As Europe races to achieve greater AI sovereignty, Yann LeCun, the visionary who spent the last decade as Meta’s Chief AI Scientist, has launched a bold initiative.
“We need a paradigm shift,” LeCun declared, and he meant every word. In November 2025, LeCun walked away from Meta to launch his audacious new venture: Advanced Machine Intelligence (AMI).
AMI arrives as a bold Paris-based player, emerging hot on the heels of Mistral AI, but with a radically different focus. While much of the AI industry continues scaling ever-larger language models through token prediction, AMI is all-in on “world models.” These AI systems aim to develop grounded, physical-world understanding (much like humans and animals learn) rather than relying solely on text. The ambition: genuine reasoning, sophisticated planning, reliable foresight, and safe, controllable behavior in complex, real-world environments.
For Europe’s AI ecosystem, AMI’s launch felt like Christmas came early. The proof? A staggering $1.03 billion seed round at a $3.5 billion pre-money valuation. That figure makes it Europe’s largest seed ever and one of the biggest in AI history for an early-stage company.
But what exactly is AMI, and will it be Europe’s ticket to true AI sovereignty? We dug into its founders, investor lineup, and bold open-source roadmap to find out more.
The Brains Behind the Bet: AMI’s Star-Studded Team
LeCun, frequently dubbed one of the “godfathers” of deep learning alongside Geoffrey Hinton and Yoshua Bengio, brings the legacy element to AMI as Executive Chairman.
Leading as CEO is Alexandre Lebrun, the French entrepreneur behind health AI unicorn Nabla, as well as Wit.ai (acquired by Meta) and VirtuOz (acquired by Nuance). Lebrun’s track record bridges research and real-world deployment.
Lebrun explained that “we are developing world models that seek to understand the world, and you can’t do that locked up in a lab. At some point, we need to put the model in a real-world situation with real data and real evaluations.“
Rounding out the founding team are elite researchers and operators: Saining Xie as Chief Science Officer (ex-Google DeepMind), Pascale Fung as Chief Research and Innovation Officer (pioneer in human-centered AI), Michael Rabbat as VP of World Models (ex-Meta research director), and Laurent Solly as COO (ex-Meta VP for Europe).
This blend of French roots and global expertise could finally deliver the AI sovereignty Europe has long sought. It creates a bridge between elite research talent and real-world ambition that stands apart from American and Chinese dominance.
World Models: The Path Beyond Token Prediction
LeCun has consistently maintained that large language models (LLMs) are fundamentally limited. Their core mechanism is predicting the next token (a word, subword, or pixel), which enables strong performance in language generation, question answering, coding, and related symbolic tasks. However, this approach is insufficient for developing broadly capable intelligent agents.
AMI offers an alternative. It creates world models that construct internal representations of physical reality from high-dimensional inputs like video, sensor data, spatial information, and interactions. These models encode object permanence, intuitive physics, cause-and-effect relationships, and outcomes in continuous, uncertain settings. These are the foundations for human-like intelligence that go beyond linguistic pattern-matching.
Central to this is LeCun’s Joint Embedding Predictive Architecture (JEPA), introduced in 2022. Unlike typical models that try to recreate every detail (every pixel or word), JEPA predicts only the important abstract patterns in a simplified “latent space,” ignoring unimportant noise. This makes learning much more efficient and helps the system build strong, reliable understanding of the real world.
Early results from JEPA-based work demonstrate strong performance in video understanding, intuitive physics prediction, and planning under uncertainty. The approach prioritizes efficiency and trustworthiness, key for embodied AI that must act safely in the real world.
Investors Betting Big on the Long Game
The $1.03 billion seed round attracted a star-studded list of backers. Co-leads included Bezos Expeditions, Cathay Innovation, Greycroft, Hiro Capital, and HV Capital. Other major participants featured Nvidia, Toyota Ventures, Samsung, Temasek, Sea, and prominent individuals like Eric Schmidt, Mark Cuban, and more.
For a company only months old with a relatively small (yet star-studded) team, this massive raise shows strong investor confidence in world models as the next big step in AI.
The funding fuels heavy R&D: massive compute power, large-scale datasets, and long-term timelines. It also supports fast hiring in Paris (HQ), New York, Montreal, and Singapore. AMI has committed to open-sourcing parts of its work and forming early partnerships, including one with Nabla in healthcare, to speed up progress and create real impact.
In short, these investors are placing a big bet on grounded, controllable AI systems rather than faster versions of today’s language models.
Industry Echoes: Praise, Skepticism, and Sovereignty Buzz
The March 2026 debut sparked fervent reactions. French President Emmanuel Macron praised AMI on X as a milestone for “the France of researchers, builders, and the bold.”
LeCun himself highlighted AMI’s unique position: “there is certainly a huge demand from the industry and governments for a credible frontier AI company that is neither Chinese nor American. I think that is going to be to our advantage.”
Backers echoed the optimism. Pierre-Éric Leibovici, founder and Managing Partner at Daphni (an early investor), declared: “AMI Labs could be the first European company to reach the scale of the GAFAM companies.” This frames AMI as a genuine shot at building a homegrown giant on par with American and Chinese AI dominance.
Critics, however, remain vocal about the risks. The long development horizon (explicitly 5–10+ years with no near-term products or revenue) combined with the eye-watering $3.5 billion pre-money valuation, has drawn skepticism.
Many observers argue that the massive funding supports a contrarian research project rather than a fast-paced commercial company. JEPA-style ideas have been around for years without becoming mainstream, so some question why this approach suddenly warrants such a large bet now.
This divide underscores AMI’s deeper significance. It represents a bold, high-profile gamble on Europe’s ability to innovate independently in frontier AI, rather than merely relying on or adapting models from the U.S. and China.
AMI as Europe’s Ticket to AI Sovereignty
Advanced Machine Intelligence arrives with unmatched momentum: record funding, a stellar team, a compelling technical alternative to LLM dominance, and early momentum via partnerships.
Whether world models become the breakthrough or a respected parallel path, AMI has already achieved something profound by planting a credible, Paris-centered flag in the global AI landscape.
For Europe, the stakes transcend technology. As AI power concentrates in a few U.S. and Chinese giants, dependency risks grow. AMI offers a genuine path to EU AI sovereignty: frontier research rooted in European values of safety, transparency, ethics, and open access. Backed by diverse investors and prioritizing controllable systems, it provides governments, industries, and developers with a credible alternative.
If LeCun’s bet pays off, AMI could mark the turning point where Europe shifts from AI follower to leader.
Author: Grace Sharp
See Also:
How Mistral’s ‘Le Chat’ Became Europe’s AI Darling
