Après avoir quitté Meta, l'entrepreneur en IA lauréat du prix Turing LeCun obtient un financement d'un milliard de dollars pour sa startup AMI

Turing Award laureate Yann LeCun, four months after leaving Meta, founded the AI startup AMI, which completed a $1.03 billion seed round, setting a record as the largest seed funding in European history. This move signifies that LeCun’s long-standing criticism of large language models (LLMs) will now be put into action. He and his team will focus on developing a World Model that surpasses LLMs and challenges existing technological barriers.

Advanced Machine Intelligence (AMI) secures billion-dollar investment

On March 10, 2026, Advanced Machine Intelligence (AMI) announced the completion of its funding round, with a pre-money valuation of $3.5 billion. Although AMI currently has no products or revenue, international investors including Bezos Expeditions, NVIDIA, Samsung, and Temasek of Singapore provided seed funding far exceeding the original target of €500 million. AMI’s high valuation reflects market concerns over current generative technology bottlenecks and trust in LeCun’s academic reputation. The leadership team includes former Meta Research Science Director Michael Rabbat and Saining Xie from Google DeepMind, forming a strong lineup.

What is AMI building?

In short, it’s a World Model—an artificial intelligence system that LeCun has been advocating for years. To understand this more deeply, it’s important to grasp why he believes AI has gone astray.

LeCun’s Criticism of LLMs Before Leaving Meta: Statistical Fallacies

LeCun’s motivation for leaving Meta was his belief that large language models are merely “statistical illusions” based on probability distributions. LLMs learn by predicting the next word in a sequence, performing well in generating fluent text but lacking understanding of the physical world. LeCun points out that this word-by-word or pixel-by-pixel prediction easily leads to “hallucinations” and cannot learn through observation and experience like humans or animals. He argues that the current development approach has fundamental limitations; relying solely on massive text training makes it difficult to achieve high-level intelligence with reasoning and physical common sense.

Core Technology of AMI: A Joint Embedding Predictive Architecture as an Alternative Path

The core technology of AMI is based on LeCun’s Joint-Embedding Predictive Architecture (JEPA), proposed in 2022. Unlike generative AI that predicts all text, JEPA aims to learn abstract representations of how the world works, ignoring unpredictable surface noise, and strives to build a system capable of understanding physical reality rather than just processing language symbols. According to plans, AMI intends to collaborate with corporate partners within the next one to two years, aiming to develop general intelligence systems deployable across various fields within three to five years, with the goal of becoming a leading provider of intelligent technology.

LeCun: AMI headquarters in Europe to counter US-China AI dominance

AMI will be headquartered in Paris, Europe, with offices in New York and Montreal, demonstrating LeCun’s ambition to challenge the AI monopolies of the US and China. He emphasizes the uniqueness of AMI as a cutting-edge laboratory with non-American and non-Chinese backgrounds. However, this ambitious project faces significant time pressures, as the new model is a long-term scientific endeavor unlikely to generate short-term profits. While the $1.03 billion provides ample funding for initial R&D, whether the JEPA architecture can effectively address LLM shortcomings and translate into competitive commercial AI remains a long road ahead.

This article about Turing Award laureate LeCun’s AI startup AMI securing a billion-dollar investment first appeared on Chain News ABMedia.