Meta Platforms is exploring a major expansion into the artificial intelligence infrastructure market, holding early-stage talks with AI company Anthropic over a computing services agreement that could be worth as much as $10 billion over two years.
The proposed deal would see Meta lease computing capacity to Anthropic, signalling a potential new revenue stream for the social media giant beyond its core advertising business.
According to reports, Anthropic initiated discussions in June, proposing to pay Meta monthly for access to its AI computing infrastructure.
While negotiations are ongoing, sources familiar with the matter said the discussions remain preliminary and may not result in a final agreement. Both companies would reportedly have the option to terminate the arrangement before the end of the proposed two-year term.
News of the talks weighed on investor sentiment, with Meta shares closing more than 2% lower after reports of the potential partnership emerged.
If completed, the agreement would mark a strategic shift for Meta, which currently does not operate a commercial business leasing computing capacity to external customers.
The move would position the company alongside specialist AI infrastructure providers such as CoreWeave and Nebius, which have benefited from surging demand for the massive computing power required to train and operate advanced artificial intelligence models.
Chief Executive Officer Mark Zuckerberg has previously indicated that commercialising Meta’s AI infrastructure is under consideration.
Speaking to shareholders in May, Zuckerberg said expanding into cloud computing was “definitely on the table,” noting that Meta receives frequent requests from companies seeking access to its AI models and excess computing capacity.
Anthropic has been aggressively expanding its computing resources as competition in generative AI intensifies.
Earlier this year, the Claude developer entered into an agreement with SpaceX to use the Colossus 1 data centre in Memphis, Tennessee, significantly increasing the infrastructure available to train and deploy its AI models.
The latest discussions with Meta reflect the industry’s growing need for high-performance computing capacity as developers race to build increasingly powerful AI systems.
The talks come during a period of significant change for both companies.
Meta recently discontinued its AI image-generation tool, Muse Image, following criticism from privacy advocates and entertainment industry groups over its use of publicly available Instagram content to generate images.
Anthropic, meanwhile, temporarily suspended international access to its flagship Fable 5 and Mythos 5 AI models after the U.S. government imposed export control restrictions.
Those restrictions were later lifted after Anthropic agreed to enhanced safety standards and monitoring requirements, allowing the company to restore access to the models.
A potential partnership between Meta and Anthropic highlights the rapidly evolving economics of artificial intelligence, where access to computing power has become as strategically important as the development of AI models themselves.
As demand for advanced AI accelerates, technology companies are increasingly investing billions of dollars in data centres, specialised chips and cloud infrastructure, turning AI computing capacity into one of the industry’s fastest-growing businesses.
