‍

Deezer says roughly half of everything now uploaded to the platform is AI-generated. Around 90,000 tracks a day. A year and a half ago that figure was a fraction of what it is now, and the direction of travel is straight up.

The trade press keeps running this as a culture story. Is the slop killing real music, is it authentic, who cares if a machine wrote it. Fine questions. But they miss the thing a catalogue owner should actually be watching, which is what a supply influx of this size does to the money.

We'd rather be honest than clever here. An astonishingly large influx of supply doesn't mean there's demand for that supply. Most of those 90,000 daily tracks will get streamed by almost nobody. That's true and it matters. But it doesn't get you off the hook, because streaming royalties don't work the way that reassurance implies.

‍

The pool isn't fixed, and that's not the point

‍

There's a common error in catalogue commentary worth correcting before making the argument.

People say the royalty pool is fixed and more tracks means everyone gets a thinner slice. That's not quite right. The pool grows. Subscription revenue goes up, platforms raise prices, new markets come online. So the denominator getting bigger doesn't automatically mean your cheque gets smaller in absolute terms.

The real mechanism is per-unit rate pressure. What you get paid per stream. And that number has been falling. Roughly 7% a year between 2021 and 2024, with a partial stabilisation in 2024 when the platforms pushed through price hikes. So the rate was already under pressure before AI supply arrived at scale.

Now add the supply. AI tracks were something like 0.5% of stream share not long ago. The trend is pointing toward 1 to 3%. Small numbers, and that sounds like a rounding error against the money real catalogues earn. But watch the direction. The share is climbing while the per-stream rate is already soft. Those two things push the same way.

‍

Why lean-back music is where it bites first

‍

Think about where AI tracks actually get streamed. Ambient. Lo-fi. Focus playlists. Sleep, study, the background of someone's day. Lean-back music, where the listener isn't choosing a specific song, they're choosing a mood and letting the algorithm fill it.

That's the category most exposed. If a playlist needs 200 tracks of rain-on-a-window piano and a machine can generate all 200 for nothing, the human-made rain-on-a-window piano is competing on price against something that costs zero to make. It might be that lean-back music is just cannibalising itself, one flavour of anonymous background audio replacing another. But even that isn't neutral for the people who own the human version, because their slice of that listening was earning something and now it's splitting with an infinite free supply.

Front-catalogue with genuine fandom is more defended. Nobody is putting on an AI track instead of the song they actually love. But more defended isn't immune, because the rate pressure runs across the whole system, not just the exposed corner of it.

‍

What we're not going to pretend

‍

We haven't got a bullish case to hand you against this. If your read is that AI is a further deluge of slop drowning quality music, we don't think you're wrong, and our position isn't counter to that thesis at all. The standard read is roughly right. We're just trying to say what it does to the numbers with less hand-waving.

So here's the useful thing. When you're underwriting a catalogue, the per-stream rate is a live variable, not a constant you can lift from last year's statements and roll forward. Build a rate decline into the base case. Build a steeper one into the downside. Ask which part of the catalogue lives in lean-back listening and haircut that part harder.

The supply shock is real, it's large, and it points one way for existing rights owners. Markets absorb shocks and reprice, and the ones that priced the shock in early tend to do better than the ones that argued it away.

‍