The American A.I. Sovereign Wealth Fund Act rests on a sweeping claim about the ownership of value created by artificial intelligence. Because AI models are trained on data generated by the public, the bill treats the resulting gains as a public resource subject to state control and redistribution.
Sen. Bernie Sanders’ (I-Vt.) proposal would require covered AI developers to transfer up to 50% of their corporate value to a new federal sovereign wealth fund. That fund would distribute “dividends” to the public and use its ownership stake to steer AI development “in the public interest.”
The bill therefore raises questions that go well beyond artificial intelligence. It implicates basic principles of value creation, property rights, corporate governance, political choice, and capital formation. Its central premise is that public data gives rise to public ownership. That premise confuses the availability of information with the entrepreneurial and technical process required to transform information into a productive asset.
This piece argues that the Sanders bill rests on four related errors. First, it treats raw data as the source of economic value, while discounting the entrepreneurial discovery and technical judgment that make data useful. Second, it assumes a federal commission can identify and impose a coherent public interest on a technology marked by conflicting preferences and rapid change. Third, it would weaken the market for corporate control and disrupt integrated firm structures that often reduce transaction costs and improve coordination. Fourth, it would distort capital formation by creating confiscation risk and encouraging firms to organize around a political threshold rather than consumer demand.
Public Data and Private Production
The proposal presents itself as a mechanism for fairness and shared progress. Its core mistake is more basic. It confuses raw information with the creative act of economic production. In doing so, it tries to build a state entitlement on a long-discredited labor theory of value, under which value comes from the inputs people contribute rather than from the entrepreneurial judgment that turns those inputs into something useful.
The problem begins with the Sanders bill’s central claim. Because AI models “ingest” publicly available text, images, and code, the public supposedly has a pre-existing ownership claim to the output. That is the economic equivalent of saying that, because a sculptor uses stone from a public quarry, the surrounding community owns half the statue.
Raw data does not become valuable by existing. Most of it is a disorderly mass of text, images, code, signals, errors, jokes, spam, and half-finished thoughts. It becomes useful only when someone finds a way to structure, filter, and recombine it into a functioning system. Economic historian Joel Mokyr calls this “useful knowledge.” In the AI context, useful knowledge does not reside in the ambient data. It comes from the architecture, training methods, engineering choices, capital investment, and commercial judgment that make the data productive.
Israel Kirzner’s work on distributive justice helps explain why this distinction matters. Kirzner argues that economic value does not emerge through the mechanical conversion of inputs into outputs. It often comes through “entrepreneurial alertness,” or the ability to notice an opportunity that others have missed. Discovery, on this view, is not a planned recipe in which inputs predictably produce outputs. It is an act of seeing and acting. “To notice an opportunity worth grasping is to have created something,” Kirzner writes, and “only she (or he) who has noticed the opportunity and has grasped it, and no one else, is responsible for and is to be credited with the discovery.”
The raw web pages, scientific papers, and social media posts that populate the internet were not “embryonic AI models” waiting for a federal incubator. They were scattered fragments of information. The value of an AI model comes from the speculative and technically demanding work needed to make that information functional. By claiming that the public deserves a 50% equity stake because the raw data was generated by “humanity,” the Sanders bill credits a passive public with an entrepreneurial act it did not perform.
Robert Nozick’s critique of “patterned” distribution in “Anarchy, State, and Utopia” exposes the same error from a different angle. Nozick rejected the idea that there is a neutral, unowned social pot of wealth waiting for planners to divide according to some preferred formula. Wealth does not fall from the sky. It enters the world tied to the labor, capital, risk, and judgment of the people who created it. Production and distribution cannot be cleanly separated.
Nozick argued that political planners often rely on “current time-slice” principles of justice. They take a snapshot of market wealth at one moment and demand that it be rearranged to match a preferred pattern, such as equal shares or state ownership. Maintaining that pattern requires constant state intervention in voluntary exchange. As Nozick put it, the state must “interfere with individual choices.”
Forcing an innovative firm to surrender half its equity to a state-controlled fund treats the creators of that technology as resources for the state to harvest. Nozick described this kind of appropriation as a form of forced labor. The state seizes the fruits of rational planning, speculative risk, and creative effort to fund political ends chosen after the fact.
The bill also collides with the legal history of American intellectual property. Adam Mossoff’s work shows why the bill’s “public resource” rhetoric would overturn a long American understanding of patents, copyrights, and other rights in creative production. Sanders’ bill relies on the progressive legal theory that intellectual property is not real property, but a “public right.” On that view, intellectual property is a regulatory privilege granted by the state, which lawmakers may alter, tax, or reclaim whenever they choose.
Mossoff’s research into the American founding and early patent law refutes that account. The Framers of the U.S. Constitution and early American jurists often treated patents and copyrights as property rights rooted in Lockean principles of self-ownership and labor, not as revocable regulatory favors. John Locke argued in “Two Treatises of Government” that each person has property in his own person, and that the labor of his body and the work of his hands are properly his.
