Meta splits its superintelligence labs into four units amid a strategy reboot

Meta is breaking up its Superintelligence Labs into four groups—research, superintelligence, products, and infrastructure—in a bid to accelerate development and sharpen accountability across its artificial intelligence stack. Leadership told staff the move is meant to streamline operations and focus after months of churn over the company's approach to advanced models. Some executives are likely to leave, and the division may see further downsizing or reassignments.

In June, Meta invested billions in Scale AI and hired its chief executive, Alexandr Wang, to lead the superintelligence push. Reports put the investment at about $14.3 billion, with up to $72 billion in capital spending this year for data centers and hiring.

Leaders are rethinking the model strategy. Internally, leaders have explored licensing third-party systems and debated whether the next flagship model should be "closed," a break from Meta's open-source philosophy. The company has abandoned a previous frontier model, Behemoth, after disappointing tests, and started fresh under Wang's team. Hiring has surged, with offers over $100 million, while also stirring some friction between new arrivals and long-time researchers.

Nat Friedman, former CEO of GitHub, and Daniel Gross, former leader at the startup Safe Superintelligence, will steer new product features and applied research. FAIR co-founder Rob Fergus will continue to lead core research. Joelle Pineau, formerly a computer scientist at Meta, and Angela Fan, formerly a research scientist at Meta, left for Cohere and OpenAI, respectively. Loredana Crisan, former vice president of generative AI, is joining Figma as chief design officer. Ahmad Al-Dahle and Amir Frenkel now report to Wang on strategic artificial intelligence work.

This reorganization follows stalled progress and a renewed push toward superintelligence. Leaders call it the last big structural change for now, meant to speed decisions, cut bloat, and ship products faster, even as spending stays high.