OpenAI Offered Washington a $42.6B Stake — Anthropic Refused and Proposed a Tax Instead

A $42.6 Billion Offer, Modeled on Alaska's Oil Money

On July 2, 2026, the Financial Times reported that OpenAI has proposed handing the US government a 5% equity stake in the company — worth roughly $42.6 billion at the $852 billion valuation OpenAI set in its March 2026 funding round. CEO Sam Altman's broader pitch, raised in conversations with President Trump, Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and Senator Bernie Sanders, envisions every leading US AI developer — including Google, Meta, and Anthropic — contributing a similar 5% slice into a sovereign-wealth vehicle modeled on the Alaska Permanent Fund, which has paid Alaska residents an annual oil-revenue dividend since 1982. The FT described the talks as conceptual and early-stage, and any formal structure would likely require an act of Congress.

Anthropic's Answer: No Equity, Tax Us Instead

Within days, Anthropic drew a hard line against the idea. According to people familiar with the matter, Anthropic is not participating in equity discussions with the administration at all, and has instead floated its own alternative: a 'digital dividend' funded by taxes on the AI sector and paid out directly to Americans, rather than a government ownership stake in any individual company. The distinction is not cosmetic — an equity stake dilutes existing shareholders and puts Washington on the cap table of a company it is also supposed to regulate; a sector-wide tax spreads the cost across every AI company's revenue without handing the government a seat at any single table.

This Isn't OpenAI's First Rodeo With Washington

The proposal lands days after Washington reportedly delayed the rollout of GPT-5.6, and follows a pattern the US government has used before: it took a 10% stake in Intel in 2025 after an $8.9 billion investment, and holds equity positions in IBM and several quantum-computing companies. Altman has framed the offer as a way to 'share the upside of AI' with the public; Trump has called the idea of the government owning stakes in AI giants 'a beautiful thing' that would make Americans 'partners in this revolution.' Whatever the framing, the timing — arriving alongside regulatory friction, not before it — reads to most observers as pressure relief as much as generosity.

The Conflict of Interest Nobody Has Resolved

The unresolved problem is structural: if the government owns a financial stake in the company whose models it is also responsible for safety-testing, export-controlling, and regulating, the incentive to flag a dangerous capability or block a risky release competes directly with the incentive to protect the value of a $42.6 billion public asset. That tension isn't hypothetical for this industry — it's the same category of conflict that turned a single flagged jailbreak technique into Claude Fable 5's 19-day global suspension via export controls last month. A government that is also a shareholder is a different kind of regulator than an arm's-length one, and nobody in this negotiation has explained how the two roles stay separated.

What to Watch If You're Building on Any of These Platforms

Three things worth tracking for engineering leads choosing a model vendor this quarter: first, watch which labs take the equity path versus the tax path — a government shareholder has a direct financial interest in that lab's commercial success, which can shape everything from feature prioritization to how aggressively safety classifiers get tuned. Second, treat any lab with direct government ownership as carrying elevated export-control and geopolitical exposure — the Fable 5 shutdown showed how fast a compliance dispute can take a model offline entirely, with zero notice to the teams depending on it. Third, keep the same scrutiny for AI-sector tax proposals: a tax funds public benefit without ownership, but it still shapes lab economics and could eventually get passed through to your API bill.

Bottom Line

The $42.6 billion number is the headline, but the real story is that the two most valuable AI labs in the world just publicly disagreed on whether the US government should own a piece of the infrastructure layer or simply tax it — and that disagreement is itself a signal that frontier AI is now treated as critical national infrastructure, not just a software category. For engineering teams, vendor risk reviews that used to stop at pricing and uptime now need a line for 'what is this company's relationship with the state,' because that relationship is increasingly what determines whether your integration survives the next regulatory cycle intact.