Yann LeCun Raises $1bn in Europe’s Biggest Ever AI Seed Round

The Man Who Helped Build Meta’s AI Just Raised $1 Billion to Prove It Wasn’t Enough

Yann LeCun spent years at the top of the artificial intelligence establishment — as Meta’s chief AI scientist, as a Turing Award winner, as one of the most cited researchers in the history of the field. And for years, he has been saying publicly, loudly, and with considerable intellectual force that the dominant approach to building AI is fundamentally wrong.

Last week, the world’s biggest investors decided he might be right. Advanced Machine Intelligence Labs — LeCun’s new European venture — has raised $1.03 billion in what is now the largest seed funding round in European history. The backers include Jeff Bezos’s Bezos Expeditions, Nvidia, Singapore’s Temasek, France’s Cathay Innovation, and Seoul-based SBVA. The company launches with a pre-money valuation of $3.5 billion and offices across Paris, New York, Singapore, and Montreal.

This is not just a funding story. It is a statement about where the most serious money in technology believes artificial intelligence is actually going.

What LeCun Is Building — and Why It Matters

To understand why a $1 billion seed round is being committed to a company with a dozen employees, you need to understand LeCun’s central argument — and why it is controversial within the AI establishment.

The dominant paradigm in AI right now is the large language model: systems trained on vast quantities of text that learn to predict and generate language with extraordinary fluency. GPT-4, Claude, Gemini — these are all variations on the same core architecture. They are genuinely impressive. They are also, LeCun has consistently argued, fundamentally limited in ways that will prevent them from ever achieving human-level reasoning.

His critique is specific. Systems trained primarily on text have no model of the physical world. They do not understand causality, physics, or the three-dimensional environment in which intelligent action must ultimately take place. They can describe a chair. They cannot understand what it means to sit in one, to balance on one, or to catch one that is falling. The race to build truly intelligent AI systems is not going to be won by scaling text prediction further — it requires a fundamentally different approach.

LeCun’s answer is what he calls “world models” — AI systems that build internal representations of the physical environment, understand cause and effect, and can reason about objects, space, and action in the way that humans and animals do from infancy. According to Meta AI Research, where LeCun developed many of these ideas before departing, world models represent one of the most promising frontiers in the field — though the path from theory to deployable systems remains genuinely difficult.

The applications LeCun’s team is targeting are telling: robotics and transport. These are precisely the domains where language model limitations are most acute and where physical world understanding is not optional but essential. The intersection of AI and physical infrastructure represents one of the largest untapped commercial opportunities in technology — and one that current AI architectures are structurally ill-equipped to address.

Why Europe — and Why Now

The decision to base the company’s primary operations in Paris is not incidental. Europe’s ambition to build sovereign AI capability has been growing steadily, backed by significant public and private investment, and France in particular has positioned itself as the continent’s AI capital with aggressive talent retention policies and state-backed research infrastructure. LeCun — a French-American scientist — choosing Paris as his operational base is both a personal decision and a signal that Europe’s AI ecosystem has reached a level of credibility that can attract world-class founders and billion-dollar capital simultaneously.

The timing also reflects a broader maturation in investor thinking about AI. According to Dealroom’s global venture data, the $1.03 billion seed round is second only to Thinking Machines Lab’s $2 billion raise in June 2025 — suggesting that the market for genuinely differentiated AI research, as opposed to incremental scaling of existing architectures, is deepening rapidly. Investors who backed the LLM wave early are now diversifying into approaches that bet against the current paradigm’s long-term dominance.

Alexandre LeBrun, former CEO of French healthcare AI startup Nabla, will lead the company as CEO, with LeCun serving as executive chair. Laurent Solly, Meta’s former VP for Europe, joins as COO — a leadership team that combines deep AI research credibility with serious operational and commercial experience.

The company that LeCun is building is a direct intellectual challenge to the organisations he has spent his career contributing to. The future of European AI leadership may depend on whether that challenge proves correct — and whether world models can deliver what large language models, for all their remarkable capabilities, ultimately cannot.

FAQ

Q: What is Yann LeCun’s new AI company and what does it do? Advanced Machine Intelligence Labs is a new AI research and development company founded by Yann LeCun, former chief AI scientist at Meta and Turing Award winner. It has raised $1.03 billion in seed funding — the largest seed round in European history — from investors including Jeff Bezos, Nvidia, and Temasek. The company is building “world models”: AI systems that understand the physical environment rather than relying primarily on text prediction, with initial applications targeting robotics and transport.

Q: Why does Yann LeCun believe large language models are not enough? LeCun has argued consistently that AI systems trained primarily on text lack a fundamental understanding of the physical world — causality, physics, spatial reasoning, and object permanence. While large language models are extraordinarily capable at language tasks, LeCun believes they will hit a ceiling when it comes to human-level reasoning and real-world action. His world models approach attempts to build AI that understands and can reason about the physical environment, which he believes is a prerequisite for genuinely general intelligence.