This TaxProf Op-Ed on whether the government should control the major AI firms is by Reuven S. Avi-Yonah (Michigan):

Should the Government Control the Major AI Firms?

Reuven Avi-Yonah¹

Senator Bernie Sanders (I-VT) has proposed recently that the government should take an equity interest in the largest AI firms and use it to fund a $7 trillion sovereign wealth fund.² The proposed American AI Sovereign Wealth Fund Act would—

1. Require the largest AI companies to pay a one-time tax of 50 percent of their equity to the American AI Sovereign Wealth Fund. The tax would apply to new AI companies when they become sufficiently large to qualify (i.e., record $200 million in annual AI sales). 
2. Require the largest AI companies that operate both AI and non-AI businesses to separate those businesses, ensuring the public receives an ownership stake in only the AI business. 
3. The American AI Sovereign Wealth Fund would be run by the Independent Commission for Democratic AI, which:
   • Consists of 7 bipartisan Commissioners nominated by the President and confirmed by the Senate, based on a list of names provided by Congress. 
   • The Commissioners would be mandated to promote the goals of worker welfare, public safety, fair competition, environmental sustainability, and financial solvency. 
   • The Fund would pay out an annual dividend of 5% of its value to be used for direct payments to the American people as well as other measures to ensure every American has a high standard of living, including access to health care, education, and housing.³

This legislative proposal is based on the work of Profs. Jeremy Bearer-Friend and Sarah Polcz, who have suggested that instead of imposing a traditional corporate tax, the government should be granted shares in the AI companies, primarily as a way of raising additional revenues to address the additional costs of mass layoffs and to reduce other public harms that can result from deploying AI.⁴ They summarize their proposal as follows:

While those are huge problems for the public, private shareholders feel little obligation to address them. How, then, to manage the rising harms of AI? Publicly share the rewards of AI with partial public ownership. Under our proposal, recently published in the Columbia Journal of Tax Law, systemically important AI firms would pay a new tax with stock, not cash. This stock would not only convey financial value to a publicly managed trust, but it would also grant limited governance powers to the public.

At its core, our AI tax would give the public fractional ownership, alongside private investors, in large AI firms. There are many appealing features to an AI tax paid with stock. The public would at last be paid back for all the data that was rightfully theirs to begin with, before it was scraped from the internet. The public would also have a greater voice in the future decisions of these firms, allowing for the public interest to be weighed in the context of corporate governance tools.

Our tax also has the added benefit of raising a lot of revenue. If taxes collected on wages plummet at the same time that the public need for services rise[s], the AI tax brings in a new public funding source.⁵

Bearer-Friend and Polcz also address some counterarguments, such as the cost of the investment and potential conflicts of interest. They argue that unlike the administration’s industrial policy through investment in, for example, Intel, the proposed tax would not require the government to bear additional expenses because the shares would be issued for free, and because it would be enacted by Congress, it would be more democratically accountable, transparent, and subject to conflict-of-interest rules.⁶

Bearer-Friend and Polcz point out that there are some precedents for this proposal:

There are clear precedents for publicly managed funds that own shares in private enterprise within a capitalist economy. Public pension funds, like CALPERS, hold hundreds of billions of ownership in private companies. Sovereign wealth funds in places as varied as Norway, Abu Dhabi and Alaska also operate like this.

Under our proposed framework, the equity paid would be in exact proportion to the share classes already outstanding, so no new forms of securities would need to be created. This also aligns the Treasury with investors who have a stake in preserving the value of their shares. No longer could a company turn to shareholders to say how much money it made, and then turn to the IRS and say it made nothing.⁷

This proposal is not entirely new. Dean Baker has argued that the corporate tax should be replaced with government ownership of the appropriate percentage of the corporation (currently 21 percent).⁸ His argument was that the corporate tax can be described as a partnership between the government and the corporation: If the corporation is profitable, the government gets its 21 percent share, and if the corporation loses money, the government gives it a net operating loss that offsets future tax liabilities. Thus, the government is already exposed to corporate losses. In the case of an equity stake, any loss does not have to be realized, and therefore there is no immediate liability and (as happened in the GM case) the government may ultimately be able to sell at a gain.

