
There is a new idea circulating in Washington that cuts across ordinary partisan lines: the federal government should take an equity interest in major artificial intelligence companies so the public can share in the gains from AI.
The Case for Capturing the Gains from AI
The instinct behind the idea is understandable. Artificial intelligence is not being created in a vacuum. AI systems draw from the accumulated knowledge of society: public research, open scientific work, software, writing, images, data, language, law, medicine, engineering, and countless other forms of human knowledge. Much of that knowledge was created by people who will not be directly compensated when AI systems use it. Some of it was supported by government-funded research. Some of it came from public institutions. Much of it came from the creative and intellectual work of millions of people over generations.
So there is a legitimate question: if AI produces enormous private wealth, should the public share in some of that gain?
Yes. But government equity ownership is the wrong way to do it.
The Case Against Government Equity Ownership of AI
The first problem is that it is not obvious where the gains from AI will actually accrue. It is possible that a few AI companies will become immensely profitable and capture a large share of the value they create. But it is also possible that competition will drive the market price of AI assistance very low. If that happens, the companies that spend the most building AI systems may not capture anything close to the full social value of what they produce.
History offers a useful warning. The railroads of the nineteenth century created enormous economic value. They opened markets, reduced transportation costs, changed settlement patterns, and increased productivity across the economy. But that did not mean every railroad investor earned extraordinary returns. In many cases, competition, overbuilding, debt, and financial instability shifted the benefits away from the original investors and toward shippers, consumers, landowners, and the broader economy. Many of these railroads went bankrupt.
AI could follow a similar pattern. The largest gains may not remain with the AI model companies. They may flow to software firms, chip companies, cloud providers, manufacturers, hospitals, banks, law firms, schools, small businesses, workers, consumers, and investors across the economy. If AI becomes a cheap general-purpose tool, much of the value may be captured by those who use AI rather than by those who build the underlying models.
This matters because a government equity stake in selected AI firms is a narrow and speculative instrument. It requires the government to decide which companies are likely to capture the future rents from AI. It also requires the government to value highly uncertain firms, negotiate ownership terms, and then manage the conflict between being a regulator and being a shareholder.
This is not a small problem. The federal government should regulate AI in the public interest. It should be concerned about safety, privacy, national security, labor-market effects, competition, misinformation, energy use, and democratic accountability. If the government also owns shares in the firms it regulates, its incentives become muddier. Will policy be written to protect the public, or to protect the value of the governmentโs portfolio? Even if the answer is โthe public,โ the appearance of conflict will be hard to avoid.
There is also a simpler point: we already have mechanisms for capturing broad economic gains.
They are called taxes.
The Case for Capturing Some of the Gains from AI Through the Tax System
If AI produces extraordinary corporate profits, the corporate income tax can capture part of those gains. If AI increases the value of publicly traded companies, capital-gains taxes can capture part of those gains when shares are sold. If AI produces great fortunes, estate and inheritance taxes can capture part of those gains when wealth is transferred across generations. If AI raises productivity and wages, individual income taxes will capture part of those gains. If AI benefits a broad range of firms and industries, the tax system can follow the gains wherever they actually appear.
This is a much better approach than trying to make the federal government a venture capitalist.
The tax system is not perfect. In fact, some of the best arguments for public participation in AI gains are really arguments for repairing the tax system.
One obvious reform is limiting the step-up in basis at death. If AI creates enormous unrealized capital gains, those gains should not simply disappear for tax purposes when an owner dies. A tax system that allows large gains to escape both income taxation during life and capital-gains taxation at death is poorly designed. If this creates a record-keeping burden for small inherited portfolios, the limit on step-up in basis can be tied to estates above some significant threshold.
A second reform is preserving, and possibly modestly increasing, the corporate income tax. The goal should not be to punish business investment. AI will require large investments in computing, energy, chips, software, and talent. But if AI substantially increases corporate profits, it is reasonable for part of those profits to support the public institutions and infrastructure that make economic growth possible.
A third reform is tightening the estate tax. A serious estate tax is one of the few mechanisms we have for limiting the permanent concentration of wealth across generations. This does not mean confiscatory taxation. It does mean that very large fortunes should not be able to avoid taxation through increasingly elaborate planning devices.
One particular issue deserves more attention: the use of charitable structures that allow wealthy individuals and families to receive large tax advantages while retaining substantial influence over the assets. Philanthropy can serve public purposes. But an unlimited charitable deduction for self-governed or family-influenced charitable entities can become a way to avoid tax while preserving social power and control. This is not the same thing as paying taxes to support public purposes through the ordinary budget process.
Conclusion
If the public has a claim to the benefits from AI, the cleanest way to recognize that claim is not for the government to take shares in a handful of companies. It is to make sure the tax system captures AI-generated gains wherever they occur.
This approach has several advantages.
It does not require the government to pick winners.
It does not require the government to decide which AI company will dominate ten years from now.
It does not entangle regulators with ownership interests.
It does not assume that AI developers will capture all of the value they create.
It preserves market competition while allowing the public to share in broad economic gains.
And it fits ordinary public-finance principles. When economic activity creates large gains, the tax system should capture a reasonable share of those gains to support public purposes. When economic activity creates risks or external costs, regulation should address those risks directly.
This distinction is important. AI regulation and AI revenue policy should not be confused.
The government should regulate AI where public risks are real. It should address fraud, discrimination, privacy violations, national-security risks, cyber risks, misuse in elections, labor-market disruption, and the concentration of market power. It should consider whether copyright law, data rules, and competition policy need to be updated for the AI era.
But if the question is how the public should share in the economic upside of AI, the answer should be broad tax policy, not public ownership of selected firms.
CIVPACโs general view is that public policy should be economically sound, fair, respectful of individual freedom, and politically realistic. A federal equity stake in major AI companies fails too many of those tests. It is economically speculative, administratively messy, politically tempting, and likely to create conflicts between regulation and ownership.
The better answer is less dramatic but more durable: preserve competitive markets, regulate AI directly where public risks are real, and fix the tax system so that AI-generated gains cannot escape taxation simply because they appear as corporate profits, unrealized capital gains, or inherited wealth.
AI may well transform the economy. If it does, the public should benefit. But the way to do this is not to make the federal government a shareholder in a few favored companies. The way to do it is to tax the gains wherever they actually appear.








