Meta employs tens of thousands of people. According to Mark Zuckerberg, the group actually building its frontier AI models could fit inside a mid-sized lecture hall.
In a recent video appearance, Zuckerberg said Meta’s frontier research team has about 100 people. The infrastructure team that keeps those researchers supplied with compute also has about 100 people.
A small team inside a big lab
The research group sits inside Meta Superintelligence Labs, or MSL. Meta launched the unit in mid-2025 to consolidate its scattered AI efforts under one roof.
MSL is led by Chief AI Officer Alexandr Wang, with Nat Friedman serving as co-lead. The broader organization has grown to around 3,400 employees.
The initial core superintelligence team numbered approximately 50 people. Zuckerberg’s latest figure of about 100 suggests that group has roughly doubled while staying within the size range he has long favored.
Back in September 2025, during a podcast appearance with Rowan Cheung, Zuckerberg laid out his thinking. He said the ideal team size for frontier AI work is somewhere between 50 and 100 researchers.
His reasoning centered on focus. A small group, he argued, lets researchers dig fully into hard problems without layers of bureaucracy or rigid deadlines getting in the way.
He compared the dynamic to a “group science project.”
Why Meta rebuilt its AI effort
MSL did not appear out of nowhere. It was a response to the mixed reception of Meta’s earlier Llama models, and Zuckerberg moved to pull its various AI teams together under the new structure.
The rebuild came with an aggressive hiring strategy. Meta courted top researchers from leading AI firms with exceptional compensation packages, and stated a goal of recruiting an additional 50 to 100 top-tier people.
Meta has since claimed meaningful progress. With its Muse Wonder release, the company asserted it had reached the “AI frontier” after heavy investment in talent and infrastructure throughout 2025.
The infrastructure half of the equation
Frontier AI research depends on enormous clusters of hardware, training pipelines that hold up over long runs, and systems that let researchers test ideas quickly.
Matching researchers and infrastructure engineers roughly one to one signals how Meta views the work. Meta has also said it plans continued compute enhancements as part of its effort to close the perceived gap with competing labs.
What this means
The structure raises a question about the other roughly 3,400 people in MSL. Their work presumably spans product integration, applied research, safety and deployment, all of which help turn frontier models into things Meta’s users actually touch.
Meta’s own recruiting playbook shows a specific vulnerability. If Meta can lure top talent away from other labs with big compensation packages, competitors can try the same move in reverse.
Does Meta hit its goal of adding another 50 to 100 elite researchers? Do independent benchmarks back up Meta’s claim that Muse Wonder put it at the frontier? Self-assessment in AI tends to be generous, and outside verification will determine whether the small-team strategy is a blueprint or just an expensive experiment.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.