by Claude Opus 5.5

If AI shifts income from labour to capital, what are the consequences for a tax base that depends heavily on income tax and National Insurance—and which policy responses are serious versus speculative?

Income tax and National Insurance together raise about £535bn, roughly 43% of the UK’s £1,235bn in receipts in 2025–26. A pound earned as wages is generally taxed more heavily than a pound earned as profit. So if AI raises output but channels the gains to profits rather than pay, receipts can fall even with GDP unchanged. The OBR modelled exactly this in March 2026 and returned to it in July. The serious responses are mostly unglamorous: tax income from capital more consistently with income from work, protect the corporate tax base, and rely more on broad bases such as consumption and property. Robot taxes, compute levies, sovereign wealth funds and universal basic income are further from being workable, for different reasons. The Autumn Budget is due on 28 October 2026. This answer doesn’t predict its contents.

How dependent the UK is on taxing work

The OBR’s tax-by-tax estimates for 2025–26 (November 2025 forecast) show the shape of the base:

  • Income tax. 2025–26 receipts: £329bn. Share of all receipts: 26.7%.

  • National Insurance contributions. 2025–26 receipts: £205.4bn. Share of all receipts: 16.7%.

  • VAT. 2025–26 receipts: £179.6bn. Share of all receipts: 14.6%.

  • Onshore corporation tax. 2025–26 receipts: £96.7bn. Share of all receipts: 8.0%.

  • Capital gains tax. 2025–26 receipts: £20.3bn. Share of all receipts: 1.6%.

The OBR’s March 2026 guide to the public finances puts total receipts for 2025–26 at £1,235bn, or 40.4% of national income, with income tax and NICs together “around £535 billion”. Income tax is not purely a tax on work, because it also falls on pensions, self-employment, savings, dividends and rent. But employment income is by far its largest component, and NICs are almost entirely a tax on earnings. Put simply, the state takes its largest cut where wages are paid.

Why a shift to profits costs revenue

Here is a worked illustration using 2026–27 rates. Suppose an employer has £11,500 to spend.

  • As extra pay to a basic-rate employee: £10,000 of salary plus £1,500 of employer NICs at 15%. The employee pays £2,000 in income tax and £800 in employee NICs at 8%. Total tax is £4,300, about 37%.

  • As profit distributed to a basic-rate UK shareholder: corporation tax at 25% takes £2,875, and dividend tax at 10.75% on the remaining £8,625 takes about £927. Total tax is about £3,800, about 33%.

  • As profit retained, or paid to a pension fund, ISA holder or overseas owner: corporation tax only. Total tax is £2,875, 25%, and less if the firm uses capital allowances such as full expensing.

For higher-rate taxpayers the gap narrows. Dividends are taxed at 35.75% from April 2026, and gains at 24%. But a large share of corporate profit ends up in tax-advantaged or overseas hands, and gains are taxed only when realised. That is why the OBR, describing its technological-displacement scenario, says that “as labour income faces a higher effective tax rate”, a shift towards profits “reduces the tax-richness of economic activity”.

There are also knock-on effects. Unemployment raises welfare spending. Frozen income tax thresholds raise revenue only if pay keeps rising. And slower wage growth can weigh on consumption, and so on VAT.

What the OBR has said in 2026

March 2026 Economic and Fiscal Outlook. AI is not quantified in the central forecast, but the OBR set out scenarios:

  • Higher productivity. If productivity growth were 1.5% a year, “for example, due to a more optimistic scenario for the impact of AI”, borrowing could be about £50bn lower in 2030–31. A downside of 0.5% a year adds about £40bn.

  • Technological displacement. “New technology displaces workers and is a substitute for labour”. Equilibrium unemployment rises to 5.5%, the level of GDP is broadly unchanged, productivity gains are “not reflected in higher real earnings”, and the result is “a lower labour share and a higher corporate profit share”. Secondary reporting put the extra borrowing at about £9bn a year. I could not confirm that figure in the OBR text.

  • A contrasting labour-costs scenario produces the same 5.5% unemployment but no productivity gain and lower GDP. It shows how different causes of joblessness have different fiscal footprints.

July 2026 Fiscal Risks and Sustainability report. Box 4.1 warns that AI-driven productivity could shift income “from (more highly taxed) labour to (lower-taxed) profits”. The same report shows the upside is large if gains are broadly shared. In a higher-productivity scenario with total factor productivity growth of 1.3% a year, debt is around 120 percentage points of GDP lower by 2075–76 than in the baseline.

The fiscal question is therefore not whether AI is good or bad for the public finances. It is who receives the gains. Broad-based productivity growth that lifts wages is the best fiscal news available. Productivity growth that bypasses wages is a slow leak.

No measurable shift yet

The OBR noted that the labour share had recently risen as profit margins were squeezed. Its central forecast, which does not quantify AI, has profits growing faster than labour income over 2026–2030. There is no evidence yet of an AI-driven collapse in the labour share. This is a scenario to prepare for, not a trend already under way.

