Last week, Senator Bernie Sanders (I-Vt.) revealed the American A.I. Sovereign Wealth Fund Act , a sweeping proposal that would give the public a direct stake in the largest AI companies. The bill text imposes a one-time 50% tax, paid in stock, on companies with more than $200 million in annual AI-related receipts. Those shares would go into a Treasury trust fund managed by a new Independent Commission for Democratic AI, whose seven members would be nominated by the president and confirmed by the Senate. Sanders estimates the fund could begin with roughly $7 trillion in assets. He would then direct 5% of its value each year toward direct payments to Americans and, over time, toward health care, education, housing, and environmental goals.

The Public Helped Build AI

In one sense, the Sanders plan is a nationalization scheme. In another, it is a recognition that AI is already a massive joint venture. The major AI companies have built impressive products, but they did so by training their systems on books, music, journalism, code, art, photographs, scientific papers, and ordinary online conversations produced by millions of people who were never asked to participate, much less paid a negotiated price. Sanders is right to criticize this. AI firms have built trillion-dollar businesses on top of a vast reservoir of human intellectual property, yet the people whose work supplied the foundation are largely expected to accept the arrangement without compensation.

Taxpayers have a similar claim. Modern AI is often described as a triumph of private genius, but its scientific foundations did not emerge from venture capital. They were built over decades with public money. The Office of Naval Research supported early neural-network research, including the perceptron , one of the first machine-learning systems. DARPA spent decades supporting many of the scientific and engineering advances that ultimately contributed to modern machine learning and generative AI. DARPA also helped move autonomous vehicles from laboratory experiments toward commercial reality. The National Science Foundation has played a similar role, using university grants and other research programs to support AI for decades and seed technologies that later became commercially valuable.

We hear a great deal about “risk takers” who finance uncertain ventures and help bring new technologies to market. In the AI economy, perhaps the biggest risk taker of all has been the taxpayer. Yet all the attention is paid to the “takers” who arrive after the hard work is done and capture a disproportionate share of the gains. These are the tech executives and financiers capturing enormous gains from an innovation wave that others primarily created. In the start-up ecosystem, luck and timing can matter as much as original invention.

The sovereign wealth fund idea rests on this simple premise. When public inputs help create extraordinary private returns, the public should share in the upside. That is also why Sanders’s plan overlaps, in important ways, with Trump’s. The Trump administration’s Intel stake , a 10% passive position tied to previously committed federal support, reflected the same instinct.

Trump’s version of a sovereign wealth fund appeared to envision smaller, more transactional, and perhaps more voluntary stakes. Sanders goes much further, effectively giving the public ownership of half of every large AI company. Yet both are circling the same basic idea. As an AP report noted earlier this month, Trump, Sanders, and OpenAI’s Sam Altman have all entertained some version of public equity in AI companies. A U.S. sovereign wealth fund is no longer a curiosity. It is becoming a bipartisan answer to one of the central economic questions of the AI age. Who owns the gains from the next technological revolution?

A reasonable middle ground is not hard to imagine. The left worries about concentrated wealth and exploitation of labor. The right worries about national security and the national debt. Both worry about automation’s social costs. A sensible compromise would be less sweeping than Sanders’s 50% stock tax, but more serious than Trump’s original executive order , which merely asked the Treasury and Commerce Departments to develop a plan but did not yet give the idea real force.

Where The Sanders Plan Goes Wrong

There are real problems with the Sanders plan. A 50% claim on the equity of any large AI company would be a sweeping government intervention, and it could chill investment in firms that are still unprofitable and dependent on outside capital. A better approach would tie public ownership to specific government contributions or have the government fund buy equity on market terms. There is also a valuation problem. It is far from clear that taxpayers should be buying into AI companies at today’s prices. AI may yet prove to be a bubble , as many leading firms are valued in ways that seem hard to reconcile with their fundamentals or current cash flows.

Another problem is governance. Sanders’s commission would be asked to promote worker welfare, public safety, competition, environmental sustainability, and financial solvency. Those are worthy aims, but they are not the same as running an investment fund. The board’s core duty should be fiduciary, earning taxpayers the best risk-adjusted return possible. Fund managers should not become shadow regulators for AI, especially when the sector is already scrutinized by an array of antitrust, securities, labor, and environmental regulators.

The plan’s proposed separation of AI and non-AI businesses is also impractical. Before long, every large company will be an AI company in some respect, just as every large company became an internet company and a software company. Trying to draw legal walls around “AI” and “non-AI” operations could invite arbitrary enforcement and constant litigation.

Sanders’s universal dividend is premature as well. He has suggested the fund could eventually pay every American roughly $1,000 a year. But many frontier AI firms are valued on expectations, not mature cash flows. Many are not profitable. Others earn profits from legacy businesses rather than AI itself. Paying out 5% of a fund’s market value each year would risk turning paper gains into hard fiscal promises, forcing asset sales when cash is not available. Normal investment funds distribute income only after returns become steady and real.

Government As Investor, Not Operator

These fixes matter, but they are not exotic. They are the basics of sound sovereign wealth fund design . A successful fund should have a clear investment mandate, professional management, political independence, and a fiduciary duty to maximize long-term returns for taxpayers. Government is better suited to being an investor than an operator.

Still, Sanders deserves credit. His proposal is serious, even when it overreaches. It recognizes that the AI boom is not only a story of heroic entrepreneurs. It is also a story of taxpayers, researchers, artists, and ordinary citizens supplying inputs that a small group of companies now monetizes. Asking the public to absorb job disruption, higher energy costs, privacy risks, and the broader social consequences of AI while a narrow class of owners captures all the gains is not a stable arrangement.

Trump made the sovereign wealth fund idea respectable on the right by tying it to federal assets and strategic national interests. Sanders has made it urgent on the left by tying it to AI and the question of who controls technological rents. Between those poles lies a workable policy. Private firms should remain free to innovate, raise capital, and earn profits. But when public resources, public science, and the labor of millions help create private fortunes, the public deserves a claim on the value it helped build. Market capitalism works, and government and taxpayers are participants in markets too. Asking for a return on their investment is about as capitalist as it gets.