On June 5, aboard Air Force One, President Trump told reporters he had been talking with AI executives about taking a public stake in their companies. It would be a voluntary and passive arrangement, essentially allowing the American public to become a partner in the giants driving the AI boom, without board seats or voting rights — along the lines of the government’s 9.9% position in Intel.
Days earlier, independent Sen. Bernie Sanders had introduced a bill that would take the same instinct much further: a one-time 50% tax on the largest AI firms, paid in stock, with the shares seeding a sovereign wealth fund and the government holding voting rights and a board seat at each company.
Two men who almost never agree on anything arrived at the same idea in the same week — one light-touch and optional, the other compulsory and hands-on.
Both approaches have been discussed as questions of wealth distribution, of whether ordinary Americans should share in the gains of a technology built on collective knowledge. There’s no doubt that these arguments are central to how AI develops, but they leave out what owning these companies would actually entail.
The companies in question own no power plants and no transmission lines. Their value is a bet on data center demand, and that demand runs entirely through a grid the federal government regulates. So owning a piece of these companies is not owning a neutral financial asset. It is taking a position in the buildout itself, in how fast the grid gets developed and how cheaply the power gets delivered, because that is what the valuation is priced on. To say the government should own AI is to say the government should hold a financial stake in how fast it lets the grid get built for AI.
The wealth debate is an energy policy question wearing a distribution policy costume.
The black box of AI company books
Anthropic filed confidentially for an IPO on June 1 at a valuation around $965 billion, and OpenAI followed a week later, off a March round that valued it at $852 billion, with some analysts projecting higher. Under the Sanders bill, a 50% position in both would approach $900 billion in federal equity.
Neither valuation rests on audited financials, because the filings are confidential; the government would be sizing a bet on companies whose books it cannot yet fully see.
The two proposals differ in how directly the government would act on its interest.
The Trump version would be passive, meaning it would come with no board seat and no voting rights, giving the government no direct say in how these companies are run. That being said, it would still have leverage as the government already regulates the grid they depend on, and a financial stake gives it a reason to use that power in their favor, to want new power connected fast and the cost spread onto everyone else
The Trump version is also voluntary, which means that companies can simply decline to do what the government wants. Anthropic has done exactly that, staying out of equity talks entirely, after a year of friction that included Trump ordering federal agencies to stop using its technology. The voluntary path, in other words, is full of holes. One of the two companies most likely to define the market has already walked through one.
The Sanders bill would leave no room for an opt-out. It would grant the government voting shares and equal representation on each company’s board, with the stated purpose of blocking decisions that hurt citizens and pushing for policies that help them. Those board seats are not a safeguard the senator neglected to remove; they are the instrument for giving the public a direct role in determining the future of the technology.
Under this model, the government — and by extension the public — engagement in the company is a key part of the package, because a passive stake cannot deliver what he is actually promising, which is direction, not just dividends. That guarantee of protecting the public requires the government to be in the room, and being in the room at an AI company means being on the board and able to weigh in when it decides where to site data centers and how to contract for the power to run them.
This is where the proposal stops being only a wealth distribution measure and becomes something larger. These companies’ largest operating constraint is power procurement, so a government holding board seats is no longer just exposed to the grid buildout, It is party to it.
The conflict is that one entity, the federal government, would now sit on both sides. It regulates the grid these companies depend on, and it owns a piece of the companies. The Sanders model would deepen that, as it deliberately increases the government’s role in running these companies in order to lock in the public benefit. The more the government steers the companies, the more of the national energy buildout it has taken onto its own books. The passive version avoids that by giving up the guarantee.
The throttle Washington already controls
The federal government is currently expanding its authority over the exact bottleneck that decides how fast these companies can grow. In October 2025, Energy Secretary Chris Wright used a rarely invoked power to direct the Federal Energy Regulatory Commission to write new rules for connecting large electrical loads to the interstate transmission system. FERC historically regulated how power plants connect to the grid, while how a data center plugs in was left to states. FERC is expected to act on the large load docket as soon as this week.
Large loads that agree to be flexible, curtailing use when the grid is stressed, could move through interconnection studies faster, possibly within 60 days, which would speed the buildout. FERC is also considering how much of the cost of the grid upgrades those loads require they should bear, with the proposal leaning toward assigning them the full amount. A government that profits when these companies grow has a reason to want the fast track easy to qualify for, and the cost burden set low enough not to deter the buildout.
Nat Purser of the advocacy group Public Knowledge warned that a government holding equity could grow less willing to enforce rules on these companies when enforcement would lower the value of its own stake. The analysts who studied the Intel stake reached the same conclusion, noting the tension between the government’s role as an investor in one company and its role as a regulator of the wider industry. With Intel that tension was about export controls. With an AI lab it is about power, because power is the binding constraint on everything these companies do.
The public has a real argument for sharing in the gains of a technology this consequential. The moment a government owns the companies driving load growth, though, is the moment its interest in cheap fast power stops being neutral.
