.png){kind=logo}
Anthropic is in early talks to lease artificial intelligence computing power from Meta. The proposed agreement could reach $10 billion over two years.
Anthropic approached Meta with the proposal in June, according to reports citing people familiar with the discussions. The companies have not reached a final agreement, and the talks may end without a contract.
Meta shares moved off their session lows after the report emerged on Friday. However, both Meta and Anthropic declined to comment on the negotiations.
Under the proposed terms, Anthropic would make monthly payments to Meta during the two years. Either company could also end the arrangement before the full term expires.
The discussions could give Meta its first large customer for rented computing capacity. Meta currently uses most of its data centers, chips, and power supply for its own services and AI models.
However, chief executive Mark Zuckerberg has discussed selling access to Meta’s infrastructure. At a shareholder meeting in May, he said cloud computing was “definitely on the table.” He also said companies approached Meta “almost every week” about buying models or spare computing power.
Meta expects 2026 capital spending of between $125 billion and $145 billion. The forecast includes finance lease payments and higher costs for components and data centers. Much of that spending supports AI systems and future computing capacity.
Anthropic proposed a structure worth up to $10 billion across two years. Still, people familiar with the matter said the talks were at an early stage.
A source also described the discussions as very preliminary. Meta must develop commercial systems for billing, customer support, security, and capacity management before operating a large cloud service.
The companies have not disclosed how much computing power Anthropic could receive. They have also not identified the chips, data center sites, pricing schedule or start date. The lack of a signed contract leaves several questions open. The final value could fall below $10 billion, while either side could withdraw before the service period ends.
Anthropic needs large amounts of computing power to train and operate its Claude models. Demand for AI services has pushed developers to secure chips, electricity, and data center space through several providers.
In May, Anthropic reached a separate agreement to use the full computing power of SpaceX’s Colossus 1 facility in Memphis. The site contains more than 220,000 Nvidia processors and provides 300 megawatts of capacity.
Later reports placed that arrangement at about $1.25 billion each month through May 2029. The contract also allows early termination, giving both parties flexibility if demand or infrastructure plans change.
Anthropic also has computing arrangements with major cloud providers. Its talks with Meta show that the company is adding more sources as usage grows across its consumer and business products. The proposal would also diversify Anthropic’s infrastructure beyond its existing cloud partners.
For Meta, a completed agreement would add revenue tied directly to its AI infrastructure. The company is also building a broader cloud operation to sell spare capacity and access to selected AI models.
No agreement has been announced. Meta and Anthropic have not provided a timeline for completing or ending the talks.
ALSO READ: Meta Adds a New Feature to Protect Teens from Self-Harm Conversations
