Lawmakers are increasingly being called upon to use the tools at their disposal — particularly tax policy — to address the public’s concerns about artificial intelligence (AI). Those concerns include a sprawling collection of different problems, some of them familiar, others novel and even unprecedented.
One set of concerns is that AI is like transformative technological changes that preceded it, meaning it may benefit society overall but may also cause massive disruptions to the environment, to families’ economic security, and to individuals’ sense of place in the world.
Another set of concerns has to do with negative externalities of AI specifically, including energy consumption and other nuisances associated with AI data centers, the uncertainty surrounding any written words produced in an educational or workplace setting, and the “slopification” of media that nearly everyone consumes.
Still other concerns focus on AI’s potential to change the economy in far more fundamental, unprecedented ways. A powerful few could profit from the fruits of human knowledge and innovation that have been accumulated over centuries and are now integrated into AI models that then build upon and improve themselves without human intervention going forward. This is potentially a radical departure from previous technological changes and could make most people economically obsolete.
Congress can use its taxing power in three broad ways to respond to AI. These categories capture the main options currently being discussed. Congress could use its taxing power to:
- Ensure a fairer distribution of the fruits of labor, innovation, and investment
- Address the specific harms (the negative externalities) of AI specifically
- Acquire a public stake in the companies that are driving the AI revolution
This policy brief does not recommend a specific tax policy approach for lawmakers concerned about AI, nor does it comprehensively identify every possible approach. Rather, this brief outlines the continuum of policy approaches and includes several illustrative examples.
Introduction
No one knows exactly how AI will transform our society, but if the changes it brings about are anything like those of previous technological revolutions, Congress will need to do a better job this time around managing those changes.
The Industrial Revolution transformed American life and sent a wildly disproportionate share of its bounty to a few wealthy tycoons at a time when the U.S. government did not have a corporate income tax or personal income tax to provide even a mild corrective to exploding inequality.
While the technological advancements of that time may have benefited society broadly in the long run, many people saw their livelihood destroyed and their communities disrupted without necessarily living long enough to enjoy any of those benefits.
Workers were replaced by machines without the opportunity to find new jobs or equivalent jobs. Advanced industry also consumed and degraded vast natural resources through mining, smelting, logging, and pollution before society had time to realize that many natural resources are not easily replaced. Some changes were violently imposed on an unwilling public. Railroads, for example, plowed through homes, farms, and towns via eminent domain instead of taking longer routes. Redundant lines were sometimes built by competing magnates vying for control.
Lawmakers were slow to react and ensure that some of the benefits of these shifts flowed to the general public who were experiencing the harms without taking part in the massive growth of wealth. Eventually, public pressure made the situation untenable, leading to the creation of the corporate income tax and the personal income tax, environmental regulations, and the establishment of national public lands.
In the coming years and decades, AI could be equally — or even more — disruptive and require equally large policy responses. We cannot afford to make the same mistakes. Workers are already being displaced from their jobs, and many more will be as AI usage grows.
The skyrocketing demand for certain raw materials during the Industrial Revolution is similar to how the current buildup of data centers, semiconductor production, and electric infrastructure is increasing the demand of raw materials like copper, silicon, and rare earth elements.
And some would argue that we are building far too many data centers far too inefficiently due to a similar competition between oligarchs as happened during the construction of the railroads.
Whether the AI revolution will be something akin to previous technological changes — or something even more dramatic and disruptive — Congress will need to respond with several types of policies, including tax policies. The potential tax policies fall into three broad categories described below.
1. Ensure Fairer Distribution of Economic Benefits
At minimum, Congress should enact policies that have been discussed in recent years to strengthen the existing federal corporate income tax and federal personal income tax, because these taxes were designed to address the unequal distribution of benefits from economic growth that all of society has contributed to in one way or another.
This is a potential problem with AI, just as it was with technological and industrial advances that preceded it. Proposals that have been discussed by progressives to reform these taxes would be necessary in the absence of AI and are now more vital because of it.