The government investment in Intel has been roundly criticized as an “assault on capitalism” because it is similar to the Chinese state-owned enterprises, and “the heavy hand of the state tends to lead to problems over time, including corruption, inefficiency and a reluctance to let bad companies fail.”⁹ Mike Schmidt and Todd Fisher, who ran the Commerce Department CHIPS program (Creating Helpful Incentives to Produce Semiconductors) in the Biden administration, also criticized the transaction because:

• Intel’s chips are not a national security priority;
• Intel has no problem raising funds in the private equity market;
• “transferring grants into equity . . . risks putting Intel at a cost disadvantage relative to other chip makers, which are manufacturing predominantly in low-cost Asian countries and receiving direct incentives in the U.S. and around the world”; and
• it raises potential conflicts of interest because “when policymakers set semiconductor strategy, will they be doing so as national stewards or as corporate shareholders?”¹⁰

The Wall Street Journal criticized the transaction because of the risk that “the government will direct Intel’s business in a way that will make it harder for the company to become competitive in chip manufacturing and design” and accused the administration of emulating China.¹¹ The Financial Times wrote that “Trump’s arbitrary interventions in the economy are more redolent of the ways that despotic regimes, such as Russia or China, operate rather than the traditional practices of the global champion of free markets.”¹² Steven Rattner, who ran the Obama administration’s takeover of General Motors, wrote that “Mr. Trump is muscling his way into our economy in ways that are alien to traditional Republican principles, alien even to what more interventionist Democrats have argued for, and wildly at odds with any sensible notion of how the relationship between government and business should be managed. . . . All of that can reasonably be called extortion.”¹³

Some of these arguments seem exaggerated.¹⁴ The loss argument is mitigated by the fact that the government need not realize its losses. The free-market argument has to contend with the fact that the chip market is not free because it is currently dominated by the Taiwan Semiconductor Manufacturing Co. (TSMC), which also makes chips for Nvidia and Intel. As the Financial Times noted:

Since its creation in 1987, TSMC has benefited from massive government support and partnered with scores of local investors, companies and educational institutions, weaving an extraordinarily intricate innovative web. TSMC is far more than a private company; it has been described as a project of the Taiwanese state.¹⁵

In addition, it can be argued that leaving a foreign corporation (especially one so vulnerable to China) in charge of such a vital ingredient is a national security risk that justifies government intervention. Moreover, Intel is already subsidized via the CHIP grants.¹⁶

How do these arguments apply to the Bearer-Friend and Polcz proposal?

AI is clearly more of a free market than chips, but it is still dominated by a few mega-corporations, and the number could shrink. National security is no less or even more important for AI. As for the concern about conflicts of interest, that is somewhat mitigated by the need for Congress to pass legislation. There are, however, some issues with the proposal.

First, the focus on revenue assumes that revenue can be generated. Even if the value of the shares goes up, because the AI companies do not typically pay dividends, to generate revenue the government will have to sell shares, reducing its percentage, and so ultimately there may be no revenue left even if the government keeps a “golden share” with control rights. That is the corollary to the government not being exposed to losses because it does not have to sell. Alternatively, the government could use the golden share to force a dividend, but that seems inconsistent with the focus on using current corporate governance rules. Or it could collect more shares as a tax each year, but that seems strange in years that the company does not have profits and could completely dilute the private shareholders.

Second, the government may be exposed to the harms caused by AI. Admittedly, as a legal matter, shareholders are generally not liable for the debts, torts, or crimes of the companies they own, and while board members can be liable, often insurance policies cover them. Moreover, rules of sovereign immunity would still apply, although they are not absolute (for example, the Foreign Sovereign Immunity Act does not apply to commercial activities).¹⁷ But the main danger is political: If significant harm arises to the public, voters may blame the government for not stopping it, which could disincentivize politicians from making the investment. The government is not like the California Public Employees’ Retirement System, because it has to face the voters.

Third, the key advantage of the proposal is its regulatory potential, which is the main reason to have a corporate tax.¹⁸ From this perspective it is crucial to give the government a “golden share” that enables it to exert regulatory control over the AI companies. That is similar to the way the United Kingdom and other countries regulate privatized public utilities and other monopolies, as well as the U.S. golden share in US Steel.¹⁹ But there is no golden share in the Bearer-Friend and Polcz proposal, just limited governance rights.

