Should the Government Own the Brain? The Case for Nationalizing AI
A senator wants the public to own 50 percent of the biggest AI companies. A bipartisan commission wants a Manhattan Project. A security legend says nationalize the labs if the money runs out. The four mechanisms that could put the state inside frontier AI — and the strongest case against each.
By MyAudioBooks.ai ·
Listen free: Should the Government Own the Brain? The Case for Nationalizing AI
On June eighteenth, twenty twenty-six, a United States senator introduced a bill that would have been unthinkable five years ago: a one-time fifty percent tax on the stock of America's biggest artificial-intelligence companies, with the shares deposited into a sovereign wealth fund owned by the public. At current valuations, the fund would be worth roughly seven trillion dollars. A five percent annual dividend from it, the sponsor calculates, could send more than a thousand dollars a year to every person in America. The bill is called the American AI Sovereign Wealth Fund Act, and its logic is stated in the plainest possible terms: the foundation of AI is the collective knowledge of humanity, and when a public resource generates wealth, the public should share in it.
The bill will not pass this Congress. But that is almost beside the point, because the bill is only the bluntest entry in a debate that has quietly moved from the fringe to the center of American policy thinking. In November twenty twenty-four, the bipartisan US-China Economic and Security Review Commission formally recommended that Congress establish a Manhattan Project-style program to develop artificial general intelligence — the language of wartime mobilization, from the government's own China watchdog. In August twenty twenty-six, one of the most respected security technologists alive, Bruce Schneier, published an essay arguing that if the markets cannot sustain the frontier AI companies, the United States should nationalize them and run them as public laboratories. And beneath all of it, a quieter school of thought argues that the nationalization of AI has already begun — not as ownership, but as control, arriving license by license, audit by audit, export rule by export rule.
The question of this article is no longer whether the state should have a role in frontier AI. That argument is over; everyone in it now accepts some version of yes. The question is which of four very different mechanisms the state will use — ownership, mobilization, receivership, or steering — and each mechanism has a serious case and a serious way of failing. This is the case for nationalizing AI, argued as strongly as its advocates can make it, and the case against, argued as strongly as its critics can. The documents have filed their testimony.
At My Audio Books dot A I, you can create your own audiobooks from prompts, turn your documents into audio, all with one subscription, and store your items in your own personal library.
The case for, first, in its strongest form. It begins with a claim about what AI actually is. Every large model is trained on the accumulated text, images, code, and science of humanity — books written by millions of authors, research funded for decades by taxpayers, a commons scraped at industrial scale. On the Sanders bill's theory, the companies did not create this raw material; they enclosed it, the way earlier generations enclosed common land. If the training data was the public's, then the value refined from it has a public claim attached, and a fifty percent public stake is not expropriation but rent collection. The mechanism is deliberately not a tax in the ordinary sense, and its architects point to a long shelf of working precedents. More than a hundred sovereign wealth funds exist around the world, from Norway's trillion-and-a-half-dollar oil fund — which owns a slice of nearly every public company on Earth and pays for the world's most admired welfare state — to the Alaska Permanent Fund, which has mailed an annual dividend check to every Alaskan for half a century, on the same theory the bill now extends: the resource under the ground, or in the commons, belongs to everyone. The United States has run this experiment before on its own soil and called it normal. The only new element is the resource.
And what is the resource? The bill's answer reaches back five hundred years, to the original enclosure movement, when the common pastures of England were fenced into private estates. The villagers kept grazing rights in law for a while, and then they did not, and the wool made other people rich. The word enclosure has carried the scent of that injustice ever since: the conversion of a shared commons into private property by the party with the power to build the fence. Every large AI model is trained on the accumulated text, images, code, and science of humanity — and the enclosure argument says the fence was built around our pasture. The mechanism is deliberately not a tax in the ordinary sense: the government does not take revenue and spend it into the general budget. It takes shares — voting shares — and holds them in a fund governed by a seven-member commission, nominated by the president, confirmed by the Senate, drawn from a bipartisan list provided by Congress. The commission would use those votes to block corporate decisions that hurt the public and push ones that help it. And the bill contains a quieter provision with enormous implications: companies operating both AI and non-AI businesses must break themselves apart, so the public's ownership attaches to the AI business specifically, undiluted by cloud hosting or advertising or retail.
