Reorganizing Meta AI Labs: The Structural Turn to Superintelligence (AI column)

I recently read a note by Alexander Wang.Superintelligence is coming", from which it appears that Meta (banned organization in Russia) The largest reorganization of its AI divisions in the company’s history, bringing all AI teams together under the new Meta Superintelligence Labs structure. (MSL). This fourth reorganization in the past six months aims to accelerate the achievement of superintelligence. The new scheme becomes centralized, where direct control is concentrated in the 28Alexander Wang, former CEO of Scale AI. He was brought in by Meta as part of a deal worth $14,3 He now leads MSL as Chief AI Officer. (And remember, joker before). If we talk about the key areas, there are four and here they are.

1) TBD LabDirected by Wang, Wang focuses on learning and scaling large models and developing an “omni” model, whatever that means. In fact, from this lab, we are waiting for new LLama models that will be better than the failed four. 2) FAIRDirected by Rob Fergus + Yann LeCun (Chief Scientist)who are now reporting to Wang. This lab is engaged in basic research and will be an innovative MSL engine. While previously there was relatively little interaction between FAIR and Meta’s generative AI division, FAIR’s research will now be directly integrated into TBD Lab’s large-scale model launches. 3) Products & Applied Research, which is run by Nat Friedman in the past from GitHub, who also reports to Wang. The goal here is simple - to integrate AI into Meta products, as well as conduct applied research. The team brings together previously disparate groups working on Assistant, Voice, Media, Trust, Embodiment and Developer pillars. This provides a closer link between product research and product development. 4) MSL Infra, which is led by Aparna Ramani, who also reports to Vangu, and also develops information for AI projects.

I wonder what. Now there are two main scientists in this structure. (chief scientists)Yann LeCun (FAIR) Shenjia Zhao (common). LeCun maintains a focus on long-term research and "building the next AI paradigms," while Zhao, co-author of ChatGPT, sets the research agenda for the entire lab. As part of the reorganization, Meta liquidated the AGI Foundations team, created in May. 2025 years. This is the second major AI team disbanded by the company this year, having previously been eliminated by the GenAI division after a tepid reception by Llama. 4. AGI Foundations employees are distributed across product, infrastructure and FAIR teams, but not within the TBD Lab.

More broadly, this reorganization is a response to Meta’s lagging in the competitive race. Persistent reorganizations contrast with the stability of competitors who face less organizational upheaval To strengthen the company actively poached talent, offering compensation packages up to $100 millions, and in one case up to $1,5 billion The company has attracted more 50 Researchers and engineers from competitors, but this campaign ended with a hiring freeze in the AI division in August. 2025 It signals the need for consolidation after a period of intense expansion.

Meta’s approach is built on offering high-quality models for free, seeking to commoditize the layer of foundational models and undermining the business models of competitors who monetize access to their models through APIs. After all, the main monetization of Meta itself occurs through increased user engagement and data generated on its platforms. It will be interesting to see how their strategy works and what MSL will give birth to. 4 The llama was weaker than everyone expected from such investments.

#AI #ML #Software #Economics #Engineering #Management #Leadership #Future