Meta is turning its enormous AI infrastructure into a business. Under a new initiative called Meta Compute, reported by Bloomberg on July 1, the company plans to sell access to raw computing power and to hosted AI models — a cloud line that would put it in direct competition with CoreWeave, Amazon and Microsoft.
Selling compute and models
The plan is twofold: offer "raw" compute capacity in the style of CoreWeave, and rent access to hosted models including Meta's recently launched Muse Spark closed-weight model. It is led by Santosh Janardhan, Meta's head of infrastructure, alongside Meta Superintelligence Labs leader Daniel Gross and company president Dina Powell McCormick.
The move follows Mark Zuckerberg's comment in May that a Meta cloud business was "definitely on the table" as a way to recoup part of the company's AI spending. Unlike Google and OpenAI, Meta has not reported significant revenue from its own AI models and services.
A Manhattan-sized buildout
The initiative is anchored by infrastructure of extraordinary scale. As of the first quarter of 2026, Meta had committed $182.9 billion to AI infrastructure, with major data-center projects underway in Louisiana and Ohio. Zuckerberg has described the Ohio facility, due online in 2026, as roughly the size of Manhattan — part of a plan to add tens of gigawatts of capacity over the decade.
Selling that capacity would let Meta defray costs during the stretches when its own workloads do not fill it. It also hands the company a foothold in the cloud market at the moment demand for AI compute is at its most acute.