Meta and Anthropic are reportedly in preliminary talks about a compute deal that could total up to $10 billion across two years, according to multiple sources cited by the New York Times on July 17. The arrangement would involve monthly payments from Anthropic to Meta with an option for either party to exit early.
Details and Industry Context
Anthropic initially proposed the deal in June. Both Meta and Anthropic have declined to comment publicly on the negotiations. CNN confirmed the discussions but noted that exact financial figures reported may be speculative. CNBC also independently verified the talks.
Following the news, Meta's stock experienced a drop of as much as 6% in intraday trading before closing down approximately 2% on July 17.
Anthropic faces significant compute challenges in 2026, having imposed usage limits on its most advanced AI models, including Claude Fable, due to insufficient processing capacity relative to demand. The company already secured a $45 billion, three-year compute agreement with SpaceX in May, which grants access to the Colossus 1 data center in Memphis. The potential Meta deal would supplement Anthropic's GPU capacity as it prepares for a potential initial public offering, reportedly targeting an October listing according to Reuters.
For Meta, the negotiations align with its strategic move toward providing cloud computing services externally under the internal project name "Meta Compute." CEO Mark Zuckerberg indicated in May the company sees cloud computing as a way to demonstrate that its AI investments can generate revenue beyond advertising.
Meta plans to invest $145 billion in capital expenditures in 2026, more than doubling the $72 billion spent in 2025. Most of these funds will support AI hardware and data center expansion. The company has cut 8,000 jobs as part of reallocating resources to AI development and recently hired Dave Brown, formerly a senior Amazon Web Services executive, signaling its cloud ambitions.
