Meta's billion-dollar bet on talent finally pays off
The new model, Muse Spark, reveals the first dividends from the biggest hiring spree in AI history
META SUPERINTELLIGENCE LABS unveiled its first major model this week, capping an 18-month hiring blitz that poached talent from OpenAI, Google DeepMind, and Anthropic with compensation packages reaching $300 million over four years. Muse Spark represents the initial output from what amounts to one of the most expensive brain trusts ever assembled in artificial intelligence.
The numbers behind Meta's talent acquisition are formidable. The company hired over a dozen elite researchers, including Tim Brooks from OpenAI's Sora team, Jack Rae from Google DeepMind's pre-training efforts, and Trapit Bansal from OpenAI's reasoning models—offering packages that industry sources say averaged $100 million per hire. Meta simultaneously committed $72 billion to AI infrastructure in 2025, including the $27 billion Hyperion data center in Louisiana that will span the footprint of Manhattan when complete. This represents the largest coordinated investment in AI talent and infrastructure by any single company.
Money talks, talent walks
Yet this "infrastructure first" approach reveals Meta's fundamental challenge in the AI race. While OpenAI and Anthropic can focus entirely on advancing model capabilities, Meta must build AI that serves multiple masters: advertising optimization, content moderation, recommendation systems, and consumer features across WhatsApp's 2.8 billion users, Instagram's 2 billion, and Facebook's 3 billion.
This article is for Vector members. Start a 7-day free trial to keep reading.
Start your free trial
