Meta is preparing to sell access to its own AI computing power and models. On Wednesday, Bloomberg reported that the company is developing plans for a cloud infrastructure business to generate revenue from these assets.
This move would place Meta in direct competition with major providers like Amazon Web Services, Google Cloud, and Microsoft Azure.
The SpaceX parallel
Meta’s decision follows weeks after SpaceX, through xAI, announced similar intentions. In early May, SpaceX signed an agreement with Anthropic to purchase all compute capacity at its Colossus 1 data center. Similar leases have since been signed with Google and Reflection AI.
The fact that Meta is adopting the same strategy signals a shift in priorities. The winners of the AI race may be the entities that own the data centers rather than those developing the best models or services.
This logic holds only if demand for compute remains strong and data centers retain their value. Skeptics warn that the rush to build AI infrastructure is creating a bubble dependent on rapidly depreciating chips. Others question whether AI firms can generate enough end-user revenue to justify trillion-dollar investments.
Massive spending continues
Those concerns have not halted Meta’s heavy investment in infrastructure. As of the end of the first quarter, the company had committed $182.9 billion to AI infrastructure over the coming years. This includes major ongoing projects in Louisiana and Ohio.
The Ohio project, which Mark Zuckerberg described as the size of Manhattan, is expected to go online this year.
Lack of internal demand
Unlike Google and OpenAI, Meta has not seen significant demand for its own AI models and services. The company does not break out revenue from Meta AI or its Llama open-weight model family in earnings reports. Executives have mostly emphasised internal corporate uses in public statements. This suggests Meta’s AI efforts do not yet represent a material standalone revenue line.
Selling raw power
To return some value on its colossal spend, Meta may adopt CoreWeave’s business model by selling access to raw compute capacity, according to Bloomberg. The outlet also reported that Meta is considering following AWS’s lead by selling access to various AI models. This includes its recently launched closed-weight model, Muse Spark, hosted on its own infrastructure.
The new business line will form part of an initiative reportedly dubbed Meta Compute. Leadership includes head of infrastructure Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and president Dina Powell McCormick.
The report confirms Zuckerberg’s May statements that a Meta cloud computing business is “definitely on the table” as a method to secure returns on massive investments into AI “superintelligence.”
TechCrunch has reached out to Meta for comment.
What it means
For people building AI applications, this creates a new option for buying compute. Instead of relying solely on the three dominant cloud providers, teams can now access Meta’s hardware. This adds a fourth major source of supply to the market.
