Commentary

Recent proposals to raise taxes on fast-growing AI companies have emerged in the United States. amid soaring market valuations, concerns over AI-driven technological disruption, and growing pressure on public budgets. This comment reviews the predicted effects of three broad types of proposals: raising the corporate tax rate, imposing a revenue tax, and imposing a tax on tokens. The proposals are briefly evaluated on how these tax structures perform as public revenue sources and regulatory mechanisms, two critical public finance dimensions.

Keywords: Artificial Intelligence (AI);AI Taxation; Technology Regulation; Digital Economy; Policy Design

Corporate Tax Increase

A number of policymakers have suggested raising corporate taxes or changing the treatment of capital investments within the tax code. Senator Elizabeth Warren (D-MA) proposed raising taxes on corporations and closing loopholes that provide tax breaks for technology investment¹. Senator Bernie Sanders (I-VT) has proposed a "robot tax" framed explicitly as both a regulatory instrument and as a funding mechanism for programs serving displaced workers². His Senate Committee on Health, Education, Labor and Pensions released a report in October 2025 warning that AI and automation could replace up to 97 million U.S. jobs over the next decade. More recently, Sanders announced the American AI Sovereign Wealth Fund Act, which would impose a one-time 50 percent tax on the largest AI companies paid in stock rather than cash and mandate public representation on company boards. The wealth fund proposal raises distinct questions of corporate governance and public ownership that are not addressed in this comment³.

The robot tax is more amenable to public finance analysis. On the revenue side, the academic literature suggests that raising corporate taxes on AI firms is a leaky instrument and firms’ tax strategies may be inefficient. IP-intensive multinationals have well-documented capacity to respond to rate increases by shifting profits to lower-tax jurisdictions rather than paying them, a behavioral response that has been extensively studied in the context of both U.S. and European corporate tax policy [1,2]. The effective tax base is therefore smaller than the statutory tax base would suggest, and revenue yield would be lower than a naive multiplication of rate times reported profits implies [3].

Evidence from the 2017 Tax Cuts and Jobs Act is instructive on this point in reverse. Researchers found that only about 20 percent of incremental cash flow from the TCJA tax cut went toward capital expenditure or R&D, while the rest went toward share buybacks, dividends, and other distributions, suggesting that large tech firms respond to tax changes primarily through financial channels rather than real investment channels [4]. More directly, a study of corporate tax cuts finds that reductions in the corporate rate lead to a relative jump in the stock market valuation of the most intangible-intensive companies, followed by a sustained increase in R&D, intangible investment, and patenting, implying that the reverse holds as well; a tax increase could disproportionately depress investment among firms at the AI frontier [5].

On the regulatory side, however, the literature offers some support for Sanders' framing. The same evidence that suggests tax increases slow R&D investment is, from a labor protection perspective, a feature rather than a bug. A quasi-experimental study using China's corporate income tax reform finds that increasing the corporate income tax rate significantly hinders corporate innovation, and that the impact is more pronounced in high-productivity firms, which are precisely the AI companies Sanders is targeting [6]. Whether slowing AI development is normatively desirable is a contested question, but the literature confirms that corporate tax increases would have a real, if partial, effect on innovation at the frontier.

Revenue Tax

Senator Mark Kelly's (D-AZ) "AI for America" proposal establishes an AI Horizon Fund financed by contributions from leading AI companies, intended to reinvest in workforce development, energy infrastructure, and responsible AI deployment⁴. The proposal deliberately leaves the tax mechanism open, floating several options including taxes on large-scale use of public resources like power, water, and land; profits from digital advertising tools powered by AI; or AI-based revenue windfalls.

The details of the funding mechanism matter enormously, and Kelly's proposal is best understood as a menu rather than a single instrument. If the Horizon Fund is built by taxing inputs to AI models such as electricity consumption, water use, or land, the tax functions more like a Pigouvian instrument that internalizes negative externalities and creates incentives for more efficient model architectures. This is arguably the most defensible design from a efficiency standpoint because it targets the social costs AI infrastructure imposes on communities and power grids rather than profits or outputs.

If, on the other hand, the fund is financed through a revenue tax, the economics resemble those of digital services taxes. The cleanest empirical evidence here comes from an event study of the EU's 2018 digital tax proposals, which finds a significant abnormal capital market reaction of -0.692 percent in the stock returns of potentially affected firms, corresponding to a market value reduction of more than 52 billion euros, 40 percent of which was attributable to U.S. firms. Investor reaction was stronger for firms with higher EU exposure and for those that engage more in tax avoidance [7]. A revenue tax on AI-based windfalls would therefore likely produce a meaningful stock market reaction and some behavioral response, but limited sustained revenue, since it faces both the avoidance problem and the structural challenge of taxing gross revenue in an industry where margins vary enormously across firms and over time.

Token Tax

Representative Greg Casar's (D-TX-35) token tax is an analytically interesting public finance proposal⁵. Casar proposes taxing AI companies based on the number of tokens processed, with an adjustment for underlying computing power to prevent gaming. The revenue would fund a federal jobs program modeled on New Deal-era employment initiatives. Casar explicitly acknowledges the dual intent, arguing that Congress should pass the tax both to slow down job loss and to fund programs ensuring that mass unemployment does not follow from AI-driven displacement.

