When AI Grows the Economy but Shrinks the Tax Base

Discussing the potential for powerful, productivity-enhancing AI systems to erode tax bases through labour displacement.

Authors: Trish Ieong, Akbar Saputra, Anuja Maniar and Deric Cheng

Governments have quietly built a fiscal assumption into their AI strategies: that productivity growth will translate into revenue growth, and that the proceeds will help meet rising public costs from ageing populations and stretched services. It is a reasonable hope. But productivity gains and tax revenues are not the same thing, and the gap between them — in an AI-driven economy — could be larger than most fiscal planning currently contemplates.

The mechanism is structural. Most OECD countries tax labour income significantly more heavily than capital income. When AI displaces a worker, the income that worker would have earned does not disappear — it shifts from wages to profits, and in doing so moves from a higher-taxed base to a lower one. If those profits accrue primarily to companies headquartered abroad — a reasonable assumption given the current geography of frontier AI development — governments lose their taxing rights on that income entirely. And where AI’s gains flow to consumers as lower prices or free services rather than as taxable income, they escape the fiscal net altogether. Meanwhile, the costs of supporting displaced workers — income transfers, retraining, expanded safety nets — rise at exactly the moment revenues are under pressure.

This is the fiscal squeeze that most AI strategies have not yet seriously reckoned with.

This paper maps out these fiscal channels and simulates how they may affect an ‘average OECD country’ under six different scenarios. Cumulative tax revenues over 10 years are roughly flat in the scenario with the most favourable assumptions, but decline by up to 15% with the least favorable ones. The net fiscal impact is even worse once support for displaced workers is factored in, with declines of up to 28% under relatively modest spending assumptions.

There is, admittedly, much uncertainty around AI’s trajectory and economic impacts, as we acknowledge in the paper. Our results are not intended to act as predictions of how severe fiscal pressures will be, and users may accordingly test out their own assumptions using our interactive simulator.

Fiscal impacts are also likely to vary considerably across OECD countries, depending on their tax structures, how AI affects their terms of trade, demographics, and existing fiscal headroom. Some countries’ tax systems may prove reasonably resilient, while others may experience pressures more severe than the averages we describe. Because the right policy response will differ across jurisdictions, we do not prescribe specific reforms. Instead, we offer high-level suggestions to help policymakers understand and prepare for the fiscal risks of AI.

Countries that recognize the risks early and take proactive steps to broaden their tax bases will generally be better placed to capture the productivity gains from AI and smooth out fiscal pressures. Countries that ignore the problem and wait for concrete evidence of mass displacement before acting may find themselves left with only the most drastic options, such as inflationary financing, capital controls, austerity or even default. Changing the structure of a tax system is never easy, but it is harder to do well in the middle of a crisis.