Overview
- Reports in early July say Meta is planning an internal cloud called Meta Compute to sell unused GPU and AI capacity or host its models for outside customers.
- The company is reportedly weighing two product paths: API access to its closed‑weight models like Muse Spark or leasing raw compute on a usage basis.
- Meta’s stock jumped about 9% after the initial reports and has since shown volatility as investors weigh whether the move justifies its unusually large capex program.
- Internally, Meta has limited some engineers’ use of third‑party models and CEO Mark Zuckerberg told staff AI agent work has not accelerated as expected, signaling operational caution.
- Execution risks remain because Meta has not set a public timeline or pricing, enterprise cloud sales require new guarantees and support, and the company continues major capacity deals such as its extension with CoreWeave.