Music at Zero Marginal Cost

In April 2026, the streaming service Deezer reported that around forty-four percent of the tracks uploaded to it each day were generated entirely by software — up from roughly eighteen percent a year earlier. The figure is worth holding still for a moment. It does not describe music made with digital tools, which has been the norm for forty years. It describes music made with no human author at any stage: melody, arrangement, voice, and the name on the release all produced by a model. Tens of thousands of such tracks now arrive daily. The cost of making one more has fallen, for practical purposes, to nothing.

THE COLLAPSE OF A PRECONDITION

For as long as recorded music has existed, scarcity was its hidden architecture. A song required a musician, time, a room, equipment, and the particular friction of getting something finished. That friction was not merely an obstacle; it was a filter. It meant the catalogue grew at a human pace, and that any recording carried the trace of a decision someone had made to bring it into the world. Generative tools remove the friction without replacing the filter. The result is not better music or worse music in the aggregate. It is more — a volume of plausible, ownerless sound expanding faster than any listener, critic, or institution can attend to it.

The case of the Velvet Sundown showed the shape of what follows. In mid-2025 the act gathered more than a million Spotify streams as an apparently new rock band before acknowledging it was an AI project. Within months its monthly audience had fallen by roughly three-quarters. The interesting fact is not that it rose, but that it could not hold. Without a person behind it — a history, a face, a reason to return — there was nothing for an audience to attach to once novelty passed. The sound was sufficient to be heard once and insufficient to be loved.

The friction was never only an obstacle. It was the filter that made a catalogue legible to the people it was for.

WHAT THE PLATFORMS ARE DEFENDING

The streaming services understand the stakes in financial terms first. Royalty pools are largely fixed: subscriber money is divided among rights-holders by share of plays. Every artificial stream of machine-made sound therefore dilutes what reaches human musicians. Deezer has estimated that the large majority of streams directed at AI tracks on its platform were fraudulent — bots listening to generated music to harvest royalties. Spotify, over a single year, removed some seventy-five million tracks it judged spam. The platforms are not defending art. They are defending the integrity of a payment system that assumes a play means a person chose to listen.

WHAT A LISTENER LOSES

The cultural cost is quieter and harder to price. Music has always done more than fill time; it has carried identity, memory and belonging precisely because someone made it and someone else recognised the making. A song that no one authored can still be pleasant. It cannot be a witness. As the catalogue fills with sound that is competent and unattributable, the act of caring which song it is — the thing that turns listening into culture — comes under a slow pressure it was never built to withstand. The danger is not that a machine writes a masterpiece. It is that the median listening experience drifts toward the ambient and disposable, and that a generation grows up for whom music is weather rather than testimony.

What endures, then, is unlikely to be any particular song. It is the human reasons to prefer one over another: authorship, scarcity, the knowledge that a recording cost someone something. Those reasons are now the scarce resource. The institutions that survive this period — labels, critics, archives, the live room — will be the ones that protect not the supply of music, which is now effectively infinite, but the conditions under which a listener can still find the music that was meant. Abundance did not solve the problem of attention. It relocated it.