What Sanders' $7 trillion AI bill would actually do

The legislation Bernie Sanders shared with AP News, as reported by Ars Technica, is not a tax on corporate profits. It is a one-time 50% tax on the stock of every AI company doing $200 million or more in annual AI sales, payable in newly issued shares. The federal government would hold those shares in a new sovereign wealth fund, a state-owned investment portfolio that distributes returns to the public, modeled loosely on Alaska's Permanent Fund. Sanders has estimated that fund could grow to $7 trillion in value, and that each American could receive more than $1,000 per year in dividends if the portfolio yields a 5% annual return.

The dividend number is the political hook. The mechanism is the policy story.

A sovereign wealth fund is a normal enough instrument. Other countries, including Norway and Singapore, run them. In the United States, the closest working analog is the Alaska Permanent Fund, which has paid an annual dividend to Alaska residents since 1982 using oil-revenue royalties. Alaska's fund owns no equity in private companies and has no governance role in them. It collects a slice of resource extraction and writes checks. The Sanders proposal is structurally different in one decisive way: the government would own voting shares in private AI firms, and a new federal body, the Independent Commission for Democratic AI, would cast those votes.

The commission, as described in the legislation Sanders released, would have seven members confirmed by the Senate on a bipartisan basis. It would not manage the day-to-day operations of any AI company. Its power, as the bill frames it, is the power to block specific corporate decisions using the government's minority voting stake. In practice, that means the commission could veto certain mergers, certain capital-allocation choices, or certain export and deployment decisions, but it would not set strategy, hire executives, or rewrite product roadmaps. Whether a blocking minority is a meaningful check or a paper tiger depends on the share of votes the commission controls, and the size of that block has not been specified in the public summary of the bill.

The $7 trillion number deserves a second look. It is an estimate of the total market capitalization of the affected companies at the time of the tax, not an appropriation, not a transfer of cash, and not a guaranteed windfall. The Sanders proposal — a forced 50% equity extraction via a sovereign wealth fund with blocking-vote governance — has no direct precedent in US policy. A 50% equity tax of that scale would be structurally novel. The Alaska Permanent Fund analogy captures the dividend-distribution model but not the governance-within-private-companies design, which is unique to this proposal.

The dividend calculation also rests on a 5% annual yield assumption. Sovereign wealth funds with concentrated equity portfolios can produce returns in that range in good years and losses in bad ones. Alaska's dividend varies with oil prices and investment performance. There is no guarantee that $1,000 per American is a floor. In a downturn, the per-capita payout could fall, and the principal could shrink. The bill, as described, treats the dividend as a target, not a promise.

Sanders intends to campaign on creating the fund, and the legislation is unlikely to advance in the current Congress. The bill would need 60 Senate votes to overcome a filibuster, and the politics of a forced 50% equity transfer from the country's most strategically important technology firms has no realistic path through a chamber where AI policy is dominated by bipartisan concerns about US competitiveness against China-based labs. Sanders' own meeting with OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei, according to AP News via Ars Technica, produced no convergence on public ownership stakes, and both executives have signaled openness to narrower public-benefit designs that stop short of forced equity.

The industry-side argument is concrete. A 50% forced equity transfer would dilute existing shareholders, including public pension funds and retail investors, and would constrain the capital available for the large-scale infrastructure buildout that current frontier AI labs treat as existential. It would also affect the talent market: equity compensation is a primary recruiting tool for AI researchers, and a 50% dilution event would reset the value of every unvested grant. Critics argue that a mechanism of this scale would impair US AI capital expenditure, weaken retention against better-resourced foreign labs, and shift the center of gravity in frontier research overseas. That critique is not a slogan. It is a second-order consequence of the same mechanism the bill proposes.

What Sanders is really putting on the table is a governance design, not a check. The dividend is the line that travels. The commission with blocking voting shares is the design that would actually change how the AI industry is run, or would not, depending on the share of votes the bill ultimately grants it. The bill's near-term effect is to set a marker in the 2026 campaign for what a public-ownership-flavored AI policy could look like, and to force a more specific answer from candidates who want to talk about AI windfalls without naming the equity mechanism that would produce one.

The next test is whether the full legislative text, when released, specifies the size of the government's voting block, the scope of decisions the commission can block, and the transition rules for already-issued shares. Until those numbers are public, the $7 trillion headline is a ceiling, not a plan, and the per-capita dividend is a yield assumption, not a benefit.