Meta has quietly built a 50-strong “Superintelligence Labs” unit, and the roster reads like an AI all-star draft. Ex-Scale AI chief Alexandr Wang will captain the effort, joined by former GitHub CEO Nat Friedman and a parade of researchers lured from OpenAI, Anthropic, and Google DeepMind. Notables include Trapit Bansal, who pioneered reinforcement learning on chain-of-thought, Jack Rae, architect of DeepMind’s Gopher and Chinchilla models, and Huiwen Chang, the imaging brain behind GPT-4o’s picture generator. Their brief is simple: push Meta past today’s chatbots toward artificial super-intelligence.

The hiring spree hasn’t gone unnoticed. OpenAI’s chief research officer likened the talent drain to a “home burglary,” while insiders say Meta offered elite engineers carte blanche to chase massive multimodal models that reason, remember, and speak as fluently as humans. Zuckerberg’s internal memo rebranded the company’s entire foundation-model group as Meta Superintelligence Labs, signaling that every Llama update, product prototype, and FAIR experiment will now funnel into one moon-shot pipeline.

Why gamble billions on brains? Meta wants an advantage bigger than ad targeting. A truly general AI could power voice-first smart glasses, real-time language tutors, and photorealistic world-building tools for its metaverse—all sticky ecosystems that keep users (and developers) under Meta’s roof. With rivals racing to the same horizon, the question isn’t whether super-intelligence is coming; it’s whose name will be on the boot screen when it arrives.