There are two hard problems in AI: building good models, and having enough compute to train and run them. The first problem is a research challenge. The second is a capital race — and it’s reshaping the competitive landscape faster than any model benchmark.
Meta and Anthropic are reportedly in talks for a potential $10 billion compute lease deal, first reported by the New York Times and confirmed by Business Times Singapore and Yahoo Finance. The structure: Meta would lease computing capacity from its massive AI infrastructure to Anthropic, which needs it to continue training and running Claude at scale.
Why this week, specifically
Timing matters here. Meta this week launched its cloud business — a deliberate shift from treating its infrastructure as internal cost center to monetizing it externally. The same week, Zuckerberg confirmed that Meta’s in-house IRIS AI chip is entering production in September, part of a plan to double Meta’s computing capacity to 14 gigawatts. That’s a staggering number — roughly equivalent to the annual electricity consumption of millions of homes.
Meta isn’t just building compute for its own models anymore. It’s building a compute surplus it can sell. That’s a fundamentally different strategic position: from AI model maker to AI infrastructure provider.
What this says about Anthropic
The deal also reveals something about Anthropic’s position. Despite raising billions and securing a major investment partnership with Amazon Web Services, Anthropic apparently still needs access to additional compute capacity to maintain its development pace. The model race is, at its core, an infrastructure race — and even well-funded companies can run into compute ceilings.
The moat has moved
What strikes me about this negotiation isn’t the $10 billion figure — it’s the logic it reveals. The most advanced AI companies are no longer just software publishers. They’re infrastructure operators who sell compute capacity the way others sell electricity or bandwidth. Meta — a social network that became an AI conglomerate — is claiming that position. If this deal closes, it’s a strong signal: in AI, whoever controls the gigawatts controls the market.