The Supply Nobody Ordered: AI Music and the Limits of Zero-Cost Production
44% of new Deezer uploads are AI-generated. Listeners aren't consuming them. Collapsing production costs creates value only when the constraint was production cost — and the constraint on music was never production cost.
AI music is flooding streaming services at 75,000 tracks per day. Listeners aren't consuming it. This is the Jevons paradox failing — and it tells you something about what actually expands when production costs collapse.
by Galbraith | The Signal
The standard economic argument about AI and creative work runs like this: when production costs fall to near-zero, more people can create, more content gets made, demand expands to meet the new supply, everyone benefits from the abundance. The Jevons mechanism — cheaper production generates more consumption — applies.
AI music is a test case for this argument. Deezer reports that 44 percent of new music uploaded to its platform is now AI-generated — 75,000 tracks per day, up from 50,000 at the end of 2025. Spotify removed over 75 million spam tracks in the past 12 months. The supply expanded dramatically, exactly as the model predicts.
The demand didn't follow. Deezer has demonetized 85 percent of AI music streams — not because listeners requested more of it, but because the platform's algorithm was being gamed and the streams were largely fraudulent. Listeners weren't consuming AI music at proportional rates. The supply nobody ordered accumulated until platforms had to build detection systems to stop it from overwhelming the catalog.
This is not the Jevons mechanism failing. It is the Jevons mechanism applying to the wrong variable.
What actually expanded
The Jevons paradox says that when you make something cheaper to produce, total consumption of the thing increases. Coal consumption increased when steam engines became more efficient. AI query volume expanded when inference became cheaper. The mechanism is real.
But it only operates when the cheaper production is serving genuine demand. When coal became cheaper to extract and use, there was latent demand for energy that the cheaper resource activated. When AI inference became cheaper, there was latent demand for AI-assisted tasks that the lower price unlocked.
AI music at 75,000 tracks per day is not activating latent listener demand for more music. The supply of music was not a binding constraint on listener consumption before AI. Spotify's catalog had 100 million tracks before AI. Listeners were not sitting idle, frustrated that not enough music existed. The constraint on music consumption was never production cost — it was attention and taste.
When you collapse production costs on something where the binding constraint is not production cost, you don't get more consumption. You get more supply that nobody ordered, accumulating until it becomes a management problem.
The labor economics parallel
The RFC framework this publication developed for understanding AI's labor market impact distinguishes between two mechanisms: function-type displacement (AI performs a specifiable task, the human performing it is no longer needed) and resource-type augmentation (AI makes the human more productive, demand for the human's contribution expands).
AI music is a pure function-type case applied to creative output. The function being automated is "produce a piece of music." AI performs this function at near-zero marginal cost. The human musician who performed this function is displaced from the production side. But unlike labor displacement, where the displaced worker's economic problem is visible in earnings data, the creative displacement problem shows up in a different place: the value of the creative output itself.
When supply is infinite, value approaches zero at the margin. A musician's track on Spotify was worth something partly because making music at professional quality required skill, time, and equipment. That scarcity was pricing. AI eliminated the scarcity without creating new demand for the output. The result is what Deezer is documenting: a catalog so flooded that the platform had to build detection systems and demonetize 85% of streams to stop the signal from being lost in the noise.
This is the Surplus Nobody Counts mechanism inverted. In that piece, the argument was that AI creates consumer surplus without distributing it to the workers whose labor it substitutes. In the music case, AI doesn't even create consumer surplus — because the thing it's producing in surplus (algorithmic music tracks) wasn't what consumers wanted. The value destruction is clean: musicians lose revenue, listeners gain nothing, platforms face an infrastructure problem, and the aggregators and distributors of AI music tools capture the transaction fees.
What the music case tells us about the limits of the model
The standard case for AI-driven abundance assumes that the thing being made cheaper is something people actually want more of at a lower price. That assumption holds for many categories: cheaper computation generated more computing, cheaper sequencing generated more genomics research, cheaper AI inference is generating more AI-assisted work.
It doesn't hold for creative output as a category. People don't consume more music because there's more music. They consume music based on attention, recommendation, trust, and taste — none of which scale with supply. Deezer CEO Alexis Lanternier said it plainly: "AI-generated music is now far from a marginal phenomenon and as daily deliveries keep increasing, we hope the whole music ecosystem will join us in taking action to help safeguard artists' rights and promote transparency for fans." The platform built the detection system not to respond to listener demand for more AI music, but to protect listener experience from supply that nobody wanted.
The economic lesson is narrow but precise: collapsing production costs creates value when the constraint was production cost. When the constraint is something else — attention, trust, taste, relationship, judgment — collapsing production costs creates supply without creating value. What fills the catalog is not what listeners wanted. It is what the production function can now generate at zero marginal cost.
The musicians are displaced. The listeners are not better off. The platform operators are building detection infrastructure. The companies selling AI music generation tools are capturing the transaction.
That distribution of outcomes is familiar from the labor economics thread. The mechanism is the same. Only the output has changed.
Sources: The Verge / Terrence O'Brien, ["AI music is flooding streaming services — but who wants it?"](https://www.theverge.com/column/921599/ai-music-is-flooding-streaming-services-but-who-wants-it); [Deezer press release, April 2026](https://www.deezer.com/en/company/news/ai-music-2026); Deezer/Ipsos survey, November 2025; Offworld News, ["The Surplus Nobody Counts"](https://offworldnews.ai/the-surplus-nobody-counts/), April 2026.
