Nations Are Angling for Ways to Tax AI. Defining How Is Elusive

Countries are champing at the bit to tax artificial intelligence—they just don’t know how.

Companies, US lawmakers, and members of Parliament have proposed a raft of levies to make up for the potential labor displacement caused by AI. But these proposals are extremely vague, starting with governments’ struggles to define the aspects of AI that would be taxed.

Douglas Holtz-Eakin, president of the American Action Forum, pointed to terms like an AI “token” or a “robot” tax. In AI parlance, a token is piece of data, like a word, that a large language model uses to process information and produce responses.

“What counts as one robot?” he said. “How do you actually identify one token in a digital world where things can be multiplied, divided, rearranged infinite number of ways? It’s hard to figure out what one token is, and if you can’t identify the base of a tax, you’re in trouble.”

Another obstacle will be reconciling two opposing positions on levying the technology emerging in the tax world.

The first camp believes that AI won’t wallop the labor markets, and governments already hold the tools to appropriately tax artificial intelligence, through, for example, additional consumption levies like value-added taxes.

“I don’t think that AI is going to broadly displace so many workers so quickly that we’re not going to be able to handle it,” Holtz-Eakin said. “There’s nothing about a new way to generate income and consumption that means that taxing income and consumption are suddenly impossible to do. You still have income, you still have consumption. You can tax them.”

The other camp believes that AI is a revolutionary tool unlike other technological advances such as the computer, the internet, or even textile machinery that will displace a disproportionate number of workers. So, new, innovative taxes are needed to address a unique problem.

“Even if you’re of the mind that there is nothing new about AI, even if you take that position, it’s still broadly understood that our corporate income tax is not working for taxing large multinational companies,” said Jeremy Bearer-Friend, a professor at George Washington University.

“So in a world where we know that corporate income tax has failed and the international tax system is broken, we still need something new.”

Looming Job Crisis

Governments are concerned they won’t be able to collect enough tax revenue to fund social welfare programs because so much of their tax bases come from labor.

The Congressional Budget Office (CBO) estimates that the US government will collect $1.8 trillion in payroll taxes in 2026, compared to $404 billion in corporate taxes. Canadian income tax makes up about 46% of the country’s total tax revenue. In France, that number increases to 52.4%, according to a report published by Windfall Trust, a non-profit dedicated to mitigating the negative economic effects of AI usage.

And the predictions on the economic impact of AI, according to the companies themselves, are alarming.

Anthropic CEO Dario Amodei predicted that AI would wipe out half of entry-level white collar jobs in the next five years. OpenAI CEO Sam Altman has also made ominous remarks about the “existential risk” AI poses to the global economy.

Bearer-Friend says the transition to AI will prompt a transfer of wealth from employees to companies because AI is doing the work and doesn’t require a salary.

“It’s the company and also its shareholders getting money, not workers,” he said.

The same report from Windfall Trust found in the worst-case scenario, tax revenue in developed countries could fall by 15% from labor displacement.

Other predictions are less decisive.

The CBO found, in a 2024 report, that if employees are permanently displaced by AI or take lower-paying jobs after their displacement, tax revenue could decrease. But the report also left open the possibility that AI may create new tasks or jobs, and offset labor and revenue lost.

“I think a lot of the AI conversation is fear-based at the moment,” said Sean Bray, vice president of global tax policy at the Tax Foundation. Thus far, AI is proving to increase employees’ productivity rather than replace them, he added.

Token Tax

Amid the uncertainty, proposals on how to tax the technology from academics, think tanks, companies, and governments are trickling into public discourse.

Most of them are high-level and undetailed.

Both Amodei and California gubernatorial candidate Tom Steyer (D) have proposed a token tax. Steyer’s token tax would charge “a fraction of a cent for every unit of data processed by large AI companies.” The tax revenue would be used for a sovereign wealth fund. But Steyer doesn’t include a numerical value to the “fraction of a cent” that would be charged.

In an emailed statement to Bloomberg Tax, a spokesperson for the Steyer campaign said the candidate’s tax would apply to “large-scale corporate AI use — not consumers.”

Last month, OpenAI released an industrial policy report detailing ideas to “keep people first” amid looming economic and social disruptions. The company suggested governments should increase taxes on capital gains, corporate income, or explore new taxes related to automated labor to fund government social programs.

There’s been no formal introduction of AI tax legislation in Congress. But Sen. Mark Kelly (D-Ariz.) suggested—similar to Steyer—that AI firms make a contribution that would then be fed into a federal fund to “provide resources to reinvest in workers.” Kelly’s proposal also doesn’t provide specifics on whether the “contribution” would constitute as a tax, or the amount AI companies would be expected to give.

Asked about the specific mechanism for companies to contribute to his proposed “AI Horizon Fund,” the senator told Bloomberg Tax that a tax was in the realm of possibility. He added that he’s had a “number of conversations” with AI companies about contribution options, and eventually those options will be narrowed down. But Kelly wouldn’t elaborate on the other options outside of a tax.

In 2024, Member of Parliament Pasquale Tridico said AI should be taxed across Europe. He admitted, however, that he was unsure how to tax the technology.

Bearer-Friend suggested that AI companies pay their tax in stock to the federal government, rather than with cash.

Old Tricks for New Dog?

Bray, Holtz-Eakin and Lee Lockwood, a professor of economics at the University of Virginia, agree that consumption-based taxes are a good option to replenish labor revenue eliminated by AI.

Lockwood said in a worst-case scenario, where labor earnings plummet and consumer spending is high, “that type of tax could raise a lot of revenue.”

Where they diverge is on whether the world should adopt new consumption taxes for AI.

Lockwood was open to the idea of a token tax, but Bray was more skeptical.

Bray said that any “AI excise tax” would hurt innovation and competitiveness.

He suggested that instead of countries applying a “new, regressive tax” that will ultimately be passed down to consumers, “a better-designed tax would be to broaden the VAT base to include digital services not yet covered by existing legislation.”

Value-added taxes are levies imposed on a product at every point in the supply chain. These charges are normally passed down to the consumer.

About 170 countries around the world have a VAT. But an additional consumption tax in the US would be politically unpopular.

Lockwood noted that there are ways to make consumption levies progressive by including exceptions for certain income brackets, but he acknowledged their lack of political viability.

Increasing taxes on capital gains would also run into a political wall. Congress won’t pass an increase on capital gains taxes, Holtz-Eakin said. “We’ve been through that thicket a million times.”