The same principle applies when an inventor or software developer uses the mind rather than the hand. When developers organize information into a working AI system, they create something private through judgment, labor, and risk. Reclassifying that creation as a “public resource” would convert a private right into a state-managed commons. That move would invite the calculational chaos and capital flight discussed in the sections that follow.
Democracy by Committee, Capture by Design
To govern this seized technology empire, the Sanders bill proposes a deceptively simple device: a seven-member “Independent Commission for Democratic AI.” This body would manage the sovereign wealth fund’s massive, non-dilutable equity stake and use its voting power to steer AI companies “in the public interest.”
That sounds tidy enough, which is usually the first warning sign. Once viewed through the Virginia School of Political Economy, which applies economic analysis to political decision-making, the bill’s “democratic” premise collapses into public-choice contradictions and practical impossibilities.
The first problem comes from Kenneth Arrow’s “Social Choice and Individual Values.” Arrow’s Impossibility Theorem shows that, when voters face three or more distinct options, no voting rule can reliably convert their individual rankings into a single, consistent “social welfare function” without violating basic democratic conditions, such as non-dictatorship and transitivity. Transitivity means that, if society prefers A to B and B to C, it should also prefer A to C. Real political preferences often refuse to behave so politely.
AI policy offers a clean example. One faction may want rapid technological deployment to accelerate medical research, scientific discovery, or productivity gains, even if the transition disrupts some jobs. Another may favor strict safety protocols, precautionary pauses, and risk avoidance. A third may want AI development structured to preserve existing jobs, even when that raises costs and reduces productivity.
Those preferences do not point to a single, coherent “public interest” for the commission to pursue. They produce trade-offs that different citizens rank differently. As James Buchanan argued in his critiques of social-choice theory, forcing a single, all-or-nothing “social choice” on a diverse public tends to empower whoever controls the agenda. The seven commissioners would not execute the “will of the people.” They would impose their own politically insulated judgments on a large part of the computational economy, while calling the result “social welfare.”
The deeper irony is that the Sanders bill seeks to “democratize” artificial intelligence by replacing consumer choice with centralized political control. A more direct and flexible democratic mechanism already exists in markets. Markets are imperfect, sometimes badly so, but they allow people to register preferences continuously through prices, purchases, subscriptions, cancellations, and switching.
Ludwig von Mises described this process as consumer sovereignty. In a market economy, consumers direct production by choosing what to buy, how much to buy, and what quality they will pay for. Political voting bundles many issues into one all-or-nothing choice. The pricing system allows more variation. It lets different groups pursue different preferences at the same time.
If some consumers want AI models that emphasize privacy and strict content controls, firms have incentives to serve them. If others want speed, flexibility, or fewer restrictions, other firms can compete for that demand. A market can support multiple models, price points, and risk tolerances. The Sanders bill would replace that process with a commission empowered to impose one politicized standard of “safe and ethical AI” on developers and users alike.
Public-choice theory also warns against the comforting fiction of the benevolent expert. Buchanan and Gordon Tullock’s “The Calculus of Consent” argues that politicians and bureaucrats do not shed self-interest when they enter government. They still respond to incentives, pressure, ambition, reputation, ideology, and organized political demands.
That matters because the Sanders commission would include representatives from specific politically organized groups, including labor unions and safety advocates. A body built around organized interests invites bargaining among those interests. Buchanan and Tullock’s analysis predicts logrolling, rent seeking, and factional bargaining. Commissioners would have strong incentives to trade support across issues, protect their constituencies, and convert long-term technological policy into short-term political payments.
The capture problem is worse. Gordon Tullock’s work on rent seeking shows that, when the state can grant or deny valuable economic privileges, firms shift resources away from production and toward political influence. Money that could fund engineers, chips, data centers, or safety testing instead funds lobbyists, lawyers, and compliance departments. The political machine does not run on fairy dust. It runs on billable hours.
A commission with voting control over up to 50% of covered AI firms would become one of the most valuable targets in Washington. Large incumbents would have the strongest incentives and best resources to shape its decisions. They already have regulatory teams, political relationships, and compliance infrastructure that smaller firms lack.
Those incumbents could press the commission to define “ethical AI safety” in ways that match their existing systems, business models, and compliance capacity. Smaller firms, open-source developers, and disruptive entrants would face higher costs and slower approvals. The result would not be democratic control of artificial intelligence. It would be a state-backed cartel dressed up as public oversight.
The Market for Corporate Control Meets the Ministry of AI
Beyond its errors about knowledge and democracy, Sanders’ AI Sovereign Wealth Fund Act would also damage corporate governance and economic calculation. The bill would intervene in two ways. First, it would take up to a 50% equity and voting stake in covered firms. Second, it would force multi-division technology companies to separate foundational AI research from their non-AI commercial businesses.
