Yann LeCun's new startup, Advanced Machine Intelligence, closed a $1.03 billion seed round on Tuesday, giving the Paris-based company a $3.5 billion pre-money valuation before it has shipped a single product. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with strategic backing from Nvidia, Samsung, Toyota Ventures, and Temasek.
That investor list also includes former Google CEO Eric Schmidt, Mark Cuban, and Tim Berners-Lee, which reads less like a cap table and more like a dinner party for people who can afford to lose a billion dollars on a hunch.
The anti-LLM thesis
AMI Labs is building what it calls "world models," AI systems that learn from visual and sensory data rather than text. The technical foundation is JEPA, the Joint Embedding Predictive Architecture that LeCun developed while still at Meta. Instead of predicting the next word (the approach behind ChatGPT, Claude, and Gemini), these models attempt to build internal representations of how the physical world actually works.
LeCun has been saying LLMs are a dead end for years now. He left Meta in late 2025 after 12 years, frustrated with the company's pivot toward closed development under new chief AI officer Alexandr Wang. In a Financial Times interview around the time of his departure, LeCun called the idea that scaling LLMs leads to superintelligence "complete nonsense."
Strong words from someone who spent a decade inside one of the biggest LLM shops on the planet. But now he's putting a billion dollars of other people's money behind the conviction.
A billion buys you what, exactly?
No product. Not yet, and not soon. CEO Alexandre LeBrun, who previously ran health-tech startup Nabla, told TechCrunch that AMI is fundamentally a research project and shouldn't be compared to applied AI startups that ship in three months. He estimated about a year before the first usable technology emerges, with commercial applications taking considerably longer.
LeBrun also made an observation that deserves more attention than it got: he predicted that "world models" will become the next buzzword, and that within six months every company will slap the label on itself to raise funding. The fact that AMI's own CEO is preemptively calling out the coming hype cycle around his own category is either refreshing honesty or savvy positioning. Probably both.
The money goes toward two things: compute and talent. AMI has assembled a leadership team heavy on Meta alumni, with Laurent Solly (Meta's former VP for Europe) as COO, Saining Xie (from Google DeepMind) as chief science officer, and Michael Rabbat (former Meta research director) as VP of world models. The company operates across Paris, New York, Montreal, and Singapore, though LeCun told AFP the immediate plan is to bring on just 20 to 30 people.
So who pays for world models?
AMI's first disclosed partner is Nabla, LeBrun's former company, which gets early access to world model technology for healthcare applications. The logic there is straightforward: LLM hallucinations in medical contexts can be dangerous, and a system that actually understands cause and effect rather than pattern-matching text could matter.
Beyond healthcare, LeCun told Reuters that manufacturers, automakers, aerospace companies, and pharmaceutical firms are the near-term targets. He also floated the idea of domestic robots and even deploying the technology in Meta's Ray-Ban smart glasses, which would make Meta a client of its former chief scientist. There's a certain poetry in that.
The $3.5 billion valuation puts AMI in the same league as Fei-Fei Li's World Labs, which raised $1 billion last month. Both are betting that the next era of AI doesn't look like a chatbot. But neither has anything close to a product yet, and the commercial path from "understands physics" to "generates revenue" remains, to put it generously, underspecified.
AMI plans to publish research openly and release code as open source, a stance that aligns with LeCun's longstanding criticism of closed AI labs. Whether that openness survives contact with a billion dollars in investor expectations is a different question entirely.
The company says it will engage with prospective customers early to test models against real data. LeBrun told TechCrunch the startup cannot build world models in isolation. Nabla is the first partner, but the presence of Toyota, Samsung, and Nvidia in the cap table suggests others are already lined up.
