Anthropic has signed compute contracts worth up to $517 billion over the past eleven months, according to a report from The Information relayed by The Decoder. Since October 2025, the company locked in at least 14.8 gigawatts of computing power, stacked on top of the one to two gigawatts it already ran, and it’s now planning its own data centers. That’s a staggering commitment from a company whose CEO spent early 2026 telling everyone else to slow down.
This is significant because it marks a clear shift in posture. Anthropic has long positioned itself as the safety-conscious lab, the one that thinks before it spends. The numbers tell a different story now.
What the deals cover
Here’s the shape of it, per The Decoder:
- Up to $517 billion in compute contracts signed in roughly eleven months.
- At least 14.8 gigawatts of new computing capacity locked in since October 2025.
- Additional plans for Anthropic’s own data centers, on top of the leased capacity.
For scale, OpenAI is targeting 30 gigawatts by 2030. Anthropic’s total planned capacity likely still falls short of that number, but the comparison isn’t clean. Many of Anthropic’s contracts run well past 2030, so the two companies are measuring different finish lines. The Decoder notes that direct comparisons are tricky for exactly this reason.
Neither company can pay for this yet
The money side is where things get interesting. Neither Anthropic nor OpenAI can cover these commitments from revenue alone right now.
- Anthropic’s annualized revenue topped $65 billion, according to Bloomberg.
- OpenAI sat above $40 billion as of July.
Those are big numbers. They’re also nowhere near half a trillion dollars in compute obligations. That gap gets filled with investment, credit, and a bet that demand keeps climbing fast enough to justify the spend. It’s the same leveraged logic driving the entire data center buildout, and Anthropic just planted a very large flag in it.
The reversal that stands out
What stands out here is the timing against Amodei’s own words. In early 2026, the Anthropic CEO warned against investing too fast, saying competitors “don’t really understand the risks they’re taking.” Months later, Anthropic is the one racing to catch up on capacity.
Meanwhile, the roles have flipped on the other side. OpenAI’s Sam Altman is now the one urging caution, warning about “unsustainable silliness” from neo-cloud providers. His argument: technical progress could make today’s expensive infrastructure projects look like bad bets tomorrow. If models get more efficient, or if a better approach arrives, all that locked-in capacity could turn into an expensive liability.
So you’ve got both leaders warning about overspending while both keep signing enormous deals. That tension is the real story.
Why it matters for practitioners
A few takeaways worth holding onto:
- Compute is the moat now. The frontier labs are competing on gigawatts as much as on model quality. Capacity commitments this size signal how they expect training and inference demand to scale over the next five years.
- The caution talk is mostly talk. When the safety-first lab commits half a trillion to compute, the market has answered the “are we overbuilding?” question, at least for now. Watch actions, not warnings.
- Pricing pressure could ease, or spike. Massive capacity buildouts can lower inference costs over time. But if these bets go sideways and capacity gets stranded, expect volatility in what you pay for API access.
What to watch next
The open question is whether revenue catches up to these obligations before the bill comes due. Both companies are betting demand grows faster than the risk. If Altman’s “unsustainable silliness” warning proves right, some of these contracts will look reckless in hindsight. If Amodei’s spending spree pays off, the caution he preached in early 2026 will read as a head fake.
Either way, the AI infrastructure race just got a new benchmark, and it has a lot of zeros. You can find the full breakdown at the original report from The Decoder.