Arlequin AI has raised €28M ($32M) in Series A funding to commercialise an AI architecture built on topological neural networks rather than the graph-based designs behind most large language models.
Redalpine and OTB Ventures co-led the round, with Bpifrance’s Defence Innovation Fund participating. Existing backers Vsquared Ventures and 10x Founders increased their stakes, and Xavier Niel joined.
Founded in 2024 by terrorism specialist Hugo Micheron and former CNRS data scientist Antoine Jardin, the Paris company builds systems for organisations making consequential decisions from fragmented datasets. Its platform analyses documents, transactions, video and operational data together, and traces conclusions back to source evidence.
Topological networks learn from how data points connect at scale, capturing multi-path interactions that grow hard to interpret as datasets expand. Arlequin applies this to counterterrorism investigations, fraud and money laundering detection, information integrity and cybersecurity, with governments across Western and Eastern Europe already using it.
Compute is part of the pitch. The company says the architecture consumes significantly less energy and hardware than frontier-scale models, a point that resonates as sovereign AI programmes confront the cost of tokens and data centres.
Arlequin works with researchers at France’s national digital science institute, the CNRS and the Max Planck Institute, and argues that further AI progress will come from different designs rather than simply larger models trained on more data.
