Built on All of Us
Bernie Sanders wants the public to own half of every big AI company. Palantir’s CEO says public control is coming whether the industry likes it or not. The fight everyone is having is the wrong one. T
In early June, at Palantir’s AIPCon 10, Alex Karp told an interviewer that he had spent the last six months privately warning the heads of the biggest AI companies about a threat they refused to take seriously. He had told them the government was going to come for their companies. According to Karp, the response from inside the labs was a shrug. They wondered why anyone would nationalize them; they’re so likable and create so much value.
Karp runs the most government-fused technology company in America. Palantir’s business is the state. When the person with the most to lose from nationalization tells you the momentum has shifted toward the people who want it, that is not an opinion. That is a weather report from inside the building.
His warning isn’t a prediction about some distant year. It’s the loud version of something that’s already been underway, quietly, on the government’s books for over a year.
Since January 2025, the federal government has gone on a buying spree. The Council on Foreign Relations counts roughly $20.9 billion across fifteen deals to take equity in private companies, which it calls the biggest such push into strategic industries since the Second World War. The Commerce Department holds about 10 percent of Intel, bought for $8.9 billion in August 2025. The Pentagon holds 15 percent of the rare-earth producer MP Materials, which makes it the company’s single largest shareholder. The government took stakes in Lithium Americas and Trilogy Metals. And in US Steel it secured what is called a golden share: not equity, but permanent veto power over whether the company can move its headquarters, shift production overseas, or close a plant.
None of this happened under a democratic socialist. It happened under a Republican president who calls himself a capitalist, in eighteen months, with the stock market at record highs.
Here is the part that should reframe the whole conversation: taking government equity stakes in semiconductor companies funded by the CHIPS Act was, originally, a Bernie Sanders proposal. Sanders proposed it, and the Trump administration went and did a version of it. The question of whether the state should hold a position in strategically important technology is not a live debate anymore. Both ends of the political spectrum have already answered yes. They just disagree about why.
The bill
On June 1, in a New York Times guest essay, Sanders announced the American AI Sovereign Wealth Fund Act. The mechanics matter, so here they are precisely.
The bill would impose a one-time 50 percent tax on the largest AI companies, including OpenAI, Anthropic, and xAI. Not a tax on profits. A tax paid in stock. Half of each company’s equity would transfer into a federally managed sovereign wealth fund. The government would hold voting shares and, in Sanders’ framing, equal representation on each company’s board, with the power to block decisions it judged harmful to the public. The proceeds would begin as direct payments to Americans and eventually fund health care, education, and housing.
The moral premise is one sentence: AI was built on the collective knowledge of humanity, so the wealth it generates must benefit humanity. Our books, our songs, our code, our journalism, our research, fed into these models without permission and without payment.
It is a clean argument. It is also doing the work of three different arguments at once, and they do not lead to the same place.
Three claims
When someone says the public should benefit from AI because it was built on all of us, they are making one of three claims, and the remedy is different for each.
The first is a claim about compensation. Specific people’s work was taken, sometimes through piracy, and those specific people are owed. This one is already being adjudicated, and the answer is more complicated than “stolen.” In June 2025, Judge William Alsup ruled that training an AI model on lawfully acquired books was fair use, which he described as among the most transformative uses he expected to see. What was not fair use was downloading pirated copies to build the training library. Anthropic settled the piracy claim for $1.5 billion, roughly $3,000 for each of about 500,000 works, paid to the identified rightsholders. That’s what compensation actually looks like: checks in the hands of specific authors. It doesn’t put one in the hands of every American.
The second is a claim about a public resource. AI is like oil, the argument goes, a common inheritance the public should share in. Sanders leans here, citing Norway’s $2 trillion oil fund and Alaska’s dividend. The analogy is weaker than it sounds. Oil is a resource the state already owns and licenses. The “collective knowledge of humanity” is not owned by the state. It is a commons, or it is private copyright, and the courts just said training on it is mostly legal.
The economist Mariana Mazzucato has spent a decade documenting how much of modern technology was paid for by the public before any venture capitalist showed up. The internet came out of DARPA. GPS came from the Navy. The touchscreen traces to CIA-funded work, and the voice assistant to DARPA. Her phrase for the pattern is socialized risk, privatized profit. The public funded the foundation, took the early losses, and watched the gains get privatized and the taxes get avoided. That’s a real grievance with a real paper trail, and it makes an honest case that the public should get back what it put in. It doesn’t obviously make the case for a 50 percent controlling stake.
The third claim is the one I care about most, and it is the one buried deepest. It is a claim about control. Who decides how the most consequential technology in history develops? That is not an economic question. It is a governance question. And it is the one Sanders answers most aggressively, because his board seats and voting shares are not about dividends. They are about control.
You can have a stake without control. The government’s Intel position is passive, with no board seat. You can have control without a stake. The US Steel golden share is pure veto, no dividend. Sanders wants both, in the same companies his own government would also be regulating.
I want to be specific about something
If you build AI, you are not the villain of this piece. If you believe the public deserves a share of what these companies become, you are not the villain either. I think the moral intuition behind the Sanders bill is correct. These companies were built on work that millions of people did, much of it without consent and some of it with public money. The instinct that the gains shouldn’t flow to five men isn’t radical. Most people get there on their own.
The mechanism is the thing I want to trace. Because the mechanism does something the moral intuition does not account for.
