Olivier Debeugny, CEO of French fintech Dragon LLM (formerly Lingua Custodia), received a call from the European Commission in June 2024: his company had won a prize from the European Commission. Along with €250,000, the real windfall was access to 2 million GPU hours, on two of Europe’s super computers (the Italian Leonardo and Germany’s Jupiter) normally worth €7–10 million.
This prize enabled Dragon LLM to tackle one of AI’s biggest headaches: energy-hungry large language models (LLMs). Instead of using transformer architectures that analyse all data, Dragon LLM consults only targeted memory blocks, making responses faster and more energy-efficient.
In October, the company unveiled a demonstration model with 7 billion parameter-equivalent performance using just 3.6 billion active parameters. Plans are in place for other models, all open source, allowing scientists and developers to train their own versions. Meanwhile, Dragon LLM will also fine-tune models for clients, particularly in the financial sector.
Active in Luxembourg since 2019, at the start of 2025, Debeugny had hoped to position Dragon LLM as Luxembourg’s sovereign AI solution but was too late. Luxembourg had already partnered with Mistral AI. Today, the company focuses on France, leveraging financial data to build and optimise its models.
“First you need good architecture, then the right data,” Debeugny says. Optimistic about Europe’s AI potential, he envisions a homegrown AI champion—an “Airbus of artificial intelligence.”
He also praised the European Commission initiative saying that for small-company scientists, access to massive computing power to publish research is more valuable than a big paycheck from Seattle tech giants.
