Jamendo vs. NVIDIA: The War for the “AI Training License” Has Begun
On June 22, 2026, Jamendo (a subsidiary of Winamp Group) reached a historic milestone in catalog protection by filing a federal lawsuit against NVIDIA. Accused of misappropriating the MTG-Jamendo dataset to train its Fugatto and Audio Flamingo AI models, the trillion-dollar tech giant faces a coordinated legal offensive across the United States and Europe. For the synchronization sector, this trial serves as the first major test regarding the legality of industrial data ingestion under Creative Commons licenses.
Klem Loden
6/23/20262 min read
The Misappropriation of the MTG-Jamendo Dataset
The offensive is built upon a documented technical fact: NVIDIA utilized the MTG-Jamendo dataset, a collection of over 55,000 tagged tracks originally created for academic research in collaboration with Pompeu Fabra University. While NVIDIA claims its models were trained on “open source” data, Jamendo highlights that the dataset’s own documentation explicitly prohibits any commercial application without prior written authorization. At The Sync Pipeline, we analyze this defense as a deliberate conflation of “public accessibility” and the “right to industrial exploitation.”
Metadata Infrastructure as Protected Work
Jamendo’s legal strategy in the U.S. is particularly sharp: it does not rely solely on the rights of individual songs, but on the copyright of the database compilation. Jamendo secured a registration with the U.S. Copyright Office (No. TX-9-606-566) for the specific structure, selection, and arrangement of its metadata. By copying this organized dataset to feed Fugatto, NVIDIA allegedly stole not just sounds, but the intellectual infrastructure that makes those sounds machine-readable. This is a direct strike against “Dark Data”: Jamendo is valuing technical curation as a sovereign asset.
A 34-Million-Euro Stake and Global Jurisprudence
NVIDIA faces a dual financial front: a €16 million claim in Belgium and a demand for at least €17.8 million ($20.3 million) in statutory damages in the U.S. If Jamendo prevails, it will establish global jurisprudence necessitating a new category of rights: the AI Training License. For synchronization libraries, this would transform every algorithm developer into a mandatory B2B client. Data ingestion would no longer be a “Fair Use” gray area but a compensable act of software synchronization.
The End of “Open Source” Plundering
The Jamendo vs. NVIDIA trial marks the end of naivety for catalog holders. In 2026, the value of a library no longer resides solely in its ability to be placed in a film, but in its structural compliance against generative models. As Alexandre Saboundjian (CEO of Winamp Group) emphasizes, AI can only transform the industry if creator rights are protected at the source of the pipeline. This dispute defines the price of algorithmic “fuel”: either tech giants pay for the excellence of human-curated data, or they face systematic exclusion from regulated markets.