Yann LeCun Told Meta He Could Do It Faster Alone. Then He Raised $1 Billion.

Sometime in November 2025, Yann LeCun walked into Mark Zuckerberg's office and told him he was leaving. LeCun had been Meta's Chief AI Scientist for twelve years, first as founding director of Facebook AI Research, then as the company's most prominent scientific voice. He had won the Turing Award. He had built one of the most influential AI labs in the world.

And for nearly all of those twelve years, he had been arguing that the technology the entire AI industry was pouring hundreds of billions of dollars into was fundamentally wrong.

"I told him I can do this faster, cheaper, and better outside of Meta," LeCun later recalled. Zuckerberg's response: "OK, we can work together."

Four months later, on March 10, 2026, LeCun's new company, Advanced Machine Intelligence Labs, announced it had raised $1.03 billion in seed funding at a $3.5 billion pre-money valuation. It is the largest seed round ever raised by a European startup, and the second-largest worldwide, behind only Mira Murati's Thinking Machines Lab.

The 65-year-old French-American scientist is not building a better chatbot. He is not building a bigger large language model. He is building what he calls a "world model": an AI that learns from reality, not language. An AI that understands physics, maintains memory, and plans actions rather than guessing the next token in a sequence.

If he is right, every major AI lab on the planet is building the wrong thing.

The Twelve-Year Argument

LeCun's departure from Meta was not sudden. It was the conclusion of a long, increasingly public disagreement about the future of artificial intelligence.

While OpenAI scaled GPT from version 2 to 5.4, while Anthropic built Claude, while Google poured resources into Gemini, LeCun kept saying the same thing: large language models cannot lead to human-level intelligence. They are statistical pattern matchers that predict words. They do not understand the world.

"An LLM doesn't understand that if you push a glass off a table, it will break," LeCun explained. "It only knows that the words 'glass' and 'break' often appear together in that context."

He was more blunt in a November 2025 public lecture: "The path to superintelligence via LLMs is complete bullshit. It's just never going to work."

Inside Meta, the tension had been building for years. LeCun championed open-source research and long-horizon fundamental science. But Meta's AI division was reorganizing around Ahmad Al-Dahle and a new leadership team that favored a more closed, product-driven approach. In October 2025, Meta laid off 600 employees from its Superintelligence Labs division, including researchers from FAIR, the lab LeCun had founded.

When asked about the organizational dynamics, LeCun told the Financial Times: "You don't tell a researcher what to do. You certainly don't tell a researcher like me what to do."

A month later, he was gone.

What AMI Labs Is Actually Building

AMI stands for Advanced Machine Intelligence. The name is also the French word for "friend," a nod to the company's Paris headquarters.

The technical foundation is JEPA: the Joint Embedding Predictive Architecture, a framework LeCun first proposed in a 2022 paper and developed further at Meta through projects called I-JEPA and V-JEPA.

The concept, stripped to its essentials, works like this: instead of predicting the next word (like ChatGPT) or the next pixel (like a video generator), JEPA predicts the next abstract representation of reality. It watches video, sensor data, and spatial information, then learns the underlying rules of how the physical world behaves. Not by memorizing specific images. By building an internal model of cause and effect.

Think of it this way. A toddler who has never heard the word "gravity" still knows that a ball dropped from a table will fall to the floor. The toddler did not learn this from reading about physics. She learned it from watching the world. JEPA aims to give machines the same kind of learning: understanding built from observation, not from text.

The key technical distinction from generative AI is that JEPA does not try to reconstruct every detail of what it observes. It learns to predict in an abstract "representation space," ignoring unpredictable details and focusing on the patterns that matter. This makes it dramatically more efficient than approaches that try to generate pixel-perfect predictions of every future frame.

LeCun describes the goal as building "an abstract digital twin of reality that an AI can use to understand the world, predict the consequences of its actions, and plan accordingly."

The applications AMI is targeting reveal just how different this approach is from the chatbot race: robotics, industrial process control, autonomous systems, wearable devices, and healthcare. These are domains where hallucinations are not just embarrassing. They are dangerous.

The Investors Who Bet Against LLMs

The investor list reads like a who's who of global technology and finance.

Five firms co-led the round: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, the personal investment vehicle of Amazon founder Jeff Bezos. Strategic investors include NVIDIA, Toyota Ventures, Samsung, and Singapore's sovereign wealth fund Temasek. Individual backers include Tim and Rosemary Berners-Lee (the inventor of the World Wide Web), venture capitalist Jim Breyer, entrepreneur Mark Cuban, former Google CEO Eric Schmidt, and French telecom billionaire Xavier Niel.

NVIDIA's presence is particularly telling. Jensen Huang's company makes the chips that power every major LLM. By backing AMI, NVIDIA is hedging: if the world moves beyond language models, NVIDIA wants to supply the compute for whatever comes next.

Denis Barrier, co-founder and CEO of Cathay Innovation, said the investment was about more than financial return: "World models represent a fundamental shift in how AI understands and interacts with reality: systems that reason about cause and effect, enabling intelligence to work in the physical world at scale."

The Team LeCun Assembled

LeCun serves as executive chairman. He remains a professor at New York University and will not run daily operations. That job belongs to Alexandre LeBrun, a French entrepreneur who previously founded Nabla, a medical AI startup where he reached the same conclusion as LeCun about the limitations of LLMs: in healthcare, hallucinations can kill.

The rest of the leadership team is stacked with researchers who left top positions to join:

The company currently has about 12 employees spread across four offices: Paris (headquarters), New York, Montreal, and Singapore. LeBrun told TechCrunch that he would prioritize quality over quantity in hiring, choosing locations for both talent access and proximity to future customers.

