Anthropic’s employees used so much of the company’s Claude AI in April that the usage, priced at the rates customers pay, was worth more than Anthropic’s monthly payroll, according to the company’s IPO filing.
There’s a catch, and the filing spells it out. It sets internal usage at list prices against monthly employee compensation with stock awards left out. Anthropic didn’t actually spend more cash on Claude than on salaries, as Alistair Barr pointed out in Business Insider. It valued what its own staff consumed at retail.
The growth is the bigger number. By September, internal use of Claude had risen more than 250-fold from January, according to the registration statement filed with the SEC. The filing says Claude wrote more than 90 percent of the code added to Anthropic’s software in September.
Revenue, meanwhile, has grown far faster than headcount.
Mr. Barr, who writes the AI Insider newsletter, wrote in March that tech companies were starting to treat AI tokens, the main unit for measuring usage, as a fourth part of employee pay alongside salary, bonuses and stock awards. Anthropic’s numbers, he wrote, show why. “Still, this feels like a watershed moment for the tech industry,” he wrote, though he acknowledged that the comparison rests on retail prices.
Not everyone reads the dependence as good news. Forbes looked at the same filing and wrote that Anthropic’s practice of relying on Claude to build Claude gives the company an advantage and also exposes it to risk.
The disclosure comes in a prospectus that has already drawn attention for other reasons. Among its risk factors, Anthropic lists the end of humanity as something that could hurt its results. Separately, Anthropic staff have said they expect AI to do 75 percent of their work by 2030, even as they forecast headcount growing 66 percent.
The April comparison leaves out stock awards, which are part of personnel expense under the filing’s own definition. With those added back in, the payroll side of the ledger gets bigger.