A Pigouvian automation tax is a proposed levy on businesses that replace human workers with robots, software, or artificial intelligence, designed to make the employer absorb the social costs of that displacement instead of shifting them onto workers, communities, and public programs. No such tax exists at the federal level in the United States. State legislators, a member of Congress, and the city of Chicago have all moved pieces of the idea into active proposals or partial measures, but a full displacement-based tax has not been enacted anywhere.
Where the Idea Comes From
The concept traces back to Arthur Pigou, a British economist whose 1920 book The Economics of Welfare argued that when a private transaction imposes costs on the broader public, the government can use a tax to close the gap between private profit and social harm. Pigou’s classic example was pollution: a factory profits while neighbors bear the health costs, so a tax on the polluter forces the price of its goods to reflect the true cost to society.
Applied to automation, the logic runs like this. When a company replaces 200 warehouse workers with robotic systems, it captures the productivity gains immediately. The displaced workers, their communities, and public safety-net programs absorb the costs: unemployment claims, retraining needs, reduced consumer spending in the local economy, and lower payroll tax revenue flowing into Social Security and Medicare. A Pigouvian automation tax would force those downstream consequences onto the company’s books, making the decision to automate more expensive and, in theory, slowing displacement enough for workers and institutions to adapt.
Why the Current Tax Code Favors Machines Over Workers
The strongest argument for the tax starts with a structural fact about the existing code. Hiring a person costs more, on a relative basis, than buying a machine to do the same job.
Employers pay a 6.2% Social Security tax and a 1.45% Medicare tax on each worker’s wages under the Federal Insurance Contributions Act, and the employee pays the same rates from their paycheck.1Internal Revenue Service. Topic No. 751, Social Security and Medicare Withholding Rates On top of that, employers owe a 0.6% federal unemployment tax on the first $7,000 of each worker’s wages, plus state unemployment insurance and workers’ compensation premiums that vary by industry and location.2U.S. Department of Labor. FUTA Credit Reductions – Unemployment Insurance Mandatory payroll costs add roughly 10% or more on top of base wages before a company pays a dime in salary.
Machines carry no such overhead. A company that buys a $500,000 robotic assembly system can deduct the full cost from its taxable income in the first year under 100% bonus depreciation, which was permanently restored by the One, Big, Beautiful Bill signed into law on July 4, 2025. Alternatively, businesses can expense qualifying equipment purchases up to approximately $2.56 million in 2026 under Section 179. Either route generates no payroll taxes, no unemployment insurance premiums, and no workers’ compensation costs, and reduces the company’s tax bill on top of that.3Internal Revenue Service. Publication 946 – How To Depreciate Property
Automation tax proponents call this a hidden subsidy for replacing people with technology. Under the Pigouvian framework, the tax neutralizes that built-in advantage so business decisions rest on actual productivity rather than tax arbitrage.
How the Tax Would Be Calculated
No single formula has won consensus. Proposals cluster around four approaches, each trying to turn the idea of “a machine replaced a person” into a dollar figure on a tax return.
The imputed-wage method taxes the business based on what a human worker would have earned doing the same job. If automation eliminates a $60,000 position, the company owes a percentage of that amount, mirroring the payroll taxes it would have paid on a human. Some proposals peg the rate to the combined FICA burden; others use the 21% federal corporate income tax rate. New York’s proposed Robot Tax Act takes this approach, imposing a surcharge equal to the state and local payroll-related taxes that would have been paid on the displaced employee’s final-year wages.4Tax Policy Center. How Does the Corporate Income Tax Work?5New York State Senate. NY State Assembly Bill 2025-A3719
The token-based method, proposed at the federal level by Representative Greg Casar, taxes AI products by the volume of computational work they perform, measured in units of text called “tokens” and the underlying computing power used to train and run the models. The tax falls on AI providers rather than end users, with higher rates for large corporate deployments than for individual consumers.6Representative Greg Casar. OP-ED: The American Prospect: Tax AI to Create Jobs
The value-added method applies the tax to the measurable productivity gain that follows automation. Accountants establish a baseline of output before deployment, then levy a percentage on the increased revenue or reduced costs. This captures the economic benefit directly but requires detailed before-and-after accounting.
The deduction removal route skips a new tax entirely. It simply eliminates the preferential depreciation and expensing treatments that make capital investment cheaper than hiring. Stripping bonus depreciation and limiting Section 179 for labor-replacing equipment would raise the effective cost of automation without adding a new line item to the code.
Every method has a weakness. The imputed-wage approach requires someone to determine which specific jobs were “replaced” rather than merely changed. The token-based method taxes computational volume without directly measuring displacement. The value-added approach demands counterfactual accounting that invites disputes. Removing deductions penalizes all capital investment, not just labor-replacing technology. These design problems explain why the debate has produced many proposals and little enacted law.
Proposals and Real-World Experiments
No country has enacted a pure Pigouvian automation tax. What exists is a patchwork of proposals and partial measures.
United States
New York Assembly Bill A3719, introduced in the 2025–2026 legislative session, is the most detailed state-level proposal. Titled the Robot Tax Act, it would impose a surcharge on any corporation with at least $1 million in New York receipts that displaces an employee by replacing their position with technology, including machinery, AI algorithms, or software applications. The surcharge equals the total state and local payroll-related taxes that the corporation or the displaced employee would have paid during the worker’s final year of employment.5New York State Senate. NY State Assembly Bill 2025-A3719 If enacted, it would take effect for taxable years beginning January 1, 2026.
