Overview
- Meta, which multiple outlets reported on Wednesday to be developing an internal unit called Meta Compute, is weighing two options: offering API access to hosted models or renting raw GPU and compute capacity to third parties.
- The report is unconfirmed by Meta beyond CEO Mark Zuckerberg’s May remark that selling excess compute is “definitely on the table,” and the project remains in development without a product launch or commercial terms.
- Investors cheered the news as Meta shares jumped roughly 9–10% on the report while specialised GPU cloud providers such as CoreWeave and Nebius and some South Korean chipmakers dropped on fears of increased supply.
- Meta’s move would target AI‑centric workloads rather than a full general‑purpose cloud and would require new enterprise sales, software and trust credentials that differ from Meta’s ad business and typically carry lower margins.
- The plan is rooted in an exceptionally large 2026 capex program for data centers and chips (company guidance cited in coverage ranges up to about $145 billion) and could help monetize surplus infrastructure if Meta overbuilds capacity.