportunity to opt out, but most people did not, because who reads the small print? And what did you get in exchange? A scientific trinket.

That scientific trinket was supposed to shed light on where someone came from, as well as insights into hereditary health, but those offerings rarely held up to scrutiny:

There is no method for identifying the geographical origin of your ancestors using genetics. Your ancestors about 50 generations back are from all over the world, and besides, biology does not bestow membership to a tribe or clan or people or country. What 23andMe was actually doing was comparing your DNA to that of other paying customers, and matching up where they live today, and inferring that you have ancestors in that location. It kind of works, but is mostly meaningless. When the data lets you believe that you are 37% German, or 18% Spanish, or whatever, it might feel fun, but of course there is no way of being 37% German. White supremacists loved this type of service too because – locked into their scientifically ignorant ideology – they believed it would reveal some sort of racial purity. Even when testing uncovered previously unknown ancestry from people they deem inferior, they would often attribute the results to a Jewish conspiracy.

As for the health information it provided, the results are also of profoundly limited use, because the tests are not designed to diagnose medical conditions, and the genetic variants analysed as part of the service are derived from population-level statistics, which are not particularly informative to individuals. I discovered that I have a genetic variant that at a population level is associated with a slightly higher risk of developing Alzheimer’s disease. Knowing this neither bothers me nor has prompted a change in my behaviour. It does not mean that I will get Alzheimer’s, and if you don’t have that same variant it doesn’t mean you won’t.

DNA is not fate. 23andMe was trading on ignorance of how the genome actually works, and perpetuating a deterministic view of genetics that is outmoded and wrong.

This is a great description of why 23andme’s business model was data-driven junk science, but doesn’t exactly tell us why it failed financially. Silicon Valley financiers, after all, will be the first to tell you that selling bullshit and making a profit are not mutually exclusive, but in fact hand in hand. Or you can simply look with your eyes at what Silicon Valley actually does with its Smuagian hoards?

Perhaps 23andMe going public via SPAC merger in 2021 should've been a giant red warning sign—as I wrote that year, the SPAC frenzy kicked off because it was a lucrative way for firms with poor financials to enter public markets and for early financiers to con suckers into investing, using the new meat as a pathway for an exit.

In February 2021, its merger with Richard Branson's SPAC gave the company a $3.5 billion value that peaked at a $6 billion market capitalization. It's now hovering at $35 million, up from $19 million when bankruptcy was announced. This still doesn’t capture how bad things have shaken out, but here’s one financial commentator on the day of 23andMe’s bankruptcy:

Branson’s SPAC went public via IPO at $10 a share in late 2020. It then acquired 23andMe at the company’s peak in revenues. The 1-for-20 reverse stock split last October turned each 20 shares [ME] into one share, and thereby turned the SPAC’s IPO price of $10 into $200. And today’s price of $0.73 would be $0.037 on a pre-reverse-split basis.

Early on, it became clear one of the major obstacles the company would have to overcome is that a customer only needs to order their genetic test once. Most of 23andMe's ventures were desperate attempts to eschew that limit.

When growth of the direct-to-consumer genetic tests stalled, 23andMe tried to generate a new revenue stream with its massive DNA hoard. The plan? Approach pharmaceutical giants interested in accelerating the development of new drugs and genetic therapies with access to 23andMe's DNA hoard. In 2018, 23andMe inked a four-year deal with GlaxoSmithKline that included a $300 million upfront investment and gave the drugmaker exclusive rights to use 23andMe's database for drug development. In 2022, GSK paid $50 million to extend the partnership for a fifth year, but 23andMe was unable to find another pharma partner afterwards—killing the hopes of recurring revenue through data licensing and IP rents, as well as the delusions about creating their own drug R&D program.

Back in 2020, 23andMe launched a premium subscription product aimed at giving customers "more ways to engage in their personal genetics journey" with "impactful new genetic health reports, and enhanced health and ancestry features." This, too, failed to materialize sustainable recurring revenue. So with a fundamentally flawed business model and persistent failure to generate sustainable revenue, the writing was on the wall.

And back in 2021, 23andMe paid $400 million for Lemonaid, a telehealth and pharmacy startup, as part of a failed pivot into primary care services bolstered by "personalized" genetic insights. As Stat+ News put it at the time:

“What are the chances 23andMe is going to be one of the last-standing, dominant players in telemedicine?” asked Erik Gordon, a health finance professor at the University of Michigan. “I would say the chances are very low.”

For one thing, Gordon said, telemedicine carries significant ongoing capital costs that 23andMe is not as well-positioned to manage as many of the bigger players, such as UnitedHealth Group, or even new, well-funded entrants like Amazon. A more fundamental challenge is linking the value of genetic testing to a primary care business where most of the care currently focuses on acute needs such as a bad cold or emergent illness.

“The idea that you can inform primary care with genetic testing — it’s pretty hard to connect those dots,” Gordon said. “Lemonaid’s telehealth is for things like the flu, not for stuff where someone’s genome is going to be really important.”

Even in this 2021 article, it's clear this is a desperate play made to "provide a lifeline for 23andMe's bottom line" and one that was likely to fail. 23andMe had at that point already laid of 14 percent of its staff in January 2020 after revenues dropped by more than $130 million since the previous year and the company projected it wouldn't make more until 2023. That milestone was never hit (the company never saw a profit), in September 2024 all seven independent directors on the company’s board resigned, and in November 2024 it laid off 40 percent of its workforce.

Revenue declined, losses grew, user growth slowed, partnerships never materialized, acquisition and R&D costs mounted—the writing was on the wall for anyone who wanted to read it. Take these charts from CNN’s Rosa de Acosta:

And this is before we even get to the constant breaches and thorny questions about privacy as hackers and corporations gained access to millions of consumers’ genetic data. Sidestepping all this, it also doesn't help that this was clearly obvious even before the SPAC merger back in 2021. Take this Financial Times op-ed from Hug Young Reinhoff Jr, who founded bio-tech startup DNA Sciences in 1998 with venture capital backing and board members like Jim Watson (one of the scientists who discovered DNA's double helix structure):

On August 1, 2000 we launched the web-based Gene Trust, a platform to recruit patients online to participate in studies to identify the genetics of common disease. Thousands participated. At our peak we were spending $3mn a month.

Six years later, long after the demise of DNA Sciences, 23andMe essentially copied our efforts but focused almost entirely on brand creation making DNA collection “celebrity cool”. Indeed, they created significant brand awareness. What they apparently have not experienced was our own early “epiphany” that doing genetics was not a business model. This revelation was true for the other significant companies sequencing and genotyping — Incyte Corp, deCode Genetics, Celera Genomics, to name a few. Though deCode has been fantastically productive doing first-rate genetic studies, it nevertheless had to find a sponsor in Amgen to sustain its efforts. Indeed, Big Pharma, most significantly GSK, has “dabbled” in genetic studies to the tune of billions with very little to show for it over the last 25 years.

