Yann LeCun Raises $1.03 Billion for AMI Labs: The "World Models" Revolution Begins

It's the largest seed round ever raised in Europe for an AI company. Advanced Machine Intelligence Labs (AMI Labs), the startup co-founded by Yann LeCun after leaving Meta, announced on Monday March 10, 2026 that it had raised $1.03 billion at a pre-money valuation of $3.5 billion. All this with no product, no revenue, and less than three months of existence.

Who is Yann LeCun and why this matters

Yann LeCun isn't just another entrepreneur jumping on the AI bandwagon. He is one of the three 'godfathers' of deep learning, alongside Geoffrey Hinton and Yoshua Bengio. His 2018 Turing Award — the Nobel Prize equivalent in computer science — recognized decades of pioneering work on convolutional neural networks (CNNs), the technology that made image recognition, computer vision, and much of modern AI possible.

During 12 years at Meta (formerly Facebook), LeCun led the FAIR (Fundamental AI Research) lab, one of the most productive AI research centers in the world. Under his leadership, Meta developed LLaMA, the open-source model family that disrupted the industry. But it was also at Meta that LeCun became the industry's most vocal dissenting voice.

Why LeCun left Meta

The break wasn't impulsive. For years, LeCun has publicly argued that large language models (LLMs) — ChatGPT, Claude, Gemini — are a dead end. Not that they're useless, but that they will never lead to true artificial intelligence. According to him, a system that merely predicts the next word fundamentally cannot understand the world.

“LLMs are incredibly useful, but they don't understand anything. A two-year-old child has an understanding of the physical world that no LLM possesses. We need a fundamentally different architecture.”

The problem for LeCun was that Meta had massively refocused on LLMs in recent years, with LLaMA's success and Meta AI's integration across all its products. His contrarian vision became increasingly difficult to defend within a company racing to win the language model competition. He left Meta at the end of 2025, while maintaining he kept 'good relations with Mark Zuckerberg.'

AMI Labs: the dream team

LeCun didn't go solo. AMI Labs brings together a team of six co-founders with complementary profiles, operating across hubs in Paris, New York, Montreal, and Singapore:

• Yann LeCun — Executive Chair, Turing Award winner, architect of the JEPA vision
• Alexandre LeBrun — CEO, former CEO of Nabla (medical AI), former Facebook researcher
• Mike Rabbat — VP World Models, former research science director at Meta
• Saining Xie — Chief Science Officer, former researcher at Google DeepMind
• Pascale Fung — Chief Research & Innovation Officer, former senior AI director at Meta
• Delphine Groll and Martin Raison — Nabla co-founders, bringing healthcare expertise

The choice of Alexandre LeBrun as CEO is strategic. A Facebook AI veteran, he led **Nabla**, a medical AI startup that will be AMI Labs' first partner. The connection between the two companies is explicit: in healthcare, LLM hallucinations can have deadly consequences. This is precisely the kind of domain where world models could make a difference.

World models: the alternative to LLMs

At the heart of AMI Labs lies a radical conviction: real intelligence does not start in language. Where LLMs learn to predict the next word in a text sequence, world models learn abstract representations of the real world from sensory data — images, videos, physical sensors.

The core architecture is called **JEPA** (Joint Embedding Predictive Architecture), proposed by LeCun in 2022. Unlike generative models that try to reproduce every detail of a scene, JEPA learns to predict in an abstract representation space, ignoring unpredictable details. It's closer to how humans and animals understand their environment.

$1.03 billion: who's investing and why

The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. But it's the strategic investors that tell the real story:

• Nvidia — The GPU maker dominating AI infrastructure is betting on diversification beyond LLMs
• Toyota — The automotive giant sees world models as the future of autonomous driving
• Samsung — The Korean electronics conglomerate is betting on robotics and wearables applications
• Temasek — The Singaporean sovereign fund confirms AMI Labs' Asian hub
• Jeff Bezos (Bezos Expeditions) — Amazon's founder diversifying his AI bets
• Eric Schmidt — Former Google CEO and prolific AI investor
• Mark Cuban — Tech entrepreneur and investor
• SBVA — South Korean investor

The roadmap: from research to applications

LeCun laid out a three-phase plan:

1. Year 1 (2026) — Fundamental R&D. Intensive hiring to reach 20-30 people 'very soon.' Building the first model: AMI Video.
2. Year 2 (2027) — First industrial applications, notably in healthcare (via Nabla), robotics, and manufacturing.
3. Years 3-5 (2028-2030) — Production of 'universal intelligent systems' for autonomous driving, robotics, wearables, and 'virtually any application requiring intelligent machines.'

The first concrete product will be called AMI Video — a world model trained on video data to understand real-world physical interactions. The company also plans to work in robotics, manufacturing, and wearables.

What this means for the AI industry

AMI Labs isn't the first company to bet on world models. But it's the first founded by a Turing Award winner with funding at this scale. The message to the industry is clear: a paradigm shift is possible, and major investors believe in it enough to bet a billion dollars.

CEO Alexandre LeBrun doesn't shy away from bold claims: 'My prediction is that 'world models' will be the next buzzword. In six months, every company will call itself a world model to raise funding.' A prediction that speaks volumes about the team's confidence — and the hype potential around this new paradigm.

A French and European bet

AMI Labs is headquartered in Paris, with hubs in New York, Montreal, and Singapore. It's a strong signal for the European ecosystem: the continent's largest seed round is led by a French researcher who chose to base his startup's headquarters in the French capital, not Silicon Valley.

For France, which has been multiplying initiatives to position itself as a global AI hub (Mistral, Poolside, Kyutai), AMI Labs' arrival is further validation. The presence of French investors like Cathay Innovation and Daphni in the round reinforces this positioning.

LLMs vs world models: who's right?

The question dividing the AI community is simple: are LLMs a step toward general intelligence, or a sophisticated dead end? LeCun is categorical — it's a dead end. The majority of the industry (OpenAI, Google, Anthropic) bets on the opposite.

The truth may lie somewhere in between. LLMs have demonstrated reasoning and generalization capabilities few researchers had predicted. But they remain fundamentally limited by their grounding in text. A robot navigating a warehouse, an autonomous vehicle anticipating pedestrian behavior, a medical system understanding human anatomy — these applications require an understanding of the physical world that text alone cannot provide.

AMI Labs isn't betting against LLMs — it's betting that the next leap will come from elsewhere. And with $1.03 billion, that's a bet taken very seriously.