The AI boom is built on one big idea: scale language models. A new billion-dollar startup is betting that this idea won’t lead to real intelligence.
Yann LeCun, the Turing Award-winning scientist who spent more than a decade building Meta’s AI research empire, has launched a startup called AMI (Advanced Machine Intelligence) and raised a $1.03 billion seed round. The Paris-based company was officially unveiled on Tuesday and is valued at $3.5 billion.
While the rest of Silicon Valley is focused on scaling large language models (LLMs), LeCun is pivoting toward “world models.” He argues that humans and animals don’t learn about reality by reading text, but by interacting with the physical world.
The funding round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with participation from several major tech companies and investors. According to AMI, the investors support the company’s goal of building “universally intelligent systems centered on world models.”
Corporate and strategic investors include Toyota Ventures, Nvidia, Samsung, Sea, Temasek, and SBVA, among others. Individual investors include Mark Cuban, Eric Schmidt, Xavier Niel, Jim Breyer, Tim and Rosemary Berners-Lee, and Mark Leslie, according to the company’s announcement.
The size of the round makes it one of the largest early-stage funding rounds in Europe’s AI sector, highlighting strong investor interest in new approaches to AI beyond today’s dominant technologies.
Why LeCun is betting against LLMs
LeCun has long argued that current AI systems, especially LLMs, cannot reach true human-level intelligence on their own. Speaking to WIRED, he criticized the idea that simply scaling language models will produce more advanced intelligence.
“The idea that you’re going to extend the capabilities of LLMs to the point that they’re going to have human-level intelligence is complete nonsense,” LeCun told WIRED in an interview.
Instead, his new company will focus on building AI systems that learn from the physical world, using data such as video, spatial information, and real-world interactions. These systems (world models) are designed to understand cause-and-effect relationships, plan actions, and reason about real-world environments.
LeCun, who is French-American, will serve as AMI’s executive chairman while continuing his role as a computer science professor at New York University. The company’s CEO is Alexandre LeBrun, a former chief executive of French digital health startup Nabla. Other co-founders include Meta’s former VP of Europe, Laurent Solly, as COO; AI researcher Pascale Fung as chief research and innovation officer; Saining Xie — formerly of Google DeepMind — as chief science officer; and Michael Rabbat, Meta’s former director of research science, as VP of world models.
Why now, and why outside Meta
LeCun announced his departure from Meta in November 2025, ending a 12-year run that included founding the company’s Fundamental AI Research lab, known as FAIR. His exit, he says, was driven by a divergence in priorities.
As world models like Joint-Embedding Predictive Architecture (JEPA) grew more sophisticated inside Meta, he told WIRED, the company “had to basically catch up with the industry on LLMs and kind of do the same thing that other LLM companies are doing, which is not my interest.” He went to CEO Mark Zuckerberg and made his case. “I told him I can do this faster, cheaper, and better outside of Meta,” LeCun said. “His answer was, OK, we can work together.”
Meta is not an investor in AMI, but both companies have indicated that a partnership is under discussion. LeCun told WIRED that AMI’s world models could one day power assistants running on Meta’s Ray-Ban smart glasses.
Also read: Another major bet on AI beyond language came through World Labs’ $1 billion funding round.