Meta is planning to launch a cloud infrastructure business to sell AI computing power and host AI models for external customers, according to a report by Bloomberg.
The proposed business would put Meta in direct competition with cloud providers such as Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. The report added that the company is exploring ways to monetise excess AI infrastructure as it continues investing heavily in data centres and AI chips.
One proposal under consideration would allow developers to access AI models hosted on Meta’s infrastructure through an API, similar to AWS’s Bedrock service. Meta would operate the underlying data centres and hardware while charging customers to use models, including its own Muse Spark models.
The company is also evaluating the possibility of renting out raw AI computing capacity, similar to cloud infrastructure providers such as CoreWeave. The effort is part of an internal initiative called Meta Compute, which oversees the company’s AI infrastructure strategy, Bloomberg reported.
AIM has reached out to Meta for a comment.
In April, Meta raised its capital expenditure forecast to $125–145 billion for 2026, up from its previous guidance of $115–135 billion, citing additional investments in AI data centres, infrastructure hardware, and future capacity requirements.
Meta is building multiple AI data centre campuses, including Prometheus in Ohio and Hyperion in Louisiana, US, which is expected to scale to 5 GW of computing capacity.
Moreover, Meta has supplemented its in-house infrastructure with external AI computing capacity through multi-billion-dollar agreements with CoreWeave and Google.
A cloud business could provide one avenue for recovering part of that spending. Established cloud providers generate billions of dollars in quarterly revenue by renting computing resources, storage, and software to enterprises. Demand for AI infrastructure has also increased as companies build and deploy large AI models.
Meta’s strategy mirrors a recent move by SpaceX, which has begun monetising excess AI computing capacity through its Memphis-based data centre. Earlier this year, SpaceX signed a deal to provide Anthropic with access to the full capacity of its Colossus 1 AI cluster and later entered a multi-year agreement with Google, under which Google will pay $920 million per month from October 2026 through June 2029 for AI compute.
Meta CEO Mark Zuckerberg had earlier indicated that he was open to commercialising surplus AI infrastructure. Speaking during the company’s call with shareholders in May, he said customers had approached Meta about purchasing compute capacity or accessing AI models through an API service.
He added that the company had not pursued those opportunities because it still expected to use the infrastructure internally, but said selling excess capacity remained a possibility if Meta eventually built more compute than it required.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.