Arlequin AI Raises €28M to Build a European Alternative to LLMs

Paris-based AI startup Arlequin AI has raised €28 million in Series A funding to scale an alternative AI architecture designed to analyse complex datasets without relying on large language models for its core reasoning.

The round was co-led by redalpine and OTB Ventures, with participation from Bpifrance's Defence Innovation Fund. Existing investors Vsquared Ventures and 10x Founders increased their positions, while French entrepreneur and investor Xavier Niel joined as a new investor.

The latest financing brings Arlequin's total funding to €32.4 million since its founding.

Founded by Hugo Micheron and Antoine Jardin, Arlequin is developing topological neural networks, or TNNs, that focus on relationships and connections within data rather than relying primarily on language patterns.

From academic research to AI infrastructure

Arlequin's origins are closely connected to Micheron's academic research into European jihadism.

Micheron learned Arabic in Syria and conducted extensive fieldwork that included interviews with hundreds of convicted terrorists to understand how radical networks form and evolve.

While teaching at Princeton between 2020 and 2023, he became increasingly critical of the limitations and biases he observed in existing AI systems.

He later teamed up with Jardin, a former research engineer at France's CNRS who specialised in big data and dimensionality reduction. Jardin had also worked with Jean Zay, France's national AI supercomputer.

Together, they set out to develop an AI architecture designed to produce conclusions that can be traced, audited, and scientifically examined rather than relying primarily on probabilistic outputs.

An alternative to LLM-based analysis

Arlequin does use a large language model, but only for its conversational interface.

The company's underlying systems process raw information including video, audio, text, images, and data from seized devices. Its unsupervised models are designed to identify relationships and connections within that information.

The company argues that this distinction becomes particularly important in high-stakes environments.

An incorrect AI-generated email may be inconvenient, but an incorrect analytical conclusion in counterterrorism could have significantly more serious consequences. Arlequin applies the same logic to areas such as energy, where analytical errors could potentially contribute to major operational or environmental consequences.

Its TNN architecture is therefore designed around relationships and structures in data, with an emphasis on traceability and verification.

Building Europe's alternative AI architecture

Arlequin believes its approach could offer Europe a different path to AI competitiveness.

Rather than attempting to match the enormous computing resources and capital being deployed by major US AI companies, the company is betting on a fundamentally different architecture.

Micheron argues that topological neural networks are still at an early stage of development, with only a relatively small global research community working in the field.

Arlequin says it is among the first companies to turn this research into a commercial product.

The startup currently has around 50 employees, including more than 35 engineers and 15 PhDs and postdoctoral researchers. Its platform is being used by more than 30 clients across Western and Eastern Europe.

Expanding across Europe

The new funding will support Arlequin's continued expansion in France, its entry into Germany, and the opening of a London office.

The company is targeting applications across safety and security as well as areas such as banking fraud and due diligence.

Its investors see AI sovereignty as extending beyond where models are trained or data is stored. The underlying argument is that European organisations may increasingly need control over the technologies used to make critical decisions.

Arlequin is positioning its architecture around that idea, offering an AI system intended to uncover relationships in complex data while maintaining traceability and auditability.

A high-risk bet on a different AI future

Arlequin's ambition comes with significant technical and commercial uncertainty.

The company is attempting to build a market around an architecture that remains far less established than the transformer-based systems behind today's dominant LLMs.

At the same time, it faces competition from highly funded AI and defence technology companies, while the possibility remains that established AI players could eventually adopt similar approaches.

The €28 million Series A gives Arlequin substantial resources to move its technology from an emerging research concept toward a broader commercial platform.

Whether topological neural networks can become a meaningful alternative to increasingly powerful LLMs - and whether Europe can build a competitive AI ecosystem around a different architectural approach - will be one of the company's biggest tests in the years ahead.