Yann LeCun, the man behind key breakthroughs in deep learning and a Turing Award winner, is not satisfied with the direction the AI industry has taken. While the rest of the world competes to build ever-larger language models, he is convinced that these will never produce human-level intelligence — and now he's putting his money where his mouth is.
World Models Versus Text Prediction
The core of LeCun's criticism is that large language models (LLMs) fundamentally only predict the next word in a sequence. They lack the ability to understand how the physical world actually works — cause and effect, spatial context, and the consequences of actions over time.
AMI Labs, the company LeCun co-founded with CEO Alexandre LeBrun, will take a radically different approach. According to research notes and the company's own descriptions, the technical foundation is LeCun's Joint Embedding Predictive Architecture (JEPA) — an approach that trains AI systems on video and spatial data to build abstract representations of real-world environments.
The goal is for the system to be able to reason about what will happen before an action is performed — a kind of mental simulation that more closely resembles how humans and animals navigate the world.
The goal is AI that can 'think through' the consequences of actions before they happen — not just predict the next word in a sentence
Record-Breaking European Seed Round
To realize this ambition, AMI Labs has raised impressive capital. The company, based in Paris, closed a seed round of $1.03 billion — which, according to available sources, is among the largest in European startup history. The pre-money valuation is set at $3.5 billion.
The investor list is not just anyone: NVIDIA, Jeff Bezos, Samsung, Toyota Ventures, Temasek, and Mark Cuban are among those who have invested. The fact that NVIDIA is involved is striking, given that the company is simultaneously developing its own world models through the Cosmos program — trained on over 20 million hours of robotics and driving data.
Competition from Tech Giants
AMI Labs is not alone in betting on world models. The field is rapidly growing and is described by industry analysts as a market with potential exceeding $100 billion.
Dr. Fei-Fei Li, often referred to as the "grandmother of AI's" counterpart in the US, leads World Labs, which reached a $1.25 billion valuation within just four months of its launch. The company focuses on true 3D understanding rather than traditional video generation.
Ambitions in Industry and Healthcare
AMI Labs will initially target sectors where the limitations of traditional LLMs are most evident: manufacturing, robotics, aerospace, and biomedicine. The company has already entered into an exclusive strategic partnership with the healthcare company Nabla — where LeBrun also serves as chairman — with the aim of integrating world model technology into future clinical products.
LeCun himself has signaled that the first year will be research-intensive, and that product development should be measured in years — not quarters. This is a deliberate tempering of expectations in a market where many companies promise more than they deliver.
Critical Perspective
It is worth noting that LeCun's claims about the fundamental limitations of language models are controversial within research communities. Many prominent AI researchers believe that scaling LLMs can still yield surprising results, and that the limits of what these models can achieve are far from being fully explored. AMI Labs has also not yet published results or demonstrations to substantiate that the JEPA approach delivers on its promises.
It remains to be seen whether the venture will convince on technical grounds, or if it primarily reflects faith in one of the industry's most influential individuals.
Sources: Digi.no, AMI Labs company information via research