Anthropic’s Enormous Overhead

There’s some concern from armchair economists and even people closer to the industry, about what Anthropic appears to be doing with its front-runner status (along with OpenAI and maybe Google) in AI.

Some of the criticism refers to a kind of “churning” where the company seems to be putting almost all of its massive gains back into computing and infrastructure costs, essentially betting big that demand will continue to grow in its general customer base, and that, so to speak, there’s nowhere to go but up.

By some accounting standards, you can see that of around $17 billion in revenue for Q3 of this year, Anthropic’s computing costs were around $9.9 billion, or over half of its entire ledger sheet. That still leave other operating costs, for an operating shortfall, again, by some accounting, of $567 million.

In other words, the company is ploughing the lion’s share of its profits back into buying compute, which it apparently does a year before it uses it. Not only that, but according to contracts, Anthropic will be on the hook to some of its vendors, notably, Amazon, Google and Microsoft, even if things go south with the AI boom.

Some Numbers

I got this from a recent article by Ansh Krishna and Charlie Sheiner at Ornn Research, where the accounting stands out rather starkly:

“Google requires at least $121.3 billion between April 2026 and July 2033 and Amazon at least $110 billion between May 2026 and April 2036, with true-up payments for any shortfall in usage,” they write. “Microsoft adds at least $44.9 billion between November 2026 and June 2033, which can be canceled only for breach or for tranches not yet delivered.”

Some back-of-the-napkin math adds up to around $273 billion for the “big three,” but apparently, that’s not all.

The piece documents how the SpaceXAI, Nscale and Lambda contracts have their various escape contingencies; Google’s, Amazon’s, and Microsoft’s, presumably, do not.

Bigger Math

And then there’s AMD. Here’s another revealing paragraph from Ornn: