AMI Labs: Yann LeCun Just Raised $1.03 Billion to Prove LLMs Are a Dead End

Yann LeCun has spent the last two years telling anyone who would listen that large language models are fundamentally limited. Now he has $1.03 billion to prove it.

The Turing Award winner’s new startup, AMI Labs (Advanced Machine Intelligence), just closed one of the largest seed rounds in AI history, valued at $3.5 billion pre-money. The bet: world models — AI systems that learn from reality itself, not just from text on the internet — are the real path to machine intelligence.

The Man Who Said No to LLMs

LeCun isn’t some outsider throwing stones. He spent 12 years at Meta leading its fundamental AI research, building systems used by billions. He’s a Turing Award winner, widely credited as one of the founding figures of deep learning. When he talks about the limits of AI, the industry listens — even when they disagree.

His argument has been consistent and blunt: LLMs are trained purely on language, so they can never truly understand the physical world. They can write poems about gravity but can’t predict that a ball will fall when dropped. They hallucinate because they don’t have an internal model of how reality works. For LeCun, this isn’t a bug that can be patched — it’s a fundamental architectural flaw.

So he left Meta. And he built AMI Labs to pursue what he believes is the right approach: world models powered by a framework he created called JEPA (Joint Embedding Predictive Architecture).

What World Models Actually Are

World models are AI systems trained on real-world data — video, sensor readings, physical interactions — rather than text corpora. The idea is to give AI an intuitive understanding of how the physical world operates: objects have mass, liquids flow, things fall.

AMI Labs’ approach centers on JEPA, which LeCun developed during his final years at Meta. Unlike generative AI that tries to predict the next pixel or token, JEPA learns abstract representations of the world. It filters out noise and focuses on the underlying rules that govern physical reality. Think of it as training an AI to build a mental model of the world, rather than memorizing patterns in data.

The practical applications are significant: autonomous vehicles, drones, robotics, industrial automation — anything that requires AI to operate safely in unpredictable physical environments. These are domains where chatbots and text generators simply don’t cut it.

A $1.03 Billion War Chest and a Star-Studded Team

The funding round, announced on March 9, 2026, was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. The individual investor list reads like a tech hall of fame: Jeff Bezos (through Bezos Expeditions), Eric Schmidt, Mark Cuban, Jim Breyer, Xavier Niel, and even Tim and Rosemary Berners-Lee — the couple behind the invention of the World Wide Web.

The team LeCun assembled is equally impressive:

Alex LeBrun (CEO) — Previously co-founder and CEO of Nabla, a health AI startup. He also worked alongside LeCun at Meta AI.
Laurent Solly (COO) — Former VP of Meta for Europe, bringing serious operational and business-scaling experience.
Saining Xie (Chief Science Officer) — A top-tier deep learning researcher, previously at Meta and NYU.
Pascale Fung (Chief Research & Innovation Officer) — A well-known AI researcher with deep expertise in multilingual and multimodal AI.
Michael Rabbat (VP of World Models) — Another Meta AI veteran who worked on distributed learning systems.

AMI Labs is headquartered in Paris, with offices planned in New York, Montreal, and Singapore.

How AMI Labs Compares to Other World Model Startups

AMI Labs isn’t the only company chasing world models. The space has become one of the hottest areas in AI funding:

World Labs (Fei-Fei Li) raised $1 billion in February 2026, with a $200 million strategic investment from Autodesk. Their first product, Marble, lets users create editable 3D environments. While World Labs focuses on visual and spatial understanding for creative and design workflows, AMI Labs is targeting a broader mission of general-purpose world understanding.

Google DeepMind launched Genie 3, a real-time interactive world model, along with SIMA and Nano Banana for 3D world modeling. DeepMind has the advantage of Google’s compute resources but is constrained by operating within a large corporate structure.

Runway and Wayve are also building world models, with Runway focusing on creative applications and Wayve on autonomous driving specifically.

What sets AMI Labs apart is the combination of LeCun’s JEPA architecture — which takes a fundamentally different approach to representation learning — and the sheer depth of AI research talent on the team. Most competitors are building on variants of transformer architectures. LeCun is explicitly betting against that paradigm.

Why This Matters Beyond the Hype

The $1.03 billion raise is significant for several reasons beyond the headline number.

First, it validates world models as a serious research direction, not just an academic curiosity. When investors like Bezos, Schmidt, and Cuban put over a billion dollars behind a thesis, the industry takes notice.

Second, it represents the largest direct challenge to the LLM-dominant paradigm. While companies like OpenAI, Anthropic, and Google continue pouring resources into scaling language models, AMI Labs is making a high-profile, well-funded argument that the entire approach has a ceiling.

Third, the practical implications could be massive. Autonomous vehicles, humanoid robots, and industrial automation all need AI that understands physics, not just language. If JEPA-based world models deliver on their promise, they could unlock applications that LLMs structurally cannot address.

The counterargument is real, though. LLMs have been improving rapidly, and multimodal models are already incorporating visual and spatial understanding. Critics argue that LeCun is fighting yesterday’s battle — that modern foundation models are already evolving beyond pure text. Whether JEPA represents a truly better architecture or just a different one remains to be proven at scale.

Who Should Watch AMI Labs Closely

If you work in robotics, autonomous systems, or any field requiring AI to interact with the physical world, AMI Labs is worth following closely. The company is still pre-product, but the combination of LeCun’s research credibility, a billion-dollar runway, and a team of Meta AI veterans means they have the resources to execute on a long-term research agenda.

For developers and engineers in the AI space, AMI Labs also signals a broader trend: the industry may be approaching a fork in the road between language-first and world-first approaches to intelligence. Understanding both paths will matter.

FAQ

How much funding has AMI Labs raised?
AMI Labs raised $1.03 billion in a seed round at a $3.5 billion pre-money valuation (approximately $4.53 billion post-money). The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions.

What are world models and how are they different from LLMs?
World models are AI systems trained on real-world data (video, sensor data, physical interactions) to understand how the physical world works. Unlike large language models that learn patterns from text, world models aim to develop an intuitive understanding of physics, spatial relationships, and cause-and-effect — enabling applications like autonomous driving and robotics.

Who is behind AMI Labs?
AMI Labs was founded by Turing Award winner Yann LeCun (Executive Chairman) after he left Meta. The CEO is Alex LeBrun, with Laurent Solly (former Meta VP Europe) as COO, Saining Xie as Chief Science Officer, and Pascale Fung as Chief Research & Innovation Officer. The company is headquartered in Paris.

How does AMI Labs compare to Fei-Fei Li’s World Labs?
Both are billion-dollar-funded startups focused on world models, but they differ in approach and focus. World Labs (led by Fei-Fei Li) focuses on spatial intelligence and 3D environments, with its first product Marble targeting creative and design workflows. AMI Labs takes a broader approach using LeCun’s JEPA architecture, targeting general-purpose world understanding for robotics, autonomous vehicles, and industrial applications.

Does AMI Labs have any products available yet?
As of March 2026, AMI Labs is still in the research and development phase with no publicly available products. Given the fundamental nature of their research — building new AI architectures from the ground up — commercial products are likely still some time away. The $1.03 billion funding gives them a substantial runway to pursue long-term research before needing to ship.

Top AI Product