BREAKINGVIEWS-Anthropic’s spending is worryingly anthropocentric

The author is a Reuters Breakingviews columnist. The opinions expressed are his own.

By Stephen Gandel

NEW YORK, Oct 29 (Reuters Breakingviews) - There’s no shortage of human angst over artificial intelligence. Anthropic, the maker of robot coder extraordinaire Claude, might seem ground zero for the displacement of workers by ever-more-proficient machines. Yet financials disclosed in its listing prospectus released on Monday imply that the company's research spending is becoming more labor-intensive over time. That’s a relief for people, and a potentially promising sign that boss Dario Amodei is pushing as hard as possible to maintain a lead over rivals. But it also complicates the broad investment case for AI.

Anthropic devotes a good portion of its prospectus to the productivity of its staff. It says it internally refers to its workers as “ants,” a description that paints them as low-ego and collaborative. The company also says pairing its ants with its Claude model lets them finish work that would normally take much larger teams. Results suggest that even if the colony isn’t growing, the cost of keeping them busy is.

The AI lab doesn’t directly disclose how much of its research spending is on salaries and the like. Still, take the $8.9 billion of spending on computing power and associated infrastructure from the first six months of this year. About $2.4 billion of that will show up in cost of revenue, “substantially all” of which is comprised of servers crunching users’ queries. The $6.6 billion remainder is likely the expense of developing new models. Take that out of the R&D budget, and it implies that practically all of Anthropic’s $3.2 billion of spending on personnel is for researchers.

That figure is up 493% from the same period a year earlier, nearly doubling the 283% growth in spending on computing power. Put another way, Anthropic spent as much as $0.48 on human research work for every dollar it spent on artificial help, up from $0.39 a year ago.

Other technology giants have similar issues. Facebook owner Meta Platforms META.O has spent mightily on AI. Despite enormous layoffs, its employee expenses, even after excluding severance, rose 21% in the first half of this year from the same period a year earlier.

https://www.reuters.com/graphics/BRV-BRV/BRV-BRV/akpellnkevr/chart.png

Anthropic's financials group together various expenses in different ways that make it difficult to pin down with exact precision how much it is spending on specific employees. What's more, stock compensation may distort the picture: a good portion of developers' pay comes from equity, and Anthropic’s has been rising rapidly in value. But presumably part of the reason people want to work there is because it’s becoming more valuable.

The bigger-picture problem for companies spending vast sums on chatbots is that they ultimately need to improve the bottom line. If productivity gains powered by Claude remain a top-line phenomenon, keeping those investment dollars flowing could become a problem.

Follow Stephen Gandel on LinkedIn and X.

CONTEXT NEWS

Artificial intelligence lab Anthropic publicly ​filed the registration statement for its planned initial public offering on October 26. According to its S-1, Anthropic spent $9.7 billion on research and development in the first half of 2026, up 427% from the same period the year before.

(Editing by Jonathan Guilford; Production by Pranav Kiran)

((For previous columns by the author, Reuters customers can click on GANDEL/ stephen.gandel@thomsonreuters.com))