In a move that signals a significant strategic shift in the global race for artificial intelligence, Yann LeCun, the Turing Award-winning scientist and a foundational figure in modern deep learning, has announced the launch of a new venture, Advanced Machine Intelligence (AMI). Based in Paris, the startup has successfully closed a seed-stage funding round exceeding $1 billion, establishing an initial valuation of $3.5 billion. The company’s primary objective is to move beyond the current paradigm of Large Language Models (LLMs) to develop "AI world models"—systems capable of understanding physical reality, reasoning through complex problems, and planning actions in the real world.
The emergence of AMI marks the first commercial endeavor for LeCun, who has served as the Chief AI Scientist at Meta for over a decade. His departure from Meta, scheduled for November 2025, follows a period of internal strategic realignment at the social media giant. While Meta has pivoted heavily toward catching up with the generative AI and LLM craze sparked by OpenAI’s ChatGPT, LeCun has remained a vocal critic of the idea that scaling language-based models will lead to true Artificial General Intelligence (AGI).
A New Philosophy for Machine Intelligence
The central thesis behind AMI is that human-level intelligence cannot be achieved through text-based learning alone. LeCun argues that the vast majority of human and animal knowledge is derived from observing and interacting with the physical world, not from reading strings of text. LLMs, which function by predicting the next token in a sequence, lack a fundamental "common sense" or an understanding of cause and effect in three-dimensional space.
"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 stated in a recent interview. He contends that while LLMs are proficient at generating code and manipulating language, they are prone to hallucinations and logical failures because they do not possess a persistent internal model of how the world works.
AMI’s mission is to build a "new breed" of AI systems. These models are designed to have persistent memory, allowing them to learn over time rather than resetting with every prompt. Furthermore, they are intended to be "controllable and safe," addressing a major concern in the industry where autonomous agents often behave in unpredictable ways when faced with novel physical environments.
High-Profile Backing and Strategic Investment
The $1 billion funding round reflects a massive vote of confidence from a diverse group of institutional and private investors. The financing was co-led by a consortium including Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, the investment firm of Amazon founder Jeff Bezos.
The investor list also features prominent figures from the tech and business sectors. Former Google CEO Eric Schmidt, who has been increasingly active in AI defense and infrastructure investments, is a notable backer. He is joined by Mark Cuban, the American billionaire entrepreneur, and Xavier Niel, the French telecommunications magnate and founder of the Station F incubator. The involvement of Niel further solidifies Paris’s growing reputation as a primary global hub for AI research and development, rivaling Silicon Valley and London.
This level of capitalization is rare for a startup in its infancy, but it underscores the immense computational and talent costs associated with building foundational models that depart from existing architectures.
The Foundation Team: A Concentration of Talent
AMI is not merely a vehicle for LeCun’s research; it is a collaborative effort involving several high-ranking veterans from Meta’s Fundamental AI Research (FAIR) lab and other leading institutions. The leadership team includes:
- Alexandre LeBrun (CEO): Formerly the CEO of Nabla, an AI healthcare startup, and a key figure in Meta’s early AI assistant initiatives.
- Saining Xie (Chief Science Officer): A former researcher at Google DeepMind and Meta, known for his work on computer vision and neural network architectures.
- Michael Rabbat: Meta’s former Director of Research Science.
- Laurent Solly: Former Vice President of Europe at Meta.
- Pascale Fung: Former Senior Director of AI Research at Meta and a renowned expert in natural language processing and empathetic AI.
The company will maintain a global footprint from its inception, establishing research and development offices in Paris, Montreal, Singapore, and New York. This geographical spread allows AMI to tap into diverse talent pools, particularly in Montreal, which is a world leader in reinforcement learning, and Singapore, a growing center for industrial AI applications.
Strategic Departure from Meta’s Consumer Focus
The decision to launch AMI as an independent entity stems from a divergence in goals between LeCun’s research vision and Meta’s corporate strategy. During his tenure at Meta, LeCun spearheaded the development of the Joint-Embedding Predictive Architecture (JEPA), a non-generative approach to AI that attempts to predict missing parts of a video or an image in an abstract representation space.
While Meta CEO Mark Zuckerberg has been supportive of LeCun’s research, the company’s recent focus has shifted toward the "Llama" series of LLMs to compete directly with OpenAI, Google, and Anthropic. LeCun noted that world models are most applicable to industrial sectors—such as manufacturing, aerospace, and robotics—which do not align with Meta’s primary business model of social media and advertising.
"I told [Zuckerberg] I can do this faster, cheaper, and better outside of Meta," LeCun explained. "I can share the cost of development with other companies." This indicates that AMI will likely operate on a B2B (business-to-business) model, providing sophisticated world models to enterprises that require high-precision reasoning and physical simulation.
Targeted Industrial Applications
Unlike current AI companies that focus on chatbots or creative content generation, AMI is targeting industries with massive datasets related to physical processes.
- Manufacturing and Aerospace: AMI aims to create realistic world models of complex machinery, such as aircraft engines. By understanding the physics of these systems, the AI can help manufacturers optimize fuel efficiency, predict maintenance needs with unprecedented accuracy, and minimize carbon emissions.
- Biomedical Research: The startup plans to apply its modeling capabilities to biological systems, helping researchers understand cellular interactions and accelerate drug discovery through predictive simulations.
- Robotics: This is perhaps the most immediate application for world models. For a robot to navigate a warehouse or perform delicate surgery, it needs more than language; it needs a "physical intuition." AMI’s models could provide the underlying intelligence for a new generation of autonomous hardware.
Chronology of the Shift in AI Research
The launch of AMI represents a pivotal moment in the timeline of artificial intelligence:
- 2012–2018: The era of Deep Learning dominance. LeCun, Yoshua Bengio, and Geoffrey Hinton (the "Godfathers of AI") receive the Turing Award for their work on neural networks.
- 2017–2022: The rise of the Transformer architecture and Large Language Models. OpenAI’s GPT series demonstrates the power of scaling data and compute.
- 2023–2024: Industry-wide "LLM mania." Companies pour billions into generative AI, but limitations regarding reasoning, reliability, and physical understanding begin to surface.
- November 2024: LeCun announces the formation of AMI, signaling a move toward "Objective-Driven AI" and world models.
- November 2025: LeCun is set to officially depart Meta, transitioning fully to his role at AMI while maintaining his professorship at New York University.
Broader Impact and Industry Implications
The formation of AMI is a direct challenge to the "Scaling Hypothesis"—the belief held by OpenAI and Anthropic that simply adding more data and more GPUs to current LLM architectures will eventually result in AGI. By contrast, AMI is betting on a "Structural Hypothesis," suggesting that a fundamental change in how AI processes information is required.
If AMI succeeds, it could shift the center of gravity in the AI industry away from Silicon Valley. The heavy involvement of French investors and the Paris headquarters highlight Europe’s ambition to lead in the "post-LLM" era. Furthermore, a successful world model could unlock the "Robotics Age," providing the software layer that has long been the bottleneck for sophisticated autonomous machines.
However, the path forward is fraught with technical challenges. Building a model that can accurately predict the complexities of the physical world requires vast amounts of video data and a different kind of computational logic than the token-prediction used by today’s chatbots.
The launch of Advanced Machine Intelligence suggests that the next decade of AI development may not be about teaching machines how to talk, but teaching them how to see, move, and understand the reality they inhabit. With over $1 billion in capital and some of the brightest minds in the field, Yann LeCun’s AMI is now the primary standard-bearer for this new frontier.
About the Author