The brand equity of 23andMe has been destroyed and leaves a stain on such efforts, given the low quality of the personal medical data and the constant threat of breaches. DNA sequence is now a commodity and large-scale efforts, sponsored by the government and foundations such as UK Biobank, render commercial efforts inadequate and irrelevant.

Genetics has an unbelievable allure, perhaps even more to non-geneticists, but most DNA data is gangue, not ore. It is the individual prospectors mining big DNA databases who occasionally hit pay dirt.

So if 23andMe, other bio-tech startups, and pharmaceutical giants can’t hit pay dirt when mining big DNA databases…who can? A few days after 23andMe declared bankruptcy, it revealed a U.S. judge ruled the company could sell customer medical and ancestry data as part of the proceedings—prompting multiple state attorney generals to remind people that they can and should delete their data if they foolishly used the service over the years.

What could that data be used for? David Choffnes, a computer science professor at Northeastern University and executive director of its Cybersecurity and Privacy Institute, told The Associated Press:

“There’s still other things that they are allowed to do with that data, including, as they mentioned, provide cross context, behavioral or targeted advertising,” he said. “So, you know, in a sense, even if they aren’t sending your personal data to an advertiser, there’s a long line of research that identifies how third parties can re-identify you from de-identified data by looking for patterns in it. And so if they’re targeting you with advertisements, for example, based on some information that they have about your genetic data, there’s probably a way that other parties could piece together other information they have access to.”

As Jason Koebler points out in 404Media, 23andMe (or whoever buys the data) could go even further:

Other genetic sequencing companies have shared customer information with police and governments, pharmaceutical companies, and health insurers. GED Match, a non-profit that once claimed it would protect customers’ genetic data, was sold to a for-profit company called Verogen, which works with the FBI and was later sold to a Dutch multinational conglomerate. Police now regularly attempt to identify suspects using information pulled from commercial genetic databases like the one that 23andMe has created.

23andMe’s bankruptcy means that the company will be put up for sale, and there’s no way of knowing who is going to buy it, why they will be interested, and what will become of its millions of customers’ DNA sequences. 23andMe has claimed over the years that it strongly resists law enforcement requests for information and that it takes customer security seriously. But the company has in recent years changed its terms of service, partnered with big pharmaceutical companies, and, of course, was hacked.

And such concerns aren’t that far fetched—Koebler also wrote about OpenSNP, an open source genetic database that has collected about 7,500 genomes over the past 14 years, that recently shut down because of growing concerns about the country’s fascist lurch:

“I’ve been thinking about it since 23andMe was on the verge of bankruptcy and been really considering it since the U.S. election. It definitely is really bad over there [in the United States],” Greshake Tzovaras told 404 Media. “I am quite relieved to have made the decision and come to a conclusion. It’s been weighing on my mind for a long time.”

Greshake Tzovaras said that he is proud of the OpenSNP project, but that, in a world where scientific data is being censored and deleted and where the Trump administration has focused on criminalizing immigrants and trans people, he now believes that the most responsible thing to do is to delete the data and shut down the project.

Who else might be interested in your genetic data? Would you be surprised to learn that professional phrenologists like Charles Murray, who suggested that one of his billionaire friends should buy it for him. Joke or not, there are 1488 reasons why Murray might want access to the genetic data of millions of people.

Subscribe now

As Quinn Slobodian and Stuart Schrader write in 2018 for The Baffler, Murray has spent the past few decades looking for data to prove that "equality was elusive because it was impossible" given racial groups "wee unequal in their endowments and would remain so indefinitely." Murray's diligent massaging of data in The Bell Curve didn't lead to us abandoning the "egalitarian premise" quite like he hoped, but perhaps troves of genetic data in the right hands might do the trick. Perhaps we might be able to use science to show the cruel but sober rationality of his pleas for "social apartheid" or a "high-tech and more lavish version of the Indian reservation for some substantial minority of the nation's population, while the rest of America tries to go about its business."

Slobodian dives a bit deeper into the influence of Murray’s academic racism in a later paper titled “The Unequal Mind” and the role genetic determinism—especially when advanced by phrenologists interested in gutting social programs—played in slashing and burning away at the welfare state. Consider this:

An irony of the revival of racial science to oppose the welfare state is that the welfare state was designed, in part, to address fears of the degeneration of the race. The modern response to fears of degeneration or the deteriorating genetic quality of a population was to increase state intervention, including through public programs of nutrition, maternal training, public health, child benefits, family planning, and access to birth control. In extreme forms, this entailed forms of negative eugenics, including sterilizations of criminals and the mentally ill and, eventually, the extermination policies of Nazi Germany.

The question for phrenologists, then, is how do you open the public’s third eye so that they realize the eugenic state, not the welfare state, is the answer to all our problems? How do you naturalize "a racial, ethnic, and gendered hierarchy of ability" as part of a larger campaign to purge society of egalitarianism?

This shit has always been junk science. Genetic data’s real value is not in discovering new drugs or gene therapies, nor in revolutionizing primary care, nor personalizing medicine, nor reducing the cost of health. The real value of genetic data is not as a commodity or a business model of its own. The real value is as an input to existing models (e.g. advertising) or reactionary projects (e.g. bolstering police or immigration authorities in a fascist polity) or undermining public empathy for others (e.g. naturalizing hierarchies that justify coercion or domination or exploitation), all as part of the desperate attempt by the least among us (skull measuring freaks, apartheid lovers, libertarians ghouls, free market jihadis, new fusionists, etc.) are undertaking to save capitalism by purging it of recent half-hearted reforms in response to social movements over the years.

In the aforementioned essay, as well as his forthcoming book “Hayek’s Bastards,” Slobodian hammers this point home well but I’ll close this out with one more quote:

All new fusionists see the welfare state as the motor of dysgenesis. The reason for this is threefold. First, it operates on the principle of equality of outcomes as well as of starting points, an unrealistic goal in a world of deeply divergent human capabilities. Second, it institutionalizes incentives that subsidize the multiplication of the mentally inferior; those unable or unwilling to find gainful employment are (supposedly) rewarded for reproduction through child benefits. Third, the provisions of the social state act as a magnet for similarly low-intelligence immigrants from populations that are already below average white intelligence, namely, all nonwhite and non-Asian populations. The attraction of low-intelligence immigrants is a special problem as all three thinkers share the conviction that economic structural change is making intelligence more important than ever.

So, genetic data has some real value. In and of itself, genetic data doesn’t seem capable of consistently yielding gold for the firms dedicated to mining it. But it offers the ability to bolster existing regimes of extraction (advertising, insurance, behavior modification, etc.) and the promise of the great white hope that someday, somehow, we will re-legitimize the ancien régime. God willing, these efforts will see the same fate as that old system.

