The dilution machine: what it means when half of new music on a streamer is synthetic

More than half of the tracks uploaded to Deezer each day in June were fully AI-generated, on the platform's own count. Those tracks drew somewhere between one and three percent of listening. That gap, most of the new supply, almost none of the demand, is not a curiosity. It is the mechanism by which human musicians lose income without anyone deciding they should.

Start with the numbers Deezer itself put out. In its 21 July press release, the platform said fully synthetic uploads peaked at a monthly average of roughly 90,000 tracks a day in June 2026, crossing the 50% line of daily deliveries for the first time.¹ TechCrunch, reporting on the same disclosure, laid out the series Deezer has published since January 2025: 10,000 tracks a day at 10% of uploads, then 20,000, then 30,000, then 50,000, up through 75,000 in April 2026 before the June peak.² Eighteen months, a fivefold rise in the share.

Deezer says AI tracks account for only one to three percent of total streams.¹ That is the headline the platform wants you to notice, because it makes the problem sound contained. It is not contained. It is the shape of the problem.

Who pays for oversupply. Streaming royalties on the major services come out of a pool. A track that gets streamed draws from that pool at a rate determined, roughly, by its share of total streams. Add a very large number of new tracks that draw almost nothing, and the pool itself does not shrink much — the fraudulent streams, at least, Deezer already excludes.¹ But every legitimate AI stream, however small the share, is a claim. And the presence of tens of millions of synthetic tracks in the catalogue is a claim on something else: attention, playlist slots, search results, the finite surface area where a human artist might be discovered.

The scale is worth stating plainly. Deezer says it detected and tagged over 13.4 million AI-generated tracks on its platform in 2025 alone, roughly 1% of the 120 million songs available to stream there.¹ That is one year of synthetic inventory equalling one percent of the total human-and-machine catalogue built up over the entire streaming era. At the current rate, that share compounds fast.

The projection, and how to hold it. The €4bn figure comes from a CISAC/PMP Strategy study that Deezer participated in, and which Deezer cites in the same press release announcing its new takedown rules.¹ That is worth naming: the number is a projection to 2028, produced with input from an industry participant that has commercial reasons to want the figure large. I would not stake an argument on the precise euros. What the projection captures directionally, that displacement pressure on creator income scales with synthetic supply, is consistent with what Deezer's own upload series shows.

What Deezer's new rules actually do. The 21 July announcement said Deezer will take down AI tracks used to generate fraudulent streams, and will remove AI tracks that have gone unstreamed for six months or more.¹ Fraud, and deadweight. Both are real problems. Neither is the displacement problem.

A track that is 100% AI-generated, uploaded in good faith, and picking up a modest but genuine listener trickle is untouched by these rules. It stays in the pool, drawing its share, competing for its slot. Deezer's CEO Alexis Lanternier, quoted in the release, framed the announcement as safeguarding the rights of artists and songwriters while keeping focus on music fans actually love.¹ The mechanism does the first only for the fraud case. For the legitimate-but-synthetic case, the harder one, the growing one, it does neither.

This is not a criticism of the specific rules. Fraud enforcement is good. Deadweight removal is good. It is a criticism of treating them as an answer to a question they were not designed to answer.

The platform choices are choices. Look across the sector as reported by TechCrunch and NME: Bandcamp has banned AI-generated tracks outright; Tidal, as of 15 July 2026, will pay no royalties to wholly AI-generated music and gives users a filter to exclude 100% AI content; Apple Music runs a voluntary tagging system; Spotify removed 75 million tracks it called spammy and built a policy calibrated to how much AI was involved; Deezer targets fraud and deadweight.²³ Five platforms, five distinct answers to the question of who the platform is for. The variation is the point. Nothing about the technology forces any single response. Each of these is a design decision being made now, with consequences for whose work gets paid.

The consent conversation is the same conversation. Last month, per NME's reporting, global music industry bodies sent an open letter demanding artist consent before AI licensing deals, and a separate campaign led by Paul McCartney, Kate Bush, Dua Lipa and Elton John pressed the UK government to protect artists' work.³ Deezer's announcement contains no consent mechanism. Lanternier's call, in the same press release, was for the industry to align on shared standards to build a sustainable ecosystem.¹ Industry alignment is a deferral. It is also, in the current landscape, an argument the platforms can afford to have slowly while the upload series keeps compounding.

I do not think there is a version of this where "let the market sort it out" ends anywhere good for working musicians. The market is sorting it out — that is what the upload chart is a picture of. Human uploads have not collapsed. They have simply been outnumbered.

The line worth watching is not whether the AI share of uploads keeps rising; it almost certainly will. It is whether any major platform starts treating legitimate synthetic supply as a distributional question rather than a fraud question, and whether the consent frameworks the industry bodies are pushing get any statutory teeth before the composition of the catalogue makes the answer moot.

Footnotes