News

Meta AI: 7 Zuckerberg moves reshaping the division

Article Highlights:
  • Meta AI split into four units: research, superintelligence, products, infrastructure
  • Aim: speed up frontier models and product rollout
  • FAIR to continue open-source fundamental research
  • Superintelligence team may pursue closed-source models
  • Possible staff reductions or internal reassignments
  • Heavy investment planned in data centers and hardware
  • Several senior researchers have left in 2025
  • Meta exploring third-party open and licensed models
  • Tensions between legacy teams and new hires
  • Reorg impacts product roadmaps and ad technology
Meta AI: 7 Zuckerberg moves reshaping the division

Introduction

Meta AI has been reorganized internally into four distinct groups: research, superintelligence, products, and infrastructure. This restructure aims to speed up frontier-model development, product delivery, and hardware scaling while addressing internal tensions.

Background

Mark Zuckerberg pushed a string of changes to keep Meta competitive in A.I., investing heavily in talent and infrastructure. Meta made large bets such as investing in Scale AI and hiring leaders from competitors; the A.I. organization grew to thousands and now requires clearer structure to be effective.

The four new units

The Meta Superintelligence Labs will split into:

  • Fundamental research (FAIR) maintaining open research and publication.
  • A superintelligence/frontier-model team focused on the most powerful A.I. systems.
  • Products and applied research for feature development.
  • Infrastructure responsible for data centers and A.I. hardware.

Short definition

Meta AI: Meta’s collective research, models and infrastructure used to build advanced artificial intelligence.

The problem / challenge

Meta faced performance setbacks (e.g., the Behemoth project) and internal friction between legacy staff and newly hired teams. Moves toward closed models mark a shift from Meta’s prior open-source stance and raise cultural and strategic questions.

Approach / solution

Splitting teams aims to preserve open scientific work within FAIR while creating a controlled pathway for a potential superintelligence project, accelerating product integration and optimizing infrastructure spending.

Concrete steps under way

  • Leadership changes: Alexandr Wang as chief A.I. officer, Shengjia Zhao as chief A.I. scientist; Nat Friedman and Daniel Gross to lead product features.
  • Potential downsizing or reassignment of roles as the company refines team sizes.
  • Exploration of third-party models, including open-source and licensed closed-source options.

Implications for products, talent and infrastructure

The reorganization should enable faster product deployment and justify high capital expenditures for data centers, but it carries risks: talent departures (already seen with Joelle Pineau and Angela Fan), internal friction, and strategic trade-offs on openness.

Conclusion

Meta AI’s restructure is a pragmatic response to a high-stakes A.I. race. It balances openness with control and aims to focus resources, yet must manage personnel, cultural shifts and technical risk to succeed.

 

FAQ

  • What is Meta AI and why did Zuckerberg reorganize the teams?

    Meta AI covers Meta’s research, models and infrastructure; the reorganization aims for clearer focus on research, products, superintelligence and infrastructure.

  • Which four areas now compose Meta AI?

    Fundamental research (FAIR), a superintelligence/frontier team, products/applied research, and infrastructure (data centers and hardware).

  • Will Meta AI use third-party models?

    Yes, Meta is exploring using third-party models, both open-source and licensed closed-source, to power some products.

  • How will the restructure affect Meta AI staff?

    Meta anticipates reorganizations that could include role eliminations, reassignments, or executives leaving as teams are consolidated.

  • Why was Behemoth abandoned in Meta AI’s roadmap?

    Behemoth’s release was delayed after disappointing performance tests and Meta decided to start a new frontier-model effort.

Introduction Meta AI has been reorganized internally into four distinct groups: research, superintelligence, products, and infrastructure. This restructure Evol Magazine