From Mexico — this afternoon I’m not rehashing another US frontier-lab product drop. Paris-based Arlequin AI just closed a €28 million Series A (about $32 million, per SiliconANGLE) to push a different bet: topological neural networks that learn from how data is connected, not only from what each record says. Per the company announce (Sep 10), the round is European-only capital.
Tech.eu and the company post line up on the leads: redalpine and OTB Ventures co-led, with Bpifrance’s Defence Innovation Fund in. Existing backers Vsquared Ventures and 10x Founders increased. New names on the company wire: Xavier Niel and ZEBOX (CMA CGM’s accelerator). redalpine’s own note frames it as sovereign safety/security tech, not a ChatGPT clone.
What they’re actually building
Founded in Paris in 2024 by Hugo Micheron (CEO) and Antoine Jardin (CTO, ex-CNRS), Arlequin already ships a platform that stitches documents, transactions, video, and operational data and tries to keep results traceable to evidence. The Series A cash accelerates the next step: proprietary TNN models meant to catch multi-way relationships that get messy as graphs scale. I’m citing that as written on arlq.ai — not inventing benchmarks.
The pitch is bluntly complementary to LLMs. Large language models changed generation. Arlequin says it wants architecture for structure and dynamics across millions of linked points — counterterrorism-style link analysis, fraud/AML, information integrity, cybersecurity, AI-system safety — where “probably” isn’t enough and auditors want a path back to source. SiliconANGLE notes the same application list and the claim of significantly less compute than transformer-scale stacks. Treat that as company claim until independent evals land.
Europe’s other AI path — and the scale still coming
The company says the stack is already used by governments and large organizations across Western and Eastern Europe, with research ties named to INRIA, CNRS, Max Planck, plus Oxford, Cornell, Princeton, and UC Santa Barbara. Headcount on the announce: about 50 people (35 engineers, 15 PhDs/researchers/data scientists). Offices in Paris, London, and Berlin; a Silicon Valley AI lab planned by end of 2026. Same primary wire: Arlequin.
I’m reading this next to this week’s giant European LLM raise (Mistral) as a reminder: sovereignty talk isn’t only “host the same architecture on EU soil.” Sometimes it’s originate a different architecture for problems where hallucination and opaque scores are unacceptable. Whether TNNs deliver that at production scale is the open question — the €28M just bought them runway to try. Primary wires: company, SiliconANGLE, Tech.eu, redalpine.
Hero image: network cables and fiber by JJ Ying on Unsplash (Unsplash License). Cropped, graded, and lightly grained by Tech & AI Pulse. Face-free still — no people, no logos.