These policies would be transformative and ambitious, but they are not particularly new to policymakers. To take some examples, Congress could:
Enact proposals to ensure wealthy shareholders do not pay lower taxes on their investment income than other Americans pay on their income from work
This includes proposals to end several types of tax breaks for capital gains, the increase in an asset’s value. These reforms are not specific to AI but could become much more important as AI directs its wealth creation disproportionately to a small group of wealthy investors, as has been the case of other technological and industrial changes.
Whereas most people live on income from work and pay taxes each year on that income, the same is not true for corporate leaders and wealthy investors. The tax code does not consider capital gains to be income until they are “realized” as the profit from selling an appreciated asset.
What happens when someone holds onto shares of an AI company (or any assets) that keep appreciating year after year without selling them? An economist would say that appreciation is income to the person who owns the asset, just as wages earned by working people are income. However, under current law, unrealized capital gains are not taxable. Extremely wealthy corporate leaders and investors often arrange their affairs so that most of their income each year is unrealized capital gains that are not taxed.
If the owner of the asset eventually sells the asset, the profit from that sale (the realized capital gain) will finally be taxed, but at lower rates than other types of income like earned income. If the owner of the asset never sells it, then the asset is passed to their heirs and the tax code forgets about the unrealized gains forever, meaning the income is never taxed.
These tax breaks for capital gains effectively subsidize people who can generate income from wealth rather than income from work and they are especially problematic any time economic change benefits the accumulation of wealth more than it benefits work. AI is the latest iteration of this dynamic.
Legislative proposals would limit these capital gains tax breaks so that very well-off people who receive most of their income from wealth rather than work pay taxes on their income each year like everyone else. For example, Sen. Ron Wyden’s Billionaires Income Tax Act would tax unrealized capital gains on tradable securities (such as corporate stocks) annually, the way other types of income are taxed. (For assets that are more difficult to appraise each year, owners would still be allowed to defer income tax until selling the asset but the tax would be increased in a way that removes the benefit of the tax deferral.)
Other proposals would tax long-term capital gains and qualified dividends at the same rates as income from labor, at least for high-income households, and sharply limit the break allowing unrealized capital gains to forever escape income tax when assets are transferred at death.
Together, these reforms would ensure that at least some of the extraordinary wealth created by AI is taxed as it accrues or when it is ultimately passed to heirs, rather than escaping taxation altogether.
Enact proposals to ensure large corporations cannot escape taxes on their profits
Policymakers should enact the reforms of the personal income tax described above while also being honest that they face certain technical and legal challenges and provide a backup plan in the form of reforms to the corporate income tax. Much of the income for well-off investors generated by AI (like most wealth generally) is generated by corporations. Reforming the corporate income tax can ensure that this income is taxed before it reaches wealthy investors who might not pay taxes on it directly.
This requires limiting different tactics used by large corporations to avoid taxes. The most important form of corporate tax dodging to be addressed is the use of accounting gimmicks that make profits appear to be earned in countries where they are taxed very little, if at all. Tech companies have long been particularly aggressive in exploiting these offshore tax havens and AI companies will likely follow this pattern if policymakers continue to allow it.
Several proposals would limit the tax benefits of these accounting maneuvers by requiring corporations to pay a robust tax on any offshore profits that are not subject to sufficient taxes by foreign governments. These include Sen. Sheldon Whitehouse’s No Tax Breaks for Outsourcing Act and Sen. Bernie Sanders’ Corporate Tax Dodging Prevention Act, as well as proposals from the Biden administration to implement the global minimum tax.
Repeal tax policies that favor investment in machines over human labor
The tax code’s bias in favor of investing in machines rather than hiring people has long been a problem and may become much more so in the age of AI.
Under current law, a business that pays $500,000 over 10 years to purchase a machine is taxed far more favorably than a business that pays an employee the exact same amount over the same period. (This is true whether the business is subject to the corporate income tax or the personal income tax.)