There is also a more recent example of the government taking equity in a class of privately held companies, namely the ones engaged in developing quantum computing. The Economist writes that— 

These days, quantum computers—which exploit quantum mechanics to perform some calculations far faster than ordinary computers—not only exist, but are attracting serious attention from investors. The market values of IonQ and Rigetti, two firms which have been listed since 2021 and 2022 respectively, are up seven- and four-fold since their debuts. In April McKinsey, a consultancy, reported that the amount of money invested in quantum startups had reached $12.6bn in 2025—a six-fold increase on the year before. The following month America’s government said it would take $2bn-worth of equity stakes in nine quantum-computing companies, including Rigetti, GlobalFoundries, a chipmaker, and Quantinuum, a firm based in Colorado that raised around $1.7bn when it went public in June.²⁰

The Economist article treats this move primarily as an investment opportunity for the government, but it is more likely to be about control, since quantum computing poses a serious threat to national security by undermining the encryption methods that currently protect the internet. The same article points out that—

In 1994 Peter Shor, an American mathematician, worked out how to reduce the time taken to break many types of encryption from billions of years to hours or less. The only snag was that “Shor’s algorithm”, as it is now known, required something that at the time only existed on university blackboards: a quantum computer.²¹

And that this is now a real threat given that quantum computing exists. 

Moreover, quantum computing and AI are closely linked, as pointed out in another article in the same issue of The Economist:

There are three ways that AI and quantum computers overlap: first, in the problems they try to solve; second, in the way each can be used to build the other; and third, in the resources for which they are competing. And in each of those, co-operation seems to be the winning strategy.²²

Moreover, the government investment in quantum computing already impacts AI as well, because some of the main companies that invest in AI (e.g., IBM and Google) also have quantum computing labs.  The Economist notes that “already in 2013 Google and NASA, America’s space agency, launched a quantum computing lab. The idea was to use a specialized quantum computer built by D-Wave, an American firm, to improve the onerous task of training of machine-learning algorithms. . . . And more recently, IBM and D-Wave were among several quantum labs in which America’s government invested $2bn in May.”²³

So, if the government is going to control quantum computing firms, it should also move to obtain a golden share in the leading AI firms, both because of the overlap and because AI poses as much of a national security risk as quantum computing. 

This proposal does have a historical precedent. As noted by William Novak, “In the late nineteenth and early twentieth centuries . . . lawyers, economists, legislators, and democratic reformers pieced together a new regime of modern business regulation. At the center of that project was the idea of the ‘public utility’ or ‘public service’ corporation.”²⁴

In Novak’s opinion, this innovation was even more important than antitrust in establishing democratic control over corporations. He writes that:

The modern American administrative and regulatory state was built directly on the legal foundation laid by the expanding conception of the essentially public services provided by corporations in the dominant sectors of the American economy: transportation, communications, energy supply, water supply, and the shipping and storage of agricultural product. In law, the original architects of the administrative state, the authors of the very first casebooks, and the teachers of the first classes on administrative and regulatory law — people like Bruce Wyman, Felix Frankfurter, and, ultimately, James Landis — basically cut their teeth on the legal, political, and economic problems posed by public service corporations and public utilities. The public utility, the public corporation, and the modern American administrative and regulatory state, in other words, all grew up together.²⁵

In some ways, the idea of treating the biggest corporations as public utilities went far beyond what is envisaged by Bearer-Friend and Polcz. As Novak explains, it was the basis for “comprehensive price and rate controls, ongoing administrative and bureaucratic supervision, municipal ownership, and, ultimately, public works. It culminated in unprecedented interventions like World War I’s Food Administration (initially justified by the idea that in times of war all businesses were ‘affected with a public interest’), World War II’s Office of Price Administration and the Tennessee Valley Authority.”²⁶ Nor was it limited to traditional public utilities: “The concept itself constantly expanded beyond early initiatives in special areas like transportation, communications, energy, and water supply to the regulation of things like hotels, warehouses, stockyards, ice plants, insurance, milk . . . you name it.”²⁷

Arguably, AI represents both a major opportunity but also a major threat, as illustrated by recent developments like Anthropic’s decision to share its Mythos program with its largest competitors because Mythos could hack into their software.²⁸ But Mythos has already been leaked, and it could be used by adversaries or even terrorists.²⁹ The litigation between Elon Musk and OpenAI is a reflection of this dichotomy between the dangers and opportunities posed by artificial general intelligence (AGI).³⁰

If as expected AI is about to rise to the level of AGI in the next few years (an event called the Singularity³¹), and AGI has the potential to inflict massive damage on humans, this problem requires a radical solution. The Bearer-Friend and Polcz proposal is a good beginning, but it may be that stronger solutions such as the public utility approach of the New Deal are necessary, and given how fast AI is developing toward the Singularity, the sooner the better.³²

