Yann LeCun Raises $1 Billion to Build AI That Understands the Physical World

Yann LeCun Raises $1B for AMI to Build AI World Models

Key Facts

• What: Advanced Machine Intelligence (AMI), cofounded by Meta’s former chief AI scientist Yann LeCun, announced it has raised more than $1 billion to develop AI world models focused on physical-world understanding.
• Valuation: The Paris-based startup is valued at $3.5 billion following the round.
• Leadership: LeCun serves as a key leader while continuing as a New York University professor; Alexandre LeBrun (former Nabla CEO) is CEO and Saining Xie (former Google DeepMind researcher) is chief science officer.
• Approach: AMI is building systems with persistent memory, reasoning, planning, controllability and safety, arguing that scaling large language models (LLMs) alone cannot achieve human-level intelligence.
• Investors: Co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions; backers include Mark Cuban, Eric Schmidt and Xavier Niel.

Lead

Advanced Machine Intelligence (AMI), a new Paris-based startup cofounded by Yann LeCun, has raised more than $1 billion to develop AI systems that understand the physical world through world models rather than relying primarily on language. LeCun, who left Meta in November 2025 after founding its Fundamental AI Research lab, argues that human-level intelligence requires grounding in physical reality and that extending LLMs to achieve it is “complete nonsense.” The funding round, which values AMI at $3.5 billion, positions the company as a direct challenge to the dominant LLM-scaling strategies pursued by OpenAI, Anthropic and Meta.

The Vision for World Models

LeCun has long advocated for world models — AI systems that learn the underlying rules, physics and cause-and-effect relationships of the physical world, typically from multimodal sensory data rather than text alone. In an interview with WIRED, he emphasized that most human reasoning is grounded in the physical world, not language.

“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.

AMI’s stated goal is to build “a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe,” according to the company’s press release. These systems are intended to move beyond the statistical pattern-matching of today’s LLMs toward predictive models capable of simulating real-world dynamics.

LeCun does not dismiss the utility of LLMs. He acknowledges their strength in code generation and many practical applications but maintains they will not lead to human-level intelligence. He describes the current industry focus on ever-larger LLMs as having created “a kind of delusion” among builders who believe scaling alone will suffice.

Founding Team and Global Ambitions

AMI was cofounded by LeCun and several former Meta colleagues, including Michael Rabbat (former director of research science), Laurent Solly (former vice president of Europe) and Pascale Fung (former senior director of AI research). Additional cofounders are Alexandre LeBrun, who will serve as CEO, and Saining Xie, who will be chief science officer.

The startup is global from day one, with offices in Paris, Montreal, Singapore and New York. LeCun will continue his role as a professor at New York University while leading the company. This marks LeCun’s first commercial venture since departing Meta.

The company plans to collaborate with enterprises in manufacturing, biomedical, robotics and other data-rich industries. As an example, LeCun cited building a realistic world model of an aircraft engine to help manufacturers optimize efficiency, minimize emissions or improve reliability.

Break with Meta and the LLM Orthodoxy

LeCun’s departure from Meta reflects a strategic disagreement over AI priorities. While at Meta he developed concepts such as the Joint-Embedding Predictive Architecture (JEPA), an early approach to world models. As these techniques matured, Meta shifted resources toward catching up with the industry on LLMs, a direction that did not align with LeCun’s interests.

He met with Meta CEO Mark Zuckerberg in November 2025 and explained that he believed he could develop world models faster, cheaper and more effectively outside the company by sharing development costs with other enterprises. Zuckerberg was supportive, according to LeCun, and the two discussed potential future collaboration. Meta is not an investor in AMI, but LeCun said the companies are in talks about possible partnerships, including using AMI’s world models to power assistants in Meta’s smart glasses.

AMI represents a notable bet against the prevailing view at leading AI labs. OpenAI, Anthropic and Meta have invested heavily in the idea that continued scaling of LLMs, combined with additional training data and compute, will eventually yield human-level or superintelligent systems. LeCun’s skepticism carries particular weight given his stature: he is a Turing Award winner (2018) and one of the most prominent voices highlighting the limitations of current LLM architectures.

Open Source Stance and AI Control

LeCun has indicated that AMI plans to build open source technology. He argues that artificial intelligence is too powerful to be controlled by any single private company. This stance echoes broader industry debates about AI safety, governance and concentration of power. The WIRED article notes that such concerns have intensified recently, including the Pentagon’s decision to blacklist Anthropic after the company attempted to set internal red lines.

Competitive Landscape and Industry Context

The announcement comes amid intense competition in AI research. Several other efforts are exploring alternatives or complements to pure LLM scaling, including work on world models at other organizations. AMI enters the field with significant credibility due to LeCun’s track record and the caliber of its founding team, many of whom held senior roles at Meta and Google DeepMind.

The $1 billion raise at a $3.5 billion valuation signals strong investor confidence in LeCun’s contrarian vision. Notable backers such as Bezos Expeditions, Eric Schmidt and Mark Cuban bring both capital and high-profile connections to the venture. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions.

Impact on Developers, Enterprises and the AI Industry

For developers and researchers, AMI could offer an alternative research direction focused on grounded, physically-aware AI rather than purely linguistic models. Enterprises in physical-world domains — manufacturing, robotics, biomedical devices, energy and aerospace — may gain access to tools that can simulate complex systems, predict outcomes and optimize processes in ways current LLMs struggle to achieve.

The company’s emphasis on controllability and safety addresses growing concerns about deploying advanced AI in real-world settings where errors could have serious consequences. Persistent memory and planning capabilities could enable longer-horizon reasoning that goes beyond the context windows of today’s LLMs.

What’s Next

AMI has not yet disclosed specific technical details, model architectures, benchmarks or release timelines beyond its high-level mission. The company is still in early stages and will likely focus initially on research and development while building out its global teams.

LeCun has suggested that world models could eventually integrate with or complement existing LLM capabilities rather than fully replace them. Potential applications range from industrial optimization to advanced robotics and scientific discovery.

The startup’s open source aspirations, if realized, could accelerate progress across the field by making foundational world-model technology available to a wider community of researchers and developers. Collaboration with Meta and other organizations could further amplify its impact.

Industry observers will watch closely to see whether AMI can translate LeCun’s long-standing research agenda into commercially viable products that demonstrate meaningful advantages over today’s dominant LLM paradigm.

Sources

• WIRED: Yann LeCun Raises $1 Billion to Build AI That Understands the Physical World
• TechCrunch: Yann LeCun confirms his new 'world model' startup, reportedly seeks $5B+ valuation
• Built In: Yann LeCun Launches AMI Labs to Build AI World Models
• Creati.ai: Yann LeCun's AMI Labs Aims to Build World Models