Jamendo sues Nvidia over alleged misuse of data and music in AI training

Nvidia just picked up its eighth copyright lawsuit related to AI training data. This time, Luxembourg-based music platform Jamendo is the one swinging, alleging that Nvidia hoovered up hundreds of thousands of audio files and metadata to train its generative AI models without proper authorization.

The federal copyright suit was filed on June 22, 2026, in the US District Court for the Northern District of California. Jamendo is seeking a minimum of 17.8 million euros, roughly $20.3 million, in damages. But the real number could climb much higher: statutory damages could reach up to $150,000 per infringed work.

What Nvidia allegedly did

At the center of the dispute is something called the MTG-Jamendo dataset. It consists of approximately 55,000 full audio tracks, developed through a collaboration between Jamendo and the Music Technology Group at Universitat Pompeu Fabra in Spain.

That dataset was released under a Creative Commons license, but specifically for non-commercial research use. Jamendo alleges Nvidia took those tracks and fed them into two commercial AI products: Fugatto, a generative audio model, and Audio Flamingo, an audio-language model.

According to the complaint, Jamendo became aware of the alleged infringement in 2024. That’s when Nvidia publicly acknowledged the MTG-Jamendo dataset as a training source for its models.

The lawsuit brings four claims: direct copyright infringement, breach of contract related to the Creative Commons non-commercial license, unjust enrichment, and unfair competition.

The bigger picture for AI and copyright

This is not an isolated event. It is Nvidia’s eighth copyright lawsuit tied to how it sources training data for AI systems.

The Creative Commons licensing angle in this case adds a particularly sharp edge. These licenses are designed to encourage sharing and collaboration, but they come with explicit conditions. The non-commercial designation is not a suggestion. It is a legal restriction.

Unlike text-based training data disputes, which often involve murky fair use arguments, audio training datasets frequently come with clearer licensing terms. That makes it harder for defendants to argue ambiguity.

The $20.3 million floor that Jamendo is seeking might sound manageable for a company of Nvidia’s scale. But multiply that logic across eight lawsuits and counting, with statutory damages potentially reaching $150,000 per work across datasets containing tens of thousands of tracks, and the numbers start to get uncomfortable.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.