At the end of the day, OpenAI’s latest viral marketing stunt—a ChatGPT 4o update that allows users to generate images, including ones in a style that barely resemble’s that of animation company Studio Ghibli—is ultimately a distraction. Let me explain.

Yes, this stuff is an “insult to art itself” as Brian Merchant points out in his invocation of Studio Ghibli co-founder Hayao Miyazaki’s earlier comments on early AI art tools (“I strongly feel that this is an insult to life itself”). Firms hawking generative AI powered are dead-set on eviscerating legal barriers obstructing their god-given right to "ingest copyrighted works into their training data" and will vomit up any answer that resembles something reasonable—much like their products. If you do not let us do this, China will surpass us is an increasingly popular and successful one.

As a result, artists are left shit out of luck. As Merchant writes:

OpenAI and the other AI giants are indeed eating away at the livelihoods and dignity of working artists, and this devouring, appropriating, and automation of the production of art, of culture, at a scale truly never seen before, should not be underestimated as a menace—and it is being experienced as such by working artists, right now.

On top of the exploitation of artists and their livelihood, we get to consider what this does to our role in cultural production. How does being flooded with slop that does not but does resemble significant cultural products—because it is superficially trained off the form of various works of art, as well as those inspired by or derived from it, as well as commentary about it—affect our own internal response to those same pieces of art, even when presented in their original slop-free context?

Erik Hoel chimes in here, warning we are entering "the semantic apocalypse" or following down the path of "semantic satiation" whereby a word loses meaning if it is said over and over and over and over and over again:

The semantic apocalypse heralded by AI is a kind of semantic satiation at a cultural level. For imitation, which is what these models ultimately do best, is a form of repetition. Repetition at a mass scale. Ghibli. Ghibli. Ghibli. Repetition close enough in concept space. Ghibli. Ghibli. Doesn’t have to be a perfect copy to trigger the effect. Ghebli. Ghebli. Ghebli. Ghibli. Ghebli. Ghibli. And so art—all of it, I mean, the entire human artistic endeavor—becomes a thing satiated, stripped of meaning, pure syntax.

This is what I fear most about AI, at least in the immediate future. Not some superintelligence that eats the world (it can’t even beat Pokémon yet, a game many of us conquered at ten). Rather, a less noticeable apocalypse. Culture following the same collapse as community on the back of a whirring compute surplus of imitative power provided by Silicon Valley. An oversupply that satiates us at a cultural level, until we become divorced from the semantic meaning and see only the cheap bones of its structure. Once exposed, it’s a thing you have no relation to, really. Just pixels. Just syllables. In some order, yes. But who cares?

So on the cultural front it is pretty bleak. AI firms are creating tools that will be used to produce good enough AI slop—stuff that’s eerily reminiscent of cultural works we’re inspired by or fond of. Cultural works and products that are stolen to fill datasets to train LLMs will be slowly sapped of meaning and value, thanks to how easy the slop is to produce and how well-capitalized its producers are.

I agree with Max Read’s take that it's "hard to be exercised over something so obviously ephemeral" but, as he points out, this is symptomatic of a larger debate: the emotional weight and reputation of Miyazaki's and Studio Ghibli's craft that excites AI shills generating these images who insist it’s democratizing creativity, at the same time frustrating critics who treasure what that art represents but also are (rightfully) skeptical of promises by the tech sector that a new technology will democratize something.

To take one small example: fintech & crypto & web3 have all been hailed as the future and have made similar promises about democratizing hard-to-enter spaces to the benefit of all. They have all amounted to little more than schemes of varying complexity to steal from investors, defraud the public, undermine consumer protections that declare a certain avenue of profit-seeking as illegal (and immoral), and use this wealth transfer to legitimize parasitic enterprises that should not exist.

Going back to Read, he builds on Merchant's essay last week with a 1935 essay from Walter Benjamin "The Work of Art in the Age of Mechanical Reproduction" that reflects on how implementing technical tools to accelerate the production of art robs us of something of its essence:

A.I., one could argue, enacts on particular artistic styles what photography and lithography enacted on particular works of painting or sculpture, rendering those styles endlessly reproducible, de-aura-fied, no longer subject to the “criterion of authenticity” from which they previously derived value.

But (without necessarily wanting to endorse this argument re: A.I. and style) we might also note that Benjamin’s main point was not so much to endlessly bemoan the withering of aura but to explore the effects of mechanical reproduction on the “social function” of art.⁴ The toothpaste is kind of out of the tube on this one, in Benjamin’s time as in ours, and there are reasons to welcome the freeing of art from its “parasitic subservience to ritual.” When art can no longer automatically obtain significance from ritual--authenticity, tradition, or even ownership--its value must be found elsewhere. “Instead of being founded on ritual,” he writes, the social function of art “is based on a different practice: politics.” What matters with art isn’t its past (where and who it comes from), but its future: what it does.

And as Read remarks, this leads us to the question: what are the politics of AI slop/art? Read quotes Benjamin’s essay and drives home the point that it is fascism which "sees its salvation in granting expression to the masses—but on no account granting them rights" and means "self-alienation has reached the point where it can experience its own annihilation as a supreme aesthetic pleasure." This is all true and well, but the use of generative AI for images or writing is ultimately a distraction.

Here's Rob Horning writing about the last viral marketing stunt from OpenAI, a piece of "creative writing" from an unreleased product:

Hari Kunzru is right to point out that “the ‘can machines do creative writing’ thing is mostly a distraction from the use of the machines to go through text and images to cancel grants and put people on deportation lists.” So the best way to understand OpenAI’s recent claims to have trained a new model that, according to CEO Sam Altman, is “good at creative writing” and “gets the vibe” of “metafiction” is that the company is running interference for the authoritarians using similar technology to automate surveillance, circumvent human scruples, and do away with due process.

What are the use cases for artificial intelligence that seem to draw up the most excitement (and potential profits)?

In December, The Financial Times reported Palantir and Anduril were in talks with a dozen other firms (including SpaceX, OpenAI, "autonomous-ship builder" Saronic, and AI data labeler Scale AI) to create a consortium that would jointly bit for Pentagon contracts.

Anduril is an arms dealer (sorry, I mean an “AI weapons manufacturer”) and Palantir is surveillance firm (sorry, I mean an “AI-driven analytics firm”). These two tech companies that have made it clear they intend to profit on death and misery by selling tools that power deportation, remote assassinations, and the privatization (sorry, I mean “digitization and modernization”) of public services and government agencies. Anduril and Palantir announced a partnership to use Pentagon data for AI training, and OpenAI completed its slow but steady military pivot with the announcement of an Anduril partnership that would build military AI products—though OpenAI’s hire of Palantir’s Chief Information Security Officer might’ve signaled the pivot had been completed much earlier.

