Meta Wants to Rent Out Its Spare AI Compute

Meta is building a cloud computing business to rent out the AI capacity its own teams cannot use, according to a Bloomberg report published July 1. The effort, known internally as Meta Compute, would put the company in direct competition with Amazon Web Services, Microsoft Azure and Google Cloud, along with the specialist GPU providers CoreWeave and Nebius.

The logic is straightforward. Meta has committed something in the region of $145 billion to AI infrastructure, and it has built more capacity than its work on Llama, recommendation systems and its Muse Spark model currently consumes. Rather than let those graphics chips sit idle, the company would sell the spare cycles. As TechCrunch noted, it is the same move SpaceX made with Starlink capacity: turn a costly asset into a revenue line while you wait for internal demand to catch up.

Two ways in

Meta is weighing two arrangements. One would sell access to AI models hosted on Meta's own hardware. The other would rent raw compute, letting customers run whatever they like on Meta's machines. The project is said to be led by infrastructure chief Santosh Janardhan alongside Meta Superintelligence Labs leader Daniel Gross and company president Dina Powell McCormick. Meta has not named a launch date, a price or a first customer.

Wall Street picked a winner immediately

The market reaction was blunt. Meta shares jumped more than 9 percent on the news, the stock's biggest single-day gain in over five months, according to CNBC. The pure-play cloud rivals went the other way. CoreWeave fell close to 11 percent and Nebius dropped more than 12 percent, as Tom's Hardware reported. Investors clearly read a well-funded new entrant with its own data centers as a real threat to companies whose entire business is renting GPUs.

There is an unspoken admission in all of this. A year ago Meta framed its spending as the cost of building world-beating models. Selling the surplus is a more pragmatic story, and a quieter one. The company still gets to keep pouring concrete and buying chips. It just needs someone else to help pay for them while its own models find their footing. For the state of those models, see our coverage of Meta's next model catching up to OpenAI.