Money & State

A robot tax is a good idea. It only works if all of Europe joins in.

Anyone who wants a robot tax also wants a stronger Europe. The two come together.

As AI agents take on more work, the state loses out on wage tax and social contributions. The obvious answer is to tax the robots instead. Bill Gates suggested exactly this in 2017, and the idea has not gone away since. The argument is ethically clean. Whoever profits from automation should help carry its social cost. Companies already pay profit tax on automation gains, but a robot tax aims at something else: replacing the lost non-wage labour costs that today fund the social system and pensions. The argument holds. The implementation is harder.

In February 2017 the European Parliament passed a resolution on robotics, brought forward by Luxembourg’s social democrat Mady Delvaux (396 in favour, 123 against, 85 abstentions). A recommendation for a robot tax that it originally contained was struck out on the floor. In 2018 South Korea trimmed an investment subsidy and was thereby counted internationally as the first country with a robot tax. Both were first attempts, and both stayed isolated.

The central objection is geographic, not ideological. A factory sits where it sits. You cannot move it to Bratislava overnight. An AI operation, by contrast, runs on servers in Dublin, billed through a holding company in Delaware. Whether value created this way is really harder to pin down than a factory is the open core question of this debate, and the experience with taxing digital corporations points that way so far. If Austria introduces a robot tax on its own, companies move their AI operations to another EU country. If the EU introduces one, the operations move to London, Singapore or Texas. The value follows the lowest tax.

A robot tax can only work EU-wide, and even then only if it hangs on something tied to a place: revenue in the EU single market, say, or group profit under OECD logic. This is exactly where there is precedent. GDPR, AI Act, OECD minimum tax: Brussels can do it. But Brussels needs time. Every time.

That is the real dilemma. AI productivity changes the tax base in months. European coordination takes years. A national solo run stays possible, but it would have to live with the kind of avoidance described here. Anyone calling for a robot tax that actually works is not calling for a tax. They are calling for a faster, more unified EU economic policy than the one we have today.

We need to talk about this

What does this topic stir up in you?

In depth

What Bill Gates actually said

In February 2017 Bill Gates told Quartz editor-in-chief Kevin Delaney, in essence: when a person in a factory creates 50,000 dollars of value, that income is taxed. When a robot does the same work, it should be taxed at a comparable level. The aim was not to stop automation but to slow the transition and free up money for care work and smaller class sizes.

It was not a request about technical detail. It was a political signal from someone who has made billions on automation: this time is different, and the state needs a new base to tax. The EU Commission rejected the idea publicly in 2017. The debate went quiet for a few years. With generative AI it is back.

Why South Korea is an ambiguous example

In 2018 South Korea became the first country to do something that was read internationally as a "robot tax". More precisely: a cut. A tax credit for investment in productivity-raising equipment was reduced for large firms from 3 to 1 percent, and for mid-sized firms from 5 to 3 percent. Small firms kept their 7 percent. No new tax. Less subsidy.

An analysis by the US think tank ITIF from February 2026 reports a drop in new robot installations in Korea relative to Japan on the order of around 28 percent. An economic study by Kang, Lee and Quach from April 2025 examines the same Korean intervention with a different method and finds employment and wage effects in the low single digits. The exact figures vary with the comparison baseline, but the direction is clear: taxation changes behaviour. That is precisely what makes Korea ambiguous, because the effect can also simply mean less investment, and whether "fewer robots at home" is the desired outcome when competitors keep automating is an open question. Korea can afford it because its industry is tied to a place. Software is easier to move than a chip factory.

What EU-coordinated regulation has achieved

Three precedents show that Brussels can push through major economic rules. The GDPR was adopted in 2016 and has applied since 25 May 2018. It binds any company that reaches EU citizens. The EU AI Act has been in the Official Journal since 12 July 2024 and entered into force on 1 August 2024, with staggered application through 2027. The global OECD minimum tax of 15 percent has applied EU-wide since the end of 2023, and from the end of 2024 it also covers group subsidiaries that pay too little elsewhere.

Each of these rules took between six and ten years from idea to application. In 2026 the global minimum tax also gained a special arrangement for the US that places the American system alongside the European one rather than overriding it. Coordination is possible, but slow and never complete.

Sources