The OpenAI Government Stake Is Capture, Not Charity

OpenAI has discussed handing the United States government an equity stake of about 5 percent of the company, worth roughly $42.6 billion at its latest valuation, the Financial Times reported on July 2, and chief executive Sam Altman has personally pitched the idea in Washington to President Donald Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent. The proposed OpenAI government stake would be the largest state shareholding ever taken in a private American technology company. Congress should make sure it stays a proposal.

Altman's ambition runs well beyond one firm. Under the framework described in the FT's reporting, every leading American AI developer, potentially including Google, Meta and Anthropic, would contribute about 5 percent of its equity to a public investment vehicle modeled on the Alaska Permanent Fund, which pays annual dividends to Alaska residents from the state's oil wealth. Returns would eventually flow to the American public. The talks are early-stage and, in the FT's word, conceptual; implementation might require an act of Congress, and OpenAI declined to comment. Strip away the civic packaging, though, and the OpenAI government stake is regulatory capture with a sticker price, offered by a company that needs Washington's permission slips more than Washington needs its shares.

A Shareholder Government Cannot Referee Its Own Portfolio

The objection is structural, not partisan. The federal government is supposed to police frontier AI: writing safety standards, timing model releases, enforcing export controls on the chips these firms depend on, and testing whether their market power warrants antitrust action. Reporting already indicates officials are moving toward standards regimes for new model releases, which makes the government a gatekeeper for every major launch. A gatekeeper holding $42.6 billion of the gated company's stock, valued off the $852 billion price set in OpenAI's March funding round, has a measurable financial interest in waving products through.

Every serious enforcement decision would then carry a portfolio cost. Halting a risky model would knock billions off a fund that mails dividends to voters. An antitrust case against a fund constituent would be a suit against the public's own asset. Export restrictions that dent OpenAI's overseas revenue would land, in part, on the Treasury's books. No institution reliably injures its own balance sheet on principle, and no sensible country designs a regulator that must.

The administration is not even pretending otherwise. Trump has called government ownership in AI companies "a beautiful thing" that would make Americans "partners in this revolution," per CNBC, and earlier confirmed discussions about concepts in which the public "essentially becomes a partner with the companies." A partner is precisely what a regulator must never be.

Look at the Calendar

The sequence of events removes any doubt about what is being purchased. Days before the FT story, Washington delayed the release of OpenAI's GPT-5.6 model, as Tom's Hardware reported, a vivid demonstration of the leverage federal officials now hold over the company's product pipeline. TechCrunch, citing the FT, reported that the proposed donation aimed to "secure good relations with the administration and … address political blowback." That is not the vocabulary of philanthropy. It is the vocabulary of protection.

Nor would this be an improvisation. The administration has run the equity playbook before: last August the federal government converted CHIPS Act grants into a 9.9 percent stake in Intel, and it turned federal support for MP Materials into a shareholding there too. What began as industrial policy has hardened into a pattern in which access to the American market, or relief from American regulators, is denominated in stock.

The financial context sharpens the incentive further. OpenAI, and reportedly Anthropic, are preparing potential trillion-dollar public offerings. A government shareholder acquired before an IPO is the strongest listing endorsement money can buy, and the surest guarantee that the state will want the offering, and every quarter thereafter, to go well. Investors will price that protection. They should also name it for what it is.

Alaska Struck Oil. Altman Struck a Deal.

The Alaska Permanent Fund analogy does heavy rhetorical work in this proposal, and it collapses on inspection. Alaska's dividends flow from oil that belonged to the state: residents are paid for a public resource extracted from public land. OpenAI's model weights are not a commons, and no citizen holds mineral rights to a transformer architecture. The Alaska fund monetized public property. Altman's fund would monetize public authority, trading a slice of equity for the goodwill of the officials who control launches, chips and merger reviews.

OpenAI has been candid about the aspiration, if not the mechanics. Its April policy paper, "Industrial Policy for the Intelligence Age," argued that returns from a public fund "could be distributed directly to citizens, allowing more people to participate directly in the upside of AI-driven growth, regardless of their starting wealth or access to capital." The instinct is defensible. The instrument is not.

Consider the offer Altman is bidding against. He has spoken in recent weeks with Senator Bernie Sanders, whose American AI Sovereign Wealth Fund Act, introduced in June, would impose a one-time 50 percent tax on the stock of systemically important AI companies and claim half their voting shares for a public fund his office values at $7 trillion. The bill has not advanced to committee, and I am not endorsing confiscation. But its existence explains the arithmetic of the counteroffer: a tenth of the equity, none of the voting control, all of it framed as a gift. With expropriation on one end of the table, a voluntary 5 percent may be the cheapest political insurance Silicon Valley has ever bought.

The Honest Version Runs Through Congress

I hold the unfashionable view that Americans do deserve a share of AI's upside. The labs were built on publicly funded research, public data and public forbearance, and the technology's gains will be radically concentrated without policy. But democracies have transparent tools for exactly this purpose: taxes debated in the open, dividend programs written into statute, obligations that bind every company identically and can be amended or repealed in daylight. A law applies to firms that never met the president. An equity handshake rewards the one whose chief executive got the meeting.

The 5 percent structure is also engineered to let both sides claim virtue while escaping accountability. The government would take upside without voting control, so it bears no governance duty when things go wrong. The company would buy goodwill without ceding power, and acquire an implicit shield: once the public's dividend depends on OpenAI's share price, every future safety rule arrives pre-lobbied. Officials who delayed GPT-5.6 last month would be marking their own fund to market the next time a release decision lands on their desk.

The FT describes these talks as conceptual, which means there is still time to stop treating them as inevitable. Congress should refuse the OpenAI government stake and say plainly why: the public's claim on artificial intelligence belongs in legislation, voted on and visible, not in shares passed across a table between a CEO seeking launch approvals and an administration that calls owning its regulated industries beautiful. A government that owns what it regulates will, soon enough, regulate to protect what it owns.