The $6 dividend: run Alaska's own payout rules on Altman's 5% pitch and the check nearly vanishes

On July 2, the Financial Times reported that Sam Altman has discussed handing the US government a 5% stake in OpenAI, about $42.6 billion of equity at the $852 billion valuationvaluationThe price tag investors put on a whole company; OpenAI's latest funding round valued it at $852 billion. the company set in March, inside a new public fund modeled on the Alaska Permanent FundAlaska Permanent FundAlaska's state-owned investment fund, built from oil money since 1977, that pays most Alaskans a cash dividend each year., the state-owned pool of oil money that mails nearly every Alaskan a check each fall. Under the version Altman has floated to President Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, every leading US AI lab would contribute the same 5% slice. He has also spoken with Senator Bernie Sanders in recent weeks.

The other labs are not obviously on board. The same day, a source familiar with the matter told Reuters that the administration and Anthropic have not discussed any government stake in that company; Anthropic declined to comment. Neither Google nor Meta has indicated it would participate.

Most coverage has debated whether the idea is socialism, industrial policy or a shakedown. We asked a narrower question that the analogy itself invites: if this fund really ran on Alaska's rules, what would the check to each American actually be, and could the fund even cash it? The arithmetic takes four inputs, all public: the labs' valuations, Alaska's payout formula, the portfolio's real cash income, and the US population. The answer starts at six dollars and gets worse.

The timing matters to the second half of this story. The proposal surfaced six days after Washington required GPT-5.6GPT-5.6OpenAI's newest AI model, which the US government required to launch only to about 20 approved customers, vetted one by one., OpenAI's newest model, to launch to only about 20 government-approved customers while officials evaluate its cybersecurity risks. We return to that below.

Alaska's fund works in two steps. First, the state constitution forces savings: at least 25% of Alaska's oil and mineral royalties must be deposited into the Permanent Fund every year, forever. Second, a payout rule caps withdrawals: each year the state may draw about 5% of the fund's smoothed average value, the 'percent of market value' or POMV draw. The legislature then splits that draw between public services and the famous dividenddividendCash a company pays out to its shareholders, usually every quarter, from its profits.. In 2025 the dividend was $1,000 per person, paid to just over 600,000 eligible Alaskans starting October 2, roughly $620 million, or about 16% of the $3.80 billion draw. The rest ran the state.

Now apply those exact rules to the deal actually on the table. Five percent of OpenAI at $852 billion is $42.6 billion. A 5% annual draw on that stake is $2.13 billion. Divide by 341.8 million Americans, the Census Bureau's July 2025 count, and you get $6.23 per person per year. If Congress behaved like Alaska's legislature and turned only about 16% of the draw into checks, the dividend would be about one dollar.

That is the answer to the question the Alaska framing poses: six dollars, before asking whether the money is even there to pay out. And to be clear about what the comparison bar shows, Alaska's $1,000 is the smallest PFD in five years and, adjusted for inflation, the smallest in the program's history. The AI dividend loses to Alaska's worst year by a factor of more than 150.

Perhaps the OpenAI stake is just the opening bid. The reported proposal covers 'every leading US AI lab', so give the fund the most generous plausible portfolio: 5% of OpenAI, plus 5% of the other three companies most often put in that category, Anthropic, Alphabet (Google's parent) and Meta.

The pieces are bigger than most readers may realize. Anthropic closed a $65 billion Series HSeries HA late-stage private funding round; the letter just counts how many big rounds a startup has done, so H means the eighth. round on May 28 at a $965 billion post-money valuationpost-money valuationA company's value measured right after new investment money comes in, counting that new money., nearly triple its February mark of $380 billion, and enough to make it the world's most valuable AI startup, ahead of OpenAI. Five percent of Anthropic is $48.3 billion. Alphabet trades around a $4.36 trillion market capitalizationmarket capitalizationThe total value of a public company's shares on the stock market, share price times number of shares., so 5% is about $218 billion; Meta, around $1.48 trillion, contributes about $74 billion. Total: roughly $383 billion, about four times the Alaska Permanent Fund's $91.2 billion.

A fund four times Alaska's sounds transformative until you divide by the population. Alaska has 738,737 residents; the United States has 341.8 million, about 463 times as many. A full 5% draw on $383 billion is $19.1 billion a year, which works out to $56 per American. At Alaska's actual 16% dividend share, it is about $9.

The dividend is not big because Alaska's fund is big; it is big because Alaska is small.

Alaska holds about $123,000 of fund per resident. The maximal AI fund would hold about $1,120 per American, a 110-fold gap that no plausible appreciation closes. Even if the whole portfolio compounded at 10% a year for a decade, the full-draw payout would reach only about $145 per person, and about $24 at Alaska's split.

The per-person arithmetic is actually the kinder half of the audit. Alaska's 5% rule works because its portfolio is built to be spent from: stocks, bonds and real estate that throw off interest, dividends and rent, plus liquid holdings the fund can sell to realize gains. The proposed AI fund fails that test in two ways.

First, nearly a quarter of it, the $42.6 billion OpenAI stake and the $48.3 billion Anthropic stake, about 24% of the total, would be private equity with no market price and no ready buyer at that scale. Neither pays a dividend. OpenAI is the opposite of an income asset: it projects about $115 billion of cumulative cash burncash burnThe money a company spends beyond what it takes in; a company that burns cash needs outside funding to survive. through 2029, a figure it raised by $80 billion last September, according to The Information. Turning those stakes into dividend checks would require an IPO or massive secondary sales, events the fund cannot schedule.

