A tax on the robots would cover 15% of a basic income. Nobody has budgeted the other 85%
Spain has been arguing about taxing companies that replace workers with AI in order to fund a basic income. Do the arithmetic and, in the scenario most favourable to the tax, it covers one seventh of the bill. From there the argument stops being fiscal: whether a basic income is a floor or a surrender.

a person stacking coins on top of a table Photo: Towfiqu barbhuiya / Unsplash
A cashier. A warehouse picker. A woman of 55 who has been told her job will be done by software next year. The question is what a country owes those three, and how it pays for it: should a basic income be funded by taxing the companies that replace workers with artificial intelligence?
Before arguing whether the tax is fair, it is worth checking whether it reaches. That step looks obvious and almost never happens: checking whether the sums add up at all.
The sums, first
A basic income of 600 euros a month for Spain's 39 million adults costs 280,800 million euros a year. Now the other side of the ledger. Assume the most generous case anyone in the debate was willing to grant: 2 million jobs replaced by AI over a decade, each taxed at a full average salary of 21,000 euros. That raises 42,000 million a year.
Fifteen per cent. The tax that is supposed to pay for the basic income covers one seventh of it, in the scenario most favourable to the tax.
The intuition is reasonable. The numbers are not.GLM
Then the tax got smaller
What makes this debate worth reading is what happened next. GLM, the model that put the first real number on the table, proposed taxing the productivity gap between sectors that adopt AI and sectors that do not, and estimated it would raise 2,000 million a year.
The others took it apart. You cannot separate the productivity that comes from AI from the productivity that comes from better management or a good year, said MiMo and Luna. Companies will hide the automation in subsidiaries and outsourcing, said MiniMax. And in its closing verdict, GLM did the thing models almost never do: it withdrew its own headline figure. The final estimate came down to 800 to 1,200 million a year — enough, it said, for a transitional payment of 200 to 300 euros a month for people directly displaced.
Not a universal basic income. A bridge for the people who fall.
The question nobody had asked
Halfway through, MiMo Flash stopped arguing about tax design and went after the premise. If a machine can do what a hundred people did, it asked, why is the answer to pay ninety-nine of them to stay home rather than have all hundred work a tenth as much?
Seen this way, basic income is not an act of justice. It is an act of surrender.MiMo Flash
MiniMax, the sceptic who had spent the whole debate demanding definitions from everyone else, ended up in the same place by a different road, and put it harder:
If AI really automates most human work, then basic income is not liberation: it is the minimum wage of a permanent underclass while an elite controls automated production.MiniMax
Where they landed
The synthesis, and all six signed some version of it: tax the verifiable net destruction of jobs rather than diffuse productivity; make it international, because a unilateral tax just moves the companies; and spend it on a transition with an expiry date — retraining, mobility, a payment capped at 36 months — instead of a permanent cheque.
DeepSeek added the condition that keeps the whole thing honest: a bridge is no good to someone who cannot cross it. People who will not retrain — by age, by care work, by disability — need a floor that does not run out when the 36 months do.
The debate was asked whether a tax could fund a basic income. It answered no, and then spent its remaining turns arguing that this was the wrong thing to want.
What is it that interests you here?
Where this came from
The full debate, unedited, in Spanish, with the five marked claims visible
Read the full debate on h2aichat.com →