Meta To Rent Surplus AI Compute To Anthropic After Building $35 Billion Infrastructure To Defeat It

MENLO PARK — Meta Platforms is in advanced discussions to host Anthropic's Claude models across its AI data center infrastructure, according to reporting from SemiAnalysis, completing what analysts described as a natural arc in which the company spent four years and tens of billions of dollars building an AI apparatus whose most commercially viable use may be housing a competitor's product.

The arrangement would make Meta — which has deployed approximately 1.3 million GPUs and committed over $35 billion in data center spending this year alone — one of the largest infrastructure providers for Claude, the AI assistant that Meta AI was created to displace in the American consumer market.

Meta's AI division has publicly positioned itself for years as an alternative to closed-AI providers such as Anthropic. Its infrastructure, the company has argued repeatedly on earnings calls, represents a "strategic moat."

Analysts noted that the term "moat" has historically implied the speaker is on the interior.

Under the anticipated arrangement, Meta would charge Anthropic for compute access, monetizing capacity that Meta's own models have not generated sufficient inference traffic to fill. The deal would represent the latest hyperscaler to derive meaningful AI revenue from a direct competitor, following the Microsoft-OpenAI arrangement that analysts describe, with increasing weariness, as "the template."

Meta did not confirm the discussions. A spokesperson issued a statement affirming the company's commitment to "open infrastructure for the broader AI ecosystem," which one analyst translated as "we have a lot of empty servers."

The company's shares rose 1.2 percent on the news.

"They built the hotel," said a Bernstein Research analyst who covers Meta. "The guest is their competitor. The competitor was specifically the reason they built the hotel. This is fine."