Meta launches Meta Compute cloud business to sell excess AI infrastructure capacity

Meta confirmed it is forming a new cloud infrastructure business, Meta Compute, to generate revenue from excess AI computing power. The move pits Meta directly against Amazon Web Services, Microsoft Azure, and Google Cloud. Meta is debating whether to offer access to AI models hosted on its infrastructure or to sell bare-metal raw compute capacity to customers. The announcement sent META shares up 9% on the day of news (July 1) and drove CoreWeave and Nebius stocks down 12–17% on competitive concerns.

The strategic motivation: Meta has committed $125–$145 billion in capex for 2026 after spending $72.2 billion in 2025 (84% YoY growth), creating massive internal capacity. Selling excess compute reframes that spending as a hedge against overcapacity and a new revenue stream, rather than pure AI R&D bet. JPMorgan estimates each gigawatt of Meta compute capacity could generate $20 billion in annual revenue.

This mirrors SpaceX's May 2026 move, when xAI sold the full capacity of its Colossus 1 data center (300 MW, 200k+ H100s) to Anthropic at $1.25 billion/month. The pattern signals a shift: hyperscaler capex winners are now monetizing idle capacity to justify continued infrastructure spending to Wall Street.

For infrastructure practitioners: the emergence of hyperscaler-backed compute services (Meta Compute, xAI capacity, CoreWeave) creates a tiered market. Bare-metal providers can now compete on price against neoclouds and traditional clouds. The real test: whether Meta can build enterprise trust and SLAs as quickly as it engineered data centers.