Who will pay when machines think for us?

The question is no longer hypothetical. In London trading floors, language models draft research notes that junior analysts were writing yesterday. In Manila call centers, conversational agents handle complaints that operators were handling last month. In New York law firms, automated reading systems sift through thousands of contracts in minutes — work that would have taken associates weeks. Artificial intelligence is not threatening employment in some distant future. It is replacing it. Here. Now. According to the Bureau of Labor Statistics, employment in the “white-collar core” sectors — finance, insurance, information, professional services — has been declining for three years despite solid GDP growth. Total employment in these sectors peaked in November 2022. Since then, it has fallen 1.9%, while the rest of the private sector grew 4.1%. Had pre-pandemic trends continued, these sectors would employ 2.3 million more people than they do today. This gap is not cyclical. It is structural.

Our tax systems were not built for this. They tax human labor. Payroll taxes, pension contributions, income tax — everything rests on the assumption that value is produced by people who earn a wage. When a European bank replaces three thousand analysts with a foundation model, it eliminates three thousand pay slips, three thousand contribution bases, three thousand contributors to the funding of hospitals and schools. The productivity gain is enormous. So is the fiscal hole. Nobody is filling it. Morgan Stanley estimates that nearly $3 trillion in AI infrastructure investment will flow through the global economy by 2028, with more than 80% of that spending still ahead. The scale of substitution coming exceeds what the current tax system was designed to absorb.

This is where a deceptively simple idea comes in: tax not the company on its capital, nor the worker on its labor, but the machine on its intelligence. Tax the token.

A token is the elementary unit of artificial reasoning. Every word generated, every inference, every response from a language model breaks down into tokens. Think of them as Lego bricks of digital thought. The idea is to impose a tax proportional to the volume of tokens consumed by businesses. A VAT on artificial intelligence. The more an organization automates, the more tokens it consumes, the more it contributes. The mechanism adjusts on its own. It does not slow down AI. It accompanies it. In January 2026, the Brookings Institution concluded that token taxes applied to end users function as consumption taxes. Companies can keep building AI systems. The state captures value where it is created.

The reply will come: this penalizes innovation. The argument surfaces every time a government dares to capture a share of value created by technology. It is weak. A few fractions of a cent per token would deter no deployment. But the revenue would be substantial through sheer volume alone, because global token consumption doubles every few months.

A more serious objection deserves a serious answer. Economists at the Tax Foundation point out that the American tax code does not technically favor automation, and that any targeted levy risks discouraging investment without protecting employment. Nobel laureate Paul Krugman has said bluntly that he is “not a UBI guy.” These reservations are not absurd. But they reason within a fiscal framework designed for an economy where value comes from human labor. That framework is becoming obsolete. Economists Anton Korinek and Lee Lockwood, in a symposium organized by Anthropic, have proposed a range of taxes covering token generation, robots, and digital services, precisely because the old framework no longer holds. The question is not whether to tax AI, but how to do it without breaking what still works.

The revenue would fund a universal basic income. Not charity. Not a bargain-basement safety net. A right, indexed to the pace of automation. The more AI replaces human functions, the higher the amount. A collective dividend on the productivity of machines. So that society as a whole captures its share — not just a handful of shareholders in California.

One could go further. Much further. Imagine the state becoming a producer of computing power. Building public data centers powered by decarbonized energy. Distributing a monthly token quota to every citizen. They could use it to start a business, get trained, consult a virtual doctor, develop a personal project. Or sell it on an open market for cash. The citizen would become the holder of a renewable cognitive capital, issued by sovereign authority.

The idea will seem excessive. That is precisely why it deserves to be raised. We accepted that the state should provide electricity, water, education — resources deemed too strategic to be left to the market alone. Computing power is joining that list. To back a currency not on gold or debt, but on a country’s sovereign computing capacity, would be to acknowledge that artificial intelligence has become infrastructure, not a luxury. OpenAI itself, in its April 2026 blueprint, proposes enshrining a “right to AI” treated like access to electricity: foundational for participation in the modern economy.

Is this utopian? Sam Altman, the head of OpenAI, funded through his foundation OpenResearch a study in which three thousand participants in Texas and Illinois received one thousand dollars a month for three years starting in 2020. A control group of two thousand people received fifty dollars. The results debunk the fantasy that guaranteed income breeds idleness. Recipients did not stop working. They searched for better jobs, more selectively. Their savings grew by 25%. Their spending on others — gifts, loans, donations — rose by 26%. As Bloomberg reported, the main benefit of cash is flexibility, not laziness.

But let us be honest about what the study also shows. Researchers observed a moderate decline in labor supply: recipients worked an average of 1.3 fewer hours per week, and their non-transfer income fell slightly. Physical health effects were negligible. The hope that participants would find higher-paying jobs did not materialize. This is not a failure. But it is not the miracle that some UBI advocates promise either. The most robust finding is more modest and more true: cash gives people room to maneuver. It does not transform lives. It loosens them.

The idea of an automation tax is no longer fringe. In 2017, Bill Gates proposed that companies deploying robots to replace human workers should pay taxes on those machines, just as they do on wages. On April 6, 2026, OpenAI published a thirteen-page document titled Industrial Policy for the Intelligence Age.

The document is blunt. Shift the tax burden from labor to capital. Create a public wealth fund that distributes returns directly to citizens. Tax automated systems at rates comparable to the workers they displace. In an interview with Axios, Altman compares the moment to the Progressive Era of the early twentieth century and Roosevelt’s New Deal in the 1930s — periods when America agreed to rebuild its institutions because the old architecture could no longer hold. One will note that it is the company building the disruption that is proposing the safety net. The paradox escapes no one. But it does not invalidate the idea.

In California, billionaire Tom Steyer has made it a centerpiece of his campaign for governor. His proposal: a state sovereign wealth fund, the Golden State Sovereign Wealth Fund, financed by a token tax — a fraction of a cent per unit of data processed by the tech giants. Steyer credits the idea to Dario Amodei, CEO of Anthropic, who raised it in 2025. Altman pushes further with his American Equity Fund, where major AI companies would contribute roughly 2.5% of their value each year into a common pool distributed to all citizens. These are no longer seminar speculations. More than 150 guaranteed income pilot programs are already running across 35 American states.

The real difficulty is not technical. It is political. Who will dare to tax artificial intelligence while the great powers wage a cognitive arms race? Who will tell the shareholders of OpenAI or Baidu that synthetic cognition owes something to the common good? The McGill Law Journal frames the problem in stark terms: up to 47% of American jobs are exposed to automation, and the tax code taxes labor more heavily than capital. The system incentivizes replacement. It was built that way. The question, put differently: which European leader will have the nerve to make the token a fiscal object?

We have taxed salt, tobacco, carbon. Every era ends up inventing the tax that corresponds to its dominant resource. Ours is artificial intelligence. Failing to tax it means accepting that the wealth it produces escapes the social contract. That the machines work, the shareholders collect, and the citizens get nothing.

There is still time. But European leaders will need to stop treating digital taxation as an engineering problem and accept that it is, at bottom, the oldest question in politics: how to distribute wealth when the means of producing it change.