In OECD countries, income tax and social contributions represent nearly half of total tax revenues. The technology that replaces these workers pays a light tax burden, sometimes even receives subsidies.

This imbalance is not new. But the acceleration of automation gives it an unprecedented dimension: if the tax base for work erodes, states lose their primary source of funding precisely at the moment when they would most need to finance retraining, protections, and transitions.

The Essential Points

  • In OECD countries, income tax and social contributions represent approximately one-third of the cost of labor and nearly half of total tax revenues, according to primary OECD publications (Revenue Statistics, Taxing Wages), notably cited in a Nature article from 2026.
  • The technology that replaces work benefits from a structurally lighter tax burden: accelerated depreciation, R&D credits, reduced rates on corporate profits.
  • The “robot tax” popularized by Bill Gates encounters a prohibitive technical obstacle: modern AI does not contain an isolatable machine to tax.
  • Economists propose a more practical approach: progressive taxation on profits relative to the number of employees, which shifts the debate from symbolism to effectiveness.

A significant portion of an employee’s cost in wealthy countries corresponds, for the employer, to taxes and social contributions, according to OECD data. A robot that performs the same tasks triggers no equivalent. It generates profits, taxed at ordinary corporate rates, with tax deductions for investment that often reduce the actual bill well below the nominal rate.

The question is not abstract. It silently structures the public finances of entire countries, and its magnitude will grow as automation spreads to new sectors.

Work Bears a Burden That Machines Avoid

To understand the gap, we must look at the raw figures. In the OECD zone, tax revenues represent on average 34% of GDP. Of this total, taxes on household income and social contributions account for approximately 45%. It is therefore wages, directly and indirectly, that finance the bulk of modern states: their armies, their hospitals, their pensions, their schools.

Corporate income tax, by contrast, represents on average about 11 to 12% of total tax revenues in OECD countries—12.0% in 2022, 11.9% according to 2023 data (Tax Foundation). A significantly lower share than that of work. And this share includes profits from companies that still employ humans.

Technology, by replacing work, mechanically shifts the value created from the first category toward the second. A cashier replaced by an automated terminal shifts fiscally productive value toward a fiscally light category. Multiplied by millions of such transactions over two decades, the macroeconomic effect becomes substantial.

Economists call this phenomenon the “automation bias” in taxation. It is not the result of deliberate choice: tax codes were written when the economy was massively human. They have not been revised since.

The “Robot Tax”: A Seductive Idea and a Technical Wall

Bill Gates called for it as early as 2017 in a now-famous interview. If a robot replaces a human who paid $50,000 in taxes per year, that robot should contribute at a comparable level. The intuition is powerful. Its implementation is another matter.

To tax a robot, you must first define it. In the industrial economy of the twentieth century, this was conceivable: a machine tool replaces an operator, you can compare. In the AI economy, it is a dead end. A language model that automates contract drafting is not an isolatable machine. It is an API accessible from any computer, hosted on servers in multiple countries, whose effects on employment diffuse progressively across dozens of professions without a one-to-one substitution ever being observable.

South Korea attempted the experiment in 2017: it reduced the tax credit granted for automation investments. This is not a robot tax, but a subsidy reduction. The fiscal effect is modest. The symbolic effect, however, relaunched the international debate for a few months before fizzling out.

The underlying problem is that a robot tax, if calibrated to employment substitution, would create considerable distortions. It would tax companies that automate low-skilled tasks more heavily than those that automate high-value tasks, producing a regressive effect opposite to the original intent. And it would be easily circumvented through accounting structures or relocations.

A More Practical Path: Tax the Rent, Not the Machine

Economists, whose work is the subject of analysis published in Nature in 2026 under the title “AI has entered the workforce: tax tech profits, not people,” propose a different approach. The issue is not to tax the employment-machine substitution, which is undetectable and undefinable. The issue is to tax the rent generated by that substitution.

The mechanism under discussion is progressive taxation on corporate profits, modulated by the profit-per-employee ratio. A company that generates 10 million in profits with 1,000 employees is taxed differently from a company that generates 10 million in profits with 10 employees. The latter captures an automation rent that is not redistributed in the form of wages or contributions. It is this rent that progressivity would target.

The advantage of this mechanism over the robot tax is that it does not require defining or identifying the substitutive technology. It applies to the observable result: the concentration of value in few hands without equivalent in wage mass. It is compatible with existing tax systems, which already know the principle of progressivity.

It also creates the right incentives: a company that generates high profits per employee is encouraged either to hire, or to pay its current employees more, or to contribute more to the common pool. All three options have social logic.

This is not a friction-free solution. Technology companies with very high value-added per employee, notably in the software sector, would be affected differently from automated industrial companies. The definition of the consolidation scope for multinational groups would pose formidable questions. And questions of tax incidence—who ultimately bears the corporate income tax—remain open in economic literature.

But this is precisely the type of reform that can be articulated with ongoing work on international minimum taxation, the so-called OECD “Pillar Two” agreement, which sets a minimum rate of 15% on profits of large multinationals. This foundation now exists. It could accommodate a layer of progressivity linked to employment intensity.

The Long Arc: When the Tax Base Shrinks While Needs Explode

The current imbalance is manageable. Its probable trajectory is not.

