Yann LeCun's AMI Labs bags $1B to build world models

Yann LeCun has spent the last few years publicly roasting large language models as a dead end on social media. Now, he has got a ten-figure war chest to prove it.

AMI Labs, the new AI venture co-founded by the Turing Award winner shortly after leaving Meta, has raised $1.03 billion in a blockbuster seed round at a $3.5 billion pre-money valuation. The massive check blows past earlier rumors of a €500 million target, instantly creating a European AI powerhouse.

A European superteam funded by global heavyweights

LeCun walked away from Meta's FAIR lab last November after a 12-year tenure, reportedly telling Mark Zuckerberg that he could build world models faster, cheaper, and better on his own.

He has set up shop in Paris instead of Silicon Valley, which he recently dismissed as thoroughly 'LLM-pilled.' The company will also maintain offices in New York, Montreal, and Singapore to source researchers and serve upcoming industrial clients in Asia.

The team reads like a frontier AI who's who. Serial entrepreneur Alexandre LeBrun serves as CEO, with former Meta VP for Europe Laurent Solly joining as COO. Research is spearheaded by chief science officer Saining Xie, chief research officer Pascale Fung, and world models VP Michael Rabbat.

Investors showed up in droves. Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions co-led the round, joined by Nvidia, Samsung, Sea, Temasek, Toyota Ventures, and French industrial giants including Dassault and Publicis. Prominent angels include Tim Berners-Lee, Eric Schmidt, and Mark Cuban.

Ditching text tokens for physical reality

AMI Labs is completely bypassing the text-prediction architecture of ChatGPT. Instead, it is betting on world models that learn how the real world operates through perception and physical cause-and-effect, aiming at robotics, manufacturing, and healthcare.

The research rests on JEPA (Joint Embedding Predictive Architecture), which LeCun introduced in 2022 to give AI persistent memory and physical reasoning.

CEO LeBrun was candid about the timeline, stating that AMI Labs is not an applied startup designed to spit out a product in three months and reach $10 million in ARR in a year. Commercial deployments could take several years, with LeBrun's digital health company Nabla lined up as the first deployment partner to tackle lethal LLM hallucinations. LeBrun also warned that 'world models' will be the next buzzword in venture capital, though AMI plans to stay distinct by open-sourcing significant code and publishing peer-reviewed papers.

Why it matters

This is the biggest validation yet that frontier AI cannot live on web text alone.

With both Yann LeCun's AMI Labs and Fei-Fei Li's World Labs landing $1 billion checks within weeks of each other, the industry is explicitly hedging against the scaling wall of generative LLMs. For robotics engineers, autonomous systems developers, and industrial manufacturers, pure language models have proven brittle and hallucination-prone in high-stakes environments. World models promise an actual understanding of physical causality.

Do not expect a consumer-ready playground or API anytime soon. AMI Labs is openly playing a long-term research game, which is refreshing, but it comes with immense capital risks. Compute and premier research talent burn billions rapidly, and unlike OpenAI or Anthropic, AMI will have zero meaningful software revenue in the near term. We think backing from physical industrial giants like Toyota and Dassault is a shrewd move that provides real testing grounds, but LeCun now faces the hardest test of his career: proving that JEPA works at scale before this ten-figure runway runs dry.