SOVEREIGN AI AND THE RETURN OF THE STATE
How AI is reshaping the liberal economic order and putting state ownership on the cap table
When I began my research into sovereign wealth over ten years ago, I did not think that this research in International Political Economy would be thrown into the centre of a global debate on sovereignty in technology. Today, as we all use AI to augment our lives (I admittedly use AI to help with my research and editing), we are also witnessing the largest challenge to the Washington Consensus and liberal economic order.
This challenge is not coming from new ideologies; not from Beijing, nor Moscow. It is coming from Washington itself and from advances in technology. It has made the question of “Who owns what” impossible to ignore. Artificial intelligence has done something that four decades of globalisation said could not happen; it has made state ownership a strategic again and a matter of national security. Once ownership is strategic, the state comes back. Not as regulator, governing the market, but as shareholder, financier and gatekeeper, standing in the middle of it.
DeepMind is a lesson for all states
As the UK debates its role in the AI race, the FT resurfaced the story of DeepMind; a sale that nobody thought twice about. In January 2014, Google bought DeepMind, a London research laboratory for a reported £400 million that was, by a distance, the most important AI asset ever built in Britain. Westminster did not blink. Nor, by the logic of the time, should it have. The Washington Consensus, epitomised by Thatcher and Reagan, held that capital should flow to the most efficient owner, that technology was a commodity like any other and that the state’s job was to regulate lightly and stay out of the way. DeepMind would be better resourced in Silicon Valley than in King’s Cross. The liberal economic order formed at the ‘End of History’ argued that this was the future of state and private sector relations.
Twelve years on, the sale reads very differently. DeepMind went on to produce the technology that underpins Google’s frontier AI models. The laboratory that might have anchored a British (and possibly European) AI champion, instead anchored an American one, for roughly the price of a football club. In 2014 no academic discourse, nor framework, existed to ask the question that now seems obvious: should states interfere in the ownership discussion on AI? American technology dominance was an inevitability. Its system of regulation, finance and market size efficiently acquires the world’s best talent and assets while the world applauded US exceptionalism.
AI Control, like Arms Control
The designers of that system are now compelled to rewrite it. In June, the US Commerce Department imposed export controls on Anthropic’s newest models, requiring a licence for any foreign national to access them, inside or outside the United States. Because enforcing a nationality test on a shared cloud service is impractical, Anthropic switched the models off for everyone. It was the first time American export control law had been applied to AI models themselves rather than to the chips beneath them. What started as semiconductor controls has moved down the technology value chain, from hardware to the model themselves.
As I now spend more of my time thinking about this in lecture halls than reviewing investment memos, I consider what it means to academic discourse in International Political Economy. A private company built a product; the state decided who may use it, as if the model were military hardware. Anthropic’s response is the more telling development. The company did not dispute Washington’s right to decide who uses its models, only the absence of process around the decision. When a tech firm treats government involvement in AI as a permanent fixture, the return of the state is no longer a forecast. It is the technology industry’s operating assumption.
As the world economy is entering an era of multipolarity, technology is hardening into a faultline of bipolarity. What began with the Entity List designation of Huawei in 2019, and escalated through the semiconductor controls of 2022, has matured into two rival stacks. Choosing a technology stack increasingly means choosing a political side. Access to frontier intelligence is becoming a foreign policy instrument, granted and withdrawn like landing rights. Every capital city watching this play out has arrived at the same conclusion; accept dependency on Washington or Beijing, or pay the cost of building your own AI sovereignty.
Gulf Capital has Picked a Side
The capital is already moving and the numbers describe the shift with unusual clarity. According to Global SWF’s latest annual report, state-owned investors deployed $66 billion into AI and digitalisation in 2025, in a year when sovereign fund assets reached a record $15 trillion. Abu Dhabi’s Mubadala alone invested $12.9 billion, followed by the Kuwait Investment Authority at $6 billion and the Qatar Investment Authority at $4 billion. Estimates of committed sovereign capital to AI infrastructure across 2025 and 2026 run to around $120 billion, including the MGX-led consortium’s $40 billion acquisition of Aligned Data Centers and the Qatar Investment Authority’s $20 billion AI infrastructure venture with Brookfield.
