What Happened
AMI Labs, co-founded by Yann LeCun and led by CEO Alexandre LeBrun (former CEO of medical AI company Nabla), secured $1.03 billion in March 2026 to develop what researchers call “world models” - AI systems designed to understand how the physical world operates. The funding round, which values the company at $3.5 billion before the investment, marks the largest seed funding round for a European AI startup in history.
The company is building on LeCun’s Joint Embedding Predictive Architecture (JEPA), a fundamentally different approach from the large language models (LLMs) like ChatGPT that currently dominate AI development. Instead of predicting the next word in a sequence, JEPA learns to understand physical relationships and predict how objects and systems behave in the real world.
LeCun, who won the Turing Award in 2018 for his pioneering work on deep learning and spent years as Meta’s chief AI scientist, has long argued that true artificial general intelligence will require machines that understand physics, not just language patterns. AMI Labs represents his attempt to prove this thesis with substantial financial backing.
Why It Matters
This development challenges the current AI paradigm dominated by language models like OpenAI’s GPT series and Google’s Gemini. While these systems excel at text generation and conversation, they lack fundamental understanding of how the physical world works - a limitation that becomes apparent when trying to build truly intelligent robots or autonomous systems.
“Current AI systems are essentially very sophisticated autocomplete,” explains the technical challenge. “They can predict what word comes next in a sentence, but they can’t predict what happens when you drop a ball or how a lever works. World models aim to change that.”
For the technology industry, this represents a potential shift toward “embodied AI” - artificial intelligence that can interact meaningfully with physical environments. This could unlock breakthroughs in robotics, autonomous vehicles, industrial automation, and scientific modeling that current language-focused AI cannot achieve.
The massive funding also signals growing investor confidence that the next major AI breakthrough may come from understanding physics rather than processing more text data. With European investors leading this round, it also represents a significant challenge to the US-China duopoly in AI development.
Background
LeCun’s skepticism of language-model-centric AI development has been growing for years. During his time at Meta, he frequently argued that human-level intelligence emerges from our understanding of the physical world, not our ability to process language. Babies, he noted, learn about gravity, object permanence, and cause-and-effect relationships long before they learn to speak.
The JEPA architecture that AMI Labs is building upon represents years of research into how AI systems can learn more efficiently by understanding relationships between objects and events rather than predicting tokens in a sequence. This approach potentially requires far less training data than current LLMs, which need massive text datasets to achieve competence.
LeBrun brings complementary experience from the medical AI space, where understanding physical biological processes is crucial for accurate diagnosis and treatment recommendations. His previous company, Nabla, focused on AI applications that required understanding of real-world medical scenarios rather than just medical text processing.
The timing coincides with growing recognition of LLM limitations. Despite impressive capabilities in conversation and text generation, these systems struggle with basic physical reasoning, spatial understanding, and causal relationships that humans intuitively grasp.
What’s Next
AMI Labs faces significant technical and commercial challenges. Building AI systems that understand physics is considerably more complex than processing language patterns. The company must develop new training methodologies, create appropriate datasets, and prove that their approach can scale to real-world applications.
The immediate focus will likely be on robotics applications, where understanding physical interactions is essential. Success could revolutionize manufacturing automation, household robotics, and autonomous vehicle navigation. However, the research timeline is expected to be measured in years rather than months.
Investors and industry observers will watch for early demonstrations of JEPA’s capabilities, partnerships with robotics companies, and publications in top AI research venues. The company’s progress could influence the broader AI research community’s priorities and funding decisions.
Competition will come from established players like Google DeepMind, OpenAI, and Anthropic, all of which are exploring multimodal AI that combines language with visual and physical understanding. However, AMI Labs’ singular focus on world models could provide a research advantage over companies balancing multiple AI approaches.
The success or failure of this approach could determine whether the AI industry continues its current trajectory of scaling language models or pivots toward embodied intelligence as the path to artificial general intelligence.