Back
Meta, Anthropic discuss potential two-year AI computing deal worth up to $10 billion

Meta, Anthropic discuss potential two-year AI computing deal worth up to $10 billion

Anthropic’s proposed lease would give it access to Meta’s excess AI computing capacity, while offering Meta a potential new revenue stream from infrastructure built for its own model ambitions.

Fact Check
Multiple outlets (PANews, BlockBeats, Phemex) report the Meta-Anthropic compute lease of up to $10 billion over two years, with PANews attributing it to the New York Times. BlockBeats supplies specific, internally consistent details (June proposal by Anthropic, monthly installments, mutual early-termination, roughly one-third of the SpaceX deal). Independent reporting confirms the broader context: Anthropic's multi-vendor compute sourcing and Meta's move into compute leasing. The claim's own wording ('discuss'/'working to finalize') and the sources both frame this as early-stage negotiations that may not close, so it is a report of talks rather than a confirmed deal. The primary NYT original was not directly retrieved, capping confidence at medium.
Summary

Meta is in early talks to lease up to $10 billion of computing power to Anthropic over two years, a potential arrangement that would help the AI developer secure scarce capacity for training and running models while opening a new infrastructure business line for Meta. The proposal, first presented in June, would have Anthropic make monthly payments and would allow either side to exit early, leaving the final value and duration subject to negotiation. The discussions underscore how aggressively leading AI companies are competing for chips, electricity and data center space, even as investors question the returns on the sector’s vast infrastructure spending. Meta has said it plans to spend as much as $145 billion this year, mostly on AI, after spending $72 billion the previous year, and a lease deal could help monetize any surplus capacity. Anthropic, which is seeking more compute as demand for its models grows, has also pursued other large infrastructure arrangements, while Meta itself has committed to major capacity agreements with providers including CoreWeave and Nebius.

Terms & Concepts
  • computing power: The data center and processing capacity used to train and run artificial intelligence models.
  • data center space: Physical facilities that house servers, networking gear and power systems needed for large-scale computing workloads.
  • infrastructure spending: Corporate investment in facilities, equipment and related capacity needed to support operations such as AI computing.