Meta Platforms launches AI cloud service to sell excess compute

Meta Platforms is preparing a cloud‑infrastructure business, dubbed “Meta Compute,” to rent out surplus AI computing power and grant access to its own AI models. The plan, reported by Bloomberg and Reuters, would pit Meta directly against Amazon Web Services, Microsoft Azure, Google Cloud and niche “neocloud” providers such as CoreWeave and Nebius. Executives leading the effort include head of infrastructure Santosh Janardhan, AI‑lab chief Daniel Gross and President Dina Powell McCormick. Investors reacted positively; Meta’s shares jumped 8‑10% on the news, while shares of competing AI‑cloud firms fell. The move is meant to monetize the company’s $125‑$145 billion AI‑related capex, diversify revenue beyond advertising and offset concerns about the heavy spending on data‑centers and GPUs.

The proposed offerings could follow two models: (1) a managed service that lets developers use hosted AI models (similar to AWS Bedrock) and (2) a raw‑compute lease that sells GPU capacity directly to customers. Analysts see the initiative as a way for Meta to turn idle infrastructure into a recurring revenue stream and to lessen reliance on ad‑driven earnings.