Yann LeCun
Renowned AI scientist Yann LeCun is raising €500 million ($586 million) at a €3 billion ($3.5 billion) valuation for his new startup AMI Labs—before launching any product—representing one of the most audacious bets yet that the AI industry's obsession with large language models is fundamentally wrong. The Turing Award winner left Meta in December after 12 years to pursue what he believes is the only viable path to artificial general intelligence: world models that understand physical reality rather than text prediction.
"We are going to have AI systems that have humanlike and human-level intelligence, but they're not going to be built on LLMs, and it's not going to happen next year or two years from now," LeCun told MIT Technology Review. "There are major conceptual breakthroughs that have to happen before we have AI systems that have human-level intelligence. And that is what I've been working on."
AMI Labs, which stands for Advanced Machine Intelligence, confirmed this week it's building world models to "create intelligent systems that understand the real world." The Paris-headquartered startup positions itself as a contrarian bet against the LLM-centric approaches dominating current AI development.
World Models vs. LLMs
The fundamental difference lies in how systems learn. LLMs predict the next word in sequences based on vast text training. World models attempt to understand environments, simulate cause-and-effect relationships, and predict outcomes in three-dimensional physical space—how objects interact, how actions lead to consequences, how forces apply in reality.
LeCun has argued publicly for years that scaling LLMs won't produce general intelligence. "LLMs are too limiting. Scaling them up will not allow us to reach AGI," he stated at NVIDIA's GTC conference. The technology's structural limitations include hallucinations, lack of physical grounding, and inability to reason about causality—problems he believes world models can solve.
The approach uses his JEPA (Joint Embedding Predictive Architecture) framework developed at Meta. "The world is unpredictable. If you try to build a generative model that predicts every detail of the future, it will fail," LeCun explained. "JEPA is not generative AI. It is a system that learns to represent videos really well."
Leadership and Funding
LeCun serves as executive chairman rather than CEO. That role belongs to Alex LeBrun, previously co-founder and CEO at medical AI startup Nabla. LeBrun sold his earlier company Wit.ai to Facebook in 2015, then worked under LeCun's leadership at Meta's AI research lab FAIR before founding Nabla.
According to Financial Times reporting, VCs in discussions include Cathay Innovation, Greycroft, and Hiro Capital (where LeCun is an advisor). Other potential investors reportedly include 20VC, Bpifrance, Daphni, and HV Capital.
The $3.5 billion pre-launch valuation appears astronomical until compared to recent AI fundraising. Former OpenAI CTO Mira Murati's Thinking Machines Lab raised at a $12 billion seed valuation. Fei-Fei Li's World Labs—a direct AMI competitor also building world models—raised $230 million at $1 billion valuation in August 2024 and is now reportedly in talks at $5 billion.
Meta as First Client
Despite leaving Meta, LeCun suggested his former employer could become AMI's first customer. "Meta might be our first client! We'll see," he told MIT Technology Review. "The work we are doing is not in direct competition. Our focus on world models for the physical world is very different from their focus on generative AI and LLMs."
LeCun criticized some of Meta CEO Mark Zuckerberg's strategic choices, particularly shutting down the FAIR robotics group. "I may not have agreed with all of them," he acknowledged diplomatically.
Target Applications
AMI Labs plans to license technology for high-stakes applications including healthcare, robotics, automation, and industrial systems—domains where LLM hallucinations pose unacceptable risks. Nabla, LeBrun's former company, has already signed a partnership to use AMI's models as they're developed.
The startup promises offices across Paris, Montreal, New York, and Singapore while contributing research through open publications and open-source initiatives—maintaining academic transparency while commercializing technology.
Whether world models deliver on their promise remains the industry's next major test. But with LeCun's scientific credibility and $586 million backing, AMI Labs has resources to find out.