LeBrun offered a candid prediction about the hype cycle his company is about to trigger: "My prediction is that 'world models' will be the next buzzword. In six months, every company will call itself a world model company to raise funding."

How It Unfolded

The Other Side: Why the LLM Industry Is Not Worried

Not everyone agrees that LeCun is onto something. In fact, the three people running the world's most valuable AI companies think he is wrong.

Dario Amodei, CEO of Anthropic, told an audience at Davos in January 2026 that AI models built on the current architecture would replace the work of all software developers within a year and reach "Nobel-level" scientific research within two. That is not the language of someone who considers his approach a dead end.

Sam Altman, CEO of OpenAI, has gone further, claiming that we are already slipping past human-level artificial general intelligence toward "superintelligence." OpenAI's $110 billion funding round in February 2026, backed by Amazon, NVIDIA, and SoftBank, represents the single largest private investment in any company in history. That money is buying bigger LLMs, not world models.

Demis Hassabis, CEO of Google DeepMind, shot back at LeCun with unusual directness on X, calling him "just plain incorrect" when LeCun dismissed the concept of general intelligence. Though notably, Hassabis has also said frontier models still lack scientific understanding and has called for renewed research into world models, planning, and memory. Even DeepMind, it seems, is hedging.

The pragmatic counterargument is simpler: LLMs work. ChatGPT has over 400 million weekly users. Claude powers classified military intelligence systems. Gemini is embedded across Google's product suite. Reasoning models built on top of LLMs are achieving expert-level performance on mathematics, coding, and scientific benchmarks. Whatever their theoretical limitations, the current generation of language models is generating hundreds of billions of dollars in revenue and producing measurable results.

LeCun's response to this argument has been consistent for years: "The fact that something works doesn't mean it's the right path. Horses worked. That didn't mean we shouldn't have built cars."

The Europe Question

AMI Labs is headquartered in Paris, not San Francisco. That detail matters more than it might appear.

Europe has struggled for years to produce an AI company that can compete at the frontier with American and Chinese labs. Mistral AI came closest, raising $2.2 billion and building competitive open-source models, but it has not matched the scale of OpenAI, Anthropic, or Google DeepMind. The European AI ecosystem has long suffered from a funding gap, a talent drain to Silicon Valley, and regulatory uncertainty.

AMI's $1.03 billion round is, by a wide margin, the largest seed financing any European startup has ever raised. The symbolism is hard to miss: the continent's biggest AI bet is not on catching up with American LLMs. It is on making them obsolete.

LeCun's dual citizenship and his insistence on keeping AMI's headquarters in Paris, with offices in New York, Montreal, and Singapore, signal a deliberate strategy. Build globally, but anchor the company in a European city where the talent pool is deep, the cost of living is lower than the Bay Area, and the regulatory environment, while complex, is at least predictable.

Whether AMI becomes the first European AI company to rival the American giants or the most expensive science experiment in French startup history will depend on a single question: is Yann LeCun right about world models?

The Bottom Line

A 65-year-old scientist who spent twelve years at one of the world's largest technology companies just walked away, raised a billion dollars in four months, and told the entire AI industry that the thing they are all building is wrong.

That is either the most consequential bet in the history of artificial intelligence or the most expensive one. There is no middle ground.

The facts are stark. OpenAI, Anthropic, and Google are spending hundreds of billions scaling LLMs. AMI Labs has 12 employees, no product, and a thesis. But that thesis comes from a Turing Award laureate who invented convolutional neural networks, the architecture that enabled modern computer vision, and who has been right about the direction of AI research more often than almost anyone alive. NVIDIA is backing both sides. Jeff Bezos and Eric Schmidt are signing checks. And the largest seed round in European history is funding an alternative to the technology that every other major AI lab considers the path to superintelligence.

AMI's CEO, Alexandre LeBrun, offered the most honest assessment of what comes next: "We are developing world models that seek to understand the world, and you can't do that locked up in a lab. At some point, we need to put the model in a real-world situation with real data and real evaluations."

In other words: the theory phase is over. Now they have to prove it works. With $1.03 billion and the most famous contrarian in AI, they are about to get their chance.

Sources

• Yann LeCun's AMI Labs raises $1.03B to build world models (Mar 9, 2026)
• Yann LeCun's New AI Startup Raises $1 Billion in Seed Funding (Mar 10, 2026)
• Yann LeCun just raised $1bn to prove the AI industry has got it wrong (Mar 10, 2026)
• Yann LeCun's AMI Labs raises $1bn in Europe's biggest seed round (Mar 10, 2026)
• Yann LeCun's new venture is a contrarian bet against large language models (Jan 22, 2026)
• Meta chief AI scientist Yann LeCun is leaving to create his own startup (Nov 19, 2025)
• "You certainly don't tell a researcher like me what to do" says LeCun as he exits Meta (Jan 3, 2026)
• Yann LeCun is targeting a $3.5 billion valuation for his new startup (Dec 19, 2025)
• Who's behind AMI Labs, Yann LeCun's 'world model' startup (Jan 23, 2026)
• Advanced Machine Intelligence (AMI) is Enabling the Next AI Revolution (Mar 10, 2026)
• AI luminaries at Davos clash over how close human-level intelligence really is (Jan 23, 2026)
• Meta lays off 600 from 'bloated' AI unit as Wang cements leadership (Oct 22, 2025)
• AI Pioneer Yann LeCun Warns of Tech Industry's "Dead End" with LLMs (Jan 26, 2026)
• V-JEPA: The next step toward advanced machine intelligence (Feb 2024)