$1.03 billion. For a seed round. For a company that's less than six months old.
AMI Labs, the Paris-based startup founded by Turing Prize winner Yann LeCun after he left Meta late last year, just closed Europe's largest-ever seed round at a $3.5 billion pre-money valuation. The investor list reads like a who's-who of global capital: Nvidia, Jeff Bezos' Bezos Expeditions, Singapore sovereign fund Temasek, plus European VCs Cathay Innovation, Daphni, HV Capital, and Hiro Capital.
The money is enormous. The thesis behind it is even more interesting.
A Billion-Dollar Bet Against LLMs
LeCun has spent years publicly arguing that large language models (the technology behind ChatGPT, Claude, and Gemini) are a dead end for reaching truly intelligent AI. His position: predicting the next word in a sentence will never produce a system that actually understands how the real world works.
AMI Labs is his attempt to prove it. The company is building what it calls "world models" - AI systems designed to understand physical environments rather than process text. Think of it this way: an LLM can write a paragraph about a ball rolling down a hill, but a world model would actually understand gravity, friction, and momentum well enough to predict where the ball ends up.
The initial target applications are robotics, manufacturing, and wearables - domains where understanding physics matters more than generating prose.
The Team LeCun Assembled
The company isn't just LeCun's name on a pitch deck. He's brought serious research talent:
- CEO: Alexandre LeBrun, former CEO of French health AI startup Nabla
- VP World Models: Mike Rabbat, formerly at Meta's AI research lab
- Chief Science Officer: Saining Xie, previously at Google DeepMind
- Chief Research & Innovation Officer: Pascale Fung, also ex-Meta
LeCun himself holds the executive chairman title. The team is spreading across four offices: Paris (headquarters), New York (where LeCun teaches at NYU), Montreal, and Singapore.
What This Means for the AI Tool Landscape
For anyone using AI productivity tools today, world models won't change your workflow tomorrow. This is deep infrastructure research, not a product launch. The practical applications - robots that can navigate warehouses, manufacturing systems that can predict equipment failures, AR wearables that understand your physical environment - are likely years away from consumer products.
But the funding signals something worth paying attention to: serious money is starting to flow toward AI approaches that go beyond text generation. This is the second-largest seed round worldwide, trailing only Thinking Machines Lab's $2 billion raise in June 2025.
The question isn't whether world models will work. It's whether LeCun - who co-invented the convolutional neural networks that power most image recognition today - is right that LLMs have a ceiling. A billion dollars says at least some very wealthy people think he might be.