1 day ago
12
Meta Platforms is developing plans for a cloud infrastructure business that would sell access to AI computing power and models, Bloomberg reported, citing people familiar with the matter.
CNBC later reported that Meta plans to sell excess computing power to outside customers.
The business would compete with Amazon Web Services, Microsoft Azure, Google Cloud, and CoreWeave, which rent computing power, storage, software, and AI infrastructure to enterprise customers.
The reported plans overlap with full-service cloud providers such as AWS, Microsoft Azure, and Google Cloud, as well as neocloud firms such as CoreWeave and Nebius, which focus more directly on GPU capacity.
Meta has been investing in data centres, chips, and other infrastructure to support its artificial intelligence work. The company is now exploring whether excess computing capacity can be sold to external customers, according to people familiar with the plans.
Infrastructure spending
Meta has projected spending of as much as US$145 billion on AI-related capital expenditure this year, including data centres and GPUs for AI workloads.
As of March 31, Meta had US$182.9 billion in future lease obligations, including commitments tied to data centre capacity, according to Bloomberg-related reporting.
A separate report said Meta’s infrastructure spending includes projects in Louisiana and Ohio, with the Ohio project expected to come online this year.
Hosted models and raw compute
One option under discussion involves giving developers paid access to AI models, including Muse Spark, hosted on Meta’s infrastructure. The approach would resemble AWS Bedrock, a managed cloud service for accessing AI models.
Meta unveiled Muse Spark in April, with Axios reporting that the company did not describe it as a new state-of-the-art model.
The plan would add a paid infrastructure service alongside Meta’s Llama strategy, which has focused on making model weights available under licence terms.
Meta is also considering selling raw computing capacity, similar to the model used by neocloud providers such as CoreWeave. Raw compute sales would give customers direct access to the specialised chips used for AI training and inference.
Meta Compute leads the effort
The plans are being developed under Meta Compute, an internal initiative focused on building and managing the company’s AI infrastructure. The effort is led by Santosh Janardhan, Meta’s head of infrastructure; Daniel Gross, a leader inside Meta Superintelligence Labs; and Meta president Dina Powell McCormick, people familiar with the plans said.
Meta has committed hundreds of billions of dollars to data centres, chips, and related systems as part of its work on AI “superintelligence.”
Those investments include computing deals with CoreWeave, Google, and Oracle, among others. Investors have questioned how Meta plans to generate returns from its infrastructure buildout.
Meta does not break out revenue from Meta AI or its Llama model family in its earnings. Executives have discussed AI use inside the company in public statements, according to a separate report.
Cloud rivals and compute deals
The reported business would involve selling infrastructure capacity that is not being used internally. AWS, Azure, and Google Cloud already generate tens of billions of dollars in quarterly revenue from cloud services.
Cloud providers have also expanded their AI offerings by renting access to specialised chips and compute capacity. The business also requires large data centre fleets, software platforms, enterprise sales operations, and customer support teams.
SpaceX has also struck compute deals with Anthropic and Google. CNBC reported that Anthropic agreed to pay US$1.25 billion per month, while Google agreed to pay US$920 million per month.
A separate report said SpaceX also signed a compute lease with Reflection AI.
Bloomberg Intelligence estimated that xAI’s infrastructure strategy could help the company generate more than US$50 billion in revenue by 2028 and US$100 billion by 2030.
Zuckerberg leaves option open
Meta chief executive Mark Zuckerberg said on a shareholder call in May that selling unused compute remains an option. He said outside companies had approached Meta about both API access and available computing capacity.
“It’s definitely on the table,” Zuckerberg said during the call. He added that Meta had not sold access yet because the company still expected to use the compute internally.
Zuckerberg has said computing capacity remains a constraint in the AI sector and that Meta should secure infrastructure before deciding how it will be used.
Meta shares rose 8.8% in New York on July 1. Shares of CoreWeave and Nebius fell 10.8% and 12.4%, respectively, following the report.
(Photo by Dima Solomin)
See also: Vodafone tests AWS cloud infrastructure for IoT network services
Want to learn more about Cloud Computing from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
Cloud Computing News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.