- The Paris-based Arlequin AI has raised €28 million in its Series A funding round to scale its AI architecture, which does not use large language models.
- Hugo Micheron, a co-founder, based the creation of his company on years of academic research, which included interviews with terrorists who had been convicted. The company now provides its services to over 30 clients in four European countries.
- Redalpine and OTB Ventures jointly led the funding round, with Bpifrance’s Defence Innovation Fund and Xavier Niel also taking part. The total amount of funding received by Arlequin is now €32.4 million.
Hugo Micheron learned Arabic in Syria and carried out extensive fieldwork on European jihadism, interviewing hundreds of convicted terrorists in order to examine how radical networks are formed. During his time as a lecturer at Princeton from 2020 to 2023, he observed that the available AI tools were, as he put it, “full of biases.” This led him to found Arlequin AI, which just raised a €28 million Series A.
The investment round was co-led by redalpine and OTB Ventures, with Bpifrance’s Defence Innovation Fund also taking part. Vsquared Ventures and 10x Founders increased their holdings, and Xavier Niel became a new investor.
“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 Hugo
Micheron, CEO and co-founder of Arlequin AI.
From interviewing terrorists to building an AI company
Micheron founded Arlequin two years ago together with Antoine Jardin, who has been his friend for 15 years and was previously a research engineer at CNRS, France’s national scientific research institute. Jardin specialised in big data and dimensionality reduction, the mathematical technique that is at the heart of Arlequin’s models, and had been involved with Jean Zay, France’s national AI supercomputer.
According to Micheron, they approached each other directly: they wanted to create an architecture with “the precision of scientific expectation,” that could prove and audit its conclusions rather than rely on the probabilistic guesswork of an LLM.
Now, two years after the first seed round, Arlequin has grown to about 50 employees, including 35 or more engineers, 15 PhDs and postdocs, and claims its platform is used by more than 30 clients in both Western and Eastern Europe.
Where LLMs fall short
Arlequin currently uses a large language model solely for its conversational interface. Its unsupervised models handle raw data such as video, audio, text, images, and seized devices, detecting connections that can be audited and traced. The large language model is used only to assist users in formulating their questions.
According to Micheron, the principal issue with using LLMs for analysis does not lie merely in accuracy but in the consequences of their making errors. For instance, an 80% confidence level would be acceptable when composing an email, but in the field of counterterrorism such a degree of uncertainty could have serious repercussions: “you end up being wrong… you have an attack.”
He applies the same reasoning to energy companies, pointing out that an error with a 90% confidence level could result in an ecological crisis. The new topological neural network, or TNN, architecture developed by Arlequin is intended to address this by focusing on relationships and connections in the data rather than solely on language patterns.
A three-way race: Europe backs its own approach
According to Micheron, the field is still in its infancy, with TNN’s research at a stage analogous to that of large language models a decade ago. He believes that only around 20 researchers globally are involved in this area and states that Arlequin is the first company to have launched a commercial product rather than merely offering a research prototype.
Arlequin is presenting this funding round within a geopolitical framework, noting that the United States has the advantage in terms of computing power and data; that China follows values which Europe does not wish to adopt; and that, although Europe’s Mistral programme is advancing, it is still falling behind in scale.
According to Micheron, TNN offers Europe an opportunity to compete in a new way rather than attempting to match Silicon Valley’s resources. The company intends to continue its growth in France, expand into Germany, as Micheron believes there will be significant investment in AI and security there, and open a new office in London, with its first employees starting this week.
Earlier in June 2025, Arlequin completed a €4.4 million seed round led by Vsquared Ventures. The Series A funding brings the total amount raised to €32.4 million. The company’s valuation has not been disclosed.
“AI sovereignty is not only about where models are built or data is hosted. It is about having control over the technologies that increasingly underpin our most critical decisions. Europe needs AI systems it can own, understand and trust. Arlequin is building a fundamentally different paradigm of AI, designed to uncover relationships in complex data without predefined ontologies, bias or hallucination, while keeping results traceable and verifiable,” stated Jeremy Teboul from OTB Ventures.
Oliver Pabst of redalpine, whose portfolio comprises N26, Mistral and Klarna, noted, “From day one, Hugo and Antoine impressed us with their exceptional founder-market fit, deep technical edge and relentless drive to solve a critical problem. Arlequin is building a truly differentiated, fully productized sovereign tech platform for safety and security in Europe and beyond, and we are convinced this exceptional team has what it takes to build a category-defining company in a market worth trillions.”
Maria Juesas Portoles of Vsquared Ventures added: “Most of the capital in AI right now is chasing the same architecture and hoping scale solves everything. Arlequin is pursuing a genuinely different architecture: developing topological neural networks as a distinct architectural approach, rather than relying solely on larger models and more compute.”
The AI and analytics market in Europe for defence purposes will rise from $4.8 billion this year to $19.27 billion by 2035; this figure doesn’t account for Arlequin’s activities in banking fraud and due diligence.
While Micheron claims the potential benefits could be “10s or 100 times more profound than those of LLMs”, this is merely his own assessment and has not been independently verified. It should be borne in mind that it is a significant risk to aim to compete with Palantir’s extensive reach and Helsing’s $18 billion in funding by employing an architecture that only about 20 researchers worldwide understand.
Whether Europe’s bet on a new architecture pays off before LLM companies adopt the idea remains to be seen.
