Policy & Regulation

OECD Data Shows Why AI Growth Won't Rescue Public Finances

OECD projections to 2060 show productivity gains barely dent a 4.7%-of-GDP fiscal gap, while AI returns may escape the labour taxes funding half of EU revenue.

The Fiscal Conversion Gap: Why AI-Led Growth May Fail to Rescue Public Finances
The Fiscal Conversion Gap: Why AI-Led Growth May Fail to Rescue Public FinancesPexels / pixabay
By James Calloway6 min read

Why it matters

  • The median OECD country needs a 4.7 percentage-point-of-GDP fiscal improvement from 2027 to stabilise debt to 2060.
  • A 10% output gain via employment cuts the projected fiscal gap by ~2.5 points of GDP; the same gain via productivity has negligible impact.
  • Labour taxes supplied 51.5% of EU-27 tax receipts in 2024, versus 26.8% from consumption and 21.6% from capital.
  • The findings come from OECD Working Paper No. 1877 (September 2026) and a VoxEU column by OECD economist Yvan Guillemette, 15 November.
  • France, Norway and Spain levy wealth taxes in differing forms, but the analysis calls relying on them to capture AI gains 'over-optimistic'.

The median OECD country must improve its fiscal position by 4.7 percentage points of GDP from 2027 just to stabilise debt ratios through 2060 — and OECD projections show that even a 10% boost to potential output from productivity gains would do almost nothing to close that gap. That is the central finding of an OECD paper dated September 2026, presented by OECD economist Yvan Guillemette in a VoxEU column for CEPR on 15 November, which argues that growth alone cannot restore sound public finances across the bloc's 35 members and accession candidates.

The stakes are straightforward. Governments facing high debt have long preferred to "grow out" of it rather than raise taxes or cut spending. Many policymakers now treat widespread AI adoption as the productivity engine that will make this possible. The OECD's own long-term model, laid out in OECD Economics Department Working Paper No. 1877, suggests otherwise — and an analysis published by The Economy on 17 November extends the argument one step further: even if AI delivers the growth, much of the resulting income may never reach the taxable wage base that funds most of the modern state.

How large is the OECD's projected fiscal gap?

The September 2026 paper calculates fiscal gaps to 2060 for 35 OECD members and accession candidates. The numbers required simply to hold debt-to-GDP ratios near current levels while maintaining existing services and transfers are large:

  • 4.7 percentage points of GDP: the median adjustment needed from 2027 onward.
  • 8+ percentage points of GDP: the required adjustment for 10 of the countries studied.
  • Zero: the allowance made for paying down existing debt stocks, although political debate routinely conflates stabilisation with repayment.

According to the paper's accompanying analysis, pensions and health care — not interest costs — drive most national fiscal gaps. An ageing population compounds the problem by raising spending while shrinking the share of the population in work.

Why does productivity growth barely help?

Several mechanical factors blunt the fiscal effect of faster output growth. Productivity growth tends to push up equilibrium interest rates, so the gap between interest rates and growth rates narrows less than optimists assume. Many expenditure items track private-sector pay and prices, and pension payments rise with earnings. Higher GDP therefore drags higher spending behind it.

The sharpest result comes from comparing scenarios that deliver identical output gains by 2060:

  • Achieving 10% higher potential output through productivity has a negligible impact on the projected median fiscal gap.
  • Achieving 10% higher output through employment cuts the projected gap by roughly 2.5 percentage points of GDP.

The reason is that additional employees pay income tax and social security contributions, and some government transfers to them fall away. Higher earnings from productivity, by contrast, arrive paired with higher indexed expenditure. Figure 2 in the analysis points to low labour-force participation among women as the clearest remaining reservoir of employment-led deficit reduction.

The takeaway in plain words: for public finances, who produces the output matters more than how much output is produced.

Can AI-led growth outrun the tax base?

Here the analysis goes beyond the OECD paper. Conventional reasoning holds that productivity gains raise wages, spending, profits and thus tax receipts across income tax, social security, VAT and corporate tax. AI may scramble that chain even without large-scale job losses.

