Yann LeCun Raises $1.03 Billion to Build an AI That Understands the Real World

Four months ago, Yann LeCun left Meta.

This week, he came back with $1.03 billion and a plan to challenge how AI is built.

His new startup, Advanced Machine Intelligence Labs or AMI Labs, just closed the largest seed funding round a European company has ever raised. No product. No revenue. About twenty employees. And already valued at $3.5 billion.

Here is why it matters. Because the story behind the money is even more interesting than the number itself.

First, Who Exactly Is Yann LeCun?

If you have used face unlock on your phone today, you have used his work.

LeCun is the scientist who pioneered convolutional neural networks back in the 1980s. That research quietly became the engine behind facial recognition, Google image search, medical scanning technology, and self-driving cars. In 2018, he won the Turing Award, the Nobel Prize of computer science, alongside Geoffrey Hinton and Yoshua Bengio.

After that, he spent twelve years at Meta building FAIR, their world-class AI research lab, from the ground up.

He is not some random tech entrepreneur with a big idea. He is one of the three or four people most responsible for the AI revolution happening right now. Which is exactly why what he is saying and building matters so much.

$1 Billion Before Shipping a Single Thing

Here is how the fundraise actually happened.

When LeCun and his team started approaching investors in late 2025, they had a modest target in mind. Around €500 million. Enough to fund serious research for several years.

Then something unexpected happened. Investors started saying yes. Fast. And in large numbers.

Demand came in so far above expectations that AMI Labs had to start turning people away. They became selective, choosing investors based on who could actually help the company, not just who had the deepest pockets. By the time the round closed, the number had more than doubled. €890 million. Just over $1.03 billion.

Jeff Bezos is in. NVIDIA, the company making the chips that power essentially every major AI system on earth, is in. Samsung is in. Singapore’s government investment fund Temasek is in. The round was also co-led by Cathay Innovation, Greycroft, Hiro Capital, and HV Capital.

Then there are the individual investors. Tim Berners-Lee, the man who literally invented the World Wide Web, put money in. So did former Google CEO Eric Schmidt. So did Mark Cuban. So did French tech billionaire Xavier Niel.

When those people all back the same company at the same time, before it has built anything, it is worth paying attention.

So What Is Wrong With ChatGPT and Claude?

This is the part most tech articles get wrong. Let me explain it simply.

ChatGPT, Claude, Gemini. All of these are built on something called large language models. They learn by reading text. Billions and billions of pages of it. They get incredibly good at predicting what word comes next, which means they get incredibly good at writing, summarising, answering questions, and holding conversations.

But here is the problem. They only learn from text.

Think about how you learned that fire is hot. You did not read about it. You felt warmth. Maybe you got too close once. You learned it through direct physical experience. Before you could read a single sentence, you already understood gravity, cause and effect, and dozens of other fundamental things about reality because you lived in the world and experienced it firsthand.

LeCun has been making this argument for years. No matter how much text you feed these systems, he says, they will always be missing something essential. They can talk about the world beautifully. But they do not truly understand it.

And there is a very real consequence to that gap. Hallucinations. When language models make things up confidently and incorrectly, it is because they have no real world anchor to check themselves against. In a chatbot, that is annoying. In a hospital, it is dangerous.

The Big Idea: Teaching AI to Experience the World

AMI Labs is built around a completely different approach. It is called world models.

Instead of learning from text, world models learn from video, audio, and sensor data. Real footage. Real sounds. Real physical information from the actual world. The goal is an AI that does not just know what things are called but genuinely understands how they work.

Think of it this way. There is a difference between someone who read every book ever written about swimming and someone who has actually been in the water thousands of times. LeCun is trying to build the swimmer.

The technology behind this is called JEPA, which stands for Joint Embedding Predictive Architecture. LeCun developed it during his last years at Meta. JEPA does not try to reproduce the world in perfect detail. Instead it learns to build a simplified, meaningful internal picture of reality, filtering out the noise and focusing on what actually matters. It is closer to how a human brain works than anything the mainstream AI industry is currently building.

The Team Behind the Bet

LeCun is executive chairman. He is also keeping his professorship at New York University, a signal that serious academic research remains central to what AMI is doing.

Running the company day to day is CEO Alexandre LeBrun, a French entrepreneur who previously built Nabla, a medical AI company, and whose earlier startup was acquired by Facebook. He and LeCun worked together directly inside FAIR. They know each other well.

The rest of the leadership team is drawn almost entirely from Meta’s AI research organisation. Laurent Solly, Meta’s former VP for Europe, is Chief Operating Officer. Saining Xie, one of the most respected young computer vision researchers in the world and previously at Google DeepMind, is Chief Science Officer. Pascale Fung is Chief Research and Innovation Officer. Michael Rabbat, a former director of research science at Meta, leads as VP of World Models.

They chose Paris as their home base. Not San Francisco. LeCun has been direct about why. He believes Silicon Valley is too deeply committed to the language model approach to be the right environment for what AMI is trying to do.

Where Will This Technology Actually Be Used?

AMI Labs is not rushing to release a product. The first full year is dedicated entirely to research. LeBrun has said openly that real-world applications of world model technology could still take years to arrive.

When products do come, the focus areas are healthcare, robotics, and manufacturing. Their first confirmed partner is Nabla, where AMI plans to build AI diagnostic tools capable of meeting clinical regulatory standards.

And there is one more detail that raised eyebrows across the industry this week. LeCun revealed that Meta, his former employer, is already in early talks about deploying AMI’s world model technology inside Meta’s Ray-Ban smart glasses. The relationship, it turns out, did not end when he walked out the door.

What This All Means for the Future of AI

The man who helped create modern AI looked at what it became and decided it was not good enough. Not close enough to real intelligence. Not worthy of the name.

So he raised a billion dollars and got back to work.

OpenAI, Google, Anthropic and Meta are all betting on the same thing. Larger models. Smarter language. More data. Yann LeCun is betting they are all building toward a ceiling without realising it.

He has $1.03 billion, a world-class research team, and fifty years of work behind him. The next few years will show whether he is the most important scientist of this generation or the man who made the most expensive wrong turn in tech history.

The future of artificial intelligence is no longer just about language. It is about understanding the world itself. And that future started this week in Paris.