Meta is developing plans for a cloud infrastructure business that will sell access to computing power and artificial intelligence models, opening a new front of competition with industry leaders such as Amazon Web Services, Microsoft Azure and Google Cloud.
Meta, which has been racing to secure expensive data centers and other structures to support its own artificial intelligence ambitions, is building a business to generate revenue from surplus computing capacity sold to outside customers, according to people familiar with the matter. They requested anonymity because the details are not yet public.
One of the plans under consideration includes selling access to different AI models hosted on Meta’s existing infrastructure, in an approach similar to AWS’s Bedrock offering, these people said. Meta would operate the data centers and chips that power the models, including its own Muse Spark models, and charge developers for access.
InfoMoney Tool
Download now (and for free)!
The company is also considering selling access to “raw” computing capacity, in line with what so-called neoclouds, such as CoreWeave Inc., do, according to these sources. The development of these new lines of business is part of Meta Compute, an internal initiative aimed at building and managing the company’s AI infrastructure, according to a person familiar with the plans. Meta Compute is led by Santosh Janardhan, head of infrastructure at Meta; Daniel Gross, one of the leaders of the Meta Superintelligence Labs AI unit; and the president of Meta, Dina Powell McCormick.
A Meta spokesperson declined to comment. The company’s plans are still in development, and it is possible that the strategy will change. Meta shares jumped 9.3% to $615.55 at 10:04 a.m. in New York on Wednesday, the biggest intraday rise since April. CoreWeave fell by up to 14%. Nebius, a Dutch AI data center company listed in New York, fell by up to 17%.
Meta has made the development of a “superintelligence” a top priority and has already committed hundreds of billions of dollars to data centers and other AI infrastructure, such as expensive chips that it considers necessary to achieve this goal. This investment — which has left investors apprehensive about how Meta intends to get a return on this spending — includes major computing deals with CoreWeave, Alphabet’s Google, and Oracle, among others.
A cloud business offers a way to recoup some of that investment. AWS, Azure and Google Cloud have spent decades building platforms that rent access to computing power, storage and software over the internet — businesses that now generate tens of billions of dollars a quarter in revenue.
The demand for computing power from large AI developers remains insatiable. Meta and other technology companies have committed tens of billions of dollars in data center capacity for their own use in recent quarters, further expanding the industry’s massive spending on artificial intelligence. Selling computing is just another way to capture value from the broader AI boom.
As the appetite for AI has exploded, these providers have also expanded their rental offerings of specialized chips and the computing power needed to train and operate AI models. It’s a complex business, requiring not only huge fleets of data centers, but also software platforms, enterprise sales teams and customer support operations.
Continues after advertising
Elon Musk’s SpaceX, which acquired his AI startup xAI in February, has recently emerged as a relevant name in this market. Earlier this year, it leased access to its massive data center in Memphis to Anthropic PBC and also struck a deal with Google. This strategy could help xAI generate more than $50 billion in revenue by 2028 and $100 billion by 2030, according to an estimate from Bloomberg Intelligence.
Despite the complexities, Meta CEO Mark Zuckerberg has already signaled to investors that he is open to selling surplus computing infrastructure, or even an API service in which customers would pay for the use of AI — a deal typically measured in “tokens,” that is, the volume of data used and generated in a customer query.
“That’s definitely on the table,” Zuckerberg said during a conference call with shareholders in May. “Almost every week there are outside companies approaching us to create an API service or asking if we have computing power that they can buy from us for some premium over the price we pay.”
Continues after advertising
“We haven’t done that yet because we think we have a use for that computing power,” Zuckerberg said at the time. “But obviously, if we get to a point where we feel like we’ve overbuilt, that’s an option we have — and that’s part of what gives us the confidence to continue investing in this expansion.”
Amid the accelerating race for AI, Zuckerberg has repeatedly suggested that he believes the industry faces a computational capacity constraint and that Meta should accumulate as much as possible now to decide its use later.
© 2026 Bloomberg L.P.
Continues after advertising
