Jamendo Sued Nvidia Over Training on 55,000 Research-Only Songs

Jamendo just sued Nvidia for training two commercial audio AI models on 55,000 songs the chipmaker downloaded from a research dataset marked non-commercial only — and Nvidia's own papers name the catalog as training fuel. The world's most valuable company is now in the same copyright crosshairs as Suno and the major labels, with a bill north of €17.8 million and a legal theory that could outlast the fair-use fight.

What happened:

• On June 22, 2026, Jamendo filed a federal complaint in the U.S. District Court for the Northern District of California against Nvidia Corporation.
• The music-licensing platform owned by the Winamp Group alleges Nvidia trained Fugatto and Audio Flamingo on the MTG-Jamendo Dataset without authorization.
• Parallel proceedings are already underway in Belgium, where the Ghent Enterprise Court confirmed jurisdiction on June 11 over a €16 million commercial claim against NVIDIA Technologies Belgium.

The dataset dispute:

• The MTG-Jamendo Dataset contains more than 55,000 full audio tracks tagged across 195 genre, instrument, and mood categories.
• It was built around 2019 through an academic collaboration between Jamendo and the Music Technology Group at Universitat Pompeu Fabra in Barcelona, then posted on GitHub for researchers.
• The dataset's own terms state it is available "solely for non-commercial research and academic use" and that "any other use, including but not limited to commercial applications, requires prior written authorization from Jamendo S.A."
• Jamendo alleges Nvidia's published research expressly identifies MTG-Jamendo as one of the datasets used to build Fugatto and Audio Flamingo.

Nvidia's open-source defense:

• Nvidia unveiled Fugatto in November 2024, calling it a "Swiss Army knife for sound" — a generative audio model in the same category as AI music tools like Suno.
• The company has described its training corpus as "exclusively comprised of open source data" spanning at least 20 million rows of text and audio.
• Jamendo's counter: open source does not mean free for any purpose. The underlying tracks carry individual Creative Commons licenses; the curated compilation and metadata carry separate non-commercial restrictions.
• When Jamendo learned of the alleged use in March 2024, it sought a commercial license. Negotiations ran through June 2025 and failed. Jamendo then invoiced Nvidia €289 per track — €16.1 million for 55,600 works — and says Nvidia replied that no contractual relationship existed.

The legal theory:

• The complaint brings six counts: two for copyright infringement, two for breach of contract, one for unjust enrichment, and one for unfair competition under California law.
• Jamendo seeks at least €17.8 million (~$20.3 million) in actual damages and Nvidia's profits — or, alternatively, statutory damages of up to $30,000 per infringed work, rising to $150,000 per work if infringement is found willful.
• The heart of the US case is a compilation copyright in the dataset itself — the original selection, coordination, and arrangement of tracks and metadata.
• Jamendo says the U.S. Copyright Office issued registration TX-9-606-566 for that compilation on June 17, 2026 — five days before the federal filing.
• That theory follows Feist Publications v. Rural Telephone (1991): raw facts aren't copyrightable, but a creative arrangement of them can be.

Why this matters beyond music:

• Nvidia is not a model lab facing a label lawsuit — it is the AI infrastructure king, with a market capitalization of roughly $5.1 trillion, building consumer-facing audio AI on top of the chips that power everyone else's models.
• Universal Music Group partnered with Nvidia on Music Flamingo earlier this year, hailing it as "responsible AI" — a branding exercise that looks awkward next to a copyright complaint from a catalog licensor.
• Jamendo publicly threatened legal action against both Nvidia and Suno last year. Suno remains in litigation with the majors; Jamendo has not yet sued Suno.
• The case lands the same week Nvidia shares fell roughly 3–4% in a broader AI chip selloff — a reminder that legal liability is joining valuation risk on the balance sheet.

What Nvidia says:

• Billboard and Digital Music News both report that Nvidia did not immediately return requests for comment.
• In Belgium, Nvidia Technologies Belgium continues to dispute the validity of Jamendo's invoice, per the stateside complaint. Oral pleadings in that case are scheduled for June 24, 2027.

The quote that frames it:

• Jamendo's lawyers write that Nvidia achieved "unprecedented financial success through the commercialization of AI technologies built on large-scale datasets" yet "declined to compensate or even meaningfully engage" — highlighting "a stark disconnect between profit and accountability."
• Winamp Group CEO Alexandre Saboundjian said Tuesday: "As artificial intelligence continues to transform the music industry, we believe it is essential that creators and rights holders are properly recognized, compensated and protected."

Convina's view: The AI industry treated "open source" training data like a public commons for a decade — download, fine-tune, ship, and let lawyers sort out fair use later. Jamendo's suit punctures that laziness with a simpler weapon: a terms-of-use page and an invoice Nvidia allegedly ignored. Whether the compilation-copyright theory holds or statutory damages scale to billions is beside the immediate point. Every enterprise buying AI audio, code, or multimodal tools from infrastructure vendors — not just model labs — now has a new diligence question: show me the license chain, not the GitHub README. The chipmakers wanted to own the full stack. The copyright bill comes with it.