Advanced Machine Intelligence Labs, the Paris-based startup founded by former Meta chief AI scientist Yann LeCun, announced Tuesday it raised $1.03 billion in seed funding at a $3.5 billion pre-money valuation. The round, first reported by the Financial Times, makes AMI Labs the largest seed-stage startup in European history, second globally only to Thinking Machines Lab's $2 billion raise last June.
Five firms co-led the round: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Amazon founder Jeff Bezos's Bezos Expeditions. Nvidia, Toyota, Samsung, and Singapore's Temasek also participated. Individual backers include former Google CEO Eric Schmidt, Mark Cuban, telecom billionaire Xavier Niel, and Tim and Rosemary Berners-Lee.
LeCun originally sought around €500 million, according to a leaked pitch deck reported by Sifted. Investor demand doubled the figure to €890 million, roughly $1.03 billion.
The Breakdown
- AMI Labs raised $1.03 billion at $3.5 billion valuation, Europe's largest seed round
- LeCun bets against LLMs with JEPA-based world models trained on sensor data, not text
- Founding team mostly ex-Meta FAIR, offices in Paris, New York, Montreal, Singapore
- No product or revenue for at least a year, targeting robotics, healthcare, manufacturing
The anti-LLM wager
AMI Labs exists because LeCun thinks the AI industry took a wrong turn. Large language models predict the next word in a sequence. They generate fluent text across an enormous range of subjects. Good enough to sell. LeCun spent a decade watching that progress from inside Meta, where he founded Facebook AI Research in 2013. He left in November 2025 convinced the approach would never produce systems capable of human-level reasoning.
"The idea that you're going to extend the capabilities of LLMs to the point that they're going to have human-level intelligence is complete nonsense," LeCun told WIRED.
His counterproposal dates to a 2022 paper. He called it JEPA, Joint Embedding Predictive Architecture, and it works nothing like an LLM. Forget predicting the next word or next pixel. JEPA tries to learn how the world actually works, building abstract representations and throwing out surface noise. A toddler catches a ball without narrating the physics. That's closer to what LeCun is after, and no chatbot can do it.
AMI plans to build on that research with a model it calls AMI Video, trained on real-world sensor data rather than text. The goal, LeCun told Reuters, is to become "the main provider of intelligent systems, regardless of what the application is." He gave a timeline of three to five years before the company produces "fairly universal intelligent systems" ready for deployment.
That places AMI in the same category as Fei-Fei Li's World Labs, which raised $1 billion last month from Nvidia, AMD, and Autodesk at a $5 billion valuation. The world models field is getting crowded and expensive fast. "My prediction is that 'world models' will be the next buzzword," AMI CEO Alexandre LeBrun told TechCrunch. "In six months, every company will call itself a world model to raise funding."
Stay ahead of the curve
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
A Meta reunion in Paris
LeBrun, who previously ran French medical AI startup Nabla, takes on the CEO role. LeCun serves as executive chairman. The rest of the founding team reads like a FAIR alumni reunion: Michael Rabbat as VP of world models, Laurent Solly as COO, Pascale Fung leading research and innovation. All three spent years at Meta under LeCun. The outlier is chief science officer Saining Xie, who came from Google DeepMind.
The company launched with about a dozen employees spread across four offices: Paris headquarters, New York, Montreal, and Singapore. LeBrun told the New York Times the startup would operate like a research lab first, exploring untested ideas before moving toward products.
"We have at least a year of research before deploying our first real-world applications," LeBrun said. "But this is not an applied AI company."
Nabla will be AMI's first partner, applying world models to healthcare, where LLM hallucinations carry life-threatening consequences. Beyond healthcare, AMI wants to put world models into factories, aircraft engines, and pharmaceutical labs. The investor list tells you where. Toyota Ventures, Samsung, and a unit of fighter jet manufacturer Dassault Aviation all wrote checks.
Neither American nor Chinese
LeCun walked into Mark Zuckerberg's office in November 2025 and told him he was leaving. By his own account, Zuckerberg took it well. Meta is not an investor in AMI but the two companies are discussing a partnership that could put world model technology into Meta's Ray-Ban smart glasses. "That's probably one of the shorter term potential applications," LeCun told Reuters.
But the pitch goes beyond technology. AMI positioned itself from the start as something rare in frontier AI, a company headquartered outside the US and China. "There is a lot of demand from the industry, from governments around the world, for a credible provider of AI technology that is neither American nor Chinese," LeCun told Bloomberg.
Some European investors sound emboldened. "AMI Labs could be the first European company to reach the scale of the GAFAM companies," said Pierre-Éric Leibovici, founder of French VC firm Daphni. Margaux Grégoir, partner at Serena, put it more bluntly to Sifted. "It's almost a political thing."
AMI plans to publish its research and open-source much of its code. No product. No revenue. No near-term prospect of either. LeBrun acknowledged the timeline doesn't fit the typical startup playbook, no product in three months, no ARR in twelve. But a billion dollars in seed money buys patience, and LeCun has bet his reputation that the AI industry's biggest companies are chasing the wrong architecture. The investors backing him are willing to wait and find out.
Frequently Asked Questions
What are world models and how do they differ from LLMs?
World models learn abstract representations of physical reality from sensor data like video, rather than predicting the next word in text. LeCun's approach, JEPA (Joint Embedding Predictive Architecture), ignores surface-level noise and builds models of how the world works. The goal is systems that can reason about physical environments for robotics, manufacturing, and healthcare.
Who are AMI Labs' investors?
Five co-leads: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Strategic backers include Nvidia, Toyota, Samsung, and Temasek. Individual investors include Eric Schmidt, Mark Cuban, Xavier Niel, and Tim Berners-Lee. A unit of Dassault Aviation also participated.
What is AMI Labs' relationship with Meta?
Meta is not an investor, but the two companies are discussing a partnership. LeCun spent 12 years at Meta founding FAIR before departing in November 2025. Most of AMI's founding team came from Meta. One potential collaboration involves deploying world model technology in Meta's Ray-Ban smart glasses.
When will AMI Labs have a product?
CEO Alexandre LeBrun said the company has at least a year of pure research ahead. LeCun gave a three-to-five-year timeline for 'fairly universal intelligent systems.' The startup plans to work with partners like Nabla in healthcare before broader commercialization.
Why is AMI Labs headquartered in Paris instead of Silicon Valley?
LeCun deliberately avoided Silicon Valley, saying it's 'LLM-pilled.' AMI positions itself as neither American nor Chinese, appealing to governments and industries seeking alternative AI providers. Paris gives access to European industrial talent and French government backing through Bpifrance.
