Source Edition · Run the Business

Munich priced both ends of the AI music model

Six songs were enough to frame both ends of the pipe: a license to train on the repertoire and another to commercially exploit model output.

Editorial timeline comparing two legal questions around model inputs and outputs
Kirk+Co editorial illustration of the model's two licensing endpoints.

Munich ruled against Suno on July 31. Six songs: “Rasputin,” “Daddy Cool,” “Mambo No. 5,” “Forever Young,” “Big in Japan,” “Atemlos.”

Three weeks earlier I wrote that Munich and the Sony hearing in Boston would price opposite ends of the same pipe: one court on what went into the model, one on what came out, with the learning itself legally invisible.

Half of that was wrong.

Munich priced both ends at once: a license to train on the repertoire, and a second license to commercially exploit what the model makes from it.

One ruling, both openings, and nobody opened the model.

The six is the part worth sitting with. Sony put 61,026 tracks in a single motion in Boston. Munich needed six, because the question was never how much the model absorbed. It was whether Suno was entitled to any of it. Volume sets a damages number. It was never evidence for the finding.

The training ran in the United States. The court applied German law anyway, to the storing and reproducing that happened in Europe. Offshore compute assumed jurisdiction attaches where the GPUs sit. It attaches where the model is served.

Damages are unset and Suno is weighing an appeal. The mechanism is already priced. You never had to see inside the model to bill both ends of it.

← Back to Blog