Advanced machine intelligence The Paris-based startup (AMI), co-founded by Meta’s former chief AI scientist Yann LeCun, announced Monday that it has raised more than $1 billion to develop AI world models.
LeCun argues that most human reasoning is based on the physical world, not language, and that developing true human-level intelligence requires an AI world model. “The idea of expanding the capabilities of the LLM [large language models] “The idea that they will have human-level intelligence is complete nonsense,” he said in an interview with WIRED.
The funding, which values the startup at $3.5 billion, was co-led by investors including Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions. Other notable backers include Mark Cuban, former Google CEO Eric Schmidt, and French billionaire and communications executive Xavier Niel.
AMI (pronounced like the French word for friend) aims to build “a new kind of AI system that understands the world, has persistent memory, is capable of reasoning and planning, and is controllable and secure,” the company said in a press release. The startup says it will be global from day one, with offices in Paris, Montreal, Singapore and New York, and in addition to leading the startup, LeCun will continue to work as a professor at New York University. AMI is LeCun’s first commercial endeavor since leaving Meta in November 2025.
LeCun’s startup represents a bet against many of the world’s largest AI labs, including OpenAI, Anthropic, and even his previous workplace, Meta. They believe that scaling up LLM will eventually lead to AI systems with human-level intelligence or even superintelligence. LLM has been behind viral products like ChatGPT and Claude Code, and LeCun is one of the most prominent researchers in the AI industry who has been vocal about the limitations of these AI models. LeCun is well known for his outspokenness, but as a pioneer of modern AI and winner of the Turing Award in 2018, his skepticism carries weight.
LeCun said AMI aims to work with companies in manufacturing, biomedical, robotics, and other industries that have large amounts of data. For example, AMI can build realistic global models of aircraft engines and work with manufacturers to help optimize efficiency, minimize emissions and ensure reliability, he said.
AMI was co-founded by LeCun and several leaders who LeCun worked with at Meta. Laurent Solly, former vice president for Europe; Pascal Huang, former senior director of AI research. Other co-founders include Alexandre LeBrun, former CEO of AI healthcare startup Nabla, who will become CEO of AMI, and Saining Xie, a former Google DeepMind researcher who will become the startup’s chief scientific officer.
World model example
LeCun does not deny the overall usefulness of the LLM. Rather, in his view, these AI models are just the latest promising trend in the technology industry, and their success has created “a kind of paranoia” among the people who build them. “That’s true [LLMs] “It’s true that we’re getting very good at code generation, and it’s probably going to become even more useful in a wide range of applications where code generation is useful, and that’s a lot of applications, but it doesn’t lead to human-level intelligence at all,” LeCun says.
LeCun has been working on world models for many years within Meta, where he founded FAIR, the company’s foundational AI lab. But he now believes his research is best done outside the social media giant. He said it became clear that the most powerful application of the global model would be to sell it to other companies, which doesn’t fit neatly into Meta’s core consumer business.
As AI world models like Meta’s Joint-Embedding Predictive Architecture (JEPA) become more sophisticated, “Meta’s strategy has basically become more about LLM and needing to catch up with the industry and do the same kind of things that other LLM companies are doing, and that’s not my interest,” LeCun says. “So, at some point in November, I went to Mark Zuckerberg and said, “He’s always been very supportive.” [world model research]but I told him we can do this faster, cheaper, and better outside of the meta. You can share the development costs with other companies… His answer was, “Okay, we can collaborate.” ”
