Meta talks to Anthropic over AI compute deal worth up to $10 billion

Meta is in discussions to lease AI compute capacity from its data centers to Anthropic in a deal that could be worth as much as $10 billion over two years, according to the New York Times. If completed, the arrangement would turn part of Meta’s AI infrastructure buildout into a revenue line while giving Anthropic more capacity for demand tied to Claude Code.

The deal is not done. The New York Times reported that Anthropic proposed the arrangement in June, Meta is still evaluating it, and either company could still walk away. The companies have not disclosed the proposed pricing structure, capacity involved, or whether the compute would use specific chips or data center regions.

For Meta, the talks test a point Mark Zuckerberg has already made to investors: if Meta builds more AI infrastructure than it needs for its own products, it could sell unused capacity. That is a different posture from the usual hyperscaler model. Meta is not a cloud infrastructure provider in the same way Amazon, Microsoft or Google are, and renting capacity to a frontier AI lab would put the company closer to the business of monetizing compute directly.

The context is Meta’s AI spending plan. The company plans to spend up to $145 billion this year, with most of that directed toward AI, according to the report. Meta is also pursuing paid AI subscriptions, including a product called Meta One, and has looked at space-based solar power as an energy source for its data centers.

Anthropic’s capacity problem

Anthropic’s interest reflects the infrastructure pressure facing AI model companies as usage grows. The company needs additional compute because demand for Claude Code has increased, according to the New York Times report. Anthropic has also signed a $45 billion deal with SpaceXAI for data center capacity.

The company’s longer-term plan is to build its own data centers. Anthropic has hired former Google executives to lead that effort, according to the report. That suggests any deal with Meta would sit alongside other supply arrangements rather than replace Anthropic’s push to control more of its infrastructure stack.

The proposed Meta deal also shows how uneven AI infrastructure markets have become. Labs with fast-growing products need capacity quickly, while companies with the balance sheet to build data centers are looking for ways to justify large capital spending. Meta’s own AI demand may absorb much of what it is building, but the reported Anthropic talks show the company is at least exploring outside customers for any surplus.

There are still material unknowns. The New York Times did not report whether the agreement would include dedicated capacity, priority access, energy commitments or operational support from Meta. It also did not report whether Anthropic would become Meta’s first major external AI compute customer under Zuckerberg’s excess-capacity plan.

A completed deal would not make Meta a conventional cloud provider overnight. It would signal that the economics of AI infrastructure are pushing even consumer internet companies to treat compute as an asset that can be sold, not just an internal cost center.

This story draws on original reporting from The Decoder.