The Turing Award winner parted ways with Meta four months ago, believing large language models were not the future. Today, he announced a record $1.03 billion seed funding in Europe to pursue a new direction.
In November 2025, Yann LeCun entered Mark Zuckerberg’s office to announce his departure. After twelve years of steering Meta’s AI research to international acclaim, he had become a vocal critic of the dominant technology.
He deemed large language models a statistical trick—remarkable but lacking true intelligence. He believed an alternative was possible and could be developed quicker outside Meta. On Tuesday, investors committed $1.03 billion to his vision.
Advanced Machine Intelligence Labs, or AMI, pronounced as “ami” in French, announced its seed funding on 10 March 2026, only four months post-establishment. The financing values the startup at $3.5 billion pre-money, marking a European record for the largest seed round.
Five firms co-led: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, along with Nvidia, Toyota, Samsung, Singapore’s Temasek, French VC firm Daphni, South Korean investor SBVA, and notable individuals such as Tim and Rosemary Berners-Lee, VC Jim Breyer, entrepreneur Mark Cuban, and former Google CEO Eric Schmidt.
According to a leaked pitch deck, LeCun initially sought around €500 million. The demand far exceeded this, resulting in €890 million, approximately $1.03 billion, allowing AMI to be selective with investors, as LeCun shared with journalists.
Headquartered in Paris, AMI plans additional offices in New York, Montreal, and Singapore. LeCun, holding dual French-American citizenship and maintaining a professorship at NYU, will be the executive chairman. Alexandre LeBrun, former CEO of medical AI startup Nabla, will be AMI’s chief executive.
The founding team primarily hails from Meta’s AI research unit. Michael Rabbat, former director of research science at Meta, joins as VP of world models. Laurent Solly, Meta’s former European VP, becomes COO. Pascale Fung, ex-senior AI research director at Meta, is named chief research and innovation officer. Saining Xie, previously with Google DeepMind, takes on the role of chief science officer.
AMI is focusing on building world models, a type of AI system LeCun has championed for years. This requires understanding why LeCun believes the industry has veered off course.
The case against LLMs
Large language models predict the next word in sequences and are trained on large volumes of text. Models like ChatGPT, Claude, and Gemini can generate fluent language across many topics. However, LeCun maintains this method has fundamental limitations.
His alternative, JEPA: the Joint Embedding Predictive Architecture, introduced in 2022, learns abstract representations rather than precise predictions, bypassing the generative AI approach. JEPA seeks systems that understand the world through lived experience