Key Highlights

• American consumers and businesses spend about USD 700 billion a year on building and running AI models, a small base next to a USD 16 trillion labour bill.

• California voters will decide in November on a 5% levy on the net Wealth of residents worth more than USD 1 billion.

• Eleven rich countries taxed fortunes in 1990, but only four do today, partly because the wealthy shift Assets offshore and private businesses are hard to value.

• A consensus is emerging among economists that changes to existing taxes on Capital income are preferable to new levies.

Slowing Adoption or Taxing the Machines

One way for fiscal systems to cope with AI is to slow or stop its adoption. Three economists, at two US universities and the IMF, have argued that America's tax system favours automation, because it taxes labour heavily while giving generous treatment to assets such as software and machinery. That, they say, gives employers an incentive to replace workers with technology.

Some economists therefore prefer to tax AI directly, through a compute tax levied on the computing power used to train models or a token tax on each use of a model. With the public increasingly hostile to AI, such a tax could prove popular even if it carried a significant cost to an economy deprived of AI's benefits.

A Small Base

The Revenue could be surprisingly small. One estimate puts spending by American consumers and businesses on building and running AI models at about USD 700 billion a year, a tiny base relative to a total labour bill of USD 16 trillion last year. AI spending may balloon from here, but by how much is anyone's guess. Total Capital Investment in AI by big technology companies is forecast to level off from 2028, while technological improvements may mean that businesses' spending on AI need not rocket. Revenue from such a tax is therefore highly uncertain.

The Wealth Tax Option

Rather than tax AI at source, governments may find it easier to go after the gains of the boom. One option is a wealth tax. In November, Californians will vote on a proposal to levy 5% on the net wealth of residents worth more than USD 1 billion. Politicians in other states like the idea, and several prominent economists have endorsed it. Lowering the threshold below USD 1 billion would capture more of the multimillionaires the AI boom is creating.

Robust academic evidence shows, however, that the rich respond to wealth taxes by moving assets offshore or relocating themselves. Privately owned businesses, where many hold their riches, are also hard to value, which makes it difficult for tax collectors to know how much to demand. These problems help explain why many countries that introduced wealth taxes have abandoned them: in 1990, 11 rich countries taxed fortunes, and today only four do.

Doubts About Yield and Fortunes

Even if AI could help value unlisted businesses, or AI tax auditors made it harder to shift assets abroad, a wealth tax might not raise nearly enough. In March a group of American lawmakers introduced a proposal for a 2% annual tax on the net worth of households and trusts valued above USD 50 million, plus a 1% Surcharge on those above USD 1 billion. On its proponents' own, extremely generous, assumptions it would raise the equivalent of 15% of current labour taxes, though commentators judge the realistic yield to be considerably lower. Most independent studies of wealth taxes also find that they slow the economy, which reduces receipts from other taxes.

The assumption that AI will create many more mega-fortunes to tax faces its own complication: many appear to be donating money as fast as they make it. All seven co-founders of Anthropic, the leading AI lab now seeking a USD 2 trillion initial public offering, have pledged to give away 80% of their wealth. An AI economy could see a large share of personal wealth held in charitable foundations rather than bank accounts and mansions, and if taxable fortunes do not grow fast, high tax rates would erode them quickly.

Reforming Capital Taxes

Among economists, a consensus is emerging that changes to existing taxes may be preferable to new levies, especially taxes on capital income, which include corporate profits, dividends and realised capital gains. If AI lives up to expectations, related capital income could be enormous.

One reform would bring tax rates on capital gains closer to those on labour income. California already applies ordinary income tax rates to realised gains. Belgium recently introduced a 10% tax on gains from financial assets, closing a loophole that had spared such income. Australia is raising taxes on realised gains but giving relief by adjusting those gains for inflation. The Netherlands is seeking to replace a system that assumes every investor earns the same gains, which favoured good investors, with one that taxes actual gains.

To limit the damage to enterprise, levies on gains above the normal Return on Investment could be paired with equally generous relief for capital losses, allowing losses to be written off against future gains and perhaps past gains and other income.

Corporate and Consumption Taxes

Corporation taxes could contribute more, especially if the AI economy generates excess profits above what a company needs to earn to keep investing. Such profits can come from Monopoly power, scarce land or unique data sets and, in theory, can be taxed with relatively little harm to incentives. A number of countries are already raising corporation taxes, reversing the trend of the 2000s and 2010s.

AI also strengthens the case for the taxes many economists prefer: those on consumption. They are difficult to avoid and can raise vast sums. Rich people spend more than poor people and so pay the largest share of consumption taxes, while governments that want to compensate poorer households, which pay more as a share of income, can do so through the welfare state. As machines take a greater role in producing income, tax authorities may have to pay less attention to how money is made and more to where it is spent.

Choices Still on the Table

The California wealth tax vote in November, the size of AI spending as a tax base, and the pace of changes to capital gains and corporation taxes are the developments that mark which route governments take.