Yann LeCun Raises $1.03B in Europe's Largest Seed Round Ever — AMI Labs Challenges LLM Dominance with World Models

On March 10, 2026, the AI world got a seismic signal: Yann LeCun, the Turing Award-winning godfather of deep learning, officially unveiled AMI Labs (Advanced Machine Intelligence) alongside a $1.03 billion seed round — the largest in European startup history and one of the biggest seed financings ever globally. The round values AMI at $3.5 billion pre-money, a staggering figure for a company founded just four months ago.

But this isn't just another mega-round in an era of inflated AI checks. AMI Labs represents something far more provocative: a direct, well-funded challenge to the large language model paradigm that has dominated AI since ChatGPT's launch in late 2022. LeCun has spent years arguing that LLMs are fundamentally limited — and now he has $1 billion to prove it.

The Founding Story: Why LeCun Left Meta

A 12-Year Run Ends

LeCun announced his departure from Meta on November 18, 2025, ending a 12-year tenure that began when he founded Facebook AI Research (FAIR) in 2013. He spent five years as FAIR's founding director and seven as Meta's Chief AI Scientist, building one of the world's most influential AI research labs.

The proximate cause of his exit was strategic. When Meta brought in Scale AI founder Alexandr Wang as its new AI chief, the company pivoted decisively toward scaling LLMs — precisely the approach LeCun had been publicly criticizing for years. The philosophical gap became untenable.

"I am creating a startup company to continue the Advanced Machine Intelligence research program I have been pursuing over the last several years," LeCun wrote on LinkedIn. "The goal is to create systems that understand the physical world, have persistent memory, can reason, and can plan complex action sequences."

The CEO: Alexandre LeBrun, France's Serial AI Entrepreneur

While LeCun serves as Executive Chairman and chief visionary, daily operations are led by Alexandre LeBrun, a remarkable serial founder in his own right. Educated at École Polytechnique and Télécom Paris, LeBrun has an enviable track record:

• VirtuOz (2003): A chatbot pioneer, acquired by Nuance in 2012
• Wit.ai (2013): An NLP platform, Y Combinator W14 batch, acquired by Facebook — which is where LeBrun first worked alongside LeCun at FAIR (2015-2018)
• Nabla: A healthcare AI startup he co-founded and continues to lead, which has become AMI's inaugural partner

LeBrun is the original author of the open-source Duckling library (time and quantity parsing), and his deep operational experience complements LeCun's research genius perfectly.

The Dream Team

AMI Labs has assembled a leadership roster that reads like an AI research hall of fame:

• Saining Xie — Chief Science Officer, formerly at Google DeepMind, expert in visual representation learning
• Pascale Fung — Chief Research & Innovation Officer, former Meta senior director, pioneer in human-centered AI
• Michael Rabbat — VP World Models, former Meta research science director, based in Montreal
• Laurent Solly — COO, technology operations veteran

The $1.03 Billion Seed Round: Breaking Down the Numbers

From €500M Target to $1B+ Close

When AMI Labs first began fundraising in December 2025, the target was a still-ambitious €500 million. But investor demand was overwhelming. By the time the round closed, AMI had secured $1.03 billion (~€890 million) at a $3.5 billion pre-money valuation.

The Investor Syndicate

The round was co-led by five firms spanning three continents:

• Cathay Innovation (France)
• Greycroft (US)
• Hiro Capital (UK)
• HV Capital (Germany)
• Bezos Expenditions (Jeff Bezos' personal investment vehicle)

Strategic investors added industrial credibility:

• NVIDIA — the chip giant whose GPUs will likely power AMI's training runs
• Temasek — Singapore's sovereign wealth fund
• Toyota Ventures — signaling automotive and robotics applications
• Samsung — consumer electronics and semiconductor play
• SBVA — South Korean venture capital

The angel roster is equally impressive: Mark Cuban, Xavier Niel (Free/Iliad founder), Jim Breyer, Eric Schmidt, and Tim and Rosemary Berners-Lee — the inventor of the World Wide Web betting on the next paradigm of world understanding.

European Context

To appreciate the magnitude: this is not only Europe's largest seed round ever but ranks among the region's biggest AI fundings at any stage. For comparison, Mistral AI's recent Series C was $2 billion, and UK-based Nscale raised $2 billion — both at far later stages. Closing $1 billion at the seed stage is virtually unprecedented.

The Technology: Why World Models Matter

The LLM Critique

LeCun's argument against LLMs has been consistent and pointed: "LLMs can't truly reason or plan, because they lack a model of the world. They can't predict the consequences of their actions."

LLMs work by predicting the next token in a sequence — an approach that has proven spectacularly effective for text and code generation. But LeCun contends this is fundamentally insufficient for tasks that require understanding physical reality: guiding a robot through a warehouse, predicting how a patient's condition will evolve, or planning a manufacturing process.

