Yann LeCun's AMI Labs Raises $1.03 Billion to Build AI That Understands the Physical World
The Turing Award winner's new startup, backed by Nvidia, Samsung, and Bezos Expeditions, is betting that world models will surpass large language models for real-world applications.
Advanced Machine Intelligence Labs, the Paris-headquartered AI startup co-founded by Turing Award winner Yann LeCun following his departure from Meta, announced on Monday that it has closed a $1.03 billion seed round at a $3.5 billion pre-money valuation — one of the largest seed rounds in history and likely the largest ever raised by a European company. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with participation from Nvidia, Samsung, Temasek, Toyota Ventures, and a range of European institutional investors and family offices.
AMI Labs is building what LeCun calls 'world models' — AI systems that learn to understand physical reality through sensory data and camera observations rather than predicting the next token in a text sequence. The approach is grounded in LeCun's Joint Embedding Predictive Architecture, which he proposed in 2022 while still at Meta's AI research division.
The startup is focused on fundamental research first, with plans to publish papers and open-source code before pursuing commercial applications.
The company's first named partner is Nabla, a healthcare AI firm, which will provide real-world data and evaluation environments for AMI Labs' models. Additional teams and offices are planned in New York, Montreal, and Singapore alongside the Paris headquarters.
Alexandre LeBrun, AMI Labs CEO, told reporters that 'world models' are set to become the next major category in AI investment, predicting that the term would attract widespread use across the industry within months.
Individual investors in the round include web pioneer Tim Berners-Lee, venture capitalist Jim Breyer, investor Mark Cuban, former Google CEO Eric Schmidt, and French tech entrepreneur Xavier Niel, reflecting broad enthusiasm for LeCun's alternative vision of AI development. LeCun has been a long-standing critic of the large language model paradigm, arguing it is insufficient for achieving human-level intelligence without grounding in physical understanding.
The raise signals a broader industry bet that the next frontier of AI capability may lie not in scaling existing transformer-based language models further but in architectures designed to model the structure and dynamics of the world — a shift with major implications for robotics, autonomous vehicles, and industrial automation.
Read the original reporting at TechCrunch.
