After leaving Meta, Yann LeCun did what everyone was waiting for.
He founded AMI Labs, completed a $1.03 billion seed round at a $3.5 billion valuation — the largest seed round in European history — and declared: "I'm going to build an AI that doesn't work by predicting the next word."
Why LeCun Dislikes Large Language Models
This is not a new position. LeCun has been criticizing the LLM approach for years. His core argument is simple: text prediction alone is insufficient for genuine understanding.
No matter how much text you feed a model, it learns statistical patterns of language, not how the world actually works. A child who sees a glass fall off a table knows it will break without ever having read a word about gravity. LLMs cannot do this because they have no direct perception of the physical world.
LeCun calls this the grounding problem: models need contact with the real world, not just words.
AMI Labs is his practical answer to that problem.
What Is JEPA and How Is It Different from LLMs
The company's core technology is JEPA — Joint Embedding Predictive Architecture — a framework LeCun proposed in 2022.
The fundamental difference lies in what is being predicted:
- LLMs: predict the next token (word fragment)
- JEPA: predicts abstract state representations
It sounds abstract, but the practical difference is significant. LLMs treat every word as a prediction target, including low-information function words like "the" and "a," wasting capacity on noise. JEPA skips these and predicts at a more abstract level: if you perform this action, what state will the world be in?
That is why it is called a world model — it attempts to build an internal model of how the world works, not just learn surface-level language patterns.
Practical implications:
- Robotics: needs to predict action consequences; JEPA is a natural fit
- Medical diagnosis: requires causal reasoning about physical processes
- Industrial control: needs continuous state tracking
AMI Labs CEO Alexandre LeBrun made a pointed remark:
"I predict world models will soon become the next funding buzzword. Within six months, every company will call itself a world model company to raise money."
The irony was not lost on someone who just raised a billion-dollar seed round.
Company Setup
AMI Labs is headquartered in Paris but has teams in New York (where LeCun teaches at NYU), Montreal, and Singapore.
Core team:
- Executive Chairman: Yann LeCun (2018 Turing Award winner, former Meta Chief AI Scientist)
- CEO: Alexandre LeBrun (serial entrepreneur whose AI company was acquired by Facebook)
- CSO: Saining Xie (visual representation learning researcher)
The round was led by Bezos Expeditions, Cathay Innovation, Greycroft, Hiro Capital, and HV Capital. Bezos's involvement signals this is not just an academic exercise.
Notable Timing
This comes one month after Fei-Fei Li's World Labs raised $1 billion. Both companies are betting on world models, both are founded by top AI academics, and both secured staggering seed rounds.
This is no coincidence. Investors sense that the current LLM scaling path may be approaching a ceiling, and the next generation of AI may require a fundamentally different architecture.
LeCun's claim that text prediction is insufficient was once just an academic debate. Now it has a billion dollars behind it. The debate has become a bet.
Why This Won't Be Easy
JEPA has existed as a paper for four years, but its track record in real systems is limited. LLMs have shown visible benchmark progress over those four years; JEPA's progress has been relatively quiet.
LeCun's criticism is correct, but pointing out a problem and building a better solution are two different things.
Even if JEPA is the right direction, AMI Labs must compete with an industry that already has a massive head start. OpenAI, Anthropic, and Google are aware of LLM limitations and are moving toward multi-modal and agentic approaches — just not as radically as LeCun.
AMI Labs is valued at $3.5 billion with no product yet. This is belief financing, not product financing.
But if LeCun is right, the $1.03 billion entering now could be the decade's most valuable bet.
Sources: Yann LeCun's AMI Labs Raises $1.03B for World Models Development (TechCrunch); AMI Labs Secures $1B Seed Round Led by Yann LeCun (Crunchbase News); Yann LeCun's AMI Labs Raises $1B in Seed Round to Develop World Model AI Systems (The AI Insider); CocoLoop; [AINews] Yann LeCun's AMI Labs launches with a $1B seed @ $3.5B to build world models around JEPA (Latent Space)