The argument everyone is having, and the one nobody is
The argument everyone is having is whether the public should benefit. That argument is effectively over. It is not just Sanders. OpenAI has proposed a public wealth fund that would give every citizen a stake in AI-driven growth. Anthropic has proposed national sovereign wealth funds holding AI assets. Elon Musk, who runs xAI, has said the government should send everyone checks to deal with AI unemployment. When the targets of a 50 percent equity tax are publicly proposing their own versions of public funds, the principle is conceded. The labs are negotiating the terms of their own partial socialization.
The argument almost nobody is having is who holds the lever after that.
Sanders describes a government that uses its board seats and voting power to block decisions that hurt people. Picture that government. Now picture the next one. Then the one after that. Board votes do not belong to a philosophy. They belong to whoever won the last election.
A lever that lets a benevolent administration block a harmful AI deployment is the identical lever that lets a hostile administration compel one. The same seat that blocks a surveillance product can mandate it. The golden share that protects a steel plant from offshoring is the golden share that, pointed at an AI lab, decides which model gets built and which gets buried. We already have the template. It is sitting on US Steel’s cap table.
This is not a hypothetical concern that lives only on the left or only on the right. David Sacks, the administration’s own AI policy lead, attacked the Sanders plan by warning it could open the door to a China-style social credit system. Cato called government equity stakes a Pandora’s box, and noted the uncomfortable mechanics: whichever party opens it hands the other party direct control of the boardroom the moment power changes hands. State-controlled wealth funds are also, around the world, a favorite instrument of authoritarian governments. The concentration of control is the danger.
The trigger
Karp’s warning had a specific causal claim inside it, and it is the part that connects this to the thing most readers actually feel. In March, at the a16z American Dynamism Summit, he told Silicon Valley that displacing all the white-collar workers while also alienating the military would lead, directly, to the nationalization of their technology. Take the jobs, lose the soldiers, and you manufacture the political coalition that takes your company.
The job numbers are contested, and they should be read carefully, because the big forecast number and what’s happening to actual workers don’t tell the same story. The World Economic Forum’s 2025 survey projects 92 million roles displaced and 170 million created by 2030, a net gain of 78 million. That is the optimistic macro frame. Underneath it, 41 percent of employers told the WEF they plan to reduce their workforce where AI can automate tasks. Goldman Sachs found a 16 percent drop in employment for workers aged 22 to 25 in AI-exposed roles, while older workers in the same roles held steady. Cornell researchers found AI-adopting firms cut junior hiring by about 13 percent.
The aggregate may well be net positive. The entry-level door is closing now. And the company whose CEO is warning that displacement triggers nationalization is itself planning to grow its revenue roughly tenfold while shrinking its headcount, having already cut its own IT team from 200 people to under 80. Karp is describing the fire while holding a match. He is right anyway.
What this looks like from where I sit
I manage technology infrastructure for a global academic community. When I think about what Sanders’ mechanism would mean, I do not have to reach for theory.
Imagine the auditor who certifies your systems also owns half your shop, sits on your board, and holds a vote on what you build. In my world, we don’t call that oversight. We call it a conflict of interest. The entire discipline of governance exists to keep those roles apart. Sanders’ bill fuses them on purpose, and calls the fusion a feature.
That is the move I cannot get past. The state would write the AI safety rules, enforce the AI safety rules, own half the companies subject to the AI safety rules, and vote on their boards. Each of those powers is defensible alone. Stacked, they are the most concentrated grip on a general-purpose technology any government has held since the railroads, and we would be installing it on the systems that increasingly mediate what people read, believe, and are allowed to say. Stack all four powers on the same body and it stops being oversight.
What I Don’t Have Answers To
I don’t know how to give the public a genuine stake in AI without handing the state a control lever I would not trust in every hand that is going to hold it. That is the whole problem, and I cannot resolve it. A passive fund that stays out of the boardroom avoids the conflict, but then it delivers a check, not a say, and the say was the entire point.
I don’t know how you compensate the millions of people whose work trained these models when the courts have ruled that the training itself was mostly legal. The piracy settlements reach the authors whose books were in a known pirated library. They do not reach the open-web writers, the forum posters, the photographers, the ordinary people whose decades of output are in the weights and will never be itemized.
I don’t know what counts as an AI company in five years. Microsoft, Google, Amazon, and Nvidia all build AI and none of them is a pure-play lab. A 50 percent equity tax has to draw a line, and every line invites a corporate restructuring designed to step over it.
And I don’t have a clean answer to Karp, which bothers me, because I do not like agreeing with him. He may be right that some form of public claim on AI is now inevitable. If he is, then the only question left is the one this piece is about. And the companies are already answering it for us, while the rest of us argue about whether it’s even a question.
What you can do
If you sit anywhere near AI procurement, governance, or risk, the useful work right now is not deciding whether you are for or against public ownership. It’s pressure-testing the mechanism. When one of these proposals lands on your desk, ask the questions. Is the stake passive or controlling? Who picks the board members, and what happens to their seats when the administration turns over? Is the regulator also the owner? And the one that decides it: does the design still hold when the person you trust least is the one holding it?
That last question is the only test that really lasts. Build for the worst hand that will ever hold the lever, because eventually it will.
Rachel Ankerholz is an IT Director and writer exploring the intersection of AI ethics, accessibility, and human-centered technology. She writes about who gets included, and who gets left behind, when we build systems.