Yann LeCun Startup AMI Raises $1.03 Billion to Build World Model AI

The Story

Advanced Machine Intelligence (AMI), a Paris-based startup founded by Yann LeCun after leaving Meta in November 2025, has raised $1.03 billion at a $3.5 billion pre-money valuation. The company aims to commercialize AI systems focused on reasoning, planning, persistent memory, and world-model architectures rather than next-token prediction used by large language models. LeCun argues that language-based systems alone will not produce broadly capable intelligent agents because much of human reasoning is grounded in the physical world. AMI is targeting industrial sectors first, with early partners and customers expected to help test the technology in real-world settings.

The Facts

• $1.03 billion funding round announced

  AMI confirmed it raised $1.03 billion to fund development and commercialization of its AI technology at a $3.5 billion pre-money valuation.

• Round backed by major investors, industry groups, and tech figures

  The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Additional named backers include Nvidia, Samsung, Sea, Temasek, Toyota Ventures, Bpifrance Digital Venture, Publicis Groupe, Groupe Industriel Marcel Dassault, and individuals including Mark Cuban, Eric Schmidt, Xavier Niel, Jim Breyer, Mark Leslie, and Tim and Rosemary Berners-Lee.

• AMI is building AI around world models, memory, reasoning, and planning

  The company says it wants to build AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe, rather than relying mainly on predicting the next word or pixel.

• Yann LeCun and AMI argue current LLM approaches are fundamentally limited

  Yann LeCun says large language models are useful for tasks such as code generation, but argues they will not by themselves lead to human-level intelligence because human reasoning is rooted in the physical world, not just language. AMI is pursuing this through world models and work linked to JEPA, or Joint Embedding Predictive Architecture, proposed by LeCun in 2022.

• Industrial customers are the near-term focus

  The company plans to work first with organizations operating complex systems, including manufacturing, biomedical, robotics, aerospace, automotive, and pharmaceutical companies.

• Early deployments will rely on partners and real-world data

  AMI says it wants to engage prospective customers early because world models need to be tested in real-world situations with real data and real evaluations. Nabla is the first disclosed partner expected to access early models.

• Yann LeCun pointed to industrial simulation as an early use case

  One example he gave is building a realistic world model of an aircraft engine to help manufacturers improve efficiency, reduce emissions, and increase reliability.

• Long-term consumer applications remain part of the vision

  Over time, the technology could support consumer products such as domestic robots that need common sense and a better understanding of the physical world.

• Possible collaboration with Meta remains on the table

  Meta is not an investor in AMI, but Yann LeCun said he is discussing potential collaboration, including the use of AMI world models in Ray-Ban Meta smart glasses.

• Yann LeCun left Meta in November 2025 after years leading world-model research

  He joined Meta in 2013 to found Facebook AI Research, later FAIR, and said Meta later reoriented more heavily toward catching up on LLMs, while his own interest remained in world models.

• AMI was founded by Yann LeCun and several former Meta and AI leaders

  Cofounders include Alexandre LeBrun, former Nabla CEO and now AMI chief executive; Saining Xie, former Google DeepMind researcher and AMI chief science officer; and former Meta leaders Michael Rabbat, Laurent Solly, and Pascale Fung.

• The startup is positioning itself as a global company from launch

  The company says it will operate from Paris, Montreal, Singapore, and New York, with hiring focused on compute and research talent across those locations.

• AMI says it will publish research and open-source code

  Alongside commercial development, the company says it plans to publish papers and make a lot of code open source, arguing that open research can help build a broader ecosystem around its work.

Background / Context

Meta has simultaneously intensified its own push into large language models. In June 2025, the company reorganized its AI efforts under Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang. AMI therefore represents not just a new startup, but a direct commercial bet on an alternative path to advanced AI. It also enters a growing world-model field that has recently attracted large funding rounds from other startups.

Why This Matters

The funding highlights a deep split inside the AI industry over whether scaling large language models is enough to reach more general intelligence. AMI is now one of the clearest and best-funded attempts to turn world-model research into commercial products, especially for companies working in complex real-world environments, even if the path from research to revenue may take years.

This article was drafted with the assistance of generative AI. All facts and details were reviewed and confirmed by an editor prior to publication.