AI nationalization is a bailout wearing a populist costume

In 2021, Sam Altman published an essay titled, “Moore’s law for everything.” In it, he imagined an abundant future where housing, food, education, and other basic needs would become half as expensive every two years. He also proposed taxing companies and land to fund what he called an “American Equity Fund,” a redistribution program for the state where every adult citizen would be given a cash dividend. Now, five years later, the future isn’t directly playing out how he predicted. Ignore the tax aspect of that essay; his company is not getting taxed and nothing resembling that redistribution program exists. What it created instead is a company-controlled foundation that makes discretionary grants to causes of its own choosing. This is standard corporate philanthropy at an unusually large scale—not a dividend, not automatic, and not owed to anyone.

In early July 2026, it was reported that Open AI had discussed handing the US government a five percent equity stake in the company worth roughly $42.6 billion based on its $852 billion valuation from a March funding round. Dario Amodei, the CEO of Anthropic, in his recent policy essay, encouraged the federal government to be more active in macroeconomic and tax policy as AI reshapes the labor market. Senator Bernie Sanders has proposed something more sweeping: a 50 percent public stake in the largest AI firms, taxed directly out of their equity. President Donald Trump has floated government equity stakes of his own and drawn a direct line to Sander’s idea. Steve Bannon has accused the AI labs of trying to buy off scrutiny; because they have built their immensely valuable systems by scraping humanity’s collective data, he demanded the government take half.

If you look at the picture from the outside, this is a genuinely strange coalition. When Sanders, Trump, Bannon, and the AI overlords themselves are all nodding along to some version of “the public should own a piece of AI,” it is worth asking what problem this actually solves, and for whom.

Why “nationalization” is misleading

On the surface, this idea sounds like a populist win. “Nationalizing AI” is something you might expect to hear about at a Democratic Socialist conference. It might look like a rare moment of consensus across all fronts, including in the philanthropic and nonprofit sector, where benefits would be distributed with almost no public cost.

When Sanders, Trump, Bannon, and the AI overlords themselves are all nodding along to some version of “the public should own a piece of AI,” it is worth asking what problem this actually solves, and for whom.

Nationalization proposals typically follow some version of a crisis like the banks in 2008 and automakers before that. Whenever a private sector fails and requires public capital to survive in exchange for partial public ownership, this is the usual mechanism. But in the case with AI, this sequence is the mirror image. The industry is proposing to hand over equity while insisting everything is going fine and before any crisis has forced its hand. In this scenario, nationalization functions more like an insurance policy.

Things look even stranger when you check the balance sheets. AI companies are financing the construction of data centers and chip buildout from debt, on the promise of unmaterialized profits. There is also a web of circular vendor-financing deals. One company invests in another company that turns around and buys the first company’s chips or cloud services, each turn of the cycle boosting the involved companies’      valuation on paper. This is not to say that the technology is not valuable or that a crash is imminent, but it has drawn comparisons to related-party arrangements where the underlying revenue is hard to verify. For example, Nvidia’s vendor-financing deals with OpenAI and others have been likened to Lucent and Enron, companies whose related-party and vendor-financed arrangements obscured how much of their reported revenue reflected genuine end-customer demand. Nvidia has pushed back on the comparison directly, stating it doesn’t use special-purpose entities to hide debt, but analysts have still flagged that the company is leaning heavily on vendor-financed demand, which creates exposure if AI growth slows.

Here, we might ask: is that ‘free’ equity being offered now, before the capital structure has been tested, a gift? It is more of a hedge, where political goodwill is converted into a friendlier regulatory and financing environment.

With public confidence in AI already shaky, an unfavorable regulatory ruling or a hostile congressional hearing does more damage to these companies than it would in a moment of public enthusiasm. Having the government as a quiet shareholder gives these firms a built-in reason for Washington to soften that response, not through an explicit quid pro quo, but through the ordinary pull of shared financial interest.

At the same time, the public mood is shifting in ways that could reshape the political ground under the industry. A Pew Research Center survey of more than 5000 U.S adults conducted earlier this year found that 40 percent expect AI to have a negative impact on society, while only 16 percent expect a positive one. Nearly half of adults under 30 expect AI to be bad for society, compared to smaller shares of older adults.

How do you reel these people in without making a direct ad campaign? By making them shareholders in the industry’s success. Shareholders have every incentive to defend their portfolio rather than regulate it.

