Advanced Machine Intelligence Labs (AMI), the AI startup founded by Turing Award winner and former Meta chief AI scientist Yann LeCun, raised $1.03 billion at a $3.5 billion pre-money valuation March 10, marking Europe's largest seed round ever and positioning the Paris-based company as a major bet against the large language model architectures dominating current AI development.
The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with participation from Nvidia, Temasek, Toyota Ventures, Samsung, and individual investors including Mark Cuban, Eric Schmidt, Tim and Rosemary Berners-Lee, and Jim Breyer. LeCun initially sought approximately €500 million when announcing the startup late last year after departing Meta, but investor demand pushed commitments beyond $1 billion.
World Models vs. Large Language Models: The Fundamental Bet
AMI Labs is building "world models"—AI systems that learn abstract representations from real-world sensor data rather than predicting the next word in a text sequence. This approach directly challenges the transformer-based large language model architecture that underpins ChatGPT, Claude, Gemini, and nearly every frontier AI system deployed today.
LeCun has spent years arguing loudly that LLMs have fundamental architectural limits. While these models can generate fluent, plausible language across enormous subject ranges, they learn by statistical pattern matching on text rather than developing genuine understanding of how the physical world operates. This limitation becomes critical in applications where AI must reason about causality, spatial relationships, physical dynamics, and real-world constraints—precisely the domains where robotics, industrial automation, and embodied intelligence require reliability.
The company's approach builds on LeCun's Joint Embedding Predictive Architecture (JEPA), a framework he proposed in 2022. Rather than predicting future states in pixel-perfect or word-by-word detail—the approach that makes generative AI both powerful and prone to hallucination—JEPA learns abstract representations of how the world works. These representations allow AI systems to predict consequences of actions, plan sequences to accomplish tasks, and operate with safety guardrails in complex physical environments.
"Real Intelligence Does Not Start in Language. It Starts in the World."
AMI Labs CEO Alexandre LeBrun, former CEO of medical AI startup Nabla who now leads day-to-day operations with LeCun serving as executive chairman, acknowledged the potential for world models to become the next AI buzzword. "My prediction is that 'world models' will be the next buzzword," LeBrun told TechCrunch. "In six months, every company will call itself a world model to raise funding."
But LeBrun believes AMI's approach is fundamentally different from trend-chasing. The company operates under the conviction that "real intelligence does not start in language. It starts in the world"—a principle directly opposing the language-first approach that drove Meta, OpenAI, Anthropic, and Google to invest billions in ever-larger text-trained models.
LeBrun's experience at Nabla reinforced this conviction. As CEO, he reached the same conclusion as LeCun about LLM limitations in contexts where hallucinations carry life-threatening consequences. Medical applications demand reliability that current generative AI architectures struggle to guarantee, creating demand for AI systems grounded in verifiable representations of reality rather than statistical text generation.
Long-Term Research Project, Not Typical Applied AI Startup
AMI Labs represents a departure from the typical AI startup playbook. The company has no product, no revenue, and no near-term prospect of either. LeCun acknowledged to journalists that AMI would spend its first year focused entirely on research and development, with commercialization timelines measured in years rather than months.
"AMI Labs is a very ambitious project, because it starts with fundamental research," LeBrun explained. "It's not your typical applied AI startup that can release a product in three months, have revenue in six months and make $10 million in annual recurring revenue in 12 months." In contrast, it could take years for world models to transition from theoretical frameworks to commercial applications delivering customer value.
Despite this extended timeline, companies developing world models have attracted massive capital commitments. Fei-Fei Li's World Labs secured $1 billion last month, while European startup SpAItial raised a $13 million seed round—unusually large for early-stage European ventures. The funding reflects investor conviction that alternative AI architectures may be necessary to achieve capabilities that LLMs fundamentally cannot deliver.
Global Talent Strategy Across Four Hubs
AMI Labs will operate across Paris (headquarters), New York (where LeCun teaches at NYU), Montreal (where VP of World Models Mike Rabbat is based), and Singapore to access global AI talent and potential customers in Asian markets. The company has assembled a high-profile research team drawn almost entirely from Meta's AI research organization, including Saining Xie as chief science officer (formerly Google DeepMind), Pascale Fung as chief research and innovation officer (formerly Meta), and Laurent Solly as COO (formerly Meta VP for Europe).
The startup plans to allocate funding toward computing infrastructure and talent acquisition rather than immediate product development. Target customers include organizations running complex systems in automotive, aerospace, biomedical, manufacturing, and pharmaceutical sectors where reliability, controllability, and safety requirements exceed what current generative AI can confidently deliver.
LeCun told Reuters the company aims to "become the main provider of intelligent systems" for industries where AI must interact with physical reality rather than just generate text or images. Potential applications include industrial process control, automation systems, wearable devices, domestic robotics requiring common sense about the physical world, and healthcare scenarios where errors have catastrophic consequences.
Open Research Principles and Early Partner Collaborations
Staying true to LeCun's principles from his tenure leading Meta's AI research (FAIR), AMI Labs will prioritize open research and open-source code. "We will also create a lot of code open source," LeBrun said. "We think things move faster when they're open, and it's in our best interest to build a community and a research ecosystem around us."
While immediate revenue generation is not the focus, AMI Labs intends to collaborate with early customers to refine models using real-world data. Nabla, LeBrun's former company, will be the first partner accessing early models, with additional collaborations expected as the technology matures. This approach mirrors successful open AI research models where community contributions accelerate development while maintaining commercial optionality for deployment.
The $1.03 billion seed round demonstrates that investors backing AMI are willing to wait for returns measured in years rather than quarters. LeCun holds one of the most credible research records in AI, sharing the 2018 Turing Award for work on convolutional neural networks that underpin modern machine vision, and his argument that LLMs have fundamental architectural limits has been consistent enough that dismissing it no longer represents the safe assumption it once did. Whether AMI can deliver on this conviction remains genuinely open, but the billion-dollar bet signals that alternatives to transformer-based language models are attracting serious capital.
