Arlequin AI raises €28 million for topological AI models

Arlequin is not competing on the same ground as the large language models most businesses already encounter. Its bet is on a different architecture — topological neural networks, which it describes as learning from how data is connected rather than from individual data points alone. The company says it has already built and implemented an easy-to-deploy, scalable platform that can use heterogeneous data, analyzing documents, transactions, video and operational information.

The close ties to public research are a notable part of the story. Arlequin says it is developing the architecture with teams at the French National Institute for Research in Digital Science and Technology, the French National Center for Scientific Research and Max Planck Institute, as well as Oxford, Cornell, Princeton and the University of California at Santa Barbara. That network, plus backing from Bpifrance's Defense Innovation Fund, points to a European sovereignty angle that CEO Hugo Micheron made explicit, framing the work as Europe competing against the United States and China to develop powerful AI models.

The outcome hinges on whether the claimed compute savings and auditable evidence paths hold up in real deployments. If they do, the appeal for security, defense, fraud and money-laundering teams could be considerable, since those areas rely heavily on predictive algorithms but need evidence and auditable paths to show what actually happened. The broader context is that capital is flowing to European AI: the round lands the same week Mistral AI SAS received €3 billion, suggesting funders see room for more than one European approach.