- Meta Platforms is developing a cloud infrastructure business — internally called Meta Compute — to sell access to AI computing power and models to outside customers, directly competing with AWS, Microsoft Azure, and Google Cloud; shares jumped as much as 8.6% in premarket trading on Wednesday before paring gains.
- One potential plan mirrors AWS’s Bedrock offering: selling access to multiple AI models hosted on Meta’s infrastructure — including its own Muse Spark models — where Meta runs the data centers and chips and charges developers per-query in “tokens”; a second option involves selling “raw” compute capacity akin to neocloud businesses like CoreWeave.
- Meta Compute is led by a high-powered trio: Santosh Janardhan (Meta’s head of infrastructure), Daniel Gross (a leader inside Meta Superintelligence Labs), and Meta President Dina Powell McCormick — a leadership structure that signals this is a board-level strategic priority, not a skunkworks experiment.
- CEO Mark Zuckerberg previewed this direction at Meta’s May shareholder call, saying external companies approach Meta “almost every week” asking for API services or compute capacity, and that if Meta concludes it has overbuilt, selling excess compute “is definitely on the table” — a signal the market had been waiting for given Meta’s hundreds of billions in data-center commitments.
What Happened?
Bloomberg reported Wednesday that Meta Platforms is building a cloud business under the internal name Meta Compute, with plans to sell AI computing capacity and model access to external customers — a direct challenge to the hyperscaler cloud market dominated by Amazon Web Services, Microsoft Azure, and Google Cloud. Two business models are under development: one mirroring AWS Bedrock (selling hosted model access, billed by token), and one selling raw compute capacity like CoreWeave. Meta’s vast infrastructure investments — hundreds of billions of dollars in data centers, chips, and AI capacity including major deals with CoreWeave, Google, and Oracle — have been a source of investor anxiety about return on capital. A cloud business offers one direct answer: monetize the excess. Zuckerberg signaled openness to this at the May shareholder call, and the company is now moving from consideration to active development.
Why It Matters?
If Meta successfully launches a cloud business, it reshapes the competitive dynamics of the AI infrastructure market in several important ways. First, it adds a fourth hyperscale competitor to the AWS/Azure/GCP oligopoly — one with access to Llama and Muse Spark models that are already widely used and a cost structure potentially below market rates if Meta is selling “excess” capacity. Second, it provides a credible revenue model for Meta’s massive capex commitments, which have spooked investors who couldn’t see how $65 billion in 2026 infrastructure spending would generate returns. Third, it puts Meta in direct competition with the very companies — AWS, Google, Microsoft — that it also relies on as cloud partners and AI infrastructure providers, creating a complex coopetition dynamic. xAI/SpaceX has already shown this model can work, generating revenue from its Memphis data center with clients including Anthropic and Google.
What’s Next?
Meta’s plans are still in development and could change. Building a credible cloud business requires not just data centers and chips but also enterprise sales teams, customer support operations, billing infrastructure, and a software platform — capabilities Meta has not historically built or sold externally. The timeline to revenue is therefore uncertain. But the market reaction — shares up sharply on the news — suggests investors were looking for exactly this kind of capex-monetization story, and the appointment of President Dina Powell McCormick to the leadership team signals Meta is serious about the commercial and government customer relationships a cloud business requires. Watch for formal announcements at Meta’s next earnings call and potential early customer deals with AI startups or enterprises that want Llama-based inference at scale.
Source: Bloomberg
