Meta CEO Mark Zuckerberg told employees at an internal town hall on July 2 that AI agent development over the prior four months “hasn’t really accelerated in the way that we expected.” The admission comes as Meta grapples with massive infrastructure bets and mixed returns on AI initiatives.

The Meta Compute Play

Just one day before Zuckerberg’s remarks, Meta announced Meta Compute, a new cloud business selling excess data center capacity to outside clients. The move revealed that Meta’s investments in AI infrastructure have outpaced its internal needs, leaving vast idle computing power.

This is unusual for Meta. The company spent billions on data centers betting that in-house AI development would consume all available resources. Instead, Zuckerberg’s admission suggests the company overestimated how fast its AI agent development would progress or how much compute those projects would actually require.

The Broader AI Slowdown Narrative

Zuckerberg’s statement feeds into a growing narrative that the AI boom may be overheating. Nvidia’s Kyber delay, Tesla’s delivery growth slowing, Samsung’s earnings beat by only 6%, and Meta’s own caution suggest that trillion-dollar bets on AI infrastructure may be betting on timelines that are slipping.

The irony is sharp: Meta is selling compute capacity because it doesn’t need it all, while other companies face shortages for their AI ambitions. This supply glut, combined with Zuckerberg’s candid admission about slower-than-expected progress, signals that the AI infrastructure race may be entering a consolidation phase.

Zuckerberg’s tone shift on AI is notable. After years of betting the company on AI, the Meta CEO sounded more cautious than celebratory about progress. That kind of honesty from a major tech leader ripples across markets and investor sentiment.