Note where the money went. The United States attracted $131.8 billion of state-owned investment in 2025, nearly double the previous year, while flows into China collapsed to $4.3 billion. Sovereign capital is not hedging between the poles; it has chosen the American side. Gulf and Asian funds hold stakes across OpenAI, Anthropic and xAI and are financing the data centres those firms depend on.
Here is the tension few have priced. Sovereign capital is funding tech companies whose products can now be switched off by a letter from the Commerce Department. The capital flows in; the control does not flow back out. Foreign states are discovering that equity in an American AI company buys exposure to the upside but not assurance of access. In Susan Strange’s terms, they have bought into the finance structure while the knowledge structure remains firmly under American command. That gap between ownership and control is a new dynamic on the diplomatic chessboard, with states, firms and SWFs thrust into a new bargaining of the AI era.
A Stake for Sovereignty
Then came the more remarkable development. A week ago, Sam Altman has proposed handing the US government a stake of 5% in OpenAI, worth roughly $42.6 billion at the company’s $852 billion valuation. Other AI developers were invited to contribute matching stakes to a proposed SWF. The talks are early, congressional approval would be required and rivals are conspicuously uncommitted. The direction of travel, however, is unmistakable; Washington already holds a ~10% stake in Intel and takes a share of Nvidia and AMD’s China chip revenues.
This is the detail that makes everything harder. The United States spent forty years telling the world that sovereign ownership of strategic companies was a distortion, a governance risk, a habit of petrostates and planned economies. SWFs were tolerated as passive money and screened as suspect the moment they wanted influence. Now the author of that doctrine is contemplating its own SWF, capitalised not by oil but by equity in the very firms whose products it is export-controlling. Whatever the merits, the ideological line has been crossed. No official in Abu Dhabi, Riyadh or Singapore will ever again sit through a lecture on the dangers of state ownership with a straight face.
The Return of the State
Strange argued in the 1980s that structural power, the power to shape the frameworks within which everyone else operates, rests on four structures: finance, production, security and knowledge. Her later work described states retreating from all four as markets advanced. What we are watching now is the retreat running in reverse and running through the market itself. Forty years into liberal economic integration, states cannot nationalise through legislation alone and must use more direct policy tools to ensure control; that control is ownership. States are buying it, financing it, licensing it and legislating around it. Sovereign capital has become the instrument through which states repurchase structural power on market terms. The knowledge structure, the one Strange considered hardest to see and most important, is the prize.
This is why the DeepMind sale could never happen today, in any major economy. Not because the economics changed, but because the framework for judging it changed. The question “who is the most efficient owner?” has been replaced by “whose sovereign capability is this?” That substitution, more than any tariff or subsidy, is the end of the Washington Consensus as an operating system.
The Sovereign of Everything
So a prediction. AI is stress-testing our idea of sovereign ownership, and the response is already visible in the institutional record. Korea is debating a Future Investment Corporation of extraordinary scale, a fund of up to $900 billion for a country running a fiscal deficit, because chips and AI are judged too strategic to leave to the private balance sheet. Canada has announced a new sovereign vehicle under Mark Carney. Indonesia built Danantara around AI, energy and food security. And the United States, of all countries, is sketching a public wealth fund seeded with AI equity. The direction is one way. We will see more sovereign funds created in the next five years than in the last fifteen, and increasingly in deficit states, funded by debt and mandate rather than surplus, because the logic has changed from recycling wealth to securing capability.
These funds, however, are only the visible instrument. The deeper shift is that the sovereignty question, once asked of oil and armies, is now being asked of everything: compute, models, chips, energy, minerals, food, payment rails. Each answer is expensive. Self-sufficiency carries a cost that comparative advantage never did, and states are choosing to pay it anyway, because the alternative, strategic dependency, is not an option. AI opened the debate. It will not close it. What began as an argument about who may use a model will end as an argument about who owns the modern economy. The age of the Sovereign of Everything has begun.