Jamendo Sues NVIDIA for Training AI Models on 55,600 Songs Without Authorization, Seeks €17.8 Million

Independent online music platform Jamendo has sued NVIDIA for copyright infringement and breach of contract, accusing the chipmaker of using tens of thousands of music tracks without authorization to train its audio AI models — and now it’s demanding at least €17.8 million in damages.

The Luxembourg-based company, part of the Winamp group, filed its complaint in US federal court in California on Monday. The move comes roughly eight months after Jamendo first sued NVIDIA in Belgium, where a judge recently rejected NVIDIA’s Belgian subsidiary’s jurisdictional challenge and ruled the case should proceed to trial on the merits.

At the heart of the dispute is the MTG-Jamendo dataset — a collection of approximately 55,600 tracks co-built by Jamendo and the Music Technology Group (MTG) at Universitat Pompeu Fabra in Barcelona. The dataset includes detailed metadata for each track and was made publicly available under a strict non-commercial license: researchers and academics could use it freely, but any commercial application required explicit written permission from Jamendo.

According to the lawsuit, NVIDIA ignored those terms entirely. The company’s own published research papers explicitly cite MTG-Jamendo as one of the training datasets used to develop two audio-focused AI models — Fugatto, a general-purpose audio generation model, and Audio Flamingo, aimed at audio understanding and reasoning.

Jamendo says it first discovered the unauthorized use in March 2024 and promptly reached out to NVIDIA to negotiate a commercial licensing deal. Over a year of back-and-forth talks — stretching into June 2025 — yielded no agreement. Frustrated, Jamendo issued NVIDIA a commercial invoice and filed the initial Belgian lawsuit.

That invoice, submitted as evidence in the US case, priced each track in the dataset at €289. For the full 55,600-track catalog, the total came to €16.1 million (approximately $17.2 million USD or ¥125 million RMB) for worldwide commercial usage rights.

“NVIDIA has refused to remedy its own breaches and infringements, has not paid the amounts set forth in the invoice, and continues to enjoy all the benefits from the unauthorized use of the MTG-Jamendo dataset,” the complaint states.

Jamendo is pursuing four legal claims: direct copyright infringement, breach of contract (for unauthorized commercial use and violation of Jamendo’s platform terms), unjust enrichment (arguing NVIDIA derived substantial commercial benefit from the dataset), and unfair competition. On the last point, the complaint alleges NVIDIA “circumvented license acquisition costs, avoided the expense of building an equivalent dataset from scratch, significantly reduced R&D expenditures, and accelerated the iteration of its AI systems.”

The minimum damages sought across all claims total €17.8 million — the original invoice amount plus overdue interest — though the final figure could climb higher depending on how the court rules. With parallel proceedings now underway in both Europe and the United States, the case could set an important precedent for how AI companies source training data and what constitutes fair compensation for rights holders in the age of generative AI.