The State Department is claiming to use artificial intelligence to revoke visas of foreign students who’ve protested Israel’s ongoing genocide in Gaza. The United States has been and is using artificial intelligence to try and perfect reliable tools of repression that it has returned to over the years in bids to crush dissents and terrorize minorities. Over the years, Big Tech firms have enthusiastically bolstered Israeli apartheid and genocide of Palestinians with tools powered by artificial intelligence. Israel created a “mass assassination factory” featuring multiple systems powered by artificial intelligence (“The Gospel,” “Alchemist,” and “Death of Wisdom,” as well as “Lavender” and “Where’s Daddy” that were used to target civilian infrastructure and kill civilians as part of a plan to cause as much death and misery as possible.

Gareth Watkins’s brilliant essay (“AI: The New Fascist Aesthetics”) spends some time with the art to flesh out some of the more nebulous connections between firms enthusiastically stealing art to generate slop and fascists deploying it to deport and assassinate civilians, but the focus is the primary things that create and sustain this technology alongside the main ways in which it is deployed:

AI is a cruel technology. It replaces workers, devours millions of gallons of water, vomits CO2 into the atmosphere, propagandises exclusively for the worst ideologies, and fills the world with more ugliness and stupidity. Cruelty is the central tenet of right wing ideology. It is at the heart of everything they do. They are now quite willing to lose money or their lives in order to make the world a crueller place, and AI is a part of this – a mad rush to make a machine god that will liberate capital from labour for good. (This is no exaggeration: there is a lineage from OpenAI’s senior management back to the Lesswrong blog, originator of the concept of Roko’s Basilisk.) Moreso even than cryptocurrency, AI is entirely nihilistic, with zero redeeming qualities. It is a blight upon the world, and it will take decades to clear up the mountains of slop it has generated in the past two or three years.

AI Art should be thought of as a trojan horse. There is much more interest and excitement in using artificial intelligence to predict human behavior, surveil groups of people, using behavioral insights to improve discrimination, synthesize and securitize new assets, innovate new forms of dispossession and extraction, terrorize migrants and dissidents, regiment work and disempower workers, and a host of other noxious deleterious social ends. Another place to turn to when thinking about this is Dan McQuillan's book Resisting AI: An Anti-fascist Approach to Artificial Intelligence:

The struggle against the fascization of AI precedes AI itself. It’s not that AI first comes into existence and we then have to tackle its dodgy politics from scratch, but rather that AI is already part of the system’s ongoing violent response to the autonomous activity of ordinary people. Instead of having to invent a plethora of new remedial measures, we can build on the long history of community solidarity generated by people’s resistance to exclusion and enclosure.

The very generalizability of AI and the way it comes to bear on different communities and constituencies creates the potential for this resistance to cut across race, gender, sexuality, disability and other forms of demographic division. If the whole of society becomes subsumed by algorithmically ordered relations and enrolled in machinic optimization, then society as a whole also becomes a site for contesting the imposition of those power relations. AI’s generalizability and its intensification of social crisis creates a position from which to question the totality of social relations.

So does OpenAI’s marketing stunt matter? That’s probably the wrong way to think about it. AI firms are interested in developing tools and marketing strategies that revolve around the allure of AGI—around a stillborn god that will transform large swaths of society into excessively profitable enterprises and incredibly efficient operations. Think of it as a desperate attempt to defend capitalism, to preserve the status quo (capitalism) while purging recent reforms that purportedly undermine it (democracy, liberalism, feminism, environmentalism, etc.). Sam Altman, OpenAI’s co-founder, has repeatedly called for “a new social contract,” though most recently has insisted the “AI revolution” will force the issue on account of “how powerful we expect [AGI] to be.” It doesn’t take much to imagine that the new social contract will be a nightmarish exterminist future where AI powers surveillance, discipline, control, and extraction, instead of “value creation” for the whole of humanity.

The subsuming of art springs out of the defense of capitalism—more and more will have to be scavenged and cannibalized to sustain the status quo and somehow, someday, realize this supposedly much more profitable horizon. The ascendance of fascism comes with the purge—the attempt to rollback institutions and victories seen as shackles on the ability of capitalism to deliver prosperity (and limiters on the inordinate power and privilege for an unimaginably pampered and cloistered elite).

Both are part and parcel to what’s going on, but one project is objectively more dangerous (and ambitious) than the other. In that way, then, all of this is a distraction. Pay no mind to the unmarked vans and plainclothes officers, the censorship and disappearances, the mass deportations or drone assassinations, the civilian assassinations or ongoing genocides. We’re just smol AI firm democratizing art!

Hello everyone! Today we are going to talk about my favorite subject: the impact of Uber and its bastard spawn. Last time we did this, we talked about the (im)moral economy of on-demand labor platforms and their transformation of care work via A Roosevelt Institute report by Katie J. Wells and Funda Ustek Spilda. Today, we’ll look at another report from Wells, Spilda, Veena Dubal and Mark Graham—this time at Fairwork US, investigating the industries corrupted by algorithmic management tools (and the firms deploying them).

As always, if you enjoy my writing then consider a free subscription. If you want to support me further so I can do this writing more regularly, consider a paid subscription. For $7 a month (a pint at the pub) or $70 a year (a new video game that you'll lose interest in after 10-15 hours), you can help me keep writing.

Subscribe now

Last time, I talked at length about my core issues with Uber and the firms taking after its business model before opening up the conversation to talk about its transmogrification of care work. I’ll summarize some key points here before we move on, but revisit my previous essay to read more:

• The so-called “gig economy” is an altar to greed and misery. Mass immiseration, starvation wages, mental health crises, horrendous working conditions, the offloading of every conceivable cost onto workers and the public, the narrowing of political imagination, racial discrimination for workers and consumers, the acceleration of surveillance pricing, these and many more social ills are key consequences of the on-demand business model.

• On-demand labor did not proliferate because it was profitable or innovative. This business model metastasized across the economy because of a stringent commitment to worker misclassification, algorithmic discrimination, anti-competitive capital-intensive strategies, impressive public relations, robust political lobbying, and shoddy journalism.

• The core ambition of the on-demand ride-hail model is to make previously illegal profits legal again. It's about subverting urban governance, breaking New Deal labor regimes that prohibit certain types of exploitation, subjecting workers and workplaces to more authoritarian forms of control and discipline, and reshaping drivers, markets, consumers, firms, and governments into forms more amenable to privatized profits and socialized losses.