1. Irwin I. Cohn Professor of Law, the University of Michigan. This paper is based in part on Reuven S. Avi-Yonah & Herbert Snitz, Taxing AI in the Wake of the Emergence of Agentic AI, 192 Tax Notes Fed. 275 (July 13, 2026). ↩︎
2. Sanders Introduces Legislation to Create $7 Trillion AI … ↩︎
3. Read the bill text here. Read a summary of the bill here. ↩︎
4. Jeremy Bearer-Friend and Sarah Polcz, “Sharing the Algorithm,” 17 Colum. J. Tax. L. 1 (2025). ↩︎
5. Bearer-Friend and Polcz, “Let the Public Share in the Rewards of Artificial Intelligence,” The Hill (Apr. 29, 2026). ↩︎
6. Id. ↩︎
7. Id. ↩︎
8. See Dean Baker, “How to Think About the Corporate Income Tax,” Center for Economic and Policy Research (Aug. 26, 2014); Baker, “Instead of Taxes, Make Corporations Give the Government Stock,” Center for Economic and Policy Research (Apr. 10, 2017). ↩︎
9. Bret Stephens, “Donald Trump’s Assault on Capitalism,” N.Y. Times (Aug. 26, 2025). ↩︎
10. Mike Schmidt & Todd Fisher, “Uncle Sam Shouldn’t Own Intel Stock,” Wall St. J. (Aug. 24, 2025). ↩︎
11. Editorial Board, “And Now a New Tax on Patents?” Wall St. J. (Aug. 24, 2025). ↩︎
12. Editorial Board, “The White House’s Flawed Foray into Chipmaking,” Fin. Times (Aug. 27, 2025). ↩︎
13. Steven Rattner, “What’s Next, Comrade Trump?,” N.Y. Times (Aug. 28, 2025). ↩︎
14. See Avi-Yonah, “Corporate Taxation and Industrial Policy,” 120 Tax Notes Int’l 155 (Oct. 6, 2025). ↩︎
15. Financial Times, supra. ↩︎
16. U.S. National Science Foundation, “CHIPS and Science”; “Two Years Later: Funding From CHIPS and Science Act Creating Quality Jobs, Growing Local Economies, and Bringing Semiconductor Manufacturing Back to America,” U.S. Department of Commerce blog, Aug. 9, 2024. ↩︎
17. Foreign Sovereign Immunities Act (FSIA) of 1976, 28 U.S.C. sections 1330, 1602-1611. ↩︎
18. See Avi-Yonah, “A New Corporate Tax,” Tax Notes Int’l, July 27, 2020, p. 497; Avi-Yonah and Lior Frank, “Antitrust and the Corporate Tax: Why We Need Progressive Corporate Tax Rates,” Tax Notes Federal, May 18, 2020, p. 1199; Avi-Yonah, “Corporations, Society and the State: A Defense of the Corporate Tax,” 90 Va. L. Rev. 1193 (2004); Avi-Yonah, “Corporate Tax: Best Tool for Taxation’s Regulation Goal,” Tax Notes Int’l, June 16, 2025, p. 1751. ↩︎
19. Rajeev Dhir, “Understanding Golden Shares: Key Benefits and Real-World Examples,” Investopedia, May 21, 2026; Steve Inskeep, “A Look at the ‘Golden Share’ Agreement in the U.S. Steel-Nippon Steel Partnership,” NPR, June 18, 2025; Naveen Thomas, “Golden Shares and Social Enterprise,” 12 Harv. Bus. L. Rev. 157 (2022). ↩︎
20. The Economist, Quantum Computers Promise Mathematical Superpowers (July 29, 2026). ↩︎
21. Id. ↩︎
22. The Economist, AI and Quantum Computers Will Be Frenemies (July 29, 2026). ↩︎
23. Id. ↩︎
24. William J. Novak, “The Public Utility Idea and the Origins of Modern Business Regulation” in Corporations and American Democracy (2017). ↩︎
25. Id. ↩︎
26. Id. ↩︎
27. Id. ↩︎
28. Brett J. Goldstein, “Your Passwords Are Probably Screwed,” N.Y. Times (Apr. 28, 2026). ↩︎
29. Thomas J. Friedman, “Anthropic’s Restraint Is a Terrifying Warning Sign,” N.Y. Times (Apr. 7, 2026). ↩︎
30. On the litigation, see, e.g., Cade Metz and Mike Isaac, “Elon Musk’s A.I. Claims of Danger Face Limits in OpenAI Trial,” The New York Times, Apr. 30, 2026. Musk lost on statute of limitation grounds, but the issue will not go away. ↩︎
31. Ray Kurzweil, The Singularity is Near (2005). ↩︎
32. See The Economist, “How to Share the AI Windfall” (May 14, 2026). ↩︎