The Billion-Dollar World Model: Yann LeCun’s AMI Labs Secures $1.03B to Challenge LLM Dominance

Yann LeCun’s campaign against the "LLM-pilled" consensus just received a billion-dollar war chest. AMI Labs (Advanced Machine Intelligence) announced today that it has raised $1.03 billion (~€890M) in a seed round that values the Paris-based startup at $3.5 billion pre-money.

The financing, co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, represents one of the largest early-stage investments in the history of the sector. It marks a decisive move by LeCun to operationalize his theory that Large Language Models (LLMs) are a "delusion" on the path to true Artificial General Intelligence (AGI). Instead, AMI Labs is betting everything on world models—systems designed to understand the physical world through observation and causal reasoning rather than just text tokens.

A Global "FAIR" Diaspora

The startup’s launch follows LeCun’s exit from Meta in November 2025, a move triggered by Meta’s strategic shift to catch up with industry leaders in generative LLMs. LeCun, who remains a professor at NYU, has effectively reconstituted the core of Meta’s Fundamental AI Research (FAIR) team under the AMI banner.

The leadership team is a significant concentration of AI talent:

• CEO Alex LeBrun: Former head of Meta’s Wit.ai and chairman of the digital health startup Nabla.
• COO Laurent Solly: Meta’s former VP for Europe.
• Chief Science Officer Saining Xie: A renowned researcher known for his work at Google DeepMind and Meta.
• Chief Research and Innovation Officer Pascale Fung: A leading expert in conversational AI and former senior director at FAIR.
• VP of World Models Michael Rabbat: Previously a director of research science at Meta.

Operating out of Paris, New York, Montreal, and Singapore, the lab intends to prioritize "quality over quantity," leveraging its funding to compete for top-tier talent and the massive compute resources required for world-model training.

The JEPA Bet

At the heart of AMI’s technical strategy is the Joint Embedding Predictive Architecture (JEPA). As previously reported, JEPA differs from generative models like ChatGPT or Sora by predicting the future in an abstract representation space. This allows the model to ignore "pixel noise"—the flickering of a light or the texture of a carpet—and focus on the underlying physics of a scene.

LeCun argues that this is the only way to grant robots the "common sense" they currently lack. While firms like Figure and Agility are finding industrial success with what LeCun calls "brittle" behaviors, AMI is looking to provide the foundational "brain" that allows for true generalization.

From Research to Reality

Despite its academic roots, AMI Labs is already lining up industrial partners. Nabla, the healthcare startup co-founded by LeBrun, is the first disclosed partner. The goal is to apply world models to high-stakes environments like medicine and manufacturing, where the hallucinations typical of LLMs can have life-threatening or catastrophic consequences.

The investor list also suggests a broad industrial reach, featuring backers like Nvidia, Samsung, Sea, and Toyota Ventures. Individual "angel" investors include former Google CEO Eric Schmidt, Mark Cuban, and World Wide Web inventor Tim Berners-Lee.

While autonomous household chores and humanoid hardware races dominate current headlines, LeCun’s massive raise suggests a growing investor appetite for the "one more big breakthrough" that DeepMind and others have long suggested is missing. AMI Labs is no longer just an academic critique; it is now the best-funded alternative to the transformer-based status quo.