The plan, according to people familiar with it, has been taking shape inside Meta for months: sell access to the company’s AI computing capacity and models to external customers, entering a market dominated by Amazon Web Services, Microsoft, and Google Cloud. The company has been stockpiling data centers and computing infrastructure at a furious pace to support its own AI development, and it is now putting together a business that would generate revenue from the capacity it does not need — an acknowledgment that the AI build-out has produced more computing than even Meta’s ambitions consume.
The move would be a departure for a company that has historically kept its infrastructure for its own use. Meta’s data centers were built to run its apps, its ads systems, and, increasingly, its AI models, and the company has never sold computing capacity at scale. The new business, according to people familiar with the plans, would rent out excess computing power and provide access to Meta’s models, giving outside developers and enterprises a way to use the infrastructure without owning it. The details of pricing, availability, and timelines have not been finalized.
The logic of the plan is economic as much as strategic. Meta’s AI spending has reached tens of billions of dollars a year, and the company has told investors that the build-out will continue. Selling unused capacity turns a fixed cost into a revenue stream, and it hedges the risk that Meta’s own demand for computing grows more slowly than its infrastructure. The plan also gives Meta a way to monetize its AI models — the Llama family, which it has distributed openly — by selling managed access to them, alongside the raw computing power to run them.
The competitive implications are significant. The cloud market’s three leaders — AWS, Microsoft, and Google — have spent years building the sales organizations, compliance frameworks, and partner ecosystems that enterprise customers require, and a new entrant would face a steep climb. But Meta brings assets the incumbents cannot match: one of the largest AI research organizations in the world, a family of models with a large open-source following, and the infrastructure to run them at scale. A Meta cloud would be the first serious challenger to the established trio to come from the AI side rather than the infrastructure side.
The timing adds pressure to an already competitive market. AWS announced this week that it is raising prices on GPU instances by 20%, citing the rising cost of Nvidia’s chips, and the other clouds are expected to follow with their own adjustments. A new entrant with fresh capacity could undercut the incumbents on price, or at least give enterprise buyers an alternative to negotiate with. Meta’s entry, if it happens, would come at the moment the cloud market most needs competition.
The strategic context is the company’s relationship with OpenAI. Microsoft, Meta’s longtime rival, has invested tens of billions in OpenAI and made its cloud the default home for the startup’s models. Meta has no equivalent anchor tenant, but it has something the incumbents lack: control over its own models. By selling access to Llama and its computing infrastructure together, Meta can offer customers a full stack that does not depend on any other company’s technology — an argument that has grown more powerful as concerns about dependence on any single AI supplier have spread.
The plan raises familiar questions about focus. Meta’s history is full of initiatives that started with fanfare and ended quietly, and the company’s executives have been disciplined about concentrating on its core businesses. Selling cloud capacity is a different business from running social networks, requiring sales teams, service levels, and legal frameworks that Meta does not have. People familiar with the plans said the company is aware of the challenge and is building the business slowly, starting with select customers, rather than attempting a full launch.
The market’s reaction will be telling. Meta’s shares have traded in part on the promise that its AI investments will eventually pay off, and a cloud business offers a concrete path to monetization — evidence that the billions spent on data centers can generate returns beyond Meta’s own products. Investors have heard such promises before, from companies that entered cloud services with less at stake, and they will be watching for signs that the business is real: contracts, customers, and revenue.
The broader significance is the direction of the AI economy. The biggest builders of AI infrastructure — the hyperscalers and, now, Meta — are all converging on the same model: build enormous capacity, use what you need, and sell the rest. If Meta’s plan proceeds, the market for AI computing will have a new supplier, the incumbents will have a new rival, and the customers who rent computing power will have a new option. Whether the plan becomes a business or a footnote will be decided by the same test every cloud faces: whether the capacity can be filled.
