Anthropic in early talks to lease $10 billion in AI compute from Meta

Anthropic is in early, preliminary talks to lease artificial intelligence computing power from Meta in a deal that could be worth about $10 billion over two years, CNBC reported July 17, 2026, citing a person familiar with the matter and an earlier report from The New York Times.

What's new

"Anthropic is in very preliminary talks to lease artificial intelligence computing power from Meta, a person familiar with the matter told CNBC's Kate Rooney," CNBC reported. Shares of Meta "climbed off their lows of the day Friday following a report from the New York Times that a potential deal was being discussed worth about $10 billion." Meta declined to comment on the talks.

The arrangement, as reported, would run about two years, with Anthropic paying Meta monthly for access to compute capacity rather than committing to a single upfront sum. Nothing has been signed, and CNBC's sourcing describes the discussions as very early stage — Meta itself has not confirmed a deal is being negotiated.

Context

The talks follow Meta CEO Mark Zuckerberg's comments in May that the company was weighing entry into the cloud-computing business, an effort, per CNBC, "to show investors that the firm can make money from AI investments beyond improvements to its current business." CNBC also notes Meta could spend as much as $145 billion on capital expenditures, including AI infrastructure, in 2026, and that former longtime AWS executive Dave Brown is set to join Meta.

This would not be Anthropic's first move to lease compute from an unconventional supplier. "The talks come weeks after Anthropic announced a similar deal with Elon Musk's SpaceX to use the computing capacity at its Colossus 1 data center to improve capacity for paid subscribers," CNBC reported — a three-year, roughly $45 billion arrangement, or about $1.25 billion a month. Anthropic has also separately committed to paying xAI for compute access. CNBC frames the pattern plainly: "The deal talks are a sign that Anthropic, one of the leading AI labs, continues to make big commitments with other companies to gain access to AI chips made by Nvidia," adding that "access to enough AI chips remains a challenge for firms like Anthropic, which places usage limits on its most advanced models like Fable."

Why it matters

If it materializes, a Meta-to-Anthropic compute lease would be a striking reversal of the usual competitive lines in AI: one of the largest AI labs paying a direct model-building rival for the raw material — GPU capacity — it needs to keep serving its own customers. For Meta, turning idle or excess data-center capacity into a paying line of business (following its overtures with SpaceX-style compute deals across the industry) gives it a way to argue its enormous AI infrastructure spend isn't purely a cost center, even as it continues to build and ship its own Llama models in direct competition with Anthropic's Claude and Fable lines.

For Anthropic, the reported talks fit a broader strategy of diversifying compute suppliers — SpaceX, xAI, and potentially now Meta — rather than depending on any single cloud provider, at a moment when Nvidia chip supply remains the binding constraint across the industry and Anthropic is already rationing access to its most capable models. Whether or not this specific deal closes, the scale of the number being discussed underscores how central compute access, not just model quality, has become to competitive positioning among frontier AI labs.