Cathay Capital led a $1.03 billion seed round financing for AMI Labs.

On March 10, AMI Labs officially announced the completion of a approximately $1.03 billion seed round of financing, which is the largest seed round in European history. This round of financing was jointly led by Cathay Innovation Fund under Cathay Capital, Greycroft, Hiro Capital, HV Capital, and Jeff Bezos Expeditions. Meanwhile, Temasek, NVIDIA, Toyota Ventures, Samsung, Bpifrance Digital Venture, the digital venture capital department of the French National Investment Bank, Eric Schmidt, Tim Berners-Lee, etc. participated in the investment.

Cai Mingpo, the founder and chairman of Cathay Capital, said, "Cathay has always believed that technologies that can truly transcend economic cycles must ultimately return to the real world, to industries, and to the real needs of people. Today, AI has demonstrated great potential in information processing and knowledge work. However, greater opportunities in the future lie in how it can enter more complex and real systems. The direction explored by AMI Labs shows us that AI is evolving from 'being able to express' to 'being able to understand', and from the digital world to the real world. For Cathay, what we focus on is not just the advancement of the technology itself, but also whether it can be integrated with industries and real scenarios in the long term to ultimately create sustainable value. AMI Labs is a team with scientific ideals, engineering capabilities, and a global perspective. We are very glad to accompany them on this longer, more difficult, but also more meaningful journey."

What exactly is AMI Labs doing?

AI has made significant progress in the past decade. Prediction and generation systems have changed the way we analyze information, acquire knowledge, and create content globally. Today, when AI steps out of the screen, intelligence no longer stops at simply generating results. It must understand the context, retain the context information, predict the results, and make more reliable decisions over time.

To achieve this goal, AMI will develop a new generation of AI systems that can understand the world, have long - term memory, be capable of real reasoning and planning, and be end - to - end secure and controllable.

AMI defines itself as a cutting - edge AI company focusing on foundational world models. Its core goal is to build a new type of intelligent system that can understand the environment, retain context, conduct reasoning and planning, and operate stably under complex constraints.

In terms of the technical path, the direction advanced by AMI is in line with the JEPA (Joint Embedding Predictive Architecture) long advocated by Yann LeCun. Different from large language models centered around 'next token prediction', JEPA emphasizes enabling the system to learn abstract representations of the real world and predict state changes and results in the latent space.

If large language models mainly deal with expressed information such as text, images, code, and speech, then what AMI focuses on is how to enable AI to form an internal representation of the real world. Because the real world is not a static collection of information, but a dynamic system that is continuously changing, full of noise and feedback. For such systems, the key issue is not just 'generating results', but 'understanding how the state changes, what consequences actions will bring, and how to make continuous decisions under uncertain conditions'.

This is exactly the significance of the world model approach. AMI attempts to make the model learn the abstract structure of the real world, model causal relationships on this basis, predict system evolution, and support higher - quality planning and execution accordingly. Therefore, it corresponds not to a single consumer - level generation scenario, but to complex application fields with higher requirements for reliability, controllability, and security, such as industrial automation, robotics, healthcare, and wearable devices.

Why did AMI Labs cause a sensation overnight?

The reason why AMI Labs quickly attracted global attention is not simply because 'this is the largest seed round in Europe'. The deeper reason is that it has simultaneously hit three key points: influential people, an important technical path, and a clear - cut contemporary problem.

1. Behind this company stands a group of people who are capable of defining the industry direction.

The founding team of AMI Labs has known Cathay Capital for many years.

AMI Labs was founded by Yann LeCun, a Turing Award winner and former chief AI scientist at Meta. For a long time, he has been one of the most representative thinkers on the core issue of 'whether LLM is sufficient to achieve true intelligence'. In a sense, AMI represents Yann LeCun's long - held technical judgment, which is for the first time being brought to the forefront in a corporate, engineering, and capitalized way.

Meanwhile, the core team of AMI is not just a single star, but a well - structured team: Xie Saining, a top expert in basic AI research and an old friend and colleague of Yann LeCun at school, has officially joined AMI as the Chief Science Officer;

Xie Saining is an absolute authority in visual representation learning and one of the co - authors of diffusion transformers (DiT). The introduction of the DiT architecture enables visual models to benefit from the Scaling Law just like large language models. By replacing the U - Net that had been used for a decade with a Transformer backbone, Xie Saining and others' work has made it possible to simulate complex and high - fidelity images/videos, laying the foundation for the launch of top - level visual generation models and tools such as Sora and SeedDance.

Alexandre LeBrun serves as the CEO of AMI. He graduated from the École Polytechnique and is a well - known serial entrepreneur. The companies he founded are all focused on bridging the gap between 'basic scientific research and the real world'. His first company, VirtuOz, was engaged in enterprise dialogue robots and was founded around 2002 when 'AI' was far from being a popular term. The company was later acquired by Nuance (which was subsequently incorporated into Microsoft). Then there was Wit.ai, a natural language understanding company, which was acquired by Facebook in 2015. Subsequently, LeBrun entered Meta to lead the engineering work at FAIR Paris. After leaving, he founded Nabla (an enterprise invested by Cathay Capital), which develops AI medical assistants and has obtained a considerable number of hospital clients.