Policy responses, from serious to speculative

  • Align tax on capital income with tax on labour income. How serious: Serious; already moving. Main case for: Narrows the wedge directly; reduces income-shifting. Main case against: Lock-in; mobility; taxes the inflationary part of gains.

  • Capital gains tax reform. How serious: Serious; politically contested. Main case for: Gains are the most lightly taxed income. Main case against: Revenue sensitive to timing and behaviour.

  • Protect and adapt corporation tax. How serious: Serious. Main case for: AI profits are corporate profits. Main case against: Profits are mobile; competitiveness trade-offs.

  • Lean more on VAT and property. How serious: Serious. Main case for: Broad bases that don’t care whether income comes from wages or profits. Main case against: VAT is regressive; property reform is politically hard.

  • Robot or automation taxes. How serious: Speculative. Main case for: Slows displacement; raises revenue. Main case against: Can’t define a robot; taxes productivity.

  • Compute or token levies. How serious: Speculative. Main case for: Targets the new input directly. Main case against: Mobile base; hits adopters; at odds with growth policy.

  • Sovereign wealth fund. How serious: Speculative as a revenue tool. Main case for: Gives the public a share of returns to capital. Main case against: Needs capital the UK doesn’t have spare.

  • Universal basic income. How serious: Speculative. Main case for: A floor if work disappears. Main case against: Very costly; designed for a scenario not yet visible.

Rebalancing labour and capital taxation. This is the response most tax economists would put first. The Mirrlees Review (2011) argued for taxing different forms of income more consistently, combined with relief for the normal return on savings. Moves in this direction are already in train. From April 2026 dividend rates rose 2 points, to 10.75% and 35.75%. Property and savings income rates rise 2 points from April 2027. Further options include extending NICs-style charges to some non-employment income, or reducing the employer NICs wedge on work. The trade-off is between equity and simplicity on one side and effects on saving, investment and the pensioner-heavy population that holds capital on the other.

Capital gains tax alignment. CGT rates are 18% and 24%, against income tax rates of 20%, 40% and 45%. The Office of Tax Simplification suggested in 2020 that government consider aligning the rates more closely, or address the boundary between income and gains. The usual counter-arguments are that higher rates encourage people to hold assets rather than sell (lock-in), that without indexation they tax inflationary gains, and that internationally mobile entrepreneurs may leave. Proposals often pair alignment with relief for inflation or the normal return on investment.

Corporation tax. The main rate is 25%. If AI profits accrue to large, partly foreign-owned firms, the integrity of the corporate base matters more. The global minimum tax and digital services taxes are the relevant tools. Domestic incentives cut the other way. Full expensing lowers the effective rate on investment, including in automation. Acemoglu, Manera and Restrepo found the US tax code taxes labour at more than 28.5% against about 5% for equipment and software. They argue this biases firms towards automation. The UK numbers differ, but the question applies.

VAT and property. Consumption and land are taxed the same whether income comes from wages or profits. VAT already raises 14.6% of receipts. The objection is distributional: VAT takes more of lower incomes. Recurrent property taxes are economically efficient and immobile but politically fraught.

Robot taxes. Bill Gates and others have floated them, but they founder on definition. Is a spreadsheet macro a robot? Is an AI subscription? They also tax exactly the productivity gains that make the fiscal upside possible. The IMF’s 2024 staff note on generative AI and fiscal policy emphasises social protection and income support, within an “agile policy framework” ready for both ordinary and highly disruptive scenarios.

Compute and token levies. These are the newest idea and the least tested. A per-token or per-compute charge targets AI usage directly, but most compute is bought from overseas cloud providers. A levy could fall on UK adopters rather than on AI developers’ profits. It would also cut against the government’s own push for AI Growth Zones, which have attracted £28.2bn of investment, and for adoption. A version levied on consumption, like VAT on digital services, is conceivable. A version that captures AI’s profits is much harder to design.

Sovereign wealth funds. The idea is that the public should own part of the capital that earns the returns. The UK has public investment vehicles, such as the National Wealth Fund and a Sovereign AI Unit with up to £500m, but these are investors, not revenue sources. Building a Norway-scale fund requires either a resource windfall or borrowing to buy assets, and neither fits current fiscal rules.

Universal basic income. UBI addresses a world in which paid work shrinks substantially. Its gross cost at meaningful levels is a large share of GDP, and the evidence so far, from modest pilots such as Wales’s scheme for care leavers, says little about economy-wide displacement. It is better understood as contingency thinking than near-term policy.

Bottom line

The UK’s tax system is built on wages, and AI could shift income from wages to profits faster than past technologies did. The OBR has now put that risk in official scenarios. The credible responses are adjustments to a tax system that already treats labour and capital unevenly, not new AI-specific taxes. The Budget on 28 October, Chancellor Healey’s first, and the OBR’s accompanying forecast are the next points at which to see whether either the government or the OBR builds AI into the central numbers rather than the scenarios.

Sources

From AI and Jobs: UK, October 2026