Meta negotiates $10B AI compute lease with Anthropic amid $130B–$145B capex push

Meta is in preliminary talks to lease up to $10 billion in AI computing capacity to Anthropic over two years, according to reporting from the New York Times and CNBC. The deal, proposed by Anthropic in June, would position Meta as a cloud infrastructure provider to one of its direct competitors in generative AI. Anthropic has been aggressively securing compute from multiple providers: a $45 billion three-year deal with SpaceX (Colossus 1) announced in May, plus a $100+ billion ten-year commitment to Amazon Web Services. A Meta arrangement would diversify Anthropic's supply chain across three major hyperscalers.

Meta CEO Mark Zuckerberg signaled the move on the company's Q2 2026 earnings call, stating that Meta receives compute inquiries 'at a significant premium over what we paid for it' almost weekly but hasn't leased capacity yet because it prioritizes internal AI development. The calculus has shifted: Meta raised its 2026 capex guidance low-end by $5B to $130B–$145B, more than double the $72B spent in 2025, and now faces investor pressure to justify that spending. By leasing spare capacity, Meta can demonstrate cash generation from its infrastructure buildout while reserving compute for its own model development under AI chief Alexandr Wang, who debuted the Muse Spark 1.1 model earlier this month.

Under the proposed terms, Anthropic would pay Meta monthly instalments over two years with bilateral early termination rights. Neither company has clarified which silicon Anthropic would access: Meta's portfolio includes Nvidia H100/H200s and Meta's own MTIA 400 accelerators. Zuckerberg acknowledged 'would be foolish to just sell all the compute and take short-term profit,' signaling caution about overbuilding. Dave Brown, a 19-year AWS veteran, recently joined Meta to lead a new 'Meta Compute' initiative, reinforcing the company's pivot toward cloud business operations alongside ads revenue.