OECD projections, conservative on the effects of automation, anticipated as early as 2019 that 14% of jobs in member countries were at “high risk” of automation, and 32% would be significantly transformed. These figures predate large language models, whose large-scale commercial deployment began only in 2023. Updated IMF projections (2024) estimate that approximately 60% of jobs in advanced economies are exposed to AI—while globally, this figure is about 40%—with a probability of partial or total substitution.

If even a third of this exposure translates into a reduction in taxable wages over 20 years, the effect on state tax revenues is on the order of several percentage points of GDP. Not a catastrophe in itself. But a structural constraint that accumulates precisely during the decade when states will need to finance the retraining of millions of workers, the adaptation of social protection systems designed for stable employment, and investments in education necessary to prepare future generations.

This is the paradox of fiscal automation: it reduces the capacity of states to manage the social effects it generates. And this compression occurs in a context where public spending needs related to demographic aging are already rising sharply in nearly all wealthy countries. The two pressures add up.

There is a generational effect not to be minimized. Workers who lose jobs at 45 or 50 in automated sectors no longer contribute, consume more social protections, and weigh on public finances for a decade before retirement. It is the young generations arriving on a transformed labor market that finance this transition, without having benefited from the jobs that existed before. The intergenerational distribution of automation’s costs is asymmetric, and current tax systems were not designed to absorb it.

The question is not whether automation will continue: the productivity gains it generates are real and documented, and slowing it would be a mistake. The question is whether tax systems can adapt their base fast enough to avoid leaving states exsanguinated facing the transition they must finance.

What States Are Doing, What They Could Do

The political response to this challenge is for now fragmented. Few countries have undertaken structural reform of the tax base linked to automation. Most adjust at the margins.

The United States, driven by the IRA in 2022, substantially strengthened tax credits for certain industrial investments, notably in clean energy. This is a growth policy that also subsidizes automation in these sectors. Europe adopted the Pillar Two agreement and is beginning to deploy it. This is a minimal safety net, not an answer to the tax base question.

A few economists and think tanks, notably at the Montaigne Institute and in OECD research work, are exploring reforms to labor taxation through a reduction in employer social contributions offset by an expanded value-added tax or corporate profits tax. The idea is to make work relatively cheaper compared to capital, without reducing overall revenues. This is a coherent path, but it assumes difficult political coordination and does not resolve the progressivity question.

The proposal for progressivity on profit-per-employee has the advantage of being technically simple and politically clear. It does not require defining AI, measuring substitution, or building a robot registry. It relies on data that tax administrations already collect: revenue, wage bill, taxable profit. It can be implemented in a coordinated manner within the OECD framework, on the model of Pillar Two.

The acceleration of AI in scientific research, which the journal recently documented through AI agents producing results in hours, gives an idea of the speed at which these transformations can spread across still-protected sectors. Professions, legal, finance, and audit professions are not safe from rapid employment compression over the next decade.

The fiscal debate is not a technical question reserved for economists. It is a question of choice: will the productivity gains from automation remain captured by capital owners, or will they be partially redistributed to finance the collective institutions that these same gains destabilize? The tools exist, or can be built. Long history of open economies shows that institutions that know how to adapt to these bifurcations traverse transitions better than those that ignore them: it is precisely this type of institutional adaptation that the fiscal question calls for today.

The real constraint is not technical. It is political: building a majority for fiscal reform in a context where technology companies have considerable lobbying power, where tax competition between states remains intense, and where governments are tempted to finance the transition through debt rather than taxation. That choice, between debt and reform, between deferral and adaptation, is being made now. Its effects will be measured in twenty years.

Sources

  1. Nature, “AI has entered the workforce: tax tech profits, not people”, 2026 — https://www.nature.com/articles/d41586-026-01877-y
  2. OECD, “Taxing Wages 2024” — https://www.oecd.org/en/publications/taxing-wages-2024_b2f91d76-en.html
  3. OECD, “The Future of Work: OECD Employment Outlook 2019” — data on employment exposure to automation
  4. IMF, “Gen-AI: Artificial Intelligence and the Future of Work”, January 2024 — data on employment exposure in advanced and global economies — https://www.imf.org/-/media/files/publications/sdn/2024/english/sdnea2024001.pdf
  5. OECD, Inclusive Framework on BEPS, Pillar Two — documentation on the 15% global minimum rate — https://www.oecd.org/en/topics/sub-issues/global-minimum-tax/global-anti-base-erosion-model-rules-pillar-two.html
  6. OECD – Revenue Statistics 2025 (official brochure) — https://www.oecd.org/content/dam/oecd/en/topics/policy-sub-issues/global-tax-revenues/revenue-statistics-highlights-brochure.pdf
  7. OECD – Revenue Statistics 2024 — https://www.oecd.org/en/publications/2024/11/revenue-statistics-2024_6e88b46e.html
  8. OECD – Taxing Wages 2026 — https://www.oecd.org/en/publications/2026/04/taxing-wages-2026_d1f39986.html
  9. OECD – Employment Outlook 2019 — https://www.oecd.org/en/publications/oecd-employment-outlook-2019_9ee00155-en/full-report/component-6.html
  10. CNBC / WEF – Bill Gates Interview (2017) — https://www.cnbc.com/2017/02/17/bill-gates-job-stealing-robots-should-pay-income-taxes.html
  11. Korea Times – South Korea Robot Tax (2017) — https://www.koreatimes.co.kr/opinion/20250908/dont-tax-the-robot-enjoy-the-windfall