The AI Bubble in 2026 (1/4)Part 1: On the coming geopolitics of the compute stack, or Our New Imperial StrategyMy 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 as I think they’re overshadowed by a myopic focus on equity prices, valuations, and capital expenditures that is in part an overcorrection by commentators and talking heads to their stubborn dismissal of 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? From Oil Diplomacy to Compute DiplomacyOne 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, 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:
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:
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.
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.
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.
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 for 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 GulfWhat 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 DealIt 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 GulfUnveiled 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.
More on PetroComputeAbdullah 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.”
(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.
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:
The Most Serene Republic of NvidiaThis 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.
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:
Another key pillar proves to be how the United States realizes extraterritorial control through legal mechanisms:
The end result being, of course, an offer you can’t refuse:
Final ThoughtsSo 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? You're currently a free subscriber to The Tech Bubble. For the full experience, upgrade your subscription. |