• On-demand nursing promises to be even worse for everyone involved. Gig nursing platforms in the health-care industry have introduced new and considerable risks: more work for less pay; unpredictable scheduling and compensation; discontinuous care that puts patients at risk; an algorithmic off-hands approach that puts workers at risk; and a general decline in the standards one can expect in care work as a practitioner or recipient.

The first entry in Uber’s Bastards talked about on-demand nursing, but there are a plethora of other Uber clones across our economy that are worth looking at, especially as they encroach into more of our daily life.

I’ll be going through key parts of Fairwork’s report “When AI Eats the Manager” in hopes of drilling into you—like with the Roosevelt Institute report—how bad things are in this sector for nearly everyone.

A key part of the on-demand business model is the deployment of algorithmic overseers. By taking in as much information as possible through pervasive surveillance and give as little information as possible, the hope is that you can create information asymmetries that give you the power to set fares and compensation at optimal levels (maximal and minimal, respectively). Veena Dubal, one of the authors of this Fairwork report, has called this algorithmic wage discrimination (also cited above). In an LPE Project blog post introducing the concept, Dubal writes:

As a labor management practice, algorithmic wage discrimination allows firms to personalize and differentiate wages for workers in ways unknown to them, paying them to behave in ways that the firm desires, perhaps for as little as the system determines that they may be willing to accept. Given the information asymmetry between workers and the firm, companies can calculate the exact wage rates necessary to incentivize desired behaviors, while workers can only guess as to why they make what they do.

In addition to being rife with mistakes that are difficult or impossible for workers to ascertain and correct, algorithmic wage discrimination creates a labor market in which people who are doing the same work, with the same skill, for the same company, at the same time, may receive different hourly pay. Moreover, this personalized wage is determined through an obscure, complex system that makes it nearly impossible for workers to predict or understand their frequently declining compensation.

You cannot understand the extent to which on-demand labor is exploitative, dehumanizing, disruptive, unpredictable, and immiserating until you appreciate the extent to which it is managed by black-box algorithms primed to extract as much as possible—and how big of a boon this has been for firms which were previously unable to achieve profits except through accounting tricks.

Editorial

Algorithmic overseers are not only deployed to extract as much from both ends of a transaction, but to flense as much off cost structures as possible. Scheduling, payment processing, paperwork, notifications, customer service, HR, Q&A, and more are all being eliminated or consolidated or automated under the pretense of algorithmic efficiency. When these systems fail, as they constantly do, you’re left with a chatbot to handle your complaints and queries.

The new algorithmic management technologies have direct implications for operational cost-cutting, efficiency metrics across the supply chains, and measurable KPIs (key performance indicators). This hyper-quantification of work can expedite decision-making, scaling, and meeting the performance indicators set by investors and funders. However, the implications for work that has been re-structured and re-imagined through these technologies are far reaching for workers, consumers, and societies. Automated technologies are displacing traditional management relationships, transparency, and accountability in the workplace. The data that platform workers produce may shape how platform companies set prices, manage workers, create personalised pay structures, and even offset financial liabilities. Indeed, one company named Argyle has amassed the employment records of 40 million platform workers in the US, and sells this data as its primary source of profits. The reality of working with AI is different from the promises associated with it. AI is, to put it descriptively, eating the managers.

To that end, the report looks at 11 on-demand platforms across ride-hail, food delivery, elder care, and healthcare to see what the impact of these algorithmic overseer technologies is shaping out to be. Are they replacing managers? Are they creating new management practices and institutions? Are they allowing these firms to develop new relations with consumers, workers, other companies, or governments? And, most important, what are they doing to our notions of work and fairness.

Fairwork's report looks at 11 platforms (CareRev, Clipboard Health, Doordash, Grubhub, Instacart, Lyft, Papa, ShiftKey, ShiftMed, Uber, and Ubereats) in the United States. They’re assessed on a 10 point scoring system that evaluates working conditions according to five principles:

• Fair Pay: Workers, irrespective of their employment classification, should earn a decent income in their home jurisdiction after taking account of work-related costs. We assess earnings according to the mandated minimum wage in the home jurisdiction, as well as the current living wage.

• Fair Conditions: Platforms should have policies in place to protect workers from foundational risks arising from the processes of work and should take proactive measures to protect and promote the health and safety of workers.

• Fair Contracts: Terms and conditions should be accessible, readable and comprehensible. The party contracting with the worker must be subject to local law and must be identified in the contract. Regardless of the workers’ employment status, the contract should be free of clauses which unreasonably exclude liability on the part of the service user and/or the platform.

• Fair Management: There should be a documented process through which workers can be heard, can appeal decisions affecting them, and be informed of the reasons behind those decisions. There must be a clear channel of communication to workers involving the ability to appeal management decisions or deactivation. The use of algorithms should be transparent and result in equitable outcomes for workers. There should be an identifiable and documented policy that ensures equity in the way workers are managed on a platform (for example, in the hiring, disciplining, or firing of workers).

• Fair Representation: Platforms should provide a documented process through which worker voice can be expressed. Irrespective of their employment classification, workers should have the right to organise in collective bodies, and platforms should be prepared to cooperate and negotiate with them.

9 companies scored 0 points, meaning they failed to adhere to any of these minimum standards for fair work. One scored one point, another scored two points. This is abysmal but unsurprising given the aggression with which these companies pursue profits illegally and immorally.

Key Findings

FAIR PAY

Fairwork found that only one company (ShiftMed) out of the 11 studied paid its workers a minimum wage. Why? ShiftMed is the only firm to classify workers as employees, which protects them under federal labor laws and provides some basic guarantees about working conditions and compensation. Every other company, however, classifies workers as independent contractors who:

are responsible for significant work-related costs and spend parts of their workdays engaged in unpaid activities, such as driving long-distances to get to a shift or waiting for a customer to receive an order.

One thing to understand about substandard wages is that they’re a great anti-competitive strategy and were understood as such back when the Fair Labor Standards Act (which establishes standards for minimum wages, overtime pay, and youth employment) was passed in 1938. As Eamon Coburn writes in Yale Law Review:

These concerns animated early minimum-wage advocates who expressed two theories of substandard wages as unfair competition: substandard wages as competition by subversion of public norms, and substandard wages as taking an implicit subsidy from workers and the public. In the 1930s, political, business, and labor leaders continued to frame labor abuses as a competition problem throughout the consideration and implementation of national wage-and-hour legislation, repeating both of these theories to justify federal action.

The first theory sees substandard wages as a violation of public standards around living wages or industry norms, allowing a firm outcompete others on prices by reducing (labor) costs. Firms like Uber might respond to this by saying—as they have in the past—that adhering to public norms would make their service too expensive and rob the public of its benefits. Little over a century ago, this was understood as such an immoral response that you had radicals like President Theodore Roosevelt saying: "We will not submit to that kind of prosperity any more than we will submit to prosperity obtained by swindling investors or getting unfair advantages over business rivals."