At the federal level, Representative Greg Casar has proposed taxing AI providers based on computational output, with the rate adjusted periodically to keep revenue in line with the pace of job displacement.6Representative Greg Casar. OP-ED: The American Prospect: Tax AI to Create Jobs No specific rate has been published as of mid-2026.
Chicago is the only U.S. jurisdiction that explicitly taxes AI services. Since 2023, the city has imposed a tax on cloud-based AI platforms used within city limits, starting at 9% and increasing to 11%. It is a sales-tax-style levy on AI usage rather than a displacement-based Pigouvian tax, but it shows that local governments are willing to treat AI as a distinct taxable category.
International
South Korea came closest to implementing a robot tax in 2018 when it reduced the automation investment tax credit by two percentage points for medium-sized and large firms. The move didn’t create a new tax; it made the existing tax advantage for automation less generous. The political lift for trimming a benefit is far smaller than for imposing a new levy.
The European Parliament considered and rejected a robot tax during its 2017 vote on robotics regulation. The final Robotics Report passed 396 to 123, but the robot tax provisions were stripped from the resolution before the vote.
The Case Against
Critics include some of the most prominent voices in international economic policy. The International Monetary Fund addressed the question directly in a 2024 staff discussion note, stating that “special taxes on AI to reduce the speed of AI investment are not recommended as they can be hard to operationalize and hamper productivity growth” and that “there should be no special tax on gen AI, robots, or other forms of labor-replacing technology.”7International Monetary Fund. Broadening the Gains from Generative AI The IMF did recommend reconsidering corporate tax incentives that encourage rapid displacement and strengthening general capital income taxes, but drew a clear line against technology-specific levies.
The practical objections are just as sharp. Defining what counts as a “robot” or “automation” is genuinely hard. An ATM replaced bank tellers, a spreadsheet replaced bookkeepers, a word processor replaced typists. Any definition broad enough to capture meaningful automation risks taxing routine software upgrades; any definition narrow enough to avoid that misses most of the displacement actually happening. Policy analysts have pointed out that a literal reading of most robot tax proposals would cover ATM machines and basic office software.
A competitiveness argument carries real weight as well. A country or state that taxes automation gives companies an incentive to move operations to jurisdictions that don’t. Goldman Sachs research estimates that AI adoption at scale will take roughly a decade, during which 6% to 7% of U.S. workers may face displacement. Opponents argue that a tax doesn’t eliminate that displacement; it exports it while reducing domestic productivity growth.
Economists who study investment incentives add a further point. Research on the investment tax credit has consistently found that tax preferences for capital equipment spur higher investment, which raises worker productivity and wages over time. Taxing automation reverses that dynamic, raising the after-tax cost of new equipment and potentially harming the same workers the tax aims to protect by making their employers less competitive.
Legal Obstacles
Even if a legislature passed the tax, it would face immediate challenges. Most state constitutions contain uniformity clauses requiring that taxes on property be applied uniformly rather than singling out specific categories. Courts have struck down taxes that use different valuation methods for similar property types and invalidated exemptions that treat certain property more favorably than comparable assets. A tax that applies specifically to robotic or AI-equipped machinery but not to other capital equipment would face scrutiny under these provisions.
Classification also matters. Courts distinguish between a “tax” subject to constitutional uniformity requirements and a “regulatory fee” that is not. An automation levy structured as a regulatory fee tied to a specific government program may survive challenges a general tax would not, but that structure constrains how the revenue can be spent.
At the federal level, a tax on AI tokens or computational output raises its own questions. Congress has broad taxing power under Article I, but the tax must be structured to avoid being classified as a direct tax requiring apportionment among the states by population, which would make it nearly impossible to administer. Most excise-style taxes avoid this problem, but the novel structure of a computation-based levy has no direct precedent.
What the Revenue Would Fund
Every serious proposal pairs the levy with a spending plan, and the spending side draws broader support than the tax itself. Even economists who oppose the tax generally agree that public investment is needed to manage workforce transitions driven by AI.
Worker retraining is the most common proposed use. Federal workforce development grants already exist; in 2025, the Department of Labor announced $30 million in grants to train workers for high-demand and emerging industries.8U.S. Department of Labor. US Department of Labor Announces Availability of $30M in Grants to Train American Workers for Jobs in High Demand, Emerging Industries Automation tax revenue could scale programs like these to fund technical certifications, community college tuition, and on-the-job training. Existing incentives such as the Work Opportunity Tax Credit, which provides employers up to $2,400 per qualifying hire, are limited in scope and target specific disadvantaged populations rather than displaced workers broadly.9Internal Revenue Service. Work Opportunity Tax Credit
Extended unemployment benefits are another common proposal. Standard unemployment insurance was designed for workers between jobs, not workers whose entire occupation has been automated out of existence. Automation tax revenue could fund longer benefit periods or higher payments for workers who can demonstrate that technology replaced their position permanently.
The most ambitious proposals direct revenue toward a universal basic income, with every citizen receiving a recurring monthly payment funded partly by the productivity gains that automation creates. Whether UBI is the right mechanism is contested, and no U.S. jurisdiction has tied UBI funding specifically to an automation tax.
The IMF, while opposing technology-specific taxes, endorsed the underlying goal: strengthening general capital income taxation and using the revenue to fund “redistribution, social insurance, and retraining programs” that help workers navigate labor market disruptions driven by AI.7International Monetary Fund. Broadening the Gains from Generative AI The disagreement is less about whether displaced workers need support and more about whether a Pigouvian tax on specific technologies is the right funding mechanism, or whether broader reforms to capital taxation would achieve the same goal with fewer distortions.