Flood, floor.

There is a flood of AI-generated music, and there is a floor under how much of it people actually listen to. The flood gets the headlines. The floor is doing most of the work of this piece.

The flood is real. Suno alone claims seven million tracks generated per day; with Udio and the rest, the market produces close to ten million a day. By 2026, 44% of everything uploaded to Deezer is AI. The numbers, by themselves, sound like a takeover.

Then look at consumption. AI’s share of streams on Deezer, the only platform that publishes a figure, is one to three percent. Strip out the bots and click-farms, and you are left with 0.15 to 0.45 percent. The flood is upstream. The floor is downstream. Most of the river never reaches the sea.

Four charts.

The platform that knows the most about how much AI music is being uploaded is also the one that has been counting the longest. In January 2025, Deezer turned on detection. Ten thousand AI tracks a day, ten percent of uploads. Sixteen months later: seventy-five thousand a day, forty-four percent. The rate has roughly doubled every six months. Linear extrapolation is unwise, but it is hard to read this curve as anything other than: the supply is unbounded.

Supply has never been the question. The question is what happens next.

Ten million tracks generated. A hundred thousand uploaded. A fraction of one percent actually heard. Each step costs two orders of magnitude.

The drop from generation to upload is the largest. Most AI music never leaves the platform it was made on; it lives in Suno’s catalog, in a Discord, on a hard drive, never licensed, never distributed. The drop from upload to consumed is a softer one; the long tail of streaming has always been long. The drop from any AI consumption to non-fraudulent AI consumption is the cleanest moral story in the dataset: eighty-five percent of AI streams on Deezer are flagged as fraud and demonetized.

Structurally: the financial threat from AI music is bottlenecked by listening, and listening is bottlenecked by fraud detection. Both are platform problems, not technological inevitabilities. Both are tractable.

Against the AI numbers, set the size of the actual royalty pool. Spotify, by far the largest payer, has gone from $1B in 2014 to $11B in 2025, an order of magnitude in eleven years. Industry-wide recorded music revenue in 2025 was $31.7B; Spotify alone is about a third of that. The pool is large, and growing.

This is the denominator. The next chart is what AI music could take from it.

The chromatic argument of this chart is the entire piece. The rust bars are big and they are demonetized. The blue bars are small and they are real money to real creators. The thing the industry has been preparing to fight, AI music swallowing the royalty pool, is mostly already neutralized by fraud detection on the consumption side. The thing actually happening is a different and quieter story: a few hundred million dollars a year, within a couple of years, flowing to whoever owns the most-streamed AI catalogs.

Not a collapse. A dilution. Small, detectable, and platforms are choosing how to respond.

Two clusters. The platforms with the most aggressive stance, Bandcamp’s ban and Deezer’s instrumented response, are also the ones least exposed to AI music as a share of catalog or revenue. The platforms with the largest catalogs and the most to lose to a hypothetical AI flood, Spotify and Apple Music, have opted for the softer touch: distributor-declared metadata, internal spam filters, no platform-level tagging.

The second cluster might be right, and AI music routes around itself through fraud and indifference. Or wrong, and the lack of a platform-level tag becomes the consequential decision of this period.