How Much a Dollar Cost?

The AI Bubble in 2026 (2/4)

Profit I (1982) by Jean-Michel Basquiat

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.