The token tax is closest in structure to a Pigouvian tax, one designed to correct a market failure by pricing an externality. In the standard Pigouvian framework, the optimal tax is set equal to the social cost of the activity being taxed [8]. If AI-driven job displacement constitutes an externality, the tax should in principle be calibrated to the marginal social cost of each unit of AI output. Casar's proposal gestures at this logic by suggesting the rate should be adjusted dynamically in response to observed labor market outcomes.

On the revenue side, the token tax has a structural advantage over both corporate income taxes and revenue taxes, as it could be considerably harder to avoid. Profit shifting or revenue reallocations across jurisdictions, the primary avoidance strategies for corporate income taxes, do not reduce token usage. The tax base is, in principle, directly observable from the operational data AI companies already collect for billing purposes, making it more robust as a revenue instrument than either of the alternatives, at least in the near term. The theoretical literature on excise taxes as Pigouvian instruments supports this general logic: when the taxable activity can be observed directly, excise taxes outperform profit-based taxes on both efficiency and revenue-yield grounds [9,10].

On the regulatory side, the token tax would operate through the price mechanism by raising the cost of AI inference and training, thereby increasing the cost of deploying AI at scale. The literature on excise taxes and technology adoption suggests this would slow deployment, particularly among price-sensitive users and use cases [11]. Whether it would meaningfully slow frontier R&D, as opposed to deployment of existing models, is less clear. AI labs invest in training runs and research capacity long before models are deployed, and a tax on tokens at inference time would not directly tax that investment. A tax that also covered computing power used in training, as Casar proposes, would have a broader effect on the R&D pipeline.

The token tax also faces genuine implementation challenges consistently associated with these types of taxes, most notably the difficulty of defining the taxable unit precisely, the risk of consumer pass-through, and the competitive distortion that arises if the tax applies only to U.S. AI providers [10]. A token tax imposed unilaterally could advantage foreign AI competitors not subject to the same levy, a concern that maps directly onto the broader literature on unilateral digital taxes and competitive distortion [3].

Conclusion

Across all three proposals, the public finance literature points to a consistent tension: taxes designed primarily to raise revenue from AI firms will yield less than expected because the avoidance elasticity of IP-intensive multinationals is high. Taxes designed primarily to regulate AI development will have real effects on R&D investment and deployment, but whether those effects are desirable depends entirely on prior beliefs about whether slowing AI is a public good. The token tax is the most Pigouvian in structure and the most avoidance-resistant, but its regulatory reach is narrowest if confined to inference. What none of the proposals fully confront is that the most powerful lever for both revenue and regulatory effect may be international coordination. Without it, unilateral U.S. taxes on AI are likely to produce behavioral responses that limit both fiscal yield and regulatory bite, as the broader literature on international tax competition consistently demonstrates [1,2].

References

1. Clausing Kimberly A (2020) Profit shifting before and after the Tax Cuts and Jobs Act. National Tax Journal 73(4): 1233-1266.
2. Devereux Michael P and Rachel Griffith (2003) Evaluating tax policy for location decisions. International Tax and Public Finance 10(2): 107-126.
3. Devereux Michael P, Alan J Auerbach, Michael Keen, Paul Oosterhuis, Wolfgang Schön, et al. (2021) Taxing Profit in a Global Economy. Oxford University Press.
4. Chodorow-Reich Gabriel, Owen Zidar, and Eric Zwick (2025) Tax policy and investment in a global economy. Working paper, Harvard University, United States.
5. Cloyne James, Joseba Martinez, Haroon Mumtaz, and Paolo Surico (2022) Short-term tax cuts, long-term stimulus. NBER Working Paper No. 30246. National Bureau of Economic Research.
6. Chang, Mengran, Xin Liu, and Yue Zhang (2024) Can changes in corporate income tax rate affect corporate innovation? PLOS ONE.
7. Nour Sandra, Erik Theissen, and Christian Theissen (2022) Taxing the digital economy: Investor reaction to the European Commission's digital tax proposals. National Tax Journal 75(1): 63-90.
8. Pigou Arthur C (1920) The Economics of Welfare. Macmillan.
9. Baumol William J (1972) On taxation and the control of externalities. American Economic Review 62(3): 307-322.
10. Goulder Lawrence H and Ian WH Parry (2008) Instrument choice in environmental policy. Review of Environmental Economics and Policy 2(2): 152-174.
11. Acemoglu Daron, Andrea Manera, and Pascual Restrepo (2022) Does the US tax code favor automation? Brookings Papers on Economic Activity 2022(1): 1-65.

foot Notes

1https://www.warren.senate.gov/newsroom/press-releases/warren-for-time-tax-ai-and-invest-in-people

2https://www.sanders.senate.gov/wp-content/uploads/10.6.2025-The-Big-Tech-Oligarchs-War-Against-Workers.pdf

3https://www.sanders.senate.gov/press-releases/news-sanders-introduces-legislation-to-create-7-trillion-ai-sovereign-wealth-fund/

4https://www.kelly.senate.gov/wp-content/uploads/2025/09/KELLY-AI-FOR-AMERICA.pdf

5https://casar.house.gov/media/press-releases/op-ed-american-prospect-tax-ai-create-jobs