AI & Music
Jamendo Refiles Its Suit Against Nvidia Over a Research-Only Music Dataset
By the Starchild Music team · October 17, 2026 · 5 min read
The short answer
Jamendo, a music licensing platform owned by Winamp Group, alleges in federal court that Nvidia trained audio AI models on a dataset of its tracks that was released for non-commercial research only. Jamendo dropped its first Nvidia suit on October 9 and filed a new one on October 12, reported on October 13. These are allegations, no court has ruled, and the case raises a question for any artist whose music sits in an open dataset.
Most AI music lawsuits so far have been brought by major labels and publishers. This one comes from a music licensing platform. Jamendo, owned by Winamp Group, says Nvidia used a dataset of its tracks to train audio models even though the dataset was released only for non-commercial research.
Everything below is an allegation unless stated otherwise. No court has decided the facts.
The dataset at the center
According to Music Business Worldwide, the MTG-Jamendo Dataset contains more than 55,000 full audio tracks with tags for genre, instrument and mood. It was created around 2019 through an academic collaboration with the Music Technology Group in Barcelona and posted to GitHub with a label saying it is for non-commercial research and academic use only. Commercial use requires prior written authorization.
Jamendo's complaint, as MBW describes it, says Nvidia's own research papers name MTG-Jamendo as a training source for its Fugatto and Audio Flamingo models. Nvidia has characterized its training data as "exclusively comprised of open source data," which Jamendo disputes.
The timeline
- June 22, 2026: Jamendo sues Nvidia in the Northern District of California. MBW reports six counts: two for copyright infringement, two for breach of contract, one for unjust enrichment and one for unfair competition. Jamendo seeks 17.8 million euros in actual damages and profits, or statutory damages of up to $30,000 per work and up to $150,000 per work if infringement is found willful.
- June 29, 2026: Jamendo sues Suno in federal court in Massachusetts, alleging about 919 hours of Jamendo audio was used to train an early open-source model called Bark.
- Early September 2026: Winamp says the Nvidia complaint was amended and served, with a first hearing set for January 12, 2027, and that the Suno complaint would be withdrawn and refiled in a more complete form. It also says it has identified six other technology and AI companies it believes use its content, and prefers commercial solutions with compensation first.
- October 9, 2026: Jamendo voluntarily dismisses its first Nvidia suit, according to the litigation tracker ChatGPT Is Eating the World.
- October 12, 2026: Jamendo files a new suit against Nvidia. Law360 reported on October 13 that an independent music licensing company had sued Nvidia days after dropping a nearly identical suit.
The Suno complaint
In the June 29 suit, MBW reports, Jamendo alleges Suno "blatantly disregarded" its intellectual property by using the catalog and data without authorization, and that roughly 919 hours of Jamendo audio were used to train Bark. Jamendo says this was willful, because the dataset was licensed only for non-commercial research and commercial use needs a paid license.
MBW also reports a pre-suit paper trail. Jamendo says it invoiced Suno on September 29, 2025, that Suno confirmed receipt by October 3 but did not pay, and that a formal default notice followed on October 14.
Winamp's September release says the Suno complaint was withdrawn so it could be refiled in a more complete version. The sources reviewed do not show the refiled Suno complaint, and Suno's response is not reported.
A parallel case in Belgium
MBW reports that the Ghent Enterprise Court confirmed on June 11, 2026 that it has jurisdiction over a separate commercial claim by Jamendo of about 16 million euros over unauthorized use of more than 55,000 musical works. Written submissions are expected from the end of 2026 into early 2027, with oral arguments scheduled for June 24, 2027.
That timetable is a reminder that these disputes move slowly. A decision that answers the research-license question could be years away.
What is not known
The sources reviewed do not explain why Jamendo filed a fresh case instead of amending the old one. The tracker called the move odd, and Jamendo has not publicly given a reason.
MBW's reporting on the June suit includes a statement attributed to Nvidia that no contractual relationship existed between the parties when Jamendo invoiced it. The sources reviewed do not include a response to the new filing, and the January 2027 hearing date applied to the earlier case.
Why this matters beyond one company
Statutory damages, which Jamendo asks for as an alternative, let a court award a set range per infringed work instead of requiring proof of actual losses. That is why the number of works in a dataset matters so much. It is also why defendants contest both the license terms and the count.
Open datasets are common in audio research, and they come with licenses. The legal question in this case, whether a non-commercial research license limits later use for commercial model training, is one many artists have not had to think about. A court answer, if one comes, could shape how datasets are labeled and how rights holders are paid.
There is also a contrast with the licensing deals made by the major labels, covered in the settlement wave. Jamendo's case is about a catalog of independent artists seeking either payment or a license, which is the situation described in the artists the settlements left behind. Also relevant is the Third Circuit's recent ruling on AI training and fair use.
What independent artists can take from it
Check the license terms of any platform or dataset where you upload music, especially terms that mention research, Creative Commons or non-commercial use. Keep a record of where each recording has been posted.
Songs licensed through a marketplace have a clearer trail than songs scattered across open archives. On Starchild, every license comes with a certificate naming the writers, licensed files carry an inaudible watermark and signed credentials, and anyone can check a certificate at /verify. See our note on the trust layer for how that works.
Sources
- Music Business Worldwide: Nvidia hit with US copyright suit claiming it trained AI models on Winamp subsidiary Jamendo's music without permission
- Music Business Worldwide: Suno sued by Winamp subsidiary Jamendo over AI training
- Winamp Group press release via Fortuneo: Winamp fait le point sur les procedures US a l'encontre de Nvidia et Suno
- ChatGPT Is Eating the World: SA Jamendo has dropped its suit v. NVIDIA
- ChatGPT Is Eating the World: SA Jamendo has filed a new lawsuit v. NVIDIA
- Law360: Nvidia hit with suit over AI training music data