Yann LeCun’s AMI Labs Raises $1.03B to Pursue World Models for Real-World AI
Yann LeCun, the Turing Award-winning AI pioneer, has launched Advanced Machine Intelligence (AMI) Labs with more than $1 billion in seed funding. The Paris-based startup announced a $1.03 billion raise at a $3.5 billion pre-money valuation, marking one of the largest early-stage rounds on record in Europe.
A Different Path from LLMs
LeCun has long argued that today’s dominant large language models, which rely on next-token prediction, are insufficient for achieving robust intelligence. Instead, AMI focuses on building world models—systems that develop abstract representations of the physical world from sensor data like video and other inputs. These models predict in embedding space rather than generating every pixel or detail, making them better suited to noisy, continuous real-world environments.
JEPA and Technical Approach
The work builds directly on LeCun’s Joint Embedding Predictive Architecture (JEPA), introduced in 2022. By learning to predict representations rather than raw observations, the approach avoids the pitfalls of forcing models to hallucinate unpredictable details. Action-conditioned variants could let agents simulate the outcomes of their decisions, supporting better planning and reasoning while incorporating safety constraints.
Team, Investors and Plans
CEO Alexandre LeBrun, who previously led digital health startup Nabla and worked at Meta’s FAIR lab, leads the company alongside LeCun. Senior team members include researchers Saining Xie, Mike Rabbat, Pascale Fung, and others with deep academic and industry experience. The investor roster is impressive: co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with participation from Nvidia, Samsung, Toyota Ventures, Temasek, Eric Schmidt, Mark Cuban, Tim Berners-Lee, and numerous European and Asian funds. AMI operates out of Paris, New York, Montreal, and Singapore. The company plans to publish research openly, release substantial code as open source, and partner with industry on applications in robotics, healthcare, industrial automation, and wearables. First healthcare collaboration is with Nabla; expectations are that meaningful products remain years away while fundamental research advances.
Why the Timing Matters
This substantial bet arrives as interest in physical AI, robotics, and embodied intelligence grows across the industry. It echoes earlier moments when new architectural ideas gained traction after periods of scaling existing techniques. LeCun’s persistence with this direction over more than a decade now has significant capital and talent behind it. Success is not guaranteed—world models must still prove superior in real-world metrics—but the effort reflects a thoughtful attempt to address fundamental limitations in current generative approaches. The coming years will reveal whether these systems can move AI closer to reliable interaction with the physical world.