Michael Rabbat, who is in charge of world model research, will be responsible for the Montreal office at AMI. Rabbat is one of the founding members of the former FAIR Montreal laboratory. After teaching at McGill University for more than a decade, he joined Meta full - time and led the research and development of three world model series: I - JEPA, V - JEPA, and V - JEPA 2. V - JEPA 2 has the greatest influence: through video self - supervised training, with less than 62 hours of robot operation data, it can control a robotic arm to complete a grasping task in a completely unfamiliar laboratory environment with zero - shot learning. The core logic of this method is directly connected to the technical path of the entire AMI company.

Pascale Fung, a well - known Chinese computer scientist, is the CRIO (Chief Research and Innovation Officer) of AMI. She is a fellow of top academic institutions such as AAAI, IEEE, and ACL, and a chair professor at the Hong Kong University of Science and Technology. She was born in Shanghai and has studied at Kyoto University, the French National Center for Scientific Research, and Columbia University. Her later work at Meta FAIR focused on embodied AI and visual - language world models, with the main application scenario being smart glasses.

Before joining AMI, COO Laurent Solly was in charge of Meta's business in France, Southern Europe, and Europe for a long time. He also worked at the TF1 Group and the French public sector. Therefore, he has rare experience in organizational building, cross - regional operation, and connecting the European industrial and public ecosystems.

The rarity of this team lies in that it has for the first time integrated cutting - edge scientific research, systems engineering, and global organizational capabilities into the same framework. This is also the biggest difference between AMI and many cutting - edge projects that 'have strong research but are still far from real - world implementation'.

2. It focuses on an ongoing transfer of capabilities.

In the past two years, large models have proven that machines can greatly improve information processing efficiency: writing, summarizing, answering questions, programming, searching, creating... These tasks essentially take place in the digital world. However, when AI is to truly enter fields such as robotics, manufacturing, experimental processes, industrial systems, and infrastructure, the problem has fundamentally changed.

The system should not only be able to 'answer', but also be able to judge the state, retain context, infer consequences, plan multi - step actions, and operate continuously in an environment full of noise and constraints. In other words, for AI to truly enter the real world, it must evolve from 'being able to generate' to 'being able to understand, infer, and act'.

The importance of AMI's world model approach lies in the fact that it directly addresses this issue. It is not concerned with how to further enhance language capabilities, but rather how to enable machines to form an internal cognitive structure of the real world.

3. AMI has for the first time seriously presented a technical path that has long remained in academic discussions to the market.

This is also what differentiates it from many AI companies that 'lead with concepts'. AMI is telling the market in a very specific way that this path requires long - term research, global talent, a heavily - invested engineering system, and patient capital support.

Since its establishment, it has simultaneously laid out operations in Paris, New York, Montreal, and Singapore; the list of investors spans financial capital, industrial capital, and heavyweights in the technology industry. The lineup, including Cathay Capital, itself is a signal: the market is not just observing an 'interesting new concept', but is seriously supporting a path that may affect the basic capabilities of the next - generation AI.

What is closer to real intelligence after large models?

What really triggers discussions in the professional circle about AMI is the fundamental question behind it: After large models, what is the direction closer to 'real intelligence'?

The answer represented by Yann LeCun and AMI is very clear: True intelligence does not start with language, but with the world. AMI's official website even writes this sentence directly - "real intelligence does not start in language. It starts in the world."

This does not mean that language is unimportant. Instead, it means that language is essentially just a projection of the world. If a machine only learns to manipulate the projection without learning to understand the world itself, its ability boundaries in the real environment will soon become apparent.

In other words, LLM has brought AI into the era of large - scale usability; while world models are driving AI from 'being able to express' to 'being able to understand, infer, and act'.

From this perspective, the reason why AMI has attracted so much attention is precisely because it has tackled the most difficult and fundamental problem in the AI industry at present.

Conclusion

From a longer - term technology evolution cycle perspective, the value of AMI does not only lie in completing a rare large - scale seed round of financing, nor in assembling a highly influential team. Its more important significance lies in enabling a technical direction that originally remained more in the academic and cutting - edge research fields to enter the industrial vision in a clearer corporate and engineering way.

In the past few years, large language models have verified the wide applicability of generative AI and reshaped people's imagination of 'intelligent systems'; however, when AI begins to move from the screen to real - world systems, the focus of the industry will inevitably change. What determines the upper limit of the next stage may no longer be just language ability, generation quality, or interaction efficiency, but rather the system's ability to model the environment, understand causal relationships, and conduct reasoning, planning, and execution under complex constraints.

In this sense, what AMI represents is not a simple supplement to the existing path, but a more fundamental extension of capabilities: it attempts to answer not 'what else the model can generate', but 'how an intelligent system can truly enter the real world'.

This is also the fundamental reason why AMI deserves continuous attention.

This article is from the WeChat official account "Cathay Communications", author: Cathay Capital. Republished by 36Kr with permission.