A company that buys the machine (perhaps on a loan that it repays over a decade) is allowed, under the 2025 Trump tax law, to “expense” the $500,000, or deduct it all in the year that the purchase is made. Meanwhile, a company that hires the employee for 10 years must deduct the $500,000 of compensation over 10 years. This tilts the scales in favor of not just AI, but any technology or machinery that can replace human workers.
Even before the 2017 Trump tax law was enacted, the tax rules allowed depreciation that is more accelerated than what would make sense economically, and they arguably created a bias in favor of investing in machines over hiring human workers. And Congress had, several times, allowed full expensing, which is the most extreme version of accelerated depreciation, allowing the cost of equipment to be fully deducted in the year it is purchased. Expensing supercharges the bias in favor of machines over people. The 2017 Trump tax law again allowed expensing, and the 2025 Trump tax law made it permanent.
There is general dissatisfaction with the 2025 Trump tax law among some lawmakers and among the public. Sen. Warner has proposed the Data Center Tax Accountability and Disclosure Act, which would tie the access of the most extreme depreciation benefits for companies building data centers to their environmental and energy efficiency design. Another approach (not currently proposed in Congress) would be to repeal the provision allowing full expensing for all companies and return to the depreciation rules in effect in 2017.
Another approach would go further by requiring something closer to “economic depreciation,” under which businesses would be required to deduct the cost of machinery for tax purposes in the same way they are deducted for accounting purposes.
This would put investments in AI and automation on a roughly neutral footing relative to labor by requiring companies to gradually deduct these investments over the period in which the asset actually wears out or generates income.
In addition, Congress should sharply limit tax deductions claimed by businesses for interest payments on debt. In the example above, the business is more likely to rely on debt financing if it chooses to purchase a $500,000 machine than pay an employee $500,000 over ten years. Even many supporters of expensing agree that when it is combined with nearly unlimited deductions for interest, it has too much potential to become a serious tax shelter.
Spend the revenue raised from these tax proposals to ensure that some part of the income generated from AI is shared broadly
Expenditures financed with the resulting revenue could take many forms. Lawmakers could redistribute the income from the wealthy (who own most of the stock in AI companies) to everyone else to curb inequality. Lawmakers could use the revenue to make public investments likely to create jobs and opportunities (infrastructure, education, research) or could devise programs more targeted to the people harmed most directly by AI. How Congress spends the revenue is not the focus of this report but is a key to how successful any policy response will be.
2. Address Negative Externalities of AI Specifically
In addition to a broad slate of progressive tax policies to help counter the expansion of inequality that is likely to result from the AI boom, Congress could enact new taxes that apply to AI specifically (or to AI and automation) to address the harms it causes. These new taxes would address AI’s negative externalities (meaning the costs borne not by the people and companies involved in the activity but instead by others in society) such as job loss and environmental harms. A major challenge for these proposals is defining exactly what is AI and what is being taxed.
Enacting a tax on companies that replace jobs with AI
In some cases, identifying job replacements due to AI or automation would be relatively straightforward. Examples might be a warehouse that installs autonomous robots and eliminates hundreds of employees, or a call center that replaces customer service representatives with chatbots. In those cases, lawmakers could tie a tax to reductions in payroll attributable to automation.
Most cases are likely to be significantly more complicated. A law firm that adopts AI tools may reduce hiring of new associates and paralegals without laying anyone off. A software company may have senior developers managing AI agents instead of managing human junior developers.
Lawmakers would also need to distinguish between layoffs directly attributable to automation and reductions caused by recessions, outsourcing, mergers, or declining demand.
Another question is how this type of policy would address newly formed companies. If an automation tax discourages an established company from replacing its employees with machines, what happens when a new competing company is formed to sell the same product or service using machines instead of human employees?
