Overview
- Bloomberg reported Friday that Meta is developing an internal unit, described as Meta Compute, to offer hosted AI models or raw GPU capacity to external customers.
- Bank of America estimated Meta could build about 19 gigawatts of AI capacity by 2030 and said commercializing half that capacity at $10–$15 billion per gigawatt could generate $100–$150 billion in revenue.
- The plan follows Meta’s April raise of 2026 capex guidance to $125–$145 billion, a sharp step up from $72.2 billion in 2025 that has raised investor concern about overbuilding.
- Analysts and investors reacted positively, with several firms reaffirming buy ratings and higher price targets, but experts warn monetizing spare compute faces margin, trust and execution challenges similar to early cloud rollouts at AWS and Google.
- A key next step to watch is whether Meta chooses to finance the build with equity or other means and whether internal demand from its own AI projects will absorb most capacity before external sales begin.