House Democrats propose taxing AI companies per token to fund jobs for displaced workers

A new bill would tax AI companies per token and use the money to hire the people AI put out of work.

What the bill proposes

House Democrats Greg Casar of Texas, Valerie Foushee of North Carolina and Sara Jacobs of California introduced the AI Tax and Work Protection Act. The mechanism is the notable part. Rather than taxing data center revenue, Casar has pitched taxing large AI companies on tokens, which means charging by the compute consumed.

The proceeds would fund a Work Protection Administration, explicitly modeled on Roosevelt’s Works Progress Administration, to create jobs for workers displaced by AI. Casar framed the intent bluntly: “We are not going to let AI company CEOs get rich by displacing millions of American workers.”

Why it is unlikely to become law

The bill’s odds are poor. The House currently sits at 218 Republicans to 212 Democrats with four vacancies, and this is not a bipartisan bill. The realistic reading is that it functions as a marker of where the political argument is heading, not as law in waiting.

Why the design still matters

Even a bill that stalls can set the terms of the next one, and this design is genuinely new. A per-token levy is a tax on usage, not profit. It lands on inference at scale rather than on the balance sheet, which makes it a very different incentive from anything proposed so far. A profit tax only touches companies that are making money. A usage tax would reach every company running large models at scale, profitable or not.

That framing raises the question the debate will likely turn on: whether taxing compute to fund employment programs is workable, or whether the cost simply gets passed through to the people and businesses using the tools. The bill puts that question on the table with a concrete mechanism attached, which is more than the displacement debate has had until now.