Paris-based Arlequin AI SAS today announced it raised €28 million, about $32 million, in new funding to accelerate development of a new AI model architecture based on a different type of network that learns complex relationships at large scale.
The new AI architecture is based on topological neural networks, rather than the graph-based neural networks that underpin most large language models. Arlequin said it has already built and implemented an easy-to-deploy, scalable platform that can use heterogeneous data, analyzing documents, transactions, video and operational information.
European investors exclusively backed the Series A round, co-led by Redalpine and OTB Ventures, alongside Bpifrance’s Defense Innovation Fund. Existing investors Vsquared Ventures and 10x Founders increased their stakes, while Xavier Niel and Zebox also joined the round.
The funding will accelerate scaling a proprietary model that learns from the structure and relationships within data by capturing complex, multipath interactions at large scale.
“Today, another revolution is taking shape: the development of new AI systems capable of understanding highly complex dynamics hidden within millions of data points,” said co-founder and Chief Executive Hugo Micheron. “At a time when we are overwhelmed by information, we need to regain control.”
Micheron added that Arlequin is building this new AI architecture as Europe competes in the market against national technological giants such as the United States and China to develop powerful AI models. The funding comes as another European open-source AI veteran, Mistral AI SAS, received €3 billion this week.
Arlequin said TNN networks are designed to learn not only from individual data points, but from how data is connected at large scale, including multiple elements at once. The company seeks to develop systems that can analyze increasingly large systems and trace end results back to root causes.
Examples include counterterrorism investigations, where the platform can analyze billions of data points generated by thousands of seized devices to identify connections among people, locations, communications and events. The company said the TNN-based model analysis approach can also apply to security and defense, criminal investigations, fraud and money laundering, information integrity, cybersecurity, and AI safety and security. All of these areas rely heavily on predictive algorithms, but evidence and auditable paths help reveal what actually happened.
Arlequin is also designing the TNN architecture to use significantly less compute. This will reduce reliance on energy, advanced hardware and compute infrastructure. As more data centers grow around the world, especially within countries developing sovereign AI, the expense of running AI models and the cost of individual tokens have become top of mind for enterprises and governments.
The company is developing the new architecture in collaboration with research 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 with teams at the universities of Oxford, Cornell and Princeton and the University of California at Santa Barbara.
With the new funding, Arlequin intends to expand its international team, which is developing its proprietary models and accelerate commercial deployment across Europe and worldwide. The company has opened offices in London and Berlin, and plans to establish an AI lab in Silicon Valley in the coming months.