The second theory, Coburn writes, sees substandard wages as a way for firms to "pad their bottom line with money owed to workers and gain the full benefit of their employees' productivity while shouldering only part of the cost." This also manifests as a tax burden shouldered by other firms as workers paid substandard wages then rely on social programs that their employers won’t pay into. Uber and Lyft have long been the gig economy’s vanguard here, using worker misclassification to avoid paying hundreds of millions of dollars in unemployment taxes, workers’ compensation, and paid family and medical leave insurance—undermining the social safety net and pushing up tax rates for other firms.

To make their unit economics work, on-demand labor platforms need to embrace anti-competitive strategies in hopes of attracting enough capital, customers, and political power to reforge relevant markets into forms more able to generate and sustain profits: price-gouging customers and underpaying (and misclassifying) workers are key.

FAIR CONDITIONS

Unsurprisingly, each of these firms all trade in horrific work conditions.

Fairwork was unable to fi nd suffi cient evidence to award a point to any of the platforms in this study. Workers report signifi cant task-specifi c risks and lack of a safety net.

Across 11 of the largest on-demand labour platforms in the US, workers reported physical assaults, verbal abuse and stressful working conditions. Fairwork finds that safety is a major issue for on-demand nursing companies, on-demand elderly care companies, on-demand delivery companies, and on-demand ride-hail companies. In healthcare, significant changes are needed to orient, train, and manage on-demand workers so that they can protect both themselves and their patients.

The gig economy is, out of necessity, built on a business model that necessitates worker misclassification. Again, they are seeking profits and returns that are locked behind rules and regulations aimed at protecting workers. Both cursory glances and in-depth examinations quickly make this feature painfully clear.

In 2022, a national survey of gig workers by the Economic Policy Institute painted a bleak picture:

• 19 percent of gig workers went hungry because they couldn't afford enough to eat (compared to 14 percent of service sector employees)

• 31 percent didn't have earn enough to pay their utility bills (vs. 17 percent of service sector employees)

• 18 percent lived in a household where someone didn't seek medical care in the last month because of the cost (vs. 13 percent of service sector employees)

• 30 percent used SNAP (vs. 15 percent of service sector employees)

In 2021, The Verge and New York Magazine collaborated on an investigative report looking at New York City’s 65,000 delivery workers who were left to fend for themselves by the apps that exploited and misclassified them, as well as a city that seemed uninterested in their safety or livelihoods, forcing them to gerry-rig solutions themselves:

Workers developed the whole system — the bikes, repair networks, shelters, charging stations — because they had to. To the apps, they are independent contractors; to restaurants, they are emissaries of the apps; to customers, they represent the restaurants. In reality, the workers are on their own, often without even the minimum in government support. As contractors and, often, undocumented immigrants, they have few protections and virtually no safety net. The few times city authorities noted the delivery worker’s changing role, it was typically with confused hostility. Until recently, throttle-powered electric bikes like the Arrow were illegal to ride, though not to own. Mayor de Blasio heightened enforcement in 2017, calling the bikes “a real danger” after an Upper West Side investment banker clocked workers with a speed gun and complained to him on The Brian Lehrer Show.

The NYPD set up checkpoints, fining riders $500, seizing their bikes, and posting photos of the busts on Twitter. The police would then return the bikes because, again, they were legal to own. It was a costly and bewildering ritual. For years, bike activists and workers pushed for legalization, though the apps that benefited from them were largely silent. It was only when another group of tech companies — hoping to make scooter-sharing legal — joined the fight that a bill moved forward in Albany. Then the pandemic hit, restaurants were restricted to takeout, and the mayor had to acknowledge that the bikes were an essential part of the city’s delivery infrastructure. He halted enforcement. The bikes were officially legalized three months later.

In 2022, the Markup reported it had independently tracked a total of 361 ride-hail and delivery drivers who'd been victims of attempted and successful carjackings since 2017—28 drivers had been killed as a result. That number was likely much higher though. Uber revealed that year there were at least 24,000 "safety incident reports" from 2017 to 2020 that involved a passenger physically assaulting a driver, but specifics were declared confidental and kept under seal:

Our data isn’t comprehensive, however. Most police departments don’t gather statistics on carjackings specifically against gig workers, and the gig companies have repeatedly declined to provide their own data on carjackings. The Markup compiled its database through police reports and local news articles that cite police reports, and by conducting interviews with drivers, family members, and lawyers representing drivers (or their families) who were victims of carjackings.

As I’ve written over the years, Uber and Lyft have repeatedly seen driver shortages at key moments despite incentives that make the pay livable because the working conditions are intolerable for most drivers!

FAIR CONTRACTS

Contracts do not fare any better among the platforms:

Two of the evaluated platforms – ShiftMed and Papa – have clear and accessible terms and conditions. But the widespread use of liability clauses on the platforms included in this year’s study place nearly all the risk of negligence on workers rather than companies.

Ethical and responsible data protection measures for worker data are needed for the 11 platforms in this study, and more transparency and accountability are needed for workers to understand how their data is collected, processed and stored. Fairwork finds that class action waivers and arbitration clauses are commonly used, and they limit workers’ ability to bring legal claims collectively or have their cases decided by a court of law.

For years, gig companies have pushed for mandatory arbitration because it is incredibly good at stymying class action lawsuits and legal precedents—again, if your business model relies on skirting the law, regulatory arbitrage, and aggressive lobbying, you need to stop angry workers in deplorable conditions from collectively demanding the right to earn a livable wage or be safe in the course of their work or get health insurance or other indications of dignified work.

Uber and Lyft recently said as much when they asked the Supreme Court in 2024 to block state officials from using their enforcement powers to seek money for workers and customers who've signed the companies' mandatory arbitration agreements.

Technically, the companies want the justices to review a 2023 California state court appellate ruling, opens new tab that allowed California’s attorney general and labor commissioner to continue litigating claims that Uber and Lyft owe money to drivers who were misclassified as independent contractors. They contend that the California Court of Appeal – like state appellate courts in five other states – misread a key 2002 Supreme Court decision when it concluded that state officials are not bound by workers’ arbitration agreements.

But make no mistake: The theory espoused by Uber and Lyft would preclude all kinds of litigation by states attorneys general, from consumer protection and unfair competition litigation to antidiscrimination suits. If Uber and Lyft are right, state AGs and other officials simply would not be permitted to bring lawsuits seeking monetary relief for anyone who signed an arbitration agreement.

The companies said as much in their petitions, arguing that in response to Supreme Court decisions allowing companies to impose mandatory arbitration on workers and consumers, states have become increasingly likely to adopt “creative devices,” in the words of Lyft’s lawyers from Munger, Tolles & Olson, to undermine arbitration. And unless the Supreme Court steps in, Uber and Lyft said, state officials will continue to expand such loopholes until the Federal Arbitration Act is effectively nullified.