That may sound like tidy administrative housekeeping. It is anything but. The bill misunderstands what corporate ownership does and why integrated firms exist. In the process, it would weaken managerial discipline, raise transaction costs, and scramble the price signals that guide investment in high-technology markets.
Henry G. Manne’s corporate-governance scholarship explains why a 50% government voting block would damage corporate efficiency. Manne showed that corporate voting shares are not political ballots for social planning. They are financial instruments that help price control over the firm and discipline managers.
The key mechanism is what Manne called the “market for corporate control.” When managers waste resources, miss opportunities, or pursue goals that reduce firm value, the company’s stock price falls. That lower price can attract outside investors who believe they can run the company better. They may launch a hostile takeover, buy undervalued shares, replace management, and restructure the firm. Even the threat of such a takeover pressures managers to control costs, serve consumers, and allocate capital carefully.
The Sanders bill would short-circuit that mechanism. A non-dilutable 50% government voting block would make hostile takeovers and proxy fights practically impossible at covered firms. No outside investor could replace management over the objection of a permanent federal shareholder with veto power.
Managers would therefore have a new audience to please. Their jobs would depend less on operational performance, cost control, or consumer demand, and more on keeping the commission satisfied. Research priorities, content-moderation policies, hiring choices, and product decisions would shift toward political approval. The corporation would become less an engine of wealth creation than a government-protected ward with a very expensive engineering department.
The bill’s forced-separation mandate would add another layer of damage. In “Universal Economics,” Armen Alchian and William R. Allen explain that firms combine complementary assets and business lines to reduce transaction costs. A transaction cost is the cost of negotiating, monitoring, enforcing, and coordinating economic activity. Firms often bring activities under one corporate roof because internal coordination can be cheaper and faster than constant contracting among separate companies.
That logic matters acutely in technology. Foundational AI research does not operate in isolation. It depends on hardware procurement, cloud infrastructure, search data, user-interface design, software distribution, and feedback from commercial products. Companies like Alphabet and Microsoft bundle these functions because separating them would require constant negotiation and contracting among units that now coordinate internally.
Forcing AI divisions into legally separate entities would destroy many of those efficiencies. It would replace internal capital allocation and technical coordination with slower, costlier, and more litigation-prone interfirm contracting. That is a particularly bad trade in a sector where speed, iteration, and cross-functional learning often determine whether a product works at all.
The final problem is economic calculation. In “Man, Economy, and State,” Murray Rothbard builds on Mises’ critique of socialism to argue that rational economic calculation depends on private property and genuine market prices for capital goods. Prices are not decorative numbers on a Bloomberg terminal. They summarize dispersed judgments about risk, scarcity, future demand, and alternative uses of capital.
In a pure market, corporate shares and capital inputs are priced through competitive bids by private owners risking their own money. Those prices let entrepreneurs calculate profit and loss. Profit and loss, in turn, guide resources toward uses that consumers value and away from projects that waste capital.
A government-backed sovereign wealth fund with a 50% equity stake in leading technology firms would distort that process. Its investment and voting decisions would not turn on expected consumer demand or profit and loss. They would turn on political mandates, distributional goals, ideological preferences, and pressure from organized groups.
That distortion would affect more than the covered firms. Stock prices help investors value related companies, suppliers, customers, and competing technologies. Once a massive political shareholder begins steering capital and control rights according to non-market criteria, those prices become less reliable guides. The result would be calculational chaos in AI investment, with capital pushed toward politically favored projects and away from technologies that consumers and businesses would otherwise choose.
The $200 Million Tripwire
To fund its redistributive scheme, the AI Sovereign Wealth Fund Act would impose a 50% equity seizure on AI firms that exceed an arbitrary $200 million capitalization or revenue threshold. The bill treats capital as if it were a pile of cash waiting to be divided. Economists sometimes call this the “fundist” fallacy, or the mistaken view that capital is a homogeneous, liquid, and durable fund that can be taxed, sliced, or liquidated without changing the structure of production.
The Austrian theory of capital, developed by Eugen von Böhm-Bawerk and refined by F.A. Hayek, starts with the opposite premise. Capital is heterogeneous. It consists of specific goods, relationships, skills, and investments arranged across time. In artificial intelligence, that means high-performance graphics-processing-unit (GPU) clusters, specialized fiber-optic networks, proprietary algorithms, cloud contracts, data-center capacity, and engineering teams with hard-to-replicate expertise.
Advanced AI development is a long and uncertain production process. Developers often commit billions of dollars years before those investments produce marketable products. Böhm-Bawerk called this “roundabout” production, meaning that firms first invest in earlier-stage capital goods that later help produce consumer-facing goods and services. Those capital goods are not interchangeable Lego bricks. A GPU cluster built for training frontier models, or a team trained to optimize a specific architecture, cannot be painlessly redeployed if the project suddenly becomes uneconomic.
