In the past year American pre-tax corporate earnings have grown by $1.2trn, or 30%. Companies in the S&P 500 index are handily beating forecasts. Taxes on corporate profits have collapsed (see chart 1).
VAST SPENDING on AI infrastructure has created a profit boom. And yet Uncle Sam is seeing little of the benefit. Government beancounters did expect lower receipts this year, owing to recent changes that let firms write off more investments against their tax bills. But the scale of the drop has surprised even them, as unusually profitable tech giants are writing off unusually high capital spending. The episode hints at how AI may upend tax systems across the world.
You might think of AI as a godsend for governments, many of which are accumulating vast debts (see chart 2). Higher economic growth could reduce the debt-to-GDP ratio after 30 years by 50% or so of GDP, relative to where it would be otherwise. The Budget Lab, part of Yale University, explores a range of scenarios in which AI boosts annual GDP growth to as much as 3.3%. GDP growth of 3.3% would be impressive, but not unprecedented.
“As AI reduces labour’s share of income, the traditional tax base erodes,” argue Anton Korinek of Anthropic and Lee Lockwood of the University of Virginia, in a recent paper. In an essay last year Philip Trammell of Epoch AI, a think-tank, and Dwarkesh Patel, a podcaster, entertained the possibility that “the share of total income paid to robot-owners” would rise to 100%. Economists expect AI to boost productivity growth, which should in turn create a bigger economy. In a recent paper Karen Dynan, Douglas Elmendorf and Louise Sheiner, three economists, model the potential effects of AI on America’s debt. This tax mix may not suit a world in which AI displaces workers or reduces their salaries, even as corporate profits rise. This produces higher tax revenues—but the extent of the boost depends greatly on the distribution of the extra income between capital and labour. The Yale economists envisage AI having fairly modest economic effects for the time being. Others envisage a deeper economic transformation. A shift need not be this extreme to produce a profound fiscal shock.
But the majority of governments that rely on labour taxes would be crippled (see chart 4). If real GDP rises by 30% and real tax revenue rises by 20%, the government might be smaller relative to the economy but it would still have more resources at its disposal. In our scenario, if the Italian government wanted to replace the lost earnings of the people that AI had displaced, it would need to find another $200bn per year as its tax revenues declined by $90bn, a fiscal deterioration of about 10% of GDP. We estimate that American consumers and businesses currently spend about $700bn a year on building and running AI models. This is a tiny tax base relative to the total labour bill ($16trn last year). Total capital investments in AI from big tech firms is expected to plateau from 2028, while technological improvements might mean that businesses’ spending on AI need not rocket. In March a group of American lawmakers introduced a proposal to establish a 2% annual tax on the net worth of households and trusts valued at over $50m, and an additional 1% surcharge on those above $1bn. It would raise the equivalent of 15% of current labour taxes.In reality, it would probably raise a lot less, according to the (extremely generous) assumptions of the bill’s proponents.
Because it taxes labour fairly heavily relative to capital and because wages and salaries form an unusually large share of GDP, america takes a smaller but still significant hit—both.
Even if AI could determine the “true” value of an unlisted firm or super-clever AI tax auditors make it harder to shift assets abroad, a wealth tax might not raise anywhere near enough money. To gauge how profound, The Economist has calculated the effects on tax revenue of AI causing labour’s share of national income to decline by ten percentage points. A few countries with relatively high taxes on capital, including Australia and South Korea, emerge relatively unscathed. In Italy, the hardest-hit big country in our model, the budget deficit could more than double as a share of GDP (assuming that spending remains constant relative to GDP). France and Germany are not far behind. A falling tax-to-GDP ratio is not necessarily a bad thing. And if the government could then deploy AI, it might also be able to spend that money more effectively. In health care, for instance, AI could improve scheduling, diagnostics and drug discovery, while reducing the overall cost of the system. It could usher in a golden age of public services. Even as it raises their cost, AI may also increase demand for public services. People displaced by AI would need retraining and unemployment support. Some governments could consider a universal basic income to ward off the mobs. Yet it might raise surprisingly little revenue. AI spending may balloon from here, but by how much is anyone’s guess. The revenue from such a tax is thus highly uncertain. Most independent studies of wealth taxes find that they also slow the economy, reducing the receipts from other taxes.