The cases in question that Uber and Lyft are petitioning trace back to a 2020 San Francisco state court case "seeking (among other relief) unpaid wages and benefits for allegedly misclassified drivers" and argues that because drivers waived their rights when signing these contracts, whether or not they read them closely, the state's officials cannot weigh in. And the success that Uber and Lyft have enjoyed with blocking drivers from costly class action lawsuit over wage theft, misclassification, liability for injuries or death, and so on have inspired others across the gig economy to emulate the forced arbitration strategy. As Jacqueline Vanacore of Washington College of Law's Arbitration Brief lays out in 2024:

The substantial damage awards won by workers who litigated misclassification claims laid the groundwork for employers’ burgeoning preference for mandatory arbitration agreements and class action waivers. In Estrada v. FedEx, a California Appeals Court held that FedEx drivers are employees instead of ICs because FedEx maintained control over drivers by requiring them to pay for business expenses such as uniforms, fuel, insurance, and truck maintenance. The workers were awarded $14 million in damages from 2005-2008 and $12 million in legal fees. Moreover, in 2014, the Ninth Circuit held that FedEx drivers were misclassified as ICs and were owed back wages for missed meals, breaks, and overtime and were entitled to worker’s compensation and unemployment insurance. FedEx settled the class action dispute with over 2,000 drivers for $228 million.

The FedEx cases illustrate the financial risk of misclassification lawsuits to Gig employers, whose workforce mainly consists of ICs. Although arbitration can attempt to address individual employment misclassification, the broader problem of misclassification for an entire workforce will persist. The outcome of private arbitrations is secret, non-precedential, and only applies to a small group or an individual worker. Statistics demonstrate that arbitrators who frequently rule in favor of an employer are more likely to be hired to resolve future employment disputes. Misclassification claims brought to arbitration will likely have a low dollar value, especially when workers are prohibited from filing a class action or must represent themselves in arbitration. An Uber driver who represented herself in arbitration received only $4,000 backpay for reimbursed expenses. The decision resulted in significant financial savings for Uber but inadequate compensation or justice for the worker.

It is a win-win situation for the firm, freeing it of the costs (fiscally and reputationally) of a court battle by disaggregating claimants, preserving misclassification, and insulating itself from laws and regulations it’s running afoul of elsewhere.

FAIR MANAGEMENT

As I’ve talked about above, algorithmic overseers are central to the gig economy business model—in fact, here and elsewhere in the world of algorithmically mediated labor, I’ve argued plantation logic is integral to disciplining and controlling a workforce that is both “independent” but persistently surveilled and highly regimented/structured/guided with insights gleaned from digital tools.

Fairwork’s assessment:

Fairwork was unable to award a score for this principle to any of the assessed platforms. We were unable to find sufficient evidence of a due process for decisions affecting workers.

Improvements are needed for workers to meaningfully appeal low ratings, report issues of non-payment, late-payment, deactivations, other penalties, and disciplinary actions. Although many of the platforms offer public statements in support of equality, diversity and non-discrimination, more evidence is needed to confirm that these policies are put in practice.

The deployment of algorithmic overseers that helps force on-demand workers to accept unfair and unchallengeable firings and dehumanizing work conditions. There are a litany of second-order consequences, but it’s important to keep in mind just how destabilizing algorithmic management is to the worker. Zephyr Teachout echo’s Dubal’s concerns in another article on the ascent of algorithmically targeted wages:

Algorithmic management also transforms the nature of supervision, and the power and sentimental relationships between supervisors and mid-level decisionmakers and locates the decision-making in a combination of the upper-level management and the results of hyperindividualized tools that rely on spying and psychological updating. Not only do the direct supervisors have little power, but the workers are then employed in a state of rational paranoia, where they know that they are being punished and rewarded and experimented upon, but they have no way of knowing whether any given decision they are faced with is a result of a game, an experiment, a punishment, or reward, or changing circumstances on the ground and changing needs at the job.

Such insecurity is necessary for highly personalized wages to work, as is an extensive surveillance apparatus that arbitrarily savages workers. It also encroaches on privacy concerns, in deep and under-appreciated ways that will come for all of us eventually. As Teachout goes on to write:

The right to privacy in one’s thoughts and actions is fundamental, a basic right that implements the democratic commitment to human dignity. Privacy implicates the right of people to have control over the boundaries of what is known about them—both what they want to protect from view, and what they want to project in the public arena. While employers have always supervised and monitored and—since the 1990s—recorded an enormous amount of data about employees, the opportunity to collect data that allows for targeted wages increases the incentive to monitor substantially, and the opportunity to collect data that allows for experimentation and extreme Taylorism does the same.

Therefore, the privacy concerns which have long attended the workplace—and never been adequately addressed—move from second order concerns to first order concerns as the scope of monitoring increases, and the arenas which are monitored move from the superficial to the intimate. As Michael Selmi has argued, privacy constitutes the person, and while “it is one thing to give an employer broad dominion over its own workplace, but it is quite another to extend that dominion wherever the employee goes.”

As Teachout, Dubal, and Sarah Jaffe also remind us—these tools of surveillance, control, domination, privacy violation, and algorithmic discrimination find earliest adoption and persistent experimentation in industries dominated by Black and brown workers—such as care work, ride-hail, and food delivery. Plantation logic through and through: we can return to older levels of extraction in modern non-white workforces by making the (digital) factory resemble the plantation with discipline, control, the treat of starvation, piece-pay, and a persistent commitment to dehumanization.

FAIR REPRESENTATION

Fairwork’s assessment:

Collective organisation and representation is a fundamental right for workers and employees. Fairwork was unable to evidence that the 11 platforms in this study assure freedom of association or expression of worker voice in line with the Fairwork Fair Representation principle thresholds.

As shown in the report, various models of contracting labour are used by digital labour platforms; these can either hinder or enable workers to act on their right to collectively organise. We were unable to evidence that the 11 platforms in this study assure freedom of association or the expression of worker voice in line with the Fairwork Fair Representation principle thresholds.

All of the elements I’ve discussed conspire to disempower workers at every possible turn. It is pretty hard to have solidarity, let alone representation, if your employer has deployed an impressive technology stack aimed at preserving your precarity with personalized starvation wages, pervasive surveillance, a lack of privacy, or arbitrary firings and punishments. Worker solidarity groups do emerge despite all of this but, as I write above, many of those groups (necessarily so) spend their energy ameliorating how unsafe and how extractive this line of work is. That is: helping bring down operating costs that the company dumps on them (bikes, scooters, cars, fuel, weather gear, maintenance of equipment, etc.), retrieving stolen equipment when cops don’t help (which is most of the time), doing patrols to deter attacks on delivery workers, and so on. Successful union drives are far and few between despite constant valiant efforts made to organize workers, take over company facilities, partner with legal advocates to push back against gig economy lobbyists, and work with lawmakers to push through pro-worker legislation.