The second mechanism is mobilization, and its document is the US-China commission's recommendation. The Manhattan Project framing is chosen with care, because it implies a specific theory of the problem: that artificial general intelligence is a strategic technology on the order of nuclear weapons, that the first country to field it gains a durable and possibly decisive advantage, and that leaving its development to quarterly-earnings logic is a category error — like leaving the atomic bomb to Westinghouse. The mobilization case does not require believing the companies are villainous. It requires believing that the stakes are national, the timelines are short, and the competitor — China, with its own state-directed labs — is not waiting for American market cycles to bottom out. On this view, the government does not need to own the labs. It needs to fund and direct the mission at a scale private capital cannot match, the way it did for fission, the space race, and the internet itself.
It is worth remembering what the original Manhattan Project actually was, because the mythology flatters it. It was not a nationalization — the government did not seize the physics industry; there was no physics industry. It was a mobilization: the state identified a mission of existential weight, gathered the world's best scientists — many of them refugees from the enemy's own universities — gave them effectively unlimited resources and a single directive, and then, crucially, got out of the way of the science. The bomb took three years and two billion dollars. The Apollo program, the nearest successor, took eight years and twenty-five billion. The pattern those programs share is the pattern the commission is invoking: when the mission is singular and the timeline is short, the state's combination of unlimited money, talent conscription, and mission clarity has historically outrun the market. The open question the commission's report does not answer is whether artificial intelligence is that kind of problem — a single device to be built once — or a general-purpose technology more like electricity, which no Manhattan Project could have produced.
The third mechanism is the darkest and, in a strange way, the most conservative: receivership. Schneier's argument starts from an observation the boosters prefer to skip: the frontier AI companies are burning capital at historic rates with no demonstrated path to the sustainable profitability their valuations assume. If the markets eventually refuse to keep funding the burn — if the bubble deflates — the capabilities do not disappear, but the companies might. And at that moment, the United States would face a choice between watching its frontier AI capacity collapse, get carved up by whoever has cash, or be acquired with strings attached by foreign capital — or taking it into public hands and running it the way it runs the national laboratories, openly, under democratic oversight, for public benefit. Receivership, on this theory, is not socialism. It is the resolution of a failed systemically important institution, the AI version of the mortgage-crisis rescues: the state steps in not because it wanted to, but because the alternative is worse.
And what would a public AI laboratory actually look like? America has run the experiment before, under a different name. For most of the twentieth century, Bell Labs sat inside a regulated telephone monopoly and produced the transistor, the laser, information theory, and the Unix operating system — arguably the densest run of foundational invention in industrial history — precisely because its researchers were insulated from quarterly earnings and mandated to pursue long problems. The national laboratories that grew out of the Manhattan Project carried the same design into physics: Los Alamos and its siblings still exist, still employ world-class scientists, still operate on decade timescales under public mission statements. The receivership scenario imagines a frontier AI lab converted to that template: the capabilities preserved, the research agenda opened, the profits — if any — flowing to the public that underwrote the rescue. Whether that template survives contact with a field whose best researchers command professional-athlete salaries is the honest unknown at the center of the plan.
The fourth mechanism is the one already running, and it needs no legislation at all. Call it steering, or soft nationalization: the accumulation of export controls on advanced chips, licensing regimes for model weights, security reviews, compute-reporting rules, and procurement standards that, taken together, give the government a hand on the industry's tiller without a single share changing hands. The reason steering works without ownership is the chokepoint — the narrow passage that everything must flow through and that a regulator can grip. The AI industry is built on several of them. The most advanced chips are made by essentially one company on one island. The machines that make those chips are made by one company in the Netherlands. The frontier models are trained by a dozen firms, all American or Chinese, all dependent on the same silicon. A state that controls the chokepoints does not need to own the river; it owns the locks. Every export license, every compute report, every security review is a hand on a lock gate — and the hands have been multiplying. The analysts who describe this path note its elegance: control without ownership, direction without the political cost of the word nationalization, and plausible deniability about who is really making the decisions. If you want to know what nationalized AI looks like in practice, they say, do not wait for a bill. Read the Federal Register.
There is also a fifth ghost in the debate, and it haunts every one of the four mechanisms: the public utility. When a technology becomes both essential and infrastructural — electricity, telephones, railroads before them — America has historically answered not with ownership but with a bargain: the provider keeps its private status and its profits, and in exchange submits to regulated prices, universal-service obligations, and oversight of its terms. The utility model is the quiet compromise between the market and the state, and pieces of it are already visible in the AI debate: the talk of compute as essential infrastructure, of grid access for data centers, of models too important to be left to terms of service. Nobody has yet proposed the formal declaration — AI as a regulated utility — but the intellectual scaffolding is assembled, and it is the compromise position everyone will discover in the end if the harder options fail.
