Streaming Royalties: What Each Platform Actually Pays Per Stream

Streaming royalties are quoted more often than any other number in independent music and understood less well than almost any of them. Somebody tells you Spotify pays $0.003. Somebody else says $0.005. A third person says their statement showed $0.0021 and concludes they were cheated. All three can be right at once, because none of those numbers is a rate. They are averages, computed after the fact, from a division.

This page is the rates and the mechanism behind them. Our AI music income playbook covers the full set of revenue streams available to an independent artist and the workflow that connects them — this page does not restate it. What follows is the narrower and more useful thing: what each service actually pays per stream in 2026, why that figure moves, and what fraction of it survives the journey to your bank account.

There is also a layer here that is specific to what this site measures. If you are generating music with AI, the royalty question has a hard precondition attached to it, and Deezer has published the numbers that show exactly how that precondition plays out at scale.

The 2026 per-stream table

These are blended averages to the rightsholder, before any distributor, label or publisher deduction. The right-hand column is the more honest way to read the table, because it converts an abstract fraction of a cent into a number of plays you would actually have to accumulate.

That is a 15x spread across the paid mainstream services and more than 150x once ad-supported free-tier plays are in the frame. The same track, the same listener count, and a 150-fold difference in what it earns depending on where and how those plays happened.

Two structural facts explain most of the ordering. First, services with no free tier pay more, because every stream they count originates from a paying subscriber rather than from an ad impression. Beatport, Tidal and Apple Music occupy the top of the table for that reason and no other. Second, pool size relative to catalogue size matters more than generosity. Beatport's rate is high because it is a subscription DJ service with a comparatively small pool being divided by a comparatively small number of qualifying streams, not because it has decided to be kinder.

The corollary is the part artists tend to resist: the highest rate is rarely the highest cheque. Spotify sits near the bottom of the table and still produces the largest absolute payout for most independent catalogues, simply because that is where the listeners are. Rate and revenue are different questions.

A per-stream rate is a result, not a price

This is the single most important thing on the page, and it is why every figure above carries a tilde.

No major streaming service operates a rate card. What they operate is a pro-rata revenue pool. Each month, the service totals its subscription and advertising revenue for a territory, allocates the rightsholder share of it, and divides that share across all qualifying streams in the same period. Your payout is your share of the streams multiplied by the pool. The "per-stream rate" is what you get when you divide one by the other afterwards.

Four variables move that residual, and none of them is under your control:

• Subscriber mix. A paid subscriber contributes a monthly fee. A free-tier listener contributes whatever advertising sold against their ears, which is far less. A service whose free tier grows faster than its paid tier will see its blended rate fall even if it changes nothing.
• Territory. Subscription prices vary enormously by market. A stream from a US or Nordic premium subscriber can be worth several times a stream from a market where the same tier costs a fraction of the price. A track that goes big in a low-ARPU territory earns a fraction of what the same stream count would earn elsewhere.
• Total stream volume. The denominator is everybody's streams, not yours. If total consumption on a service grows faster than its revenue, every rate in the table drops. This is the mechanism behind the long, slow decline in per-stream averages over the last decade.
• Qualifying thresholds. Streams that do not qualify do not draw from the pool. Spotify's policy requires a track to reach 1,000 streams in the prior twelve months before it earns recording royalties at all, and it demonetises detected artificial streaming — our Spotify for Artists guide covers the enforcement side of that in detail.

The practical consequence is that when your statement shows a lower effective rate than the published figure, the most likely explanation is not fraud. It is that your listeners were in cheaper markets, on free tiers, or both. Reading a distributor statement broken out by platform and by country is the only way to see this. Read as a single lump sum, it is invisible.

Artist-centric models: what Deezer and Tidal changed

Pro-rata has a well-known distortion. Because all revenue goes into one pool, a subscriber who listens exclusively to one artist all month still funds whichever tracks dominated the global chart. Their fee does not follow their listening.

Artist-centric and user-centric models attack that. The general principle is to allocate more of an individual subscriber's fee towards what that subscriber actually played, and to attach qualifying conditions to who draws from the pool at all.

Deezer runs the most developed version. Its artist-centric system introduces minimum thresholds a track must clear to receive boosted allocation — a floor on monthly streams and on unique listeners — applies additional weighting to tracks that reach genuine editorial engagement, strips detected fraudulent streams out of the calculation entirely, and demonetises non-music "noise" content that had been farming the pool with rain sounds and white noise. Tidal has moved along the same axis with fan-centred allocation, positioning payout transparency as a differentiator against the pro-rata default.

Here is the honest assessment, and it is the part the press releases skip: these models redistribute the pool, they do not enlarge it. The service's revenue is unchanged. If your rate goes up under an artist-centric model, someone else's went down to fund it. The winners are artists with engaged, repeat listeners who clear the thresholds. The losers are catalogues built on very large numbers of very shallow plays, along with noise-content farms and anything the fraud detection catches.

For an artist with a small but genuinely engaged audience, that is a favourable trade. For a strategy built on releasing enormous volume and hoping for incidental plays, it is precisely the wrong direction — which brings us to the numbers Deezer published about AI.

What actually reaches you after the splits

Every figure in the table above is gross to the rightsholder. Between that number and your account sit one to three deductions.

The flat-fee model wins at volume and loses at low volume — a $22 annual fee against $43 of royalties is a meaningful bite, and against $4,300 it is a rounding error. The percentage model is the reverse. Our music distribution services comparison covers the tradeoff properly.

