This is the announcement the ecosystem of AI-model publishers had been waiting for since Yann LeCun declared his departure from Meta and his intention to build Advanced Machine Intelligence (AMI) Labs.
Questions remained: With whom? With what funds? When?
The former head of Facebook AI Research (FAIR) unveiled the news on March 10, announcing a seed round of $1.03 billion (about €890 million) at a pre‑money valuation of $3.5 billion. A record for Europe’s seed rounds, surpassed only by the American Thinking Machines Lab, founded by Mira Murati, the former OpenAI chief technology officer, which had raised $2 billion in June 2025.
A team of researchers from Meta
We also now know the team that will build AMI Labs’ business: six co-founders, all former Meta colleagues. Alexandre Lebrun, the former CEO of Nabla—the health-tech start-up acquired by Meta in 2015 under the Wit.ai umbrella—will lead AMI Labs as chief executive, while Yann LeCun will chair the board.
Laurent Solly, the former Meta head for Europe, will serve as chief operating officer. Alongside him, Pascale Fung is director of research and innovation, Saining Xie is chief scientific officer, and Michael Rabbat is responsible for the “world representations.”
AMI Labs is currently said to count around ten employees and aims to grow to 30–50 within six months, split between its headquarters in Paris and offices in New York, Montreal and Singapore.
World models versus LLMs
At the heart of AMI Labs’ project is a proclaimed break with large language models (LLMs) like ChatGPT, Claude or Gemini. For Yann LeCun, these systems trained primarily on text will never reach human-level reasoning, not even that of a six-year-old or a cat. “The generative architecture trained through self‑supervised learning imitates intelligence; they don’t truly understand the world,” summarizes Alexandre Lebrun.
The alternative AMI Labs is developing is called “world models”: AI architectures capable of representing the physical environment in an abstract and conceptual way, capable of storing information and planning sequences of complex actions—such as predicting that an object will fall if it reaches the edge of a table.
AMI Labs will build on Yann LeCun’s work at Meta around the JEPA (Joint Embedding Predictive Architecture) architecture, trained on videos and spatial data rather than text.
Industrial and robotics ambitions
AMI positions itself first and foremost as a fundamental research project, but with clearly identifiable commercial avenues: manufacturers, carmakers, aerospace firms, and players in the biomedical and pharmaceutical fields. “We want to become the leading supplier of intelligent systems, regardless of the application,” LeCun told Reuters.
Robotics stands out as one of the priority applications. “Why today, we have systems that pass the bar exam, prove theorems, write code, but we still don’t have home robots or self-driving cars?” he stressed on France Inter.
He also envisions public‑facing uses with the deployment of this technology in its Ray-Ban Meta smart glasses. “This is likely one of the near-term applications,” he told Reuters.
AMI Labs will offer its models in two formats: API access (a pay‑as‑you‑go model) and a downloadable version, potentially open source. A differentiating asset, especially for attracting top-tier research talent.
A global funding round with a European anchor
Initially targeted at $500 million, investor demand vastly exceeded expectations. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions (Jeff Bezos’ fund). Nvidia, a familiar participant in AI startup funding, also took part, alongside Temasek of Singapore, SBVA (SoftBank), Toyota Ventures and Samsung.
On the European and French side, Daphni, Bpifrance, Groupe Industriel Marcel Dassault, the Mulliez Family Association, Aglaé Ventures (LVMH), Zebox Ventures (CMA CGM) and Xavier Niel, personally, were among the backers.
“There are investors very diverse, with a little over a third European, a large portion French, a bit under a third coming from Asia and the Middle East, and about a third American. When a company surpasses a valuation of a billion, that’s called a unicorn; we’ve surpassed three billion, so we’re a triceratops,” rejoiced Yann LeCun.
AMI Labs’ first strategic partner will be Nabla, the health-AI start-up co‑founded by Alexandre Lebrun, who will concurrently remain its chief scientific officer. This partnership will allow testing world models on real data in a sector of high stakes from the outset.
