A reported plan to sell surplus computing power would put Meta up against AWS, Google Cloud and Azure, and give investors a reason to look past its spending.
Meta has spent two years buying every scrap of AI computing power it can lay hands on. Now it appears to be working out how to sell some of it back.
Bloomberg reported on Wednesday, citing people familiar with the matter, that the company is building a cloud business to offload excess AI capacity, and investors treated the news as a small relief after months of anxiety about how much Meta is spending and what it will get for it.
The plans are early and could still change, according to the report. What Meta is said to be weighing is the shape of the offering rather than whether to pursue it. One option is to sell access to AI models hosted on its own infrastructure, roughly the way Amazon’s Bedrock works.
The other is to sell raw computing capacity, the model that neocloud providers such as CoreWeave have built entire businesses on. Either path would put Meta head to head with Amazon Web Services, Google Cloud and Microsoft Azure, the three incumbents that dominate the market it would be entering.
The effort is reportedly gathered under a new unit called Meta Compute, led by the company’s head of infrastructure Santosh Janardhan alongside Meta Superintelligence Labs figure Daniel Gross and Meta president Dina Powell McCormick. The logic is straightforward enough.
Meta has guided to capital spending of $115bn to $135bn in 2026, an enormous outlay on chips, land and power, and a cloud business is one of the few ways to turn idle capacity from that build-out into revenue instead of a sunk cost.
The market read it that way immediately. Meta shares jumped more than 10% on the report, a sharp move for a stock that had been having a poor year, down close to 15% as of the day before and lagging the S&P 500 as investors questioned the pace of its AI outlay.
A credible route to earning money back from the infrastructure, even a still-hypothetical one, was apparently enough to shift the mood.
Meta would not be the first to spot the opportunity. SpaceX has been renting spare capacity from xAI’s Memphis data centre to Anthropic, an arrangement Bloomberg Intelligence estimates could bring in more than $50bn by 2028 and $100bn by 2030.
The pattern is becoming familiar across the industry: build far more compute than you can use today, on the bet that you will need it tomorrow, and rent out the surplus in the meantime to defray the bill.
For Meta the surplus is real and growing. The company has a 2,250-acre hyperscale campus in Louisiana, a gigawatt-scale data centre under construction in the American Midwest, and a web of external deals layered on top, including new capacity from Crusoe worth roughly 1.6 gigawatts across two sites.
That appetite has run into limits elsewhere, with Google recently rationing Meta’s access to its Gemini models because it could not spare the compute.
There is an irony in a company that has been scrambling for compute now positioning itself to sell it, and it points to how lumpy this build-out has become.
Capacity arrives in enormous, indivisible chunks, timed to projections rather than current demand, which leaves even the hungriest buyers holding more than they can immediately use.
Selling the overflow is how the neoclouds, and now apparently their customers, plan to make the arithmetic work. Deals like Jane Street’s $6bn contract with CoreWeave show how much money is moving through that layer.
For now it remains a report rather than a product. Meta has not confirmed the plan, no pricing or launch timing has surfaced, and the people describing it stress that the strategy could still shift. What is not in doubt is the incentive.
When you have committed well over $100bn a year to infrastructure, finding buyers for whatever you are not using stops being a side project and starts looking like a necessity.
Be the first to comment