At My Audio Books dot A I, you can listen to this story and thousands of others that explore the hidden science and mechanics behind the headlines.
Now the strongest case against nationalizing AI, and it must be stated at full strength, because it is formidable. The first objection is constitutional and practical at once: a fifty percent stock seizure is a taking, and the litigation would consume a decade; meanwhile the very announcement would crater the valuations being taxed, so the seven trillion dollar fund would begin by shrinking itself — the tax eats its own base. The second objection is about competence: the state is a good funder of missions and a poor operator of products, and a nationalized AI lab would inherit every pathology of public administration — salary caps that cannot compete for the hundred researchers who matter, procurement rules that turn a chip order into a biennium, and political appointees directing research priorities. The third objection is the one that keeps the nationalists honest: China is not pausing. A United States that tied its frontier labs into regulatory knots while Beijing's labs run flat out has not nationalized AI; it has merely relocated it. And the fourth objection is the deepest, because it applies to the gentle mechanisms too: capture. A government powerful enough to steer the most important technology of the century is powerful enough to be steered by it, and the line between the state regulating the labs and the labs staffing the state is the line that every previous technology regime — railroads, banks, telecoms, defense — has eventually blurred beyond recognition. Soft nationalization, the critics note, has all of capture's risks and none of ownership's accountability: decisions made invisibly, by nobody anyone elected, under authorities nobody voted on.
One more number belongs in the debate before the synthesis, because it disciplines every mechanism: the money is already public-adjacent in ways few appreciate. The data centers the frontier depends on are financed by debt raised against the credit of a handful of companies, much of it ultimately held by pension funds and index funds owned by ordinary savers; the chips are subsidized by legislation whose stated purpose is national security; the research lineage runs through decades of taxpayer-funded science at public universities. The public is already inside the machine — as creditor, as subsidizer, as the unpaid source of the training commons. The question the four mechanisms circle is not really whether the public gets involved. It is whether the public's involvement ever shows up on the asset side of its own ledger.
So which mechanism, if any, actually fits the problem? It helps to notice that the four are not really answers to the same question. Ownership answers the distribution question: who gets the wealth? Mobilization answers the security question: who gets there first? Receivership answers the stability question: what happens if the money stops? Steering answers the control question: who sets the rules? A serious policy could answer all four at once — and that is, in fact, the direction events are drifting, piecemeal, without anyone choosing it. Chips are already controlled. Funding is already flowing to domestic compute. Safety review is already arriving. The only piece missing is the public's share of the upside, and that is the piece the Sanders bill puts on the table, however far from passage it sits today.
Three findings would disprove one mechanism or crown another, stated as concretely as the arguments. First, if the frontier companies reach durable profitability at anything like their current valuations, the receivership scenario evaporates and the enclosure argument loses its urgency — the market will have answered the ownership question by making the tax base real and taxable in the ordinary way. Second, if a frontier lab actually fails or requires rescue, the receivership debate stops being theoretical overnight, and the terms of that rescue — equity, control, public access — will set the precedent for everything after. Third, if China's state-directed labs demonstrably pull ahead on capability while America debates, the mobilization argument absorbs all the others, and the question shifts from whether to nationalize in some form to how fast the mobilization can be assembled. Each of these is observable, and none requires waiting long.
It is worth saying what this article has not claimed. It has not claimed the Sanders bill is good policy, or that it will pass; it is a marker of where the debate has moved, not a forecast. It has not claimed the companies stole the training data in a legal sense — the copyright courts are separately deciding what was licensed, what was fair use, and what was piracy, and those are different questions from the political theory of enclosure. It has not claimed nationalization has a good track record; the record is mixed enough to fuel both sides, which is exactly why the mechanism matters more than the slogan. And it has not claimed the quiet option is sinister; steering may be the least-bad available path, which is a different thing from being good.
Which returns to the seven trillion dollars, because the number is the tell. Whether or not a share of it ever lands in a public fund, the fact that a sitting senator can propose public ownership of half of it — and have the proposal treated as a serious contribution to the debate rather than a curiosity — measures how far the Overton window has traveled in three years. The Manhattan Project comparison has been made by the government's own commission. The receivership scenario has been gamed out by the security community's most careful voices. The steering is already underway in the registers and the rulemakings. The debate is no longer about whether the state touches the most powerful technology ever built. It is about which hand, which grip, and whose benefit — and that debate is, at last, happening in public.
At My Audio Books dot A I, you can create fiction, non-fiction, and turn your documents into audio, all stored in one place with a single subscription — plus get instant access to thousands of audiobooks and deep-dive investigations. Learn more today at My Audio Books dot A I.