Zuckerberg explains why he chose Alexandr Wang to rebuild Meta’s AI team

Mark Zuckerberg chose Alexandr Wang to rebuild Meta’s AI operation for his judgment and ability to recruit researchers—not a record of training its most advanced models. An October 17 Fortune report recounts their discussions and Wang’s memo, revealing how a small-team strategy won Zuckerberg’s backing amid dissatisfaction with Meta’s existing direction.

A recruiting mandate, not a research résumé

The account draws on Zuckerberg and Wang’s appearance on Cheeky Pint, the podcast hosted by Stripe cofounder John Collison. They described a partnership that began with a WhatsApp message in April 2025, when Zuckerberg was seeking advice after the reception to Meta’s Llama 4 models.

Zuckerberg recalled that their first call lasted about 44 minutes. Two days later, Wang sent a long memo explaining what he would do in Zuckerberg’s position. Its first principle was talent density: the quality of the people mattered more than having a large team. Another was a willingness to start again.

That proposal met a receptive audience. Zuckerberg said he was unhappy with Llama 4’s trajectory and prepared to rebuild the team. His requirement was someone who could attract leading researchers and persuade them to leave other labs. Wang had built Scale AI into a data-labeling business, but Collison noted that he had never trained a frontier model—one competing at the leading edge of AI capability.

An expensive campaign for a small core

Zuckerberg said he had spoken with lab leaders and research chiefs before deciding Wang was the right choice. He described Wang as the sharpest entrepreneur of his generation among those he had met. That was Zuckerberg’s assessment of a prospective leader, rather than a claim about Wang’s technical training.

Securing Wang and assembling the team required substantial money and executive attention. Meta paid $14.3 billion for a 49% stake in Scale AI to bring him aboard. Fortune also cited reported $100 million compensation offers to OpenAI researchers. Zuckerberg said he cleared his calendar for two months during the recruiting push.

The resulting research group is deliberately compact. Wang said Meta Superintelligence Labs’ core has fewer than 200 people, closer to 100, supported by larger infrastructure and product teams. He described cultivating “insecure overachievers” who under-promise and over-deliver. The headcount refers to that research core, not everyone supporting Meta’s AI work.

The financial pressure behind the rebuild

A small research staff does not mean a small spending commitment. Meta expects capital expenditures of $130 billion to $145 billion in 2026, compared with $72.2 billion in 2025. Those are company capital-spending figures, distinct from the size of Wang’s core team and the compensation offers used to recruit researchers.

Wang also recounted repeated warnings from Meta CFO Susan Li at an executive gathering in Hawaii late last year. In his telling, Li stressed that a bad model would have severe financial consequences. Her message made model quality a financial requirement, not simply a research ambition.

Wang, meanwhile, credited Zuckerberg with an unusual willingness to change direction once convinced of a new approach. His description emphasized both the speed and scale of those pivots. It offers the leadership counterpart to the memo: Wang proposed a reset, and Zuckerberg was willing to commit resources to carry it out.