Anthropic signs $517 billion in compute deals despite Amodei's warning

Anthropic has signed compute contracts worth up to $517 billion over the past eleven months, according to a report by The Information. Since October 2025 the company has locked in at least 14.8 gigawatts of computing power on top of the one to two gigawatts it already held, and it is also planning its own data centers. That total planned capacity likely still falls short of OpenAI's 30 gigawatt target for 2030, though The Information notes many of Anthropic's contracts run well past that date, which makes a direct comparison between the two companies difficult. Neither company can cover these commitments from current revenue alone. Anthropic's annualized revenue has topped $65 billion, according to Bloomberg, while OpenAI's stood above $40 billion as of July. The contracts land only months after Anthropic CEO Dario Amodei warned in early 2026 that competitors "don't really understand the risks they're taking" by investing too fast in AI infrastructure. Anthropic is now the one racing to match the scale of that spending. OpenAI CEO Sam Altman, for his part, has been warning about "unsustainable silliness" from neo-cloud providers, arguing that technical progress could turn today's expensive infrastructure bets into bad ones.
Key facts
- Anthropic has signed compute contracts worth up to $517 billion over eleven months, per The Information
- Since October 2025 it locked in at least 14.8 gigawatts of computing power, on top of one to two gigawatts already held, and is planning its own data centers
- That still likely falls short of OpenAI's 30 gigawatt target for 2030, though many Anthropic contracts extend past that date, complicating comparison
- Anthropic's annualized revenue has topped $65 billion (Bloomberg); OpenAI's was above $40 billion as of July, and neither can cover its compute commitments from revenue alone
- The deals follow Dario Amodei's early 2026 warning that rivals 'don't really understand the risks they're taking' by overinvesting in compute
Why it matters
The $517 billion figure puts a number on how fast the AI infrastructure race is escalating, and the irony is pointed: the same Anthropic CEO who warned in early 2026 that rivals were taking reckless investment risks has now committed his own company to a compute buildout on a scale that neither Anthropic nor OpenAI can currently pay for out of revenue. It signals that both leading labs now treat massive, multi-year compute commitments as the price of staying competitive, regardless of what either CEO has said publicly about restraint.
Who it affects
Anthropic and OpenAI are the direct parties, since both are locking in compute far beyond what their current revenue supports. The comparison also touches the cloud and neo-cloud providers supplying that capacity, whom Sam Altman has separately criticized. Investors in either company, and in the data center and chip supply chain behind these contracts, are exposed to whether the spending pays off.
How to use it
For anyone tracking the AI infrastructure race, the useful comparison here is capacity, not just dollars: Anthropic's 14.8 gigawatts locked in since October 2025 versus OpenAI's stated 30 gigawatt target for 2030, with the caveat that contract timelines differ enough to make the two numbers hard to line up directly. The revenue figures ($65 billion annualized for Anthropic per Bloomberg, over $40 billion for OpenAI as of July) are the yardstick for judging whether either company can plausibly grow into these commitments.
How solid is it
The $517 billion figure and the gigawatt numbers come from a report by The Information, relayed here by The Decoder; the revenue figures are sourced separately to Bloomberg. Neither Anthropic nor OpenAI is quoted confirming the specific numbers directly in this account, so the figures rest on that reporting rather than on company disclosure.
Risks and caveats
Neither Anthropic nor OpenAI can currently cover these compute commitments from revenue alone, meaning both are betting on future growth to make the spending pay off. The report does not break down how much of the $517 billion is already spent versus merely committed, nor does it name the counterparties behind the contracts. Sam Altman's warning about 'unsustainable silliness' from neo-cloud providers points to a real risk on the other side of these deals: if technical progress reduces the compute needed to run frontier models, today's expensive, long-dated infrastructure commitments could turn into bad bets for whoever signed them.
“don't really understand the risks they're taking.”
— Dario Amodei, Anthropic CEO