Anthropic is in early discussions with Meta over a proposed agreement to lease artificial intelligence computing capacity that could be worth up to US$10 billion over two years.
If completed, the deal would see Meta provide computing infrastructure to Anthropic despite the two companies competing in the market for large language models, where Meta develops the Llama family of models and Anthropic offers its Claude platform.
Discussions remain at an early stage
According to Reuters, Anthropic proposed the arrangement in June and Meta is still evaluating the proposal.
The reported structure would involve Anthropic making monthly payments over a two-year period, although the terms remain under negotiation and either company would be able to terminate the agreement before the contract expires.
Reuters noted that the discussions remain preliminary and may not result in a final agreement.
The negotiations are reportedly complicated by the fact that Meta does not currently operate a commercial business dedicated to selling cloud computing capacity to external customers.
Growing demand for AI infrastructure
Anthropic has increasingly relied on third-party infrastructure providers to support growing demand for its Claude and Claude Code products.
Earlier this year, the company signed a multi-year agreement with SpaceX for AI computing capacity at its Colossus 1 and Colossus 2 data centers in Memphis, Tennessee.
That arrangement, reportedly valued at approximately US$45 billion over three years, is substantially larger than the proposed Meta agreement.
The potential Meta contract would further diversify Anthropic’s sources of AI infrastructure as competition for high-performance computing resources continues to intensify.
Meta expands AI infrastructure ambitions
Meta has been investing heavily in AI infrastructure while exploring new revenue opportunities beyond its core advertising business.
During the company’s shareholder meeting in May, Chief Executive Officer Mark Zuckerberg said offering AI computing services to third parties remained a possible future business.
“Definitely on the table.”
Zuckerberg also said companies regularly approach Meta seeking access to its AI models or excess computing capacity.
Meta expects capital expenditure to reach as much as US$145 billion in 2026, more than double the approximately US$72 billion invested the previous year, with much of that spending directed toward AI infrastructure.
A commercial computing business could position Meta alongside specialized AI infrastructure providers such as CoreWeave and Nebius.
Competitors increasingly share infrastructure
The discussions reflect a broader trend within the artificial intelligence industry, where competing model developers increasingly buy and sell computing resources to one another as demand outpaces available capacity.
While the reported valuation of up to US$10 billion highlights the scale of potential demand, Reuters noted that negotiations remain ongoing and there is no assurance that the companies will finalize an agreement.
If completed, the deal would mark another example of AI developers relying on competitors’ infrastructure to accelerate the deployment of increasingly compute-intensive foundation models.
Source: DataMagz