Yann LeCun
Yann LeCun, one of artificial intelligence's most decorated researchers and Meta's former Chief AI Scientist, confirmed in December 2025 his departure from the company to launch Advanced Machine Intelligence (AMI) Labs, a startup seeking a valuation exceeding $5 billion before releasing any products.
The venture represents one of the most significant bets yet on "world models"—AI systems that understand physical reality through simulation rather than statistical language patterns—as an alternative pathway to artificial general intelligence. LeCun has argued publicly for years that scaling large language models will not achieve human-level intelligence, describing LLMs as "too limiting" and fundamentally constrained by their lack of embodied understanding.
AMI Labs is reportedly seeking to raise €500 million (approximately $586 million) at a €3 billion ($3.5 billion) valuation right out of the gate, according to Financial Times sources. Given the unprecedented capital flowing toward AI startups founded by prominent scientists, the target appears achievable. Former OpenAI CTO Mira Murati's Thinking Machines Lab raised at a $12 billion seed valuation in 2025, while Fei-Fei Li's World Labs became a unicorn immediately upon launching.
LeCun serves as AMI Labs' Executive Chairman rather than CEO. That role belongs to Alex LeBrun, previously co-founder and CEO at Nabla, a medical AI startup that has raised $120 million. LeBrun worked under LeCun's leadership at Meta's AI research laboratory FAIR after Facebook acquired his previous startup, Wit.ai. The duo will be joined by Laurent Solly, who stepped down as Meta's vice president for Europe in December 2025.
The startup's focus on world models builds on LeCun's I-JEPA (Image Joint Embedding Predictive Architecture) research developed at Meta. I-JEPA learns by predicting representations of image regions from other regions, developing abstract understanding of visual scenes without explicit labels—paralleling how humans develop intuitive physics through observation rather than instruction.
World models differ fundamentally from generative AI architectures dominating current development. Instead of predicting next tokens in text sequences, world models predict future states in physical reality by learning cause-and-effect relationships, spatial reasoning, and physics constraints from video and sensor data. The approach promises AI systems that understand how the world actually works rather than inferring physics from text patterns.
AMI Labs' first deployment will occur in healthcare through a partnership with Nabla. The initial product targets scheduling, documentation, and billing while maintaining context throughout patient workflows—capabilities where current AI systems struggle due to lack of persistent memory and causal reasoning. LeCun invested in Nabla previously, alongside Tony Fadell's Build Collective, HV Capital, Highland Europe, and Cathay Innovation.
The startup faces formidable competition. Google DeepMind released Genie 3 in August 2025—the first real-time interactive world model generating navigable 3D environments at 24 frames per second. Fei-Fei Li's World Labs launched Marble in January 2026, making world model generation commercially available with pricing from free to $95 monthly. General Intuition raised $134 million in October 2025 specifically for spatial reasoning capabilities.
LeCun is deliberately building AMI Labs with strong European presence centered in Paris, though he maintains his professor position at New York University teaching one class annually. He told MIT Technology Review that "Silicon Valley is completely hypnotized by generative models" and believes geographic distance from Bay Area groupthink enables genuine architectural innovation.
The contrarian bet challenges the scaling hypothesis driving hundreds of billions in LLM infrastructure investment. While OpenAI and Anthropic pour resources into larger language models, LeCun argues breakthroughs will come from how models learn rather than parameter counts. His departure from Meta after 12 years—five as founding director of Facebook AI Research and seven as chief AI scientist—signals deep conviction in this alternative path.
French President Emmanuel Macron publicly welcomed the news, expressing pride that LeCun, a Turing Prize winner, chose to headquartere AMI Labs in Paris as part of France's push to position itself as a competitive AI hub.
