Overview
- Bloomberg and follow-up reports say Meta is building an internal unit called Meta Compute to either host its models for external API use or rent raw GPU and compute capacity to third parties, though the company has not announced a product, pricing or launch date.
- The plan is being driven by senior leaders including Santosh Janardhan, Daniel Gross and Dina Powell McCormick and contemplates two clear offers: hosted access to Meta models such as Muse Spark or wholesale access to raw compute on a usage basis.
- Meta sharply raised 2026 capital spending to a $125 billion–$145 billion range and disclosed roughly $107 billion in new infrastructure purchase commitments, a buildout the company says could support third-party demand.
- The market reacted with high volatility after the cloud reports, with shares jumping about 9% on July 1 and then pulling back, and CEO Mark Zuckerberg told staff in an internal town hall that AI agent development has not progressed as expected over the prior months.
- Key uncertainties remain because Meta has not confirmed how much capacity is truly surplus to its own AI roadmap, enterprise sales and service structures would be required to win customers, and investors will look for concrete updates in Meta’s July 29 earnings report.