Anthropic has held early talks to rent AI computing power from Meta

Anthropic has held very early discussions about renting computing power from Meta, CNBC’s Kate Rooney reported, citing a person familiar with the matter. For retail investors, the talks point to a bigger shift inside the AI trade: the companies spending the most on chips and data centers are looking for ways to turn that infrastructure into revenue.

The New York Times reported that a potential arrangement under discussion could be worth about $10 billion. CNBC reported that Meta shares moved up from their session lows on Friday after that report.

Compute is the industry term for the processing capacity needed to train and run artificial intelligence models. In practice, that usually means data centers packed with specialized AI chips, including chips made by Nvidia, that can handle the large workloads behind products such as chatbots, coding tools and image systems.

CNBC reported that access to enough AI chips remains a problem for companies such as Anthropic. The company has placed usage limits on its most advanced models, including Fable, according to CNBC.

Anthropic is looking beyond its own infrastructure

The Meta discussions follow another compute arrangement Anthropic announced with Elon Musk’s SpaceX. Under that deal, Anthropic said it would use capacity at SpaceX’s Colossus 1 data center to improve availability for paid subscribers, according to CNBC.

That pattern matters because AI labs need enormous amounts of computing capacity before customers ever see a product. A model may be expensive to build, but it can also be expensive to serve once people start using it at scale. If demand rises faster than available chips, companies can limit usage, pay outside providers or build more of their own infrastructure.

Anthropic is one of the leading AI labs, CNBC reported, and its willingness to work with other companies for capacity shows how tight the market for high-end AI infrastructure remains. CNBC described the Meta talks as very preliminary.

Meta is testing a new way to monetize AI spending

The talks also fit with Meta’s own public comments about cloud computing. CNBC reported that Meta CEO Mark Zuckerberg said in May that the company was considering entering the cloud computing business. The goal, according to CNBC, was to show investors that Meta can earn money from AI investments outside its existing advertising-driven business.

Cloud computing means renting computing resources to other companies instead of requiring them to own all the servers themselves. For Meta, selling spare or dedicated AI capacity could create another use for infrastructure it is already building for its own AI products.

The Wall Street Journal reported that Meta hired Dave Brown, a former senior Amazon Web Services executive, as its head of infrastructure. Amazon Web Services is Amazon’s cloud unit and one of the biggest players in renting computing capacity to businesses.

Meta could spend as much as $145 billion on capital expenditures in 2026, including AI infrastructure, according to CNBC. Capital expenditures, often called capex, are long-term investments in assets such as data centers, servers and chips.

Zuckerberg said last October that companies regularly were “asking if we have compute that they could buy from us at some premium to what we’ve bought it at,” according to CNBC.

Meta declined to comment to CNBC.

This story draws on original reporting from CNBC.