Sanders’ bill would force an artificial restructuring of these capital arrangements. Firms facing the loss of half their corporate value would see their cost of capital rise sharply. Investors would demand higher returns to compensate for confiscation risk, or they would put their money elsewhere. To satisfy the new constraints, developers would scale back, delay, or abandon long-term research projects. Specialized investments in chips, infrastructure, and engineering talent would lose value. The damage would appear not only in projects killed today, but in discoveries never attempted tomorrow.
The bill compounds that problem by shifting control to surrogate decision-makers. In “Knowledge and Decisions,” Thomas Sowell distinguishes between decisions made by people who bear the costs of their choices and decisions made by insulated actors spending other people’s money. In a market, venture capitalists and technology entrepreneurs face a harsh feedback loop. If they misread demand, choose the wrong software architecture, or misallocate capital, they lose money. That risk forces them to revise expectations, correct errors, and shut down failing projects before they consume still more scarce resources.
The seven commissioners managing the AI Sovereign Wealth Fund would face no comparable discipline. They would make centralized bets on the future of computation with other people’s money. If they pushed capital toward a failing model architecture or forced a company to adopt an inefficient safety protocol, they would not bear the financial losses. The usual market signals of profit and loss would weaken. Errors would last longer, spread further, and impose costs on firms, workers, investors, and consumers who had no meaningful say in the decision.
The bill’s $200 million threshold adds another defect. Gordon Tullock’s work on rent seeking shows that firms waste resources when political decisions determine economic rewards. Instead of investing in production, firms invest in protection. They hire lobbyists, seek exemptions, and design their conduct around political risk. The cost includes not only the money spent on lobbying, but also the innovations that never happen because the expected return has been taxed, threatened, or bargained away.
Under Sanders’ bill, an AI startup valued at $199 million remains a private enterprise. Once it reaches $200 million, it faces a 50% equity seizure. That kind of cliff creates perverse incentives. Firms would slow growth, refuse capital, split into smaller entities, or hold back commercial deployment to avoid crossing the line.
The same threshold would redirect scarce talent. Instead of competing to build better models for customers, growing firms would spend more time lobbying the commission for exemptions, waivers, and favorable treatment. Engineers and executives would be pulled away from product development and into regulatory survival. The sector would trade discovery in the lab for supplication in the hearing room. Not exactly the future of artificial intelligence anyone should be rushing to beta test.
The Fatal Conceit, Now With Equity Shares
The American A.I. Sovereign Wealth Fund Act is a near-perfect specimen of what Hayek called the “fatal conceit.” It assumes that a small committee of political appointees can possess the dispersed, tacit, constantly changing knowledge needed to direct a complex technological system. It also reflects what Sowell called “the vision of the anointed,” or the belief that selected experts have both the moral authority and the practical capacity to make high-stakes choices for everyone else.
The bill’s central claim that artificial intelligence is a “public resource” because models train on public data confuses inputs with production. Raw information does not become valuable by existing. Value emerges when entrepreneurs, engineers, investors, and firms discover ways to organize information into tools people actually use. Treating that discovery as public property would weaken the private rights that support innovation in the first place.
The promise of a “democratically directed” AI commission fares no better. Arrow’s Theorem shows why no committee can convert conflicting public preferences into one coherent “social welfare” choice without smuggling in agenda control. Public-choice theory explains what follows. The commission would not embody the “public interest.” It would become a forum for interest-group bargaining, logrolling, and capture by incumbents with the money and lawyers to work the system.
The bill’s structural mandates would deepen the damage. A permanent 50% government voting block would freeze the market for corporate control, weaken managerial discipline, and turn executives toward political survival rather than consumer demand. Forced separation of AI divisions would break apart business structures that firms created to reduce costs, coordinate research, and move quickly. The result would be slower decisions, higher contracting costs, and worse economic calculation..
The bill’s equity seizure also treats capital as if it were a liquid fund waiting for redistribution. High-technology production does not work that way. AI development depends on specific, time-sensitive investments in chips, infrastructure, software, and engineering talent. A confiscatory threshold would raise capital costs, encourage firms to stay small or split themselves apart, and redirect scarce talent toward lobbying rather than building.
Real progress in artificial intelligence does not require central planning by a federal committee. It requires private property, freedom of contract, open competition, and legal rules that let entrepreneurs test ideas, fail cheaply, and scale when consumers find value in what they build.
The future of artificial intelligence will not be discovered by seven commissioners armed with voting shares and a mission statement. It will be discovered by people free to build, buy, reject, improve, and try again.
The state should not claim ownership over artificial intelligence’s future. It should protect the freedom that makes that future possible.
