Who Is Behind AMI Labs? Inside Yann LeCun’s $3.5 Billion Bet on World Models

For years, Yann LeCun has been the AI industry’s most vocal contrarian. While Silicon Valley poured billions into Large Language Models (LLMs) like ChatGPT and Gemini, LeCun argued they were insufficient for achieving true machine intelligence. He has described them as "off-ramps" on the highway to AGI, useful tools, but incapable of the reasoning required to navigate the physical world. Now, he is putting his own reputation on the line to prove there is a better way.

According to reports from the Financial Times and Sifted, LeCun has officially launched AMI Labs (Advanced Machine Intelligence). The Paris-based startup is reportedly seeking funding at a valuation of $3.5 billion. But LeCun isn't doing it alone, and his goal is to move the industry beyond the era of the chatbot.

The Power Players Running AMI

While LeCun is the visionary face of the company, he has structured the leadership to allow him to remain deeply involved in research without getting bogged down in management. LeCun serves as the Executive Chairman and Founder, providing the scientific north star known as the "World Model."

To run the business, he has tapped Alex LeBrun as CEO. LeBrun is an operational powerhouse who previously founded Wit.ai, which was acquired by Facebook in 2015. Most recently, LeBrun served as CEO of Nabla, a healthcare AI startup. His transition to AMI Labs was part of a strategic deal where Nabla gained early access to AMI’s technology. While other former Meta executives have been rumored to be circling the project, LeBrun’s appointment confirms that AMI Labs is prioritizing leaders with deep experience in shipping applied AI products.

Why LLMs Are Not Enough

AMI Labs is founded on the premise that text generation is not the same as thinking. Current AI models work by predicting the next word in a sentence, making them excellent at mimicking language but unreliable at understanding reality. They frequently struggle with facts, fail at basic physics, and cannot plan for the future because they lack an internal concept of cause and effect.

LeCun has been vocal about this limitation for some time. As we reported earlier, Yann LeCun says Meta is burning cash on dead-end AI models, arguing that simply training bigger LLMs will never lead to human-level intelligence.

The Solution Is World Models

AMI Labs is building "World Models" specifically based on an architecture LeCun calls JEPA (Joint Embedding Predictive Architecture). Instead of training primarily on internet text, AMI’s systems are designed to learn by observing the physical world.

This involves analyzing video and sensor data to understand geometry, gravity, and object permanence. The system focuses on prediction, guessing what will happen next in a video rather than just predicting the next word. This capability allows for planning, using an internal simulation of the world to plan complex actions without relying on trial and error.

The Roadmap: Intelligence for the Real World

Because AMI Labs is focusing on physical understanding, its potential market extends far beyond writing emails. The company is positioning itself to solve problems in industries where precision and reliability are non-negotiable.

The appointment of Alex LeBrun, with his background in medical AI at Nabla, signals a strong interest in high-stakes fields. LeBrun has previously noted that the opportunity to apply world models to sectors like healthcare—where "hallucination" is a safety risk—was a key factor in his decision to join. By building systems that understand cause-and-effect, AMI Labs aims to create machines that can eventually navigate complex environments, from industrial automation to advanced robotics.

With a reported $3.5 billion valuation goal and backers like Cathay Innovation and Greycroft reportedly circling, AMI Labs is entering the arena as a heavyweight. The startup represents a massive divergence in the AI timeline. If LeCun is wrong, AMI Labs is an expensive science experiment. But if he is right, and current LLMs truly hit a wall, then AMI Labs might be the only company building the actual future of Artificial General Intelligence.