Silicon Valley’s New Partnerships

A key insight offered by this report is that these digital labor platforms are partnering with major institutions to embed their services. There are three strategies at play here:

Step 1: Ignoring the law

A fundamental element of the on-demand business model. Ignoring the law doesn't only allow firms to roll back the clock on labor laws in hopes of realizing profits that are otherwise illegal, but by insisting the law itself is outdated—it wasn’t made with The Digital Economy in mind. As Fairwork’s researchers write:

In this phase, the rule of law is secondary to the holy aura of innovation. To advance this worldview and to subvert governance structures, labour platforms exploit real problems caused by years of austerity, from decaying public transit infrastructure and neighborhood disinvestments to struggling social services and wage stagnation. Platform firms argue that they, rather than a recalcitrant government or any of its under-resourced programmes, should be at the center of solutions. When Obama’s 2008 campaign manager and White House senior advisor, David Plouffe, joined Uber as its new senior vice president for policy and strategy, Plouffe said Uber would help workers put money “back in their pocket” and receive the “pay raise that they’ve been denied for years.” At the same time, he offered assurances that the company was self-reliant. “We are not asking for special tax breaks like those who want to build a factory or headquarters in a city often do.” Legal precedents and government institutions, Plouffe suggested, were an impediment to progress, not evidence of it. Other platforms have mimicked these Silicon Valley ideas about how change happens – with powerful outsiders – and followed suit with their own arguments against regulation as a common good.

Step 2: Securing permanent exemptions by rewriting local and state laws

As Fairwork writes:

One of Uber’s greatest innovations is its argument that technologically-mediated business models are so unique that they merit brand new business categories. What’s so important about having a new business category? The new categories are the very tool that helps platforms win carve-outs from existing rules. After Uber or DoorDash convince policymakers that they deserve their own category, the platforms then argue that they should operate wholly free of government interference or any standing regulatory body. To wiggle themselves out of this public oversight, labour platforms draw on campaign language that is eerily similar to the Koch brothers’ deregulatory efforts in the 1990s, and the contemporary efforts of groups like the ultraconservative American Legislative Exchange Council. In this phase, platform companies act as “regulatory entrepreneurs,” companies for which rewriting laws, as opposed to simply currying favor through traditional lobbying, is a significant part of their development plans.

That wildly successful strategy has already started to see action with on-demand care workers, with firms arguing that "Digitally-dispatched healthcare workers" and their employers should be exempt from local and state laws much like Uber's legal category of "Transportation Network Company" allowed the firm to flirt established regulations in place for ride-hail dispatch operators that did not operate through an app. For nurses and nursing assistants, this would mean any worker deployed through an app or website would be misclassified as an independent contractor—a move that, again, removes the host of protections and guarantees getting in the way of profitability for these firms.

In this phase, if a platform meets city-level resistance in these category-making efforts, the platform turns to state preemption – the nullification of municipal ordinances by state legislatures. Since 2017, Uber and its peers have pressured 34 state legislatures to prohibit governments at the city and county level from setting labour standards such as a minimum wage, raising tax revenues on ridehailing services, or mandating safety or accessibility measures. Hawaii’s law, for instance, preempts “any ordinance or other regulation adopted by a political subdivision that specifically governs transportation network companies, transportation network company drivers, or transportation network company vehicles.” An economic development expert pointed out the irony of these state-level interventions: “I frankly think it’s hypocritical of Uber and Lyft to say ‘We are partners of cities’ while systematically undermining the ability of their elected officials to actually manage how these services fit into the milieu.”

Step 3: Embrace government and institutional partnerships

In this phase, platforms from Uber and DoorDash to Instacart and Papa secure partnerships with a host of institutions, from insurance providers and non-governmental social service providers to the federal government itself. Starting with Arizona in 2019, a handful of southern Republican states changed their laws to allow patients to use Medicaid funds to pay Uber and Lyft for rides to nonemergency medical appointments. In 2021 Joe Biden’s administration partnered with Uber to provide free rides to Covid-19 vaccination appointments and installed Seth Harris, who wrote an influential study about the benefits of Uber’s worker treatment, in a top labour position. Uber has also worked with traditional unions to legislate sectoral or industry-wide bargaining for rideshare drivers while exempting workers from established labour protections, like the right to strike. Last year, Uber issued $30 million to one of California’s largest single-funded PACs, while a partnership between Uber and the Minnesota Department of Human Services to provide transit for disabled and elderly residents, especially in rural areas, threatened to derail minimum wage campaigns in that state. After a decade of disregarding laws and deceiving policymakers, now the company is, as a spokesperson told Bloomberg, “pitching proposals to state legislators that add benefits while protecting flexibility.”

DoorDash has built partnerships with food banks, churches, Meals on Wheels, veteran non-profits, hundreds of anti-hunger organisations, and major grocery chains. Instacart has not only followed suit by emulating DoorDash's move to accept platforms or Uber's work with public transit agencies, but has gone even further: it enjoys partnerships with universities to back PR talking points with research it funds, as well as partnerships that go so far as to use public funding for stipends for Instacart groceries. Papa partners with Medicare Advantage plans that let public funds go to Papa, works with insurers like Allstate to provide itself as a benefit for employee plans, and partners with Uber to offer ride-hail for its workers and their clients.

What is wrong with gig firms choosing collaboration instead of antagonism? Well they aren’t. They’re still impoverishing their workers and subjugating them to starvation wages, eroding labor standards for people outside their industries that find inspiration in Silicon Valley’s fight for the right to exploit workers, and persisting as a public parasite on multiple fronts:

• they evade taxes that pay into our social safety net while pushing more of their workers onto it

• they degrade the quality of public goods and services in cities they purport to partner with

• they degrade the working conditions of employee and independent contractor working conditions outside of industries infected by gig economy logic

• they introduce algorithmic governance tools that use surveillance to enthusiastically violate basic norms around privacy, fairness, dignity, and morality when it comes to worker agency and autonomy, prices, wages, and the boundary between our personal lives and our work lives

• they consolidate market power and political power off a project that is explicitly aimed at starving and disempowering workers, price gouging consumers, and eviscerating obstacles to monopolies and the rents that come with them—whether that be public institutions, market competitors, or their workers’ livelihoods

Or, as Fairwork writes:

Collectively, the past 15 years have seen a strategic move among Silicon Valley’s platforms and their relationships with government, institutions, and civic organisations. It is a shift from antagonism to collaboration. Platform partnerships can generate dependencies, help companies gain institutional legitimation, and secure market power.