# Yann LeCun Raises $1B for AMI Labs: World Models Will Replace LLMs > Turing Award winner Yann LeCun leaves Meta and raises Europe's largest AI seed round — $1.03B — to build world models with JEPA architecture, backed by Bezos, Cuban, and Schmidt. Author: AI Mastermind Lab Published: 2026-03-09 Canonical: https://aimastermindlab.com/news/yann-lecun-ami-labs-billion-world-models-vs-llms --- ## The Biggest Contrarian Bet in AI Yann LeCun, the Turing Award-winning pioneer of deep learning, has raised **$1.03 billion in seed funding** for AMI Labs — making it Europe's largest AI seed round in history. The valuation: $3.5 billion pre-money. The thesis: large language models are a dead end. LeCun left Meta in late 2025 after years of publicly arguing that LLMs cannot achieve genuine understanding because they only process text. His new venture bets everything on **world models** — AI systems that learn how physical reality works by observing video and sensor data rather than reading text. ## JEPA: The Architecture Behind the Bet At the core of AMI Labs is **JEPA (Joint Embedding Predictive Architecture)**, a framework LeCun developed at Meta. Unlike generative AI models that try to predict every pixel or token, JEPA operates at a higher level of abstraction — learning to represent the structure and dynamics of the physical world without getting lost in low-level details. This isn't just academic theory. JEPA-based models can understand how objects move, interact, and transform in 3D space — capabilities that are essential for robotics, autonomous vehicles, and industrial simulation but fundamentally out of reach for text-based LLMs. ## A Dream Team and Heavyweight Backers The AMI Labs team reads like an AI research all-star roster: - **Laurent Solly** (ex-Meta VP Europe) as COO - **Saining Xie** as Chief Science Officer - **Pascale Fung** as Chief Research and Innovation Officer - **Michael Rabbat** as VP of World Models The investor list is equally striking: Bezos Expeditions, Cathay Innovation, Greycroft, Hiro Capital, HV Capital — plus personal investments from **Jeff Bezos**, **Mark Cuban**, **Eric Schmidt**, and **Tim Berners-Lee**. ## Why World Models Matter in 2026 AMI Labs isn't operating in a vacuum. The world models paradigm exploded in early 2026: - **Google DeepMind** released Genie 3, the first real-time interactive world model generating persistent 3D environments at 24 fps - **Fei-Fei Li's World Labs** launched its Marble product commercially - **Ineffable Intelligence** in London raised $1B led by former DeepMind scientist David Silver These developments signal a genuine architectural shift — from models that manipulate language to models that understand physical reality. For applications in robotics, manufacturing, healthcare simulation, and autonomous systems, world models may prove far more useful than even the most capable LLMs. ## The Enterprise Play AMI Labs isn't building a chatbot competitor. Its target applications are industrial: building realistic simulations of aircraft engines, factory processes, and logistics systems. A world model of an engine can help manufacturers optimize for efficiency, minimize emissions, and predict failures — tasks where LLMs provide surface-level answers at best. ## TL;DR Yann LeCun raised $1.03B (Europe's largest AI seed round) for AMI Labs to build world models using JEPA architecture, arguing LLMs are fundamentally limited. Backed by Bezos, Cuban, and Schmidt at a $3.5B valuation. — Learn more about model architectures in our [AI Fundamentals course](/courses/ai-fundamentals) and explore [how transformers work](/glossary/transformer) in the glossary. --- Source: https://aimastermindlab.com/news/yann-lecun-ami-labs-billion-world-models-vs-llms — AI Mastermind Lab (https://aimastermindlab.com)