The potential impact of AI on the labour share and tax receipts

An important area of uncertainty in our long-term projections is the impact that the increased use of artificial intelligence (AI) could have on the economy and on future tax receipts. Tax revenue is driven both by overall nominal GDP growth and by the composition of GDP, particularly the shares relating to labour and profits – both because the effective tax rate on each is relatively high and because it is currently higher on labour than on profits.^a

AI could significantly affect overall GDP growth through its impact on productivity growth. But the size of this effect is extremely uncertain, with a very wide range of estimates in the economic literature. In Briefing paper No.9: Forecasting productivity, based on this literature and our own analysis, we produced a central estimate that AI could boost annual UK productivity growth by 0.2 percentage points over the next decade, but also considered alternative scenarios where this effect was less than 0.1 percentage points or reached up to 0.8 percentage points. In Chapter 2, we explain that a larger long-term impact from AI is one motivation for the higher productivity scenario we use in this report.

AI could also significantly affect the relative shares of GDP going to labour and profits, which is the focus of this box. The effective tax rate on labour income is higher than on profits, so a lower wage and salary share would, other things equal, decrease tax receipts (and vice versa). The framework presented by Acemoglu (2024) suggests that the long-term impact of AI adoption on the labour share will depend on the degree to which AI becomes incorporated into economic activity, whether it functions as a substitute or a complement to existing labour, and the degree to which new jobs are created by AI.^b,c In more detail, Acemoglu suggests that:

• If AI is used as a substitute for labour, so that work tasks previously done by workers will be done by AI, then the income from this work will largely accrue to the owners of AI capital rather than workers (although wages may still increase).^d This could be associated with a reduction in the share of labour income in GDP.
• If AI functions as a complement to labour, it will make workers more productive and raise their wages. However, this would have an ambiguous effect on the overall share of labour compensation in GDP as it will also raise returns to capital.^e
• AI will likely create new tasks and jobs, either directly as a result of the emergence of new tasks that are complementary to AI or indirectly as higher productivity from AI raises real incomes and creates new demand. Where these tasks are labour intensive, this raises the labour share under Acemoglu’s framework.

The net impact of AI on the labour share is therefore not clear as it depends on the balance of these factors, alongside the degree to which they affect the relative bargaining power of employees and employers. To date, most studies find significant scope for AI to both complement and substitute for labour, especially in more developed economies.^f,g,h The estimates we produced in Briefing paper No.9 for the UK economy suggest that, over the next 10 years, around 10 per cent of the UK labour force could be exposed to substitution from AI and a further 30 per cent could have their work complemented.ⁱ There is some support in the literature for future developments in AI capability leading to a greater share of work being substituted beyond the next 10 years – though this is very uncertain.^j

Our baseline long-term scenario in this report implicitly assumes that the labour and profit shares remain similar to the shares at the end of our medium-term forecast in 2030-31. This would be in line with the stable trend seen over the previous 40 years – a period which included the adoption of the previous general-purpose technology (GPT) through the ICT revolution. A long-term shift in the labour share would require generative AI to be significantly different to previous GPTs, where job destruction has resulted in higher productivity and higher wages for existing workers, and a limited overall impact on the labour share. Drozd and Tavares (2024) have suggested that this could occur because AI differs from previous new technologies which were automating tasks that were already heavily mechanised.^k To demonstrate the potential fiscal consequences of changes in the labour share, we present an illustrative scenario where the share of GDP accounted for by wages and salaries declines significantly:^l

• In the baseline scenario, wages and salaries are unchanged at 40 per cent of GDP. Tax as a share of GDP remains roughly constant at 38 per cent (Chart E). This would be consistent with Acemoglu and Restrepo’s (2019) finding that historically, displacement effects have largely been offset by task creation and labour reinstatement.^m
• In the reduced labour share scenario, AI – in net terms – leads to a significant reduction in jobs in the economy and/or large changes in relative wage bargaining power away from workers to employers. The wage and salary share falls from 40 per cent in 2030-31 to 20 per cent in 2075-76. Much of the literature supports AI putting downward pressure on the labour share,ⁿ but the scale of reduction we present here is at the upper end of the scenarios explored in it, such as Benzell et al. (2015) or Drozd and Tavares (2024).^o,p In our reduced labour share scenario, total GDP is unchanged relative to the baseline but tax falls to under 35 per cent of GDP over the same period, nearly 4 percentage points lower than in the baseline. This occurs because the fall in taxes paid on wages and salaries, which have an average effective tax rate (ETR) of 46 per cent over the projection, is only partly offset by a rise in taxes paid on profits, for which the average ETR is 27 per cent.^q

This reduced labour share scenario is highly stylised. For example, it assumes the effective tax rates on profits and wages and salaries are unchanged from the baseline. However, ETRs could change – for example, if AI affects the earnings distribution, or if future governments respond with changes to policy. The scenario also assumes that additional corporate profits from AI would be captured by existing taxes. However, it may be that this income cannot be efficiently taxed using existing instruments, and so some commentators have suggested that increased taxes on consumption or wealth, or new taxes directly on AI use, could be needed.^r,s

There are also wider areas of potential fiscal risk associated with AI, including: the gains from AI accruing offshore or in parts of the economy that are hard to tax; increased public spending pressures due to structurally higher unemployment (see Box 2.2 in the March 2026 Economic and fiscal outlook) and/or the need to retrain workforces and support the transition to new occupations; or other countries building a dominance in AI infrastructure and attracting capital away from the UK, leading to higher interest rates.^t AI also presents potential opportunities for improved public service delivery that could reduce fiscal pressure.

Chart E: National Accounts taxes projection under AI-driven labour share assumptions

Note: This scenario focuses just on tax receipts using the National Accounts definition, and so differs to other measures of receipts shown in this chapter.

Source: OBR