Second, the listed half yields almost nothing. Alphabet paid about $10.0 billion of dividends in 2025, a 0.24% yield, and Meta paid $5.32 billion. Five percent of both comes to about $766 million a year. That is the fund's entire real cash engine: $2.24 per American, before a single administrative cost.

So a 5% draw rule would demand about $19 billion a year from a portfolio collecting under $0.8 billion in cash. The gap could only be filled by selling roughly $18 billion of Alphabet and Meta shares annually, a permanent, telegraphed seller that markets would price in as overhangoverhangThe price-depressing effect of a big known seller: if the market expects the government to unload billions of a stock every year, buyers pay less for it today., because the private stakes cannot be sold at will. Alaska's rule assumes liquidityliquidityHow easily an asset can be sold for cash at a fair price; a private, unlisted stake is illiquid because there is no ready market for it.. This portfolio's defining feature is the lack of it.

Set the two funds side by side and the analogy thins further. Alaska's fund has a constitutional refill: at least a quarter of every year's mineral royalties flows in automatically, in perpetuity. The AI fund, as described, is a one-time endowment, four correlated bets on the same technology cycle, with no mechanism to replenish it after a drawdown. In an AI downturn, all four holdings would fall together; Alaska's diversified portfolio is designed so that they don't.

None of this makes the idea worthless. A $383 billion public endowment would be a real asset, and Intel's example below shows government stakes can appreciate handsomely. But it would be a sovereign wealth fundsovereign wealth fundA big investment pool owned by a government, which invests money on behalf of the public. in the Norwegian sense, slow, compounding, spent on budgets, not an Alaska-style dividend machine. The check-in-the-mail framing is the one part of the analogy the arithmetic cannot support.

If the fund cannot deliver meaningful dividends, why offer it? The calendar suggests one answer. On June 26, Washington required GPT-5.6, the first OpenAI model rated 'High' capability in both biology and cybersecurity, to launch to only about 20 customers, each individually approved by the federal government. OpenAI complied and publicly bristled, writing that this kind of government access process 'should not become the long-term default'. Six days later, the 5% proposal surfaced. Equity offered in that context looks less like a dividend and more like an option premiumoption premiumIn markets, the price paid up front for a valuable right later; used here as an analogy for equity handed over now in exchange for regulatory goodwill.: value paid up front for regulatory goodwill later.

The market record of the administration's four traded equity deals since July 2025 is consistent with that reading. In every case, the announcement-day gains went to existing shareholders, and the gains scaled with the sweeteners bundled into the deal rather than with any public payout.

MP Materials jumped 51% the day the Pentagon bought $400 million of preferred stockpreferred stockA class of shares that gets paid before common shareholders and often carries special rights; the Pentagon bought preferred stock in MP Materials. and, more importantly, guaranteed a $110-per-kilogram price floorprice floorA guaranteed minimum price, in MP Materials' deal, the Pentagon promised to ensure $110 per kilogram for its key rare-earth output. for its key rare-earth output, roughly double the prevailing Chinese price, with a decade of offtake commitments attached; the one-day market-cap gain was several times the government's investment. Trilogy Metals, a microcapmicrocapA very small listed company, worth a few hundred million dollars or less; its stock can move violently on news in ways a giant company's cannot., rose 211% on a $35.6 million stake packaged with warrants and a reversal of the permit block on the road to its Alaskan mine. Lithium Americas nearly doubled on the mere report that a stake was being discussed. Intel, where the $8.9 billion, 9.9% stake mostly converted CHIPS Act grantsCHIPS Act grantsFederal subsidies Congress approved in 2022 to fund US semiconductor factories; Intel's government stake was largely a conversion of grants it had already been awarded. the company had already been awarded, with no new sweeteners, rose just 5.5%.

The Intel case also shows the other side of the ledger: by late April 2026 the government's stake was worth about $35 billion, a $26.5 billion paper gain for taxpayers. Equity stakes can be good investments. But note what the market was pricing on each announcement day: benefits to the company, floors, permits, implicit federal backing, not cash to the public. And note the contrasting design at US Steel, where Washington took a 'golden sharegolden shareA special single share giving the government veto power over key company decisions, as the US holds in US Steel, control without an ownership claim on profits.' in the Nippon Steel acquisition: veto power over key decisions with no ownership claim on profits at all. Control and cash flow are separable; every deal so far has chosen one.

The cleanest available signal arrives within weeks. In an internal memo reported by Engadget, Altman said federal leaders would be approving GPT-5.6 access 'customer by customer during this preview period', hopefully followed 'a couple of weeks later' by a general release. The backdrop is stark: Anthropic's frontier cybersecurity model Claude Mythos was restricted to vetted partners in April and forced offline entirely by the Commerce Department in June. OpenAI badly wants a different outcome.

So watch the two tracks together. If the broader GPT-5.6 release advances alongside progress on the stake talks, the option-premium reading strengthens; if the release proceeds while the fund idea quietly dies, the proposal looks more like the genuine benefit-sharing scheme Altman describes. Either way, the first payers are identifiable now: OpenAI's existing shareholders, who would absorb the dilutiondilutionWhen new shares are created or given away, shrinking each existing owner's slice of the company. of a donated 5%; and smaller labs, for whom a government that holds equity in the incumbents has a new reason to like the incumbents.

What the public would get is the one number this whole exercise pins down. Run Alaska's own rules on the deal on the table and it is six dollars a year, about one dollar, if Congress splits the draw the way Alaska actually does. Whatever the 5% is buying, it is not a dividend.