The bailout model

Under this administration, this foray into equity stakes is not an outlier. The Trump administration built what the Council on Foreign Relations has called the largest push by the US government into equity stakes in private industry since World War II. There have been more than a dozen deals concentrated in rare earth, semiconductors manufacturing, quantum computing, nuclear power, and defense systems, a portfolio that has since grown to approximately 30 companies including MP materials, Intel, US. Steel, and IonQ. In a lot of these cases, there is a consistent pattern to set your watch by: a deal is announced and the target company’s stock along with its unannounced peers jumps. USA Rare Earth’s stock rose on the day a $1.6 billion government deal was announced and was up 125 percent for the year by that point. Wall Street analysts have instrumentalized this basic pattern recognition by curating publishing lists of which companies they expect the government to buy into next.

Having the government as a quiet shareholder gives these firms a built-in reason for Washington to soften that response, not through an explicit quid pro quo, but through the ordinary pull of shared financial interest.

While speaking with PBS New Hour about Trump taking a 10 percent stake in Intel last year, Peter Harrell who worked on international economics for the Biden White House commented on the unusual nature of this: the United States has taken equity in companies before but always in the context of bailouts, with the government expected to exit once the firm is recovered. In this case, however, they are buying into healthy companies, seemingly justified by the legal rationale that nothing in the law clearly forbids it and the companies have agreed.

This is the frame the AI equity proposals are getting into. It’s the newest and largest application of a template that has already played out in other sectors, where the main documented effect has been to reward whoever holds the stock and whoever gets advance word, not the public at large.

On this topic of nationalization, it is also relevant to ask why frontier AI was ever treated as a purely private-sector project in the first place. Nuclear weapons, for example, weren’t left to the rough hands of capitalists and which of their labs got there first. It was treated as a state program, and civilian nuclear power was built under public ownership and licensing from day one. On the other hand, while the research work that serves as the basis traces back to public universities and federal funding, AI and its models—and governance and revenue—sit entirely in private hands.

Control usually moved from the states outward, in cases such as nuclear weapons and the internet. In the AI sector, private firms that built market dominance on a foundation of public research are handing back a tiny slice of the revenue cake and calling it the same kind of arrangement.

Nationalization simply means public ownership paired with public power; the people would have a say in directing priorities or overruling a company’s decisions in their own interest. A five percent stake or a 50 percent stake sitting inert in a commission a future administration could staff however it likes, does none of that.

Accountability is critical

The clearest illustration of the gap between this current proposal and structural accountability is a live piece of legislation in Congress. In June, Senator Bernie Sanders introduced the American AI Sovereign Wealth Fund Act. Its contrast with OpenAI-style proposals is instructive precisely because both get described in headlines as “AI nationalization.”

Sanders’s bill would impose a one-time 50 percent tax paid in stock rather than cash, on any AI company with more than $200 million in annual AI-related gross receipts. The resulting shares would go into a Treasury-held fund managed by a new independent commission for Democratic AI consisting of seven members, nominated by the president and confirmed by the Senate. Furthermore, the fund would pay out five percent of its value directly to US taxpayers, and channel the rest towards priorities like healthcare and housing. Sanders has framed the underlying logic bluntly, arguing that because AI is built on humanity’s collective knowledge and decades of taxpayer-funded research, the wealth it generates should be broadly shared rather than captured by what he calls a handful of Big Tech oligarchs.

This design, however, might appear a bit exuberant to some parties. Even Norway’s sovereign wealth fund, often cited as the gold standard for a nation, has a limit of owning no more than 10 percent of any single public company; Sanders’s proposal would go well beyond that.

The Sanders bill is unlikely to become law as written. It faces the same problem that has dogged other versions of this idea: it depends on Congress standing firm without being budged by the Executive Branch that has been exercising authority unilaterally. Still, its existence matters because it shows a version of AI nationalization that is not just a dividend but a lever.

This does not mean that civil society should treat public AI ownership as inherently suspect. Any government equity stake would need enforceable conditions attached, especially on how the company treats workers and communities, and the consequences if the company reneges. The terms could be negotiated in public, through Congress, and with input from the workers facing AI-driven displacement and the communities hosting the data centers these companies are building. The result of all that would need the kind of transparent, continuously updated public accounting that the government’s existing 30-company stake portfolio still lacks.

What’s at stake in getting this distinction right goes beyond AI policy. If a government equity stake without governance rights, transparency, or enforceable terms gets accepted as “nationalization,” it sets the template for how every future negotiation between the state and concentrated corporate power gets conducted—as a voluntary gift on the company’s terms, dressed up as a public win. The next industry seeking political cover, whatever it is, will have a precedent to point to. Civil society and policymakers have a narrow window, before these arrangements become normal practice, to insist that public ownership come with public power attached.