The Brief

Advanced Machine Intelligence Labs, the AI startup co-founded by Turing Award winner Yann LeCun after his departure from Meta, has closed a $1.03 billion seed round at a $3.5 billion pre-money valuation — the largest seed round ever raised by a European company. The round, co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions with backing from Nvidia, Toyota, Samsung, and more than twenty other investors, will fund development of “world model” AI systems built on LeCun’s JEPA architecture rather than the large language models he has publicly dismissed as “a dead end.”

The Report

Yann LeCun’s AMI Labs has raised $1.03 billion in seed funding to build AI systems that learn from physical reality rather than text prediction, nearly doubling the €500 million the company originally sought. The round values the less-than-six-month-old company at $3.5 billion.

The fundraise drew an unusually broad investor base spanning three continents. Nvidia, Toyota, Samsung, and Singapore’s Temasek joined alongside Bezos Expeditions, former Google CEO Eric Schmidt, web inventor Tim Berners-Lee, and Mark Cuban. French institutional capital showed up in force: public investment bank Bpifrance, the Mulliez family, Marcel Dassault’s industrial group, LVMH’s Aglaé Ventures, Publicis Groupe, and telecoms billionaire Xavier Niel all participated. Some investors characterised the round as a “Coconut” round — a label acknowledging that the scale bears little resemblance to traditional seed funding.

LeCun, 65, spent twelve years at Meta — five as founding director of Facebook AI Research and seven as chief AI scientist — before departing in November 2025. His exit followed a period of internal friction: Meta’s acquisition of Scale AI for $14 billion in June 2025 installed 28-year-old CEO Alexandr Wang as the company’s chief AI officer, a reporting structure LeCun found untenable. In a Financial Times interview, he described the Llama 4 benchmark controversy bluntly — “Results were fudged a little bit” — and said Mark Zuckerberg “basically lost confidence in everyone who was involved.”

AMI’s technical bet centres on JEPA, the Joint Embedding Predictive Architecture LeCun first proposed in 2022. Where large language models predict the next token in a sequence, JEPA learns abstract representations of how the physical world behaves — training on video and spatial data rather than text. The company’s flagship system, AMI Video, is intended to serve as the foundation for applications in robotics, manufacturing, healthcare, and wearable technology. CEO Alexandre LeBrun, a serial AI entrepreneur who previously ran medical AI startup Nabla, said the company expects its first usable products within a year, with “fairly universal intelligent systems” targeted within three to five years.

AMI is not alone in the world-model space. Fei-Fei Li’s World Labs raised $1 billion in February for its “spatial intelligence” platform, and former OpenAI CTO Mira Murati’s Thinking Machines Lab closed a $2 billion seed round — the largest on record. LeBrun appeared aware of the herd forming behind him: “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.”

LeCun has positioned AMI explicitly as a European challenger. “Silicon Valley is completely hypnotized by the current models of generative AI,” he told Fortune. “To pursue this kind of new research, you have to go outside the Valley — to Paris.” The company operates from four hubs: Paris, New York, Montreal, and Singapore. French President Emmanuel Macron texted LeCun personally after the news broke.

AMI has no product and no revenue. LeCun acknowledged the company will spend its first year entirely on research and development.

The Angle

The interesting number here is not the billion dollars. It is the zero. Zero products, zero revenue, zero near-term prospect of either — and a valuation north of three billion on the strength of one man’s conviction that an entire industry is digging in the wrong direction.

What LeCun is selling, and what two dozen investors across three continents are buying, is an architectural thesis: that intelligence built from language prediction will never cross the gap into genuine understanding of the physical world, and that the systems which eventually do cross it will be built on something closer to how biological minds actually learn — from observation, not narration. The JEPA framework is the specific technical claim. The broader bet is that the hundreds of billions currently pouring into scaling language models are producing diminishing returns that the market has not yet priced in.

That bet has structural consequences regardless of whether LeCun is right. The sheer weight of capital now flowing toward world models — AMI’s billion, World Labs’ billion, Thinking Machines’ two billion, all within a few months — is creating a second front in AI development that did not exist a year ago. The industry was, until very recently, a monoculture. It is becoming something more like a species with a hedge.

LeBrun’s prediction that “world models” will become the next buzzword in six months is almost certainly correct, and it is the most important thing either founder said. The cycle is predictable: a credible alternative emerges, capital floods toward it, dozens of imitators dilute the category, and the signal of genuine architectural difference gets lost in the noise of rebranding. The question for AMI is whether it can establish technical lead before the label it is building under becomes meaningless. LeCun has a research record that buys him more runway than most. He shared the Turing Award for work on convolutional neural networks — a bet on a specific architecture that the rest of the field eventually came around to, years after he made it. Whether that pattern repeats is unknowable. That it happened before is the reason a company with no product just raised a billion dollars.

The European framing is worth noting for what it reveals about the current geography of AI confidence. AMI’s chairman lives in New York. Its investors span Abu Dhabi to Seoul. Calling it a European company is a choice — and the choice is being made because Europe needs it to be true. The French state bank invested. Macron texted. The narrative of a European frontier lab is being constructed in real time, and the enthusiasm is running slightly ahead of the evidence. AMI is headquartered in Paris the way a holding company is headquartered wherever the tax regime suits it. The talent and capital are global. The flag is French.

None of which diminishes the underlying wager. A sixty-five-year-old scientist walked away from one of the most powerful research positions in the world because he believed the dominant paradigm was wrong, and the market just handed him a billion dollars to prove it. The last time the AI field had a serious architectural schism, the dissidents turned out to be right and the consensus turned out to be a decade-long detour. The billion dollars is not buying a product. It is buying the possibility that this is one of those moments again — and the price of finding out too late that it was.

The first serious capital reallocation away from the architecture that built this era of AI has begun. The market is not yet sure it is a correction. It is sure it cannot afford to miss one.