Meta eyes cloud business to sell excess AI computing power

Meta Platforms is building a cloud business to sell excess artificial intelligence computing capacity, a move that would push the Menlo Park, California, company deeper into the infrastructure layer of the AI race. The plan would let outside developers tap Meta-hosted models, including Muse Spark, and pay for the compute needed to run them.

The market responded fast. Meta shares rose more than 10% on the news, while neocloud names such as CoreWeave and Nebius fell on concern that Meta could one day need less outside capacity. The reaction showed how investors are reading the strategy: not just as a new product line, but as a possible shift from a company built on ads into one that can sell AI infrastructure alongside Amazon Web Services, Microsoft Azure and Google Cloud.

That shift matters because Meta’s spending is still enormous. In January, the company said it expected capital expenditures of $115 billion to $135 billion for 2026, up from $72.22 billion in 2025. Mark Zuckerberg said at the time that 2026 would be a big year for delivering personal superintelligence and accelerating business infrastructure, underscoring how aggressively Meta is still building out the systems behind its AI push.

The timing is also telling. Meta has repeatedly delayed a wider developer launch for Muse Spark, and as of early June there was still no scheduled launch date, even though the company was testing the API with early partners. Selling cloud access could help Meta recover some of the cost of those delays and turn a heavy internal buildout into a second revenue stream beyond advertising.

Meta is already entangled with the specialists it could eventually compete against. CoreWeave said in April that it expanded its agreement with Meta to provide AI cloud capacity through December 2032 for about $21 billion. Meta also has a Nebius contract worth up to $27 billion over five years. If Meta can commercialize its own infrastructure, it may reduce reliance on those providers and gain more leverage over the market for AI compute.

Analyst Gil Luria said the bigger effect may be on neoclouds than on the biggest hyperscalers, because Meta may no longer need as much outside help if it monetizes its own capacity. That would make Meta’s cloud move less a defensive sideline and more a test of whether one of the most expensive AI buildouts in the industry can also become a platform business.