Meta weighs selling “spare” AI computing power, a potential new cloud push that targets AWS and Azure

Meta Platforms is reportedly moving to turn its excess AI computing capacity into a business, with plans that could put the company in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud. The idea, described as a buildout of a “cloud computing division,” would allow other companies to tap into AI horsepower that Meta already accumulates to train and run its own models.

According to the report, Meta’s approach is two-pronged. One option would let external developers pay to use AI models hosted on Meta’s servers, framed as a service similar to Amazon’s Bedrock, which gives customers access to multiple foundation models through an API rather than requiring them to run models themselves. The second option would sell access to raw computing capacity itself, an infrastructure strategy more akin to specialized GPU hosting providers.

The effort is said to sit under an internal program called Meta Compute, led by Meta infrastructure chief Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and Meta President Dina Powell McCormick. The reporting also ties the concept to the broader position Zuckerberg has been emphasizing for roughly two years: that Meta prefers to own more computing capacity than too little in order to keep pace with rapidly scaling AI workloads.

Meta’s infrastructure buildout has been a central theme in its recent operations, with the report pointing to a sustained push to construct data centers, buy chips, and lock in power agreements. The thrust is that the company is trying to ensure it can train and serve AI models at scale, even as demand for compute grows across the industry.

If Meta follows through, the competitive effect would be significant. AWS and Azure have spent years and large capital commitments building cloud services, including managed compute and developer-accessible AI tooling, while Meta would be using a different starting point: its own model and chip supply chain plus dedicated power and data center capacity. The report frames this as a potential shift in the cloud pecking order by making Meta a seller of capacity alongside a provider of AI models.

For Amazon investors and cloud market observers, the key question is whether Meta’s offering becomes a credible alternative for customers who currently source AI workloads from AWS, Azure, or Google Cloud. The report does not provide details on pricing, customer targets, or the timing of any launch. It also does not describe what level of demand Meta expects to capture by bundling access to its models with underlying compute, or whether Meta would restrict access to certain customers or use cases.

Meta shares reportedly rose sharply after the news, but the same reporting indicates the plan’s status remains uncertain. The post suggests Meta is “quietly building” the division, not that it is already operating at scale, and it stops short of outlining service availability or formal customer contracts. What to watch next is whether Meta names a launch timeline, discloses which models or services would be offered, and clarifies how Meta Compute would integrate with external developer workflows and enterprise procurement cycles.