The artificial intelligence chessboard is shifting again: Yann LeCun is considering leaving Meta to launch his own technology company, according to the Financial Times. The potential departure of the French-American scientist, one of the pioneers of deep learning, would come amidst the race for generative AI and with the sector adjusting its pace between research and product.
Sources close to the matter indicate that it has already is holding preliminary financing talks for the new project. In parallel, within Meta, the organizational shift towards more immediate results has accelerated, a context that would have pushed LeCun to explore an independent path with his own vision of AI.
Why now: The new balance of power in Meta's AI
The Menlo Park company reshaped its AI strategy with the creation of Meta Superintelligence Labs (MSL)A unit focused on accelerating product launches. The move elevated Alexandr Wang (Scale AI) to head of the area and placed LeCun reporting directly to him, a profound change in the traditional research organizational chart.
The key operation of the summer was Meta's entry into Scale AI with a relevant stakewhich catapulted Wang to the center of the strategy. This same push also brought in Nat Friedman (formerly of GitHub) to strengthen the new division, giving more traction to business-oriented approaches over long-term projects.
The shift had internal consequences: in the fall, approximately 600 jobs linked to MSLwith a particular impact on FAIR, the historic lab founded by LeCun in 2013. In contrast, the elite team known internally as TBD, backed by Mark Zuckerberg and bolstered by recruits from OpenAI, Anthropic, Google, and DeepMind, It was left safe and reinforced.
Since then, Meta has prioritized deploying models and services at a faster pace. The lukewarm reception of Call 4 and the Meta AI chatbot In the face of industry rivals, pressure has intensified to accelerate the conversion of research into product, reducing the space for more exploratory scientific lines.
What LeCun is after: a commitment to “world models”
The new startup would revolve around the so-called world modelsArchitectures designed to enable machines to construct a coherent representation of physical and social reality. This approach aims to go beyond simply predicting the next word: it seeks an AI with contextual understanding, learning from experience, and the ability to act with common sense.
In LeCun's vision, the next generation of systems should incorporate four pillars: understanding of the environmentpersistent memory, genuine reasoning, and complex planning. The goal is to enable agents capable of relating vision, language, and spatial cues to make informed decisions, not just generate plausible text.
This field already has some powerful rivals: World Labs, powered by Fei Fei LiLeCun recently secured a significant funding round for world model-based technologies. His new company would operate in that same league, leveraging scientific traction and offering a distinct narrative compared to large, purely generative models.
According to leaks, the initial team would be made up of former FAIR researchers and European academicsAnd initial discussions with funding bodies are already underway. Operational details—location, structure, and roadmap—have not yet been made public.
Impact on Meta, the market and the European ecosystem
For Meta, the departure of their scientific role model would be a symbolic and strategic blowFew figures combine LeCun's academic prestige with the experience of managing a leading laboratory. His departure could spark an internal debate about the importance of fundamental research at a time of intense pressure to scale up product development.
In the market, recent adjustments and sensitivity to computing costs have put [the issue] under scrutiny. the sustainability of AI spendingThe news of LeCun's plans came at a time of volatility for Big Tech, with investors waiting to see if investments translate into real traction and revenue.
Europe is watching with particular interest. A project led by LeCun could attract talent and capital from the continentThis strengthens research centers independent of US Big Tech. For Spain and other European economies, riding this wave—in training, supercomputing, and technology transfer—is an opportunity to scale their role in the AI value chain.
Neither Meta nor LeCun have made official comments Regarding the timeline or the details, the shift in the cycle is clear: if the project materializes, the competition for top-tier engineers and computing infrastructure will intensify, and we will see new alliances between universities, laboratories, and venture capital on both sides of the Atlantic.
What's coming in the short and medium term
In the short term, the focus will be on the seed funding and founding core of the startup. In the medium term, the challenge will be to demonstrate that world models can be translated into products with tangible advantages in robotics, advanced assistants, and systems capable of planning and executing complex tasks.
If Meta accelerates its roadmap in large models and LeCun materializes its vision with convincing prototypes, the sector could bifurcate between general-purpose, consumer-oriented AI and agents with a deep understanding of the environment. Two complementary speeds that will define the next phase of the race.
LeCun is planning his exit and a new company focused on world models, Meta is redoubling its focus on product under MSL and The race for talent intensifiesEurope has options to gain influence if it effectively coordinates investment, computing, and scientific transfer.