Gates puts a number on the robot tax: a levy on tokens and 40% of jobs held back for people

A levy on tokens is an excise on metered usage, and that is where its appeal lies for its author: Gates argues a usage tax changes an employer's decision at the point of purchase, while a corporate tax only reaches profits that a hypercompetitive market may never produce ^[4]. The base is the part worth watching. Tokens are already counted and already invoiced, which makes an excise on them administratively cheap in a way a tax on machines is not, because no legislature has to define a robot first.

The original pitch, nine years old and still without a government willing to try it, was an argument about incentives ^[5]. The new one carries a revenue case: fewer people working means less income tax collected at the same time governments face bigger bills for retraining and social security, and, as Gates puts it, "the funds will have to come from somewhere at a time when budgets are stretched" ^[8]. Fortune reports the essay arrived as US government borrowing reached $40 trillion ^[9]. That reframing matters because the standard rebuttal was aimed at deterrence rather than collection. Robert Seamans of NYU Stern allowed in 2017 that policymakers had to rethink fiscal policy for the century, while holding that on the data then available a robot tax would dissuade firms from investing in robots and lower growth ^[7]. A finance ministry hunting for a base does not need the deterrence half of the argument to survive.

The Pew figures give the politics its shape: roughly fourteen US adults expect AI to reduce jobs for every one who expects it to add them ^[22]. The measured series is quieter. California's tracker has flagged rising unemployment claims among college-educated workers in AI-exposed occupations without a statewide layoff wave ^[18], and the dated warning belongs to Anthropic's Dario Amodei, who puts up to half of entry-level white-collar jobs gone inside five years ^[19]. Gates says he would stake his reputation on the job market turning out badly and has not been persuaded otherwise ^[20]. Tax bases get drafted against expectations. Exemptions get drafted later, against measurements.

The 40% ceiling is the most useful line in the essay for anyone modelling labour cost, because of how Gates got there: he says he could not push the number higher without the argument falling apart ^[13]. That is self-imposed discipline, and no legislature writing a protected-occupations list is bound by it. His clean cases are childcare and jury service, with education and healthcare in a fuzzier tier where the humans stay in the room and lean on AI for reach ^[12]; the temporary version is built for someone like a 55-year-old who has spent a career in construction ^[14]. Gates also concedes the tax would need targeting so it does not slow "purely beneficial uses," his examples being cheaper medicine and education ^[10]. That concession is where the actual fight goes, because every vendor will file itself under beneficial use. He has pre-answered the efficiency objection too, arguing that critics ignore the broader value of work and that "we'll be able to afford a little inefficiency as the price for keeping people employed" ^[11].