As LeBrun put it: "Predicting tokens, though powerful, works best for discrete and low-dimensional tasks. Factories, hospitals, and robots require AI that grasps continuous, high-dimensional reality."

JEPA: Joint Embedding Predictive Architecture

AMI Labs' core technology is based on JEPA (Joint Embedding Predictive Architecture), outlined in LeCun's influential paper "A Path Towards Autonomous Machine Intelligence." The key innovation is deceptively simple yet profound: instead of predicting outputs directly from inputs (like LLMs generating text), JEPA makes predictions in an abstract representation space.

Think of it this way: humans don't predict every pixel of what they'll see next. Instead, we form compressed mental models of the world and predict at the level of abstract concepts — "if I push this cup, it will fall." JEPA works similarly, learning compressed representations of reality that enable efficient reasoning about physical dynamics.

This approach, based on self-supervised learning and energy-based models, aims to capture the causal structure of the physical world — something that statistical pattern-matching on text fundamentally cannot achieve.

AMI Video: The First Product

AMI's first publicly announced product is the AMI Video model, which learns physical-world dynamics from video data. Target applications span robotics, manufacturing, and wearable devices — domains where understanding the 3D physical world is essential.

Market Analysis: The World Model Opportunity

Competitive Landscape

AMI Labs isn't alone in the world models space. Stanford professor Fei-Fei Li's World Labs has also raised $1 billion pursuing a similar vision of AI that understands physical reality. But a key distinction exists: AMI is built specifically around the JEPA architecture, which LeCun has been developing and refining for years, giving it a strong theoretical foundation.

Critically, as one VC noted: "US firms dominate the LLM space, but world models are still nascent and the playing field is still open." This positions Europe — and AMI Labs specifically — to claim leadership in the next major AI paradigm rather than perpetually chasing American LLM leaders.

The TAM: Beyond Text, Into Physical Reality

The markets AMI is targeting are enormous. The global AI robotics market is projected to grow from $24.7 billion in 2026 to $124.3 billion by 2034. The AI-driven industrial robotics segment alone is expected to reach $49.1 billion by 2034, growing at a 20.8% CAGR. Healthcare AI, autonomous vehicles, and smart manufacturing represent additional multi-billion-dollar opportunities.

The fundamental insight is that LLMs have captured the "digital" AI market — text, code, and digital content generation. But the physical world represents a far larger economic surface area that remains largely unaddressed by current AI approaches.

European AI's New Chapter

One investor made a bold claim: "AMI Labs could be the first European company to reach the scale of the GAFAM companies." While that remains to be seen, the combination of world-class talent, record funding, and a genuinely differentiated technological approach gives the thesis real substance.

If Mistral AI proved that Europe could compete in LLMs, AMI Labs may prove that Europe can lead in the next paradigm.

Strategic Outlook: What Comes Next

Global Operations, Selective Hiring

AMI Labs operates from four strategic hubs: Paris (HQ), New York (near LeCun's NYU professorship), Montreal (home to VP Michael Rabbat and Canada's deep AI talent pool), and Singapore (gateway to Asian markets and Temasek's home turf). LeBrun has emphasized "quality over quantity" in hiring, focusing on recruiting the world's best researchers rather than scaling headcount rapidly.

Capital Deployment

The $1.03 billion will fund three priorities: massive compute infrastructure for training world models, elite talent acquisition, and R&D on JEPA-based architectures. The inaugural partnership with Nabla suggests healthcare will be an early application domain, with industrial robotics and manufacturing close behind.

Why Investors Wrote the Check

Three factors drove this historic investment. First, the team: a Turing Award winner surrounded by top-tier researchers from Meta, DeepMind, and NYU is as close to a "can't miss" founding team as exists in AI. Second, technological differentiation: JEPA addresses problems that LLMs structurally cannot solve, creating a new market rather than competing in a crowded one. Third, timing: as LLM scaling laws show signs of diminishing returns, appetite for the next paradigm has never been higher.

The strategic investor composition tells its own story. When NVIDIA, Samsung, Toyota, and Temasek all invest in a seed-stage company, they're signaling that the technology has real industrial demand — not just academic promise.

What to Watch

LeBrun offered a prediction that captures the moment perfectly: "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." Bold, perhaps — but with a billion dollars in the bank, a dream team assembled, and industrial giants placing their bets, AMI Labs has the resources to make that prediction self-fulfilling. The next 12–18 months will be critical: watch for the first real-world deployments of AMI Video, expansion of the research team, and whether the JEPA architecture delivers on its theoretical promise at industrial scale. The LLM era isn't over — but the post-LLM era may have just begun.

---

Sources: TechCrunch, Sifted, Crunchbase News, Cathay Innovation, The Next Web, CNBC, Fortune