If a relatively small number of businesses capture most of the AI gains, the analysis argues, the result would be substantial increases in profits and share prices for those firms relative to a much more limited rise in total pay bills. The additional returns — profits, IP-related income, shareholder distributions and unrealised share-price appreciation — would concentrate among few individuals and companies, shaped by capital, skills, flexibility and market power. Some of the payments would flow to overseas technology providers, IP holders and foreign shareholders. Each of these income forms is taxed at different rates, through different mechanisms, on different timelines.

The EU's revenue structure shows what is at risk. In 2024, taxes on labour accounted for 51.5% of total tax receipts in the EU-27, against 26.8% from consumption and 21.6% from capital. Most labour tax is deducted at source — an extremely reliable funding mechanism. A growth model that shifts value from payrolls to capital gains and unrealised wealth strikes directly at the largest pillar.

One diagnostic the analysis proposes is "buoyancy" (the ratio of the percentage increase in tax receipts to the percentage increase in nominal GDP). Values persistently below one would signal that output growth is no longer converting into revenue growth. Actual readings would be clouded by discretionary tax changes and cyclical effects — a caveat the author concedes.

In plain words: the question is not whether AI creates wealth, but whether governments can tax where it lands.

What cuts against the pessimistic case?

The analysis is careful to flag counter-evidence. Nothing in the argument implies total tax revenues would fall, and significant sums could be raised from the very highest earners alone. Task-level studies point the other way on distribution: a paper by Brynjolfsson, Li and Raymond in the Quarterly Journal of Economics (2025, vol. 140, no. 2) found generative AI produced significantly greater performance gains for lower-tenure call centre operators than for experienced staff. Efficiency gains inside government departments themselves could also temper expenditure growth linked to rising living standards.

There is also a partial offset running the other way. If profits rise relative to wages, wage-indexed public spending grows more slowly, easing some OECD-identified pressure. But that saving would be countered by reduced employment, lost social security contributions and new spending to address unemployment. Broad assumptions about productivity growth, the analysis notes, simply cannot resolve questions this granular.

Why won't wealth taxes close the gap?

Taxing the AI winners directly looks attractive and runs into practical walls. Wages are taxed continuously; asset appreciation can remain unrealised indefinitely. Profits may arise across multiple countries, complicating allocation. The existing policy patchwork illustrates the difficulty:

  • France confines its wealth tax to real estate following its 2018 reform.
  • Norway continues to levy a fairly wide-ranging annual net wealth tax.
  • Spain operates a limited temporary tax on very substantial wealth.

An IMF How To Note (2024) catalogues possible routes through income, capital gains and inheritance taxes, while OECD work on capital gains taxation (Tax Policy Studies, 2025) documents the trade-offs each creates. The analysis's verdict is blunt: it would be "over-optimistic" to suppose all or most of AI's productivity dividend could be tapped in an acceptable and efficient manner.

What options remain for heavily indebted states?

For countries with substantial debt, ageing-driven pension and health costs, defence pressures and higher interest payments all arrive together. If AI gains concentrate among a few firms and individuals, GDP growth alone will not suffice. The analysis lists the plausible levers:

  • More effective collection of taxes already due.
  • Wider international apportionment of profits and rents.
  • Greater reliance on taxes on investment income and distributions — with attendant risks of postponement and avoidance through relocation.
  • Measures to support employment, given the quantified superiority of employment-led growth.
  • Spending reductions, weighed against possible damage to growth and equality.

One option the author rejects: taxing AI use itself, which would tax the cause of the gains rather than the resulting rents.

The open question is whether today's tax systems can capture an appropriate share of AI-created value without deterring its use. Until that is answered, the analysis concludes, increased national wealth does not automatically translate into adequate resources for government — and the debt question stays exactly where it was.

Source: economy.ac

Share this article:

More from James Calloway

James Calloway

Show full bio

News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.

722 articles

Related articles

  1. Trump's AI Growth Promise Echoes Reagan's Tax Cut Fantasy
  2. Dutch government plans 25,000 job cuts in AI efficiency push
  3. OpenAI's 2026 revenue projection drops $20bn short of investor guidance
  4. Fed Chair Warsh Sketched 3 AI Futures. One Is a Valuation Trap
  5. OpenAI's Lehane Frames 2028 Election as Defining Test for AI Policy

« Previous articleNext article »