On July 21, 2026, the streaming service Deezer said that more than half of the tracks uploaded to its platform each day are now fully AI-generated. The figure works out to roughly 90,000 synthetic tracks a day, up from about 10,000 a day just eighteen months earlier. It is the first time a major streaming service has reported that machine-made music outnumbers human-made music in its daily intake.
The headline is striking, but the more useful story is in the gap between two numbers. AI tracks dominate what gets uploaded, yet they remain a sliver of what gets played. Understanding why those two figures diverge so sharply says more about the economics of streaming than the milestone itself.
From 10% to more than half in eighteen months
Deezer has published periodic snapshots of AI upload activity since January 2025, which makes it one of the few places to watch the curve bend in near real time. In January 2025, AI-generated tracks were about 10% of daily uploads. By April 2025 that share was 18%, by September 2025 it was 28%, and by November 2025 it had reached 34%. In April 2026 the company reported 44%, and by June 2026 the share had crossed 50%.
AI-generated music crossed half of Deezer's daily uploads in 18 months
Share of new tracks uploaded to Deezer each day that the platform classifies as fully AI-generated, per Deezer's own disclosures.
The June 2026 figure is Deezer's "more than 50%," shown here as 50%. Daily volume rose from roughly 10,000 tracks to about 90,000 over the same period.
The slope matters as much as the level. This is not a one-off spike tied to a single viral tool; it is a steady climb across six disclosure points, which suggests a structural shift in how music reaches streaming catalogs rather than a passing novelty. Text-to-music tools have become cheap enough and good enough that generating and uploading a track costs almost nothing, and the supply has responded exactly the way any near-zero-marginal-cost supply does.
Why uploads and listens tell opposite stories
Here is the counterweight to the headline. Deezer says AI-generated tracks still account for only 1% to 3% of total streams on the platform. In other words, machine-made music is the majority of what arrives and a rounding error in what people actually choose to hear.
A flood of uploads, a trickle of listens
AI-generated tracks dominate what is uploaded to Deezer but remain a small share of what people actually stream.
Streams shown at the 2% midpoint of Deezer's stated 1% to 3% range. Deezer says up to 85% of the AI-music streams it does see are fraudulent, driven by bots rather than listeners.
That divergence is the analytically interesting part. Uploading is effectively free, so the upload count measures how easy generation has become. Streaming reflects demand, which is still overwhelmingly for human artists and established catalogs. A catalog can fill with synthetic tracks without those tracks ever finding an audience, because discovery, playlisting, and fandom are not automated the way generation now is.
Generation has become nearly free. Attention has not. The gap between what gets uploaded and what gets heard is the price of that asymmetry.
Metir AI analysis
There is a second, less comfortable reason the streaming share is not zero. Deezer says that up to 85% of the AI-music streams it does detect are fraudulent, meaning they are generated by bots rather than listeners. This reframes a slice of the AI-music phenomenon as a royalty-fraud problem rather than a taste shift: flood a catalog with cheap tracks, then use automated streams to siphon royalty payouts from the same pool that pays human artists. Deezer has said it excludes AI-generated content flagged for fraud from royalty calculations and does not recommend such tracks in editorial playlists.
The detection and labeling response
Deezer's public numbers exist because the company built an internal tool to flag fully AI-generated tracks, and in 2025 it began tagging them so listeners can see when a track was machine-made. Labeling is the pragmatic middle path between two extremes: banning synthetic music outright, which is hard to enforce and arguably overbroad, and ignoring it, which erodes trust and invites fraud.
The harder question is durability. Detection of fully synthetic audio is more tractable than detection of AI-assisted human work, where a producer uses generative tools inside an otherwise human track. As tools blur that line, a clean binary label becomes less meaningful, and platforms will likely move toward disclosure regimes closer to nutrition labels than to a single AI-or-not stamp.
What this signals beyond music
The music case is a preview of a pattern now visible across every medium where generation has collapsed to near-zero cost. When making a thing becomes free, the binding constraint moves from production to everything downstream: distribution, discovery, trust, and the ability to tell provenance. The volume of content stops being a useful proxy for value, because volume is now trivial to manufacture.
That is why the interesting metrics are shifting. In a world where anyone can generate a plausible track, article, or image in seconds, the questions that matter are which of those things earn attention, which are authentic, and which are simply gaming a payout formula. For anyone building products on top of generative models, the lesson from Deezer is that raw output is the easy part, and the systems that filter, attribute, and rank that output are where the real work now sits. Model-agnostic platforms such as Metir AI reflect the same logic on the tooling side: the model that generates content is increasingly a commodity, and the durable value is in the workflow wrapped around it.
Looking ahead
Deezer's crossing of the 50% line is a clean marker of how fast generative supply can scale, but the 1% to 3% streaming share is the more honest measure of impact so far. The two numbers will not necessarily converge. If synthetic tracks keep flooding uploads while listeners keep choosing human artists, the gap could persist for years, with fraud rather than genuine demand doing most of the work to keep the streaming share above zero. The metrics worth watching next are whether the streaming share moves at all, and whether the industry settles on labeling standards that survive the blurring of the human-versus-machine line.
Build on models without betting on one
The Deezer story is a reminder that generative models are becoming a commodity input, and the value is in the systems around them. With Metir AI you get unified access to leading models from OpenAI, Google, Anthropic and others in one workspace, so you can route each task to the right model without locking your product to a single provider. Try Metir AI free.
Sources:
- Music streamer Deezer says more than 50% of daily uploads are AI-generated | TechCrunch
- AI Music Tops 50% of Daily Uploads on Deezer | Deezer Newsroom
- AI-generated music is now more than half of Deezer's uploads | Music Ally
- 75,000 AI-generated tracks now flood Deezer daily, representing 44% of all new music uploaded | Music Business Worldwide
- Deezer says 44% of songs uploaded to its platform daily are AI-generated | TechCrunch
- Deezer Cracks Down on AI-Generated Music, Fake Streaming on Its Platform | iTech Post
Image credits
Header image: a recording engineer mixing on a Rupert Neve Designs 5088 console at Jackpot! Recording Studio in Portland, Oregon, by VACANT FEVER via Wikimedia Commons, licensed under CC BY-SA 2.0. In-body photograph of a mixing desk at the Hipposonic Recording Studio, by Hipposonic Studios via Wikimedia Commons, licensed under CC BY-SA 4.0. Both photos depict conventional music production and do not show any AI tool or Deezer itself.