AMI Labs raises €890 million to build a new generation of AI based on “world models.”

As generative artificial intelligence becomes increasingly embedded in real-world applications, a scientific debate is now emerging within the research community: can current language models, based on statistical text prediction, truly lead to an artificial intelligence capable of understanding the world? For Yann LeCun, the answer is no. One of the pioneers of modern AI has for several years advocated a different approach built around what he calls “world models,” systems designed to model the dynamics of the real world rather than simply generate text.

It is around this vision that AMI Labs has been launched, a new artificial intelligence laboratory that has raised €890 million in funding at a €3 billion pre-money valuation to develop a new generation of systems capable of learning from their environment and anticipating the consequences of their actions. By the scale of its financing and the scientific ambition it displays, the initiative immediately positions itself among the most ambitious AI projects launched in Europe.

An AI laboratory structured as a global scientific organization

AMI Labs is chaired by Yann LeCun, a leading figure in artificial intelligence research and a Turing Award laureate. The operational execution of the project is entrusted to Alexandre LeBrun, an experienced AI entrepreneur who serves as CEO.

Around them, the organization is structured like a truly international research laboratory. The role of Chief Science Officer is held by Saining Xie, a specialist in visual representation learning in machine learning. The direction of research and innovation is entrusted to Pascale Fung, a pioneer of human-centered AI approaches. Work on world model architectures will be led by Michael Rabbat.

Operational organization is overseen by Laurent Solly, who has been appointed COO. His role will include structuring the laboratory’s development, its international operations, and its industrial partnerships.

The governance structure reflects the project’s ambition to combine fundamental research, advanced engineering, and the capacity for industrial deployment.

“World models”: an alternative to dominant architectures

Since the rise of generative models, most major advances in AI have been driven by large language models (LLMs). These systems primarily learn by analyzing vast volumes of text in order to predict the most probable words in a given sequence.

For Yann LeCun, however, this approach reaches structural limits. Generative models may produce coherent responses, but they remain unable to understand the causal relationships that govern the physical and social world.

World models aim precisely to address this limitation. The objective is to design systems capable of building an internal representation of their environment and anticipating how situations will evolve.

Such an approach could enable major breakthroughs in areas such as:

• autonomous robotics
• scientific simulation
• intelligent agents
• decision-making systems

In other words, while LLMs specialize in content generation, world models seek to model reality itself.

An oversubscribed funding round

AMI Labs has secured €890 million in financing, an exceptional amount for a project still at the launch stage. According to several sources close to the deal, investor interest significantly exceeded the final size of the round, with commitments far surpassing the amount ultimately retained.

This oversubscription led several participants to reduce their initial allocations in order to allow a broader group of strategic investors to participate.

The funding round brings together a mix of capital sources including venture capital firms, institutional investors, and industrial partners. Participants include Temasek, SoftBank, Greycroft, New Legacy Ventures, and Nvidia, whose presence reflects the growing importance of compute infrastructure in the development of AI models.

Among European investors are Cathay Innovation and Daphni, which appear as the leading French institutional investors, followed by Serena, Eurazeo, and Bpifrance.

The diversity of the investor base reflects the hybrid nature of the project, which sits at the intersection of science, industry, and strategic infrastructure.

The global race for AI infrastructure

Beyond fundamental research, a significant share of the capital raised is expected to be devoted to the computing power required to train AI models.

Modern AI increasingly relies on heavy infrastructure: GPU clusters, high-performance networks, and specialized data centers. In this context, access to compute capacity is becoming a decisive factor of competitiveness.

A symbolic project for the European ecosystem

The ambition of AMI Labs is part of a broader effort to strengthen Europe’s capacity to fund and develop large-scale artificial intelligence research laboratories.

Until now, initiatives of this scale have largely emerged in the United States, through organizations such as DeepMind, OpenAI, and Anthropic. The creation of AMI Labs therefore represents a European attempt to build a laboratory capable of competing on both scientific and technological grounds.

While the success of such a project remains uncertain, the scale of the funding secured and the caliber of the team assembled signal a shift in ambition within the European ecosystem. In an industry where the race for computing power and scientific talent is intensifying, AMI Labs already stands as one of the boldest bets in European artificial intelligence.