Meta’s Superintelligence Labs and the Fight for AI Researchers

Mark Zuckerberg has been recruiting AI researchers from rival companies, especially OpenAI, and a reported internal memo gives us a clearer picture of the new Meta Superintelligence Labs. I want to walk through how Meta got here, how OpenAI has reacted, and who is joining the effort.

A few months earlier, Meta released Llama 4 Scout and Maverick, while its larger Behemoth model was still in training. I thought the released models were good, but good may not be enough for Meta to compete at the frontier. Zuckerberg reportedly began personally pursuing leading researchers with exceptionally large offers.

In an excerpt, OpenAI CEO Sam Altman says Meta made offers to people on his team including $100 million signing bonuses. That is Altman’s characterization, not independent confirmation of each offer: Meta’s CTO later disputed the signing-bonus description, saying the largest packages were structured differently ([TechCrunch](https://techcrunch.com/2025/06/27/meta-is-offering-multimillion-dollar-pay-for-ai-researchers-but-not-100m-signing-bonuses/)).

Meta also invested approximately $14.3 billion for a 49% minority stake in Scale AI, which supplies data services including labeling for AI development. Scale founder and CEO Alexandr Wang joined Meta to lead the superintelligence effort. The transaction was not an outright purchase of Scale; the company said it would remain independent ([Scale AI](https://scale.com/blog/scale-ai-announces-next-phase-of-company-evolution)).

I suspect speed and the regulatory review associated with a full acquisition helped explain the structure, but those motives are my interpretation. I also suggest that rival customers might pull away from Scale because Meta is a competitor. Reports did describe customer concerns and changes in plans, but that does not establish that every OpenAI or Google contract was cancelled or that Scale’s business would die.

I claim that Meta got all of Scale’s data in the transaction. That claim is not established: Scale stated that Meta would not get access to its internal systems or customers’ confidential information, and that Wang would continue to be bound by confidentiality obligations as a Scale board member ([Scale’s customer statement](https://scale.com/blog/customer-trust-scale-meta-deal)).

Altman had initially said OpenAI’s best people were not accepting Meta’s offers, but subsequent reporting described several departures. I discuss researchers including Shengjia Zhao, Jiahui Yu, Hongyu Ren and Shuchao Bi, whose earlier work covered reasoning models, perception and multimodal systems. The video describes the Zurich office as entirely recruited; the reports cited here do not establish that the entire office moved.

OpenAI chief research officer Mark Chen addressed the departures in a reported internal memo. According to WIRED, he compared the feeling to a break-in and said OpenAI was working to retain researchers, including by recalibrating compensation and finding ways to reward leading talent ([WIRED](https://www.wired.com/story/openai-meta-leadership-talent-rivalry/)).

Chen also stressed fairness to other employees while saying he would fight to keep people. I see a financial imbalance: Meta is a large public company with substantial resources, while OpenAI was still private. That does not by itself establish what either company could or would pay an individual researcher.

OpenAI leadership also warned that fast-expiring offers could pressure employees during a scheduled week off. The advice in the reported messages was to ask for more time and reach out to leadership rather than make a major career decision in isolation. The alleged pressure is OpenAI’s account of the recruiting tactics.

Chen urged colleagues to keep their attention on turning more compute into intelligence, treating the contest with Meta as a side quest. I think that focus on superintelligence captures why the battle for researchers has become so intense, even as labs continue to ship more incremental products.

Zuckerberg’s memo, obtained by WIRED, called the overall organization Meta Superintelligence Labs. It included Meta’s foundation model, product and FAIR teams and a new lab focused on its next generation of models. I wonder whether Yann LeCun’s role might have been deprioritized, but the memo’s description of FAIR does not establish that claim.

The memo identified Wang as Meta’s chief AI officer and MSL leader. Former GitHub CEO Nat Friedman would co-lead the new lab with Wang, focusing on AI products and applied research. The announced group brought together people with experience at several competing labs.

The reported list includes Trapit Bansal, associated with reinforcement learning and OpenAI’s o-series, and Shuchao Bi, who worked on GPT-4o voice and multimodal post-training. Huiwen Chang worked on GPT-4o image generation, while Ji Lin contributed to OpenAI models and the Operator reasoning stack. These are the backgrounds described in Zuckerberg’s reported memo, not evidence that any one person built a product alone.

Other hires in the memo include Joel Pobar, who worked on inference at Anthropic; Jack Rae, a Gemini pre-training and reasoning lead; Hongyu Ren, who worked on several GPT and o-series models; and Johan Schalkwyk, a former Google Fellow and contributor to Sesame’s Maya. Pei Sun worked on Gemini post-training and reasoning, while Jiahui Yu and Shengjia Zhao brought experience across OpenAI’s multimodal and reasoning models ([WIRED’s list](https://www.wired.com/story/mark-zuckerberg-welcomes-superintelligence-team/)).

Taken together, I see an unusually strong collection of researchers aimed at the goal Zuckerberg calls superintelligence. The personnel list shows Meta’s ambition and spending power; whether the new organization can deliver a better model is a separate question that remains to be seen.