Anthropic is in preliminary talks to lease artificial intelligence computing power from Meta, a deal that could be worth roughly $10 billion and would mark one of the largest capacity-sharing agreements between rival AI labs to date. The discussions were first reported by The New York Times and confirmed by CNBC, which cited a person familiar with the matter.

The reported arrangement would see Meta rent out a slice of its hard-won GPU capacity to Anthropic, the maker of the Claude assistant — a striking reversal for two companies that compete head-on for the same enterprise customers and developer mindshare. For continuing coverage of the deals reshaping the industry, follow our AI industry coverage as it develops.

What the reporting says

According to CNBC, which attributed the scoop to correspondent Kate Rooney, the talks remain "very preliminary" and no agreement has been finalized. The New York Times put the potential value of the lease at around $10 billion. Meta declined to comment.

Key details emerging from the reporting:

  • The structure is a lease, not a partnership. Anthropic would pay to use Meta's compute rather than the two firms co-developing models, keeping their respective AI products and research agendas separate.
  • It follows a similar SpaceX arrangement. The talks come weeks after Anthropic announced a deal to use computing capacity at Elon Musk's Colossus 1 data center to expand capacity for paid subscribers.
  • Meta has been openly courting such buyers. Chief Executive Mark Zuckerberg told analysts in October that companies were "regularly asking if we have compute that they could buy from us at some premium to what we've bought it at."

The negotiations underscore that access to Nvidia's AI chips remains the single biggest constraint for frontier labs like Anthropic, which has imposed usage limits on its most advanced models, including Fable, to manage demand against finite supply.

Meta's pivot toward selling compute

For Meta, a lease of this magnitude would validate a strategy Zuckerberg began signaling publicly in May, when he said the social media giant was considering entering the cloud computing business as a way to show investors it can monetize its enormous AI investment beyond its core advertising engine.

Meta is on track to spend as much as $145 billion on capital expenditures, including AI infrastructure, in 2026 alone — a figure that has placed mounting pressure on the company to demonstrate returns beyond incremental improvements to its apps. Selling spare capacity to a well-funded buyer like Anthropic would convert a cost center into a revenue line almost overnight.

The company has also been building the leadership to execute on that ambition. Dave Brown, a former longtime senior executive at Amazon Web Services, is set to join Meta, CNBC confirmed — a hire that signals Meta intends to run its compute like a hyperscale cloud provider rather than purely an internal resource.

Why Anthropic needs the chips

Anthropic's hunt for capacity reflects a broader reality: the cost of staying competitive at the frontier of AI has escalated far beyond what any single cloud provider can supply. The company, backed heavily by Amazon, has layered multiple compute arrangements on top of its AWS relationship — including the SpaceX deal and now potentially Meta — to feed the insatiable training and inference demands of its Claude model family.

Industry observers note that such cross-rival leasing was almost unheard of two years ago, when labs hoarded every GPU they could acquire. The emergence of compute-as-a-commodity deals suggests the market is beginning to treat AI chips the way it once treated electricity: as a fungible resource that can be bought, sold, and leased across company lines when the price is right.

The bigger picture

A $10 billion lease would also recalibrate how investors judge Meta's AI spending. Rather than a sunk cost tied to engagement metrics, Meta's data centers would begin generating predictable, multi-year revenue from a blue-chip tenant — a narrative that could reshape the debate over whether the industry's capital expenditure boom will ever pay off.

For Anthropic, the appeal is more existential. Every additional tranche of compute is another increment of headroom before its models are once again throttled by supply, and another reason for enterprise customers to believe the company can scale alongside their needs.

Neither company has confirmed a final agreement, and sources caution that the talks could still fall apart or shrink in scope. But the mere fact that two of the most watched names in AI are negotiating over compute access — rather than models, products, or talent — points to where the real bottleneck, and the real money, now sits in the industry.

Stay Ahead of AI

Anthropic's reported $10 billion compute talks with Meta are the latest sign that AI infrastructure has become the industry's defining constraint. For more on the deals, models, and policy moves reshaping artificial intelligence, follow our breaking AI news as it happens.

Read more AI news →