Anthropic is in preliminary talks to lease as much as $10 billion in computing power from Meta over a two-year period, The New York Times reported on July 17, citing people familiar with the discussions. Under the terms being discussed, Anthropic would pay Meta monthly, with either company able to walk away before the deal concludes. Both companies declined to comment on the talks.
An unlikely infrastructure partner
The arrangement would be notable because Meta and Anthropic compete directly in the model market — Meta’s Llama family against Anthropic’s Claude line. But Meta has told investors it is exploring becoming a cloud-computing supplier in its own right, after CEO Mark Zuckerberg said in May that outside companies had already approached the firm asking to buy its spare capacity. Meta expects to spend as much as $145 billion on AI infrastructure this year, giving it computing capacity beyond what its own model training and products need.
Part of a broader pattern
Anthropic has assembled a patchwork of compute deals as it competes for scarce Nvidia chips, including arrangements with Google, Microsoft and Amazon. Its largest so far is a three-year, $45 billion agreement with SpaceX for capacity at the Colossus 1 data center — signed only weeks before the Meta talks emerged, and following a similarly sized $19 billion lease with TeraWulf. Anthropic has previously capped usage of its premium Claude tiers, citing capacity constraints tied to the AI compute shortage that has gripped the sector this year.
Neither company has confirmed a deal will be signed, and a person briefed on the matter told CNBC the negotiations remain “very preliminary.” If finalized, the lease would mark one of the clearest signs yet that even well-capitalized AI labs are turning to competitors — not just cloud computing giants — to secure enough hardware to keep training and serving their models.