Arlequin AI Secures 28 Million Euros for Its Neural Networks

The startup founded by Hugo Micheron and Antoine Jardin aims to provide a European alternative to decision-making platforms like Palantir. It is shifting its technological focus to develop proprietary topological neural networks and has secured €28 million to train its models and accelerate deployments.

Arlequin Targets the Decision-Making Platform Market Like Palantir

Arlequin AI is not primarily looking to replace ChatGPT, Claude, or Le Chat. The company targets platforms that connect fragmented data to assist in sensitive decision-making, a segment built over the past twenty years by Palantir. At Palantir, an ontology links an organization’s data to real-world objects, serving as an operational layer to analyze, apply rules, or trigger actions. Arlequin embraces this comparison and presents itself in its recruitment as "the sovereign European alternative to Palantir." The distinction from LLMs positions the project but does not define its market, and Palantir does not market an identical model.

An Architectural Bet: Topological Neural Networks Instead of LLMs

Arlequin claims that its approach differs from the current trend of favoring large models. According to Antoine Jardin, the next major steps in AI will depend on the emergence of new architectures rather than the continuous increase in the number of parameters, the amount of data, or computing power; for his part, Hugo Micheron mentions systems capable of understanding complex dynamics within millions of data points. While LLMs learn sequence regularities and graph networks learn relationships through edges, topological models aim to represent simultaneous higher-order relationships, including groups, cycles, and hierarchies. Examples include multi-entity configurations in anti-money laundering or, in logistics, chains linking supplier, port, ship, geopolitical incidents, and delays. Arlequin emphasizes a promise of traceability and seeks to understand phenomena through their structuring relationships. The translation into computable architecture involves transforming raw data into topological structures and determining which elements to connect, at what threshold, and how to distinguish a real group from a coincidental proximity.

Initial Publications but No Public Validation of Models Yet

The startup has begun to make its scientific activity visible. Antoine Jardin and Rémi Devaux co-authored an article in August 2026 with the Geometric Intelligence Lab presenting TopoExplorer, a tool that visualizes the transformation of a dataset into a topological structure before training. Correlations between certain upstream indicators and the performances obtained afterward have been observed across twenty benchmark datasets. This work aligns with the promise of interpretability advanced by Arlequin, but it does not yet constitute a validation of the performance of the proprietary models the company aims to develop.

Arlequin AI Raises €28 Million from European Funds

Arlequin AI has completed a Series A funding round of €28 million co-led by Redalpine and OTB Ventures, with participation from the Fonds Innovation Défense of Bpifrance. Vsquared Ventures and 10x Founders, already investors, have increased their stakes, and Xavier Niel has joined the round. Redalpine invests at the intersection of software and science, OTB Ventures targets AI and new architectures, and the entry of the Fonds Innovation Défense positions the project within dual-use technologies. This round brings the total funds announced since the company's inception to at least €32.4 million. Unlike the previous round, the funding's goal is now to design proprietary models based on topological neural networks, rather than building an analytical platform. This shift in focus requires investing in longer-term research, establishing benchmarks, and attracting rare scientific profiles. The capital will be used to strengthen the team, train the models, and support commercial deployments.

From the HuDEx Prototype to First Clients and Seed History

In June 2025, Arlequin raised €4.4 million from Vsquared Ventures, 10x Founders, Kima Ventures, Better Angle, and individual investors. This initial funding aimed to industrialize HuDEx, enhance on-site installations, and double a research team that then consisted of about twenty people. The company indicated it was working with ministries, Radio France, BNP Paribas, and several European organizations. This step preceded the shift towards a layer of proprietary models.

Founders with Field Experience and Expanding Presence

Co-founder Hugo Micheron conducted eighty interviews between 2014 and 2019 with actors linked to jihadism, both in France and abroad, to reconstruct trajectories and systems, a work extended by a thesis at ENS-PSL and a book. He passed through the Near Eastern Studies department at Princeton and continues his work at Sciences Po, where he leads a seminar on artificial intelligence, democracy, and the informational environment. He also testified at the trial for the November 13 attacks. Alongside Antoine Jardin, a former research engineer at CNRS specializing in data science and human behavior, he founded Arlequin AI in 2024 to automate connections while maintaining traceability for the analyst. Two years later, the company claims around thirty employees, including eight in R&D, has opened offices in London and Berlin, and plans to establish a lab in Silicon Valley.