Trump's AI Profit-Share Proposal Introduces Sovereign Extraction Risk Into AI Business Models

# The Government Wants to Be Your Largest Shareholder. Price It.

A White House proposal to take equity stakes in frontier AI companies, first floated by Sam Altman and confirmed by Trump on Air Force One on June 5, is being read as a populist gesture or a negotiating tactic [1]. The actual story is that Washington is reclassifying AI from a tech sector into something closer to an extractive industry, and that reclassification, once made, is very hard to undo.

The reclassification, not the percentage

Washington has fixated on the number: Sanders wants 50%, industry whispers 1–5%, OpenAI prefers a voluntary donation of unspecified size [2]. That is the wrong axis. The number is negotiable; the framing is not.

For thirty years, US tech policy rested on a settled premise: software businesses generate returns to capital and labour, taxed through ordinary corporate income and payroll mechanisms. What is being discussed now treats frontier AI revenue the way Norway treats North Sea oil or Saudi Arabia treats Aramco production: as a rent flowing from a strategic national asset, in which the sovereign holds a residual claim. Trump's own language on Air Force One, "the American public essentially becomes a partner," is the vocabulary of resource royalties, not corporate taxation [3].

Once a sector is publicly reclassified this way, the reclassification persists across administrations. The UK has been trying to walk back the windfall taxes it imposed on North Sea operators in 2022 for three years; the fiscal architecture is sticky because the political coalition that built it does not dissolve when the original justification fades. A Democratic successor administration inheriting an AI equity framework will not dismantle it. They will expand it.

Margin compression at exactly the wrong moment

Investment committees have been modelling frontier AI on a familiar software trajectory: heavy capex now, gross margins of 70%-plus once inference workloads scale, terminal multiples in line with hyperscaler cloud. OpenAI's $850 billion private valuation and the $1 trillion-plus IPO talk for both OpenAI and Anthropic assume this curve [4] [5].

The curve is already under pressure. Inference economics are deteriorating as model sizes grow and competition compresses per-token pricing. The investment case rests on operating returns arriving before the cash burn breaks the balance sheet. A sovereign claim on the equity, even a notional 5% donated stake, changes two things that matter to a public-market investor. First, the float available to private shareholders shrinks, and any future capital raise dilutes against a government holder who cannot be bought out. Second, the precedent establishes that the federal share is the floor, not the ceiling. The Sanders bill exists. Bannon is publicly demanding 50% [6]. The ratchet only turns one way.

For the buy side, the practical implication is that AI revenue should be discounted at something closer to a regulated utility's cost of equity than a software company's. On a $1 trillion valuation, that re-rating is not a rounding error. It is the difference between an IPO that prices and one that does not.

The conflict of interest as a feature

Nat Purser at Public Knowledge has flagged the obvious problem: a government that holds equity in OpenAI is simultaneously regulating OpenAI, which "creates substantial conflicts of interest" [7]. The PitchBook read is more cynical and more accurate. Harrison Rolfes called it a "regulatory insurance policy": OpenAI buys forbearance by handing Washington a financial stake in its success [8].

This is why the deal is attractive to Altman and dangerous to Anthropic. Anthropic refused Pentagon use of its models without safety guardrails; in February 2026, Trump ordered all federal agencies to stop using its technology [9]. Anthropic is not in the equity conversation. The competitive picture this creates is sharp: one frontier lab with the federal government as anchor shareholder and federal procurement as a captive customer, and a second frontier lab cut out of both. Enterprise buyers procuring AI for regulated workloads, banks, defence contractors, healthcare systems, will read those signals correctly.

The second-order effect is that the "AI safety vs. AI acceleration" debate, which had been a values argument inside the industry, becomes a balance-sheet argument outside it. Labs that align with sovereign objectives get cheaper capital and regulatory cover. Labs that hold the line on independent safety governance get frozen out of federal procurement and face a higher cost of capital. This is the mechanism by which AI safety as a corporate strategy dies, without anyone in Congress having to vote against it.

What the EU and China see

The ORF analysis on technology sovereignty captures the foreign view succinctly: frontier AI dependencies are not merely commercial but reflect the politics of the vendor's home state [10]. A US government with an equity stake in OpenAI is, for a European or Gulf buyer, a different counterparty than a US government merely regulating OpenAI. The dependency is no longer commercial with political overlay. It is political with a commercial wrapper.

For Brussels, this accelerates the case for sovereign European model capacity that European policymakers have been arguing for two years without much traction. For Beijing, it confirms the framing that Chinese policymakers have used domestically about US tech firms as instruments of American state power, and it provides cover for symmetrical Chinese measures. For the Gulf sovereign wealth funds that co-led OpenAI's March 2026 funding round through MGX [11], it raises an awkward question: are they minority co-investors alongside a strategic partner, or minority co-investors alongside a rival sovereign with a regulatory veto?

The counter-case: theatre

The strongest version of the dismissive read is the one David Sacks made before leaving the administration: government equity stakes accelerate corporate-government fusion, and the libertarian right will block them. The White House declined to comment on record. Multiple sources told NOTUS the deal "may ultimately not come together." The legal vehicle is unclear and may require congressional authorisation that does not exist [12]. Jennifer Huddleston at Cato is making the orthodox free-market case publicly [13]. On this reading, the proposal is dealmaking pressure, Trump extracting concessions from Altman in exchange for regulatory forbearance, with no statutory residue.

The problem with this read is the Intel precedent. The administration has already taken direct equity stakes in roughly ten companies this term, including Intel at 10% and IBM [14]. Intel's stock is up roughly fourfold from the government's entry price, and the White House has been touting the gain as "a direct windfall for American taxpayers" [15]. The political economy this creates is self-reinforcing: every successful sovereign stake makes the next one easier to propose and harder to oppose. California's Newsom signed an executive order in May 2026 directing the state to study universal basic capital [16]. Hawley is pushing parallel extraction on data centre infrastructure [17]. The mechanism may shift, from equity donation to data tax to infrastructure levy, but the political demand for a sovereign cut of AI economics is now bipartisan and multi-jurisdictional. That is not noise. That is a regime.

What to watch

1. Whether OpenAI's S-1, when filed, contains a disclosed government equity holder or a contingent liability for one. If it does, every subsequent AI IPO prospectus will follow the template, and the sovereign claim becomes a standard line item in AI corporate structures. If OpenAI files clean, the proposal is closer to theatre.

2. Whether Anthropic's next enterprise procurement cycle shows measurable share loss in federally regulated sectors (defence primes, federal contractors, federally insured banks) over the next two quarters. A widening gap between OpenAI and Anthropic in regulated-industry win rates would confirm that the equity discussion is already functioning as de facto industrial policy, regardless of whether any equity actually changes hands.

3. Whether the European Commission or a major EU member state issues guidance, formal or informal, treating US frontier models as subject to foreign-state-influence review under existing investment-screening or critical-infrastructure frameworks within the next six months. This would be the clearest signal that the sovereign extraction framing has been received abroad as a permanent reclassification rather than a domestic American negotiation, and it would mark the point at which US AI export economics change in kind.