Meta Platforms is developing plans for a cloud infrastructure business that would sell outside developers access to its AI computing power and hosted models, according to Bloomberg, which cited people familiar with the matter who were not named because the plans are not yet public. The move would turn Meta from a pure consumer of data-center capacity into a seller of it, competing directly with Amazon Web Services, Microsoft Azure and Google Cloud.
The logic is straightforward. Meta has spent multiple billions of dollars securing data centers, chips, and other infrastructure to support its own AI ambitions, from Llama model training to the Meta AI assistant. Bloomberg’s sourcing indicates the company now intends to generate revenue from whatever computing power it is not using internally, selling that excess to outside customers rather than letting it sit idle.
That is a familiar playbook. Amazon built AWS out of infrastructure it originally provisioned for its retail business, and Google Cloud followed a similar path from Google’s internal systems. Meta has never run a public cloud business before. Doing so would require it to build billing systems, customer support, compliance tooling, and reliability guarantees that a purely internal infrastructure operation does not need.
The scale of the opportunity tracks the scale of Meta’s spending. The company has been among the most aggressive hyperscalers in committing capital to data centers and custom silicon, and any unused capacity at that scale represents a large potential revenue base if Meta can convert it into a sellable product. Bloomberg’s report does not specify a timeline, a target customer segment, or whether Meta intends to sell raw compute, hosted model access, or both.
That absence of detail matters. Selling raw GPU capacity is a commodity fight against AWS, Azure and Google Cloud, all of which have years of enterprise sales relationships and global data-center footprints Meta lacks. Selling hosted access to Meta’s own models is a different business, closer to what OpenAI and Anthropic already do through their APIs, and would put Meta in more direct competition with the model providers it currently trails on enterprise adoption.
Bloomberg’s report does not disclose whether Meta has set a launch date or named specific hyperscaler contracts it would compete for. Those specifics will determine whether this becomes a genuine third revenue stream alongside advertising and Reality Labs, or a smaller-scale offering aimed mainly at recouping capital costs.
For enterprise buyers currently negotiating multi-year compute contracts with the existing hyperscalers, Meta entering the market as a seller is a reason to wait for pricing details before locking in 2027 commitments.
Reported by Bloomberg on July 1, 2026.
