Is the Age of Purely Private-Sector AI Coming to an End?

The debate on AI nationalization is often framed as though the question is whether governments will seize OpenAI, Anthropic, or Google DeepMind—a thought famously speculated upon by ex-OpenAI employee Leopold Aschenbrenner in an essay. That is unlikely in most democracies, but nationalization rarely begins with a flag being planted on private property. It begins with equity stakes, subsidies, procurement dependence, export controls, compute allocation, public wealth funds and national-security vetoes; and by that definition, the nationalization of AI has already begun.

In the United States, OpenAI has reportedly discussed giving the government a 5 percent equity stake framed as a way for citizens to share in the upside of AI. Washington has also taken a stake in Intel and experimented with revenue-sharing in strategic chip exports. This is not old-fashioned socialism. Instead, it is strategic capitalism: public money, public risk and, increasingly, a public claim.

India is moving through a different route. The IndiaAI Mission—New Delhi’s effort to foster a homegrown AI stack—is using public funding and compute access to support domestic AI capacity. Reports that the government could take a small stake in Sarvam AI, a leading AI company, as a result of public support are significant not because the stake is large, but because the principle is new. If public compute helps create sovereign AI capability, the state may ask for public equity in the value created.

Europe, the United Kingdom, the Gulf, and China point in the same direction. The European Union is mobilizing capital for AI factories and gigafactories. The UK has treated data centers as critical national infrastructure. Gulf states are building AI through sovereign wealth capital. China is pursuing AI infrastructure as an industrial and geopolitical project. Their models differ, but the impulse is the same: no serious state wants its economy, defense systems, and public services to depend entirely on private or foreign intelligence infrastructure.

Yet government demands for equity in AI ventures may also prove useful for AI companies. The largest AI firms face a legitimacy crisis. In much of the West, public sentiment toward big technology companies has hardened. People worry about jobs, misinformation, copyright, surveillance, concentrated wealth, energy use, and the influence of unelected technology executives. AI companies are asking societies to trust them with systems that may shape education, health, courts, labor markets, media, and national security. In that sense, “trust us” is no longer enough.

A limited public stake could help solve these legitimacy issues. The stake need not mean the government running AI companies, nor does it immediately require AI dividend checks to every citizen. It could take several forms.

At one end, governments could hold small, non-voting equity stakes or warrants in frontier AI firms that benefit from public compute, procurement, defense contracts, energy support, or chip access. Those returns could flow into a sovereign AI fund rather than into the general budget. At the next level, the fund could finance public compute, worker retraining, AI access for schools and small businesses, or energy infrastructure strained by data centers. Only at the most direct end would it resemble a citizen dividend model, similar in principle to Alaska’s Permanent Fund Dividend, which distributes resource-linked investment returns to eligible residents. OpenAI itself has proposed a Public Wealth Fund that would give citizens, including those outside financial markets, a stake in AI-driven growth.

The point is that if citizens share in the upside, AI companies can argue that they are not merely extracting value from society, but also returning part of it. If public institutions have a transparent, non-controlling claim, firms may gain political cover, social legitimacy, and a stronger license to operate. For companies heading toward enormous valuations, giving up a small slice of equity may be cheaper than facing punitive regulation, antitrust action, public backlash or structural separation later.

But that is also the danger. Public ownership must not become a way for AI companies to purchase political indulgence. A state that owns part of an AI company may become less willing to investigate it aggressively, impose product-liability rules, challenge anti-competitive conduct or force structural remedies. The conflict seems obvious: the government cannot credibly act as shareholder, customer, regulator and competition authority without clear institutional firewalls.

In the United States, regulators have already scrutinized Microsoft, OpenAI, and Nvidia over competition concerns in the AI market. In Europe, the new Product Liability Directive explicitly recognizes software, including AI systems, as products capable of causing harm and brings them within a no-fault liability framework. Any public stake should therefore come with a hard rule stipulating no immunity from antitrust, no waiver from liability, no preferential procurement, and no protection from model-safety regulation. Public participation should widen accountability, not allow firms to buy their way out of it.

That might prove to be the biggest paradox. In other words, AI companies may resist nationalization rhetorically while embracing partial public ownership pragmatically. A 1, 2, or 5 percent state stake, especially through a public wealth fund or sovereign AI fund, could become political insurance. It could tell citizens that the AI windfall is not reserved only for founders, investors and cloud providers. It also tells governments that these companies recognize the public character of the infrastructure they are building.

The danger, however, is also obvious. Bad nationalization would politicize models, protect favored firms, create opaque state-company bargains and turn AI into an instrument of surveillance or ideological control. The state should not decide what artificial intelligence is allowed to say, nor use public stakes to shield incumbents from competition. That risk can be limited only if public ownership is structurally passive.

Any state stake in an AI company should be held through a ring-fenced public wealth fund or sovereign AI fund, not by the ministry that regulates, procures or deploys AI systems. Voting rights should be capped or suspended, and board seats should either be avoided or held by independent fiduciaries with a narrow financial mandate. The government should have no special power to direct model outputs, suppress lawful speech, or protect the company from competitors. Content, safety and liability rules should be made through laws of general application, enforced by independent regulators and courts, not through shareholder pressure. Procurement should remain competitive, with open tenders and audit trails, and competition authorities should retain full power to investigate mergers, cloud partnerships, data sources and repositories, and anti-competitive conduct.

Nor should public benefit be understood only as equity upside or future AI dividend checks. A more credible model would treat public participation as a bargain. That is, if AI firms benefit from public compute, energy infrastructure, defense procurement, research grants, chip access, public datasets, or regulatory support, society should receive something in return. That return could include affordable compute credits for startups and universities, open technical standards, public-interest research access, model-safety testing, worker retraining funds, AI tools for schools and hospitals, and investment in the energy and data-center infrastructure strained by AI growth. This would allow public claims to be partly financial, but also institutional and developmental.

This also matters if a major frontier developer fails to become profitable. Many leading AI companies are valued on expectations of future dominance, while their current economics remain capital-intensive and uncertain. A public stake must not become a backdoor bailout or a way to socialize losses after private investors have captured the upside. If governments participate, they should do so on clear terms of no automatic rescue, no exemption from insolvency or antitrust law, no protection from product-liability rules, and no guarantee that public procurement will keep an unviable firm alive. Where public money, compute or infrastructure support is provided, the state should negotiate downside protections such as warrants, revenue-sharing rights, access rights, continuity obligations for critical public-sector deployments, and escrow arrangements for safety documentation or essential model infrastructure. Ideally public participation should widen accountability, not make the taxpayer the buyer of last resort. But the alternative to that is not pure private AI, as that world seems to be disappearing. The better settlement is that public participation should look more like a passive sovereign wealth holding plus a public-benefit compact than a right to political control. The state may share in the upside of AI, but it should not become the editor, gatekeeper, rescuer, or protector of the companies it partly owns.

The age of purely private AI is quickly coming under pressure. The real question now is whether public participation arrives deliberately and democratically or through crisis, coercion, and backroom deals.