Congress could instead define the types of work that companies could rely on human labor to provide and impose a tax when they rely on machines. This would treat established companies and newly formed ones the same and tax both if they use machines when they could use human labor. The problem with this approach is that it would be challenging for lawmakers to settle on such legislative language. What’s more, society’s ideas about what tasks should be performed by machines or humans are constantly changing.
Enacting a tax on the consumption of AI, such as a “token” tax.
Congress could consider imposing a tax on AI “tokens,” which are the unit of AI usage, a way that the owners of AI companies measure the data processed by their models. A token tax is less like an income tax and more like an excise tax or consumption tax. AI providers already track token usage for billing purposes, so this approach might be relatively easy to administer.
Tokens are not standardized across AI companies, however. Rather, they are defined by individual AI companies. Even more problematic, different models tokenize the same data in different ways, and they may change their token metric over time. The same exact prompt could generate different token counts depending on whether it was submitted to ChatGPT, Claude, Grok, or another model.
Taxing tokens does not seem to be a straightforward solution if it means that Congress would effectively be taxing a measurement that private companies themselves control. This would be a challenge for Congress and the IRS. They would need to clearly define what they are taxing and ensure that the metric applies uniformly across all AI companies.
Enacting a tax on energy consumption
For those concerned particularly about AI’s energy usage, this could also be taxed. Sen. Elizabeth Warren proposes this approach with a fee on the electricity used by large data centers supporting AI systems.
An energy-based tax would also align with several public concerns beyond just generating new tax revenue. Data centers stress grids, requiring substantial new generation and transmission, and can increase electricity prices for households and other businesses.
A tax on AI-related energy consumption could help offset some of these public costs while encouraging companies to invest in more energy-efficient computing infrastructure.
From a revenue perspective, though, lawmakers would need to consider how to treat facilities powered by on-site generation that is not monitored by utility companies.
3. Acquire Partial Public Ownership of AI Companies
Another option before Congress would use its taxing power in a way that has not been attempted before. Some scholars and lawmakers propose using Congress’ taxing power to acquire a partial ownership stake in AI companies for the public, so that the public can share the profits and have a say in how the companies operate. While non-cash taxes (or “in-kind” taxes) have existed throughout most of human history, Congress has never used its power of taxation to take partial ownership in a company.
Under this approach, the federal government would apply a tax to AI corporations which the companies would pay with their own stock. The stock would be held on behalf of the public by an independently managed investment fund, similar to the way some governments have created sovereign wealth funds. This public ownership would allow taxpayers to share directly in the industry’s long-term financial returns.
Last month, Sen. Sanders introduced a bill that would follow this route. His bill would levy a one-time excise tax on the biggest AI companies, depositing 50 percent of their stocks into a sovereign wealth fund.
This proposal would establish an independent board to oversee the use of the sovereign wealth fund, similar to the Federal Reserve Board. Sen. Sanders’ bill would require that the fund would pay dividends to the public each year equal to five percent of the value of the fund. Estimates show that this fund would be initially valued at around $7 trillion and those 5 percent dividends would provide a payment of around $1,000 to every American in the first year.
Perhaps more importantly from the viewpoint of proponents, this approach would give the public a say in how the companies’ decision-making.
Conclusion
While the explosion of AI may have similarities to all technological advances throughout history, in some ways it is entirely different. AI does not simply increase the productivity of humans. It often imitates us, regurgitating our shared information without compensating or crediting us. The cynical view, which is perhaps justified, is that AI companies want to sell our own intellectual property and the results of decades of public infrastructure investment back to us while simultaneously displacing us from our jobs.
From this standpoint, the public is entitled to the fruits of this new technology and should share directly in the long-term appreciation of companies creating wealth based on decades and centuries of human knowledge that was individually produced and disseminated under different terms.
Lawmakers should not be passive as AI reshapes our economy and our society. We cannot wait to react to those changes once they are already entrenched. Instead, policymakers must anticipate these changes and be proactive in ensuring that AI works for the betterment of humanity rather than its detriment.
