Meta is about to sell off excess AI capacity, and it's a sign the AI race is changing direction
For many years, Meta has always said their data center is only for internal service, feeding Facebook, Instagram and now the company's own AI models. But according to Bloomberg news on July 1, 2026, Meta is quietly preparing a different direction, turning that same infrastructure into a rental business. The project is internally named Meta Compute, and if implemented, Meta would sell access to its AI computing power, and even its AI models, to other companies. In other words, the social networking giant is planning to enter the realm of Amazon Web Services, Google Cloud and Microsoft Azure.
How Meta Compute will work
According to what Bloomberg gathered, Meta Compute is headed by three people, including Santosh Janardhan, who is in charge of the company's infrastructure, Daniel Gross, head of Meta Superintelligence Labs, and Dina Powell McCormick, current president of Meta. This business is expected to go in one of two directions, or both. The first direction is to sell "raw compute", which means directly selling the computing power of GPU clusters and graphics processors used to train and operate AI models, without anything else, like the way CoreWeave is doing.
The second direction is to sell access to hosted AI models, which are stored and operated on Meta's infrastructure, including Muse Spark, the latest closed source model from Meta Superintelligence Labs, similar to how AWS is selling rights to use many different AI models through its Bedrock service.
Meta is not the leader
It's worth noting that Meta is not the first large company to think about selling excess computing power. As early as May 2026, SpaceX, through its subsidiary xAI, signed a contract to lease the entire Colossus 1 data center to Anthropic. After that, xAI continued to sign similar contracts with Google and another AI company, Reflection AI. The fact that a company known for its rockets and satellites is reselling its AI capabilities shows that when the entire AI industry is pouring money into building infrastructure faster than it can figure out how to use it all, selling the surplus to competitors or partners becomes a financially sound choice, even if it sounds competitively contradictory.
Why does Meta need to do this?
The reason Meta is considering this direction is not difficult to guess when looking at the scale of money they have poured into AI. According to filings with the US Securities and Exchange Commission as of the end of the first quarter of 2026, Meta has committed to spending up to $182.9 billion on AI infrastructure in the coming years, including large-scale projects in Louisiana and Ohio. As for the project in Ohio, according to Mark Zuckerberg once shared, it has an area equivalent to the entire island of Manhattan and is expected to go into operation this year. The problem is, unlike Google or OpenAI, Meta has not shown really great user demand for its AI products. The company does not separately disclose revenue from Meta AI or from Llama, its open source model line, and in public statements, Meta leaders often emphasize using AI for internal improvements rather than turning AI into a standalone revenue source. To put it bluntly, Meta is sitting on a mountain of expensive infrastructure that cannot be fully utilized, so selling the surplus is the most realistic way to recover part of the capital spent. This also matches what Zuckerberg said in May 2026, that Meta establishing a cloud business is "definitely under consideration".
The AI race is changing the rules of the game
More broadly, the fact that Meta and SpaceX are moving in this direction suggests a change in the way people evaluate who is winning the AI race. In the past two years, the story often revolved around which company had the strongest model and best product. But when even companies that don't have commercially successful AI products like Meta can make money just by renting out hardware, it shows that the real winner may lie in who owns the most data centers, not who writes the smartest model. Of course, this view is not lacking in skeptics. There is an opinion that the whole industry is creating a kind of investment bubble, based on chips that lose value very quickly with each new generation. There is also the question of whether AI companies can generate enough revenue from end users to justify these hundreds of billions of dollars in investments. Meta has not yet officially confirmed Meta Compute. But if this project comes to fruition, it won't just be a side business for Meta to make some extra money, but also a signal that the companies that are spending the most on AI aren't entirely sure their AI business model will bring in adequate revenue, and they need a backup plan now.