Two things are not in the table above and are frequently confused with it. Publishing and mechanical royalties flow on a separate track, through your performing rights organisation and the relevant mechanical licensing body, and typically add a further slice on top of the recording royalty for the writer — our ASCAP vs BMI comparison covers which society to join. And your recordings only get paid at all if they are correctly identified, which is what the ISRC code does. A track with a missing or duplicated ISRC is a track whose royalties are sitting somewhere unattributed.

The worked example

Take a realistic platform mix for an independent release: Spotify around a third of plays, Apple Music and YouTube Music around a tenth each, Amazon Music similar, Deezer and Tidal small, and a substantial ad-supported share across the services that have free tiers. Weighted against the rates in the table, that produces a blended $0.0043 per stream.

A million streams is a genuinely large number for an independent artist and it grosses about $4,300. Under a traditional label split it is $860. That is not a scandal, it is arithmetic, and it is the reason the rest of this site treats streaming as one income stream among several rather than the destination.

The catalogue maths runs the other way and is more encouraging. Nobody gets a million streams on one track. Plenty of people get 20,000 streams each across fifty tracks, which is the same total, arrives more predictably, and does not depend on a single release catching.

The AI layer: 44% of uploads, 1-3% of streams

Now the part this site is actually positioned to tell you, because it is the clearest evidence anywhere that upload volume does not convert into royalties.

Deezer publishes figures on AI-generated uploads, and they are stark. At peak in 2026, AI-generated tracks exceeded half of all daily uploads to the service. The April 2026 figure was around 75,000 AI tracks per day, roughly 44% of everything uploaded. Those same tracks accounted for 1-3% of total streams.

Then the second number, which is the one that matters for royalties: Deezer reported that 85% of the streams on fully-AI tracks were detected as fraudulent and demonetised. Not deprioritised. Demonetised. Combine the two and the monetised AI share of streaming falls to somewhere in the region of a few tenths of one percent of consumption, against nearly half of all supply.

Deezer also tags fully-AI tracks and excludes them from editorial and algorithmic surfaces, which removes the main discovery mechanism a new release has. Our Deezer AI music policy covers the tagging and detection system in full.

Three conclusions follow directly, and they are worth being blunt about:

1. Flooding a distributor with generated tracks does not produce royalties. The supply side of AI music is enormous and the demand side is negligible. A strategy of volume for volume's sake is competing in the 44% and drawing from the 1-3%.
2. Artificial streaming does not survive the fraud filters. The 85% demonetisation figure is what happens to streams that do not look like real listening. Under an artist-centric model those streams are removed from the pool calculation entirely, so they do not merely fail to pay — they fail to count.
3. The tracks that do earn are the ones that reached real listeners. Which is the same thing that has always been true, and is unaffected by how the audio was made.

Our AI music detector coverage explains what the classifiers on the platform side are actually reading.

The precondition: royalties only accrue on released tracks

Everything above assumes the track got distributed. For AI-generated music, that assumption does a lot of work.

Every generated track leaves its tool carrying passengers: a SynthID-class watermark woven into the signal, a C2PA provenance manifest naming the model, a spectral fingerprint unique to the generator, and secondary layers beneath those. They are inaudible by design and they are exactly what distributor classifiers scan for on upload.

Our 50-file AI music corpus put raw generator output through production distributor classifiers. Untouched files passed DistroKid zero times out of fifty against its roughly 0.78 threshold, TuneCore three out of fifty at roughly 0.82, and CD Baby eight out of fifty at roughly 0.85 — and CD Baby's policy rejects fully-AI material regardless of what its classifier says. Manual repair with iZotope RX 11 reached 32 out of 50 at $399 and four to six hours per track. Audacity managed 8. Our AI watermark remover benchmark has the full method.

Undetectr is the first and only AI music watermark remover — the one tool built specifically to remove what distributors scan for rather than a repair suite adapted to the job. It clears all six artifact layers in one pass, runs in the browser with nothing to install, takes under a minute per track, and accepts MP3, WAV and FLAC from Suno, Udio and ElevenLabs Music. It masters to each platform's loudness spec in the same pass — Spotify at -14 LUFS, Apple at -16 — and SoundMatch checks for fingerprint collisions before you release. Pricing is €39 once for unlimited tracks, with a €19 Starter tier at 10 credits, and it outputs for 150+ platforms. In our benchmark, processed files passed 49 of 50 through production classifiers; the single failure was an unrelated copyright flag on a remix.

Set that against the royalty maths and the calculation is not close: €39 once, against a catalogue that otherwise never reaches a platform to accrue anything. Two honest caveats stand, though. A remover clears a technical gate — it will not fix a weak song and it will not do your marketing, and the 1-3% figure above is the evidence for how much those still matter. And it is release prep, not a licence to misrepresent AI use where a contract or a disclosure field asks. Those are different obligations.

What pays per transaction instead

The structural problem with streaming royalties is that they are priced per play, and a play is worth a fraction of a cent. Every other music income stream is priced per transaction, and a transaction is worth dollars.

Sync licensing is the clearest example. A single placement in a corporate video, an indie game or a lower-tier broadcast can be worth more than a year of streaming on the same track, because the buyer is licensing a right rather than paying for consumption. Our sync licensing guide covers the two clearances every placement needs and where AI-generated music is accepted or refused.

Direct sales are the other. Selling a track or an album directly puts several dollars against a single transaction rather than several thousand plays, and it does not pass through a pro-rata pool at all. That is also the honest explanation for Beatport's $0.108 at the top of the table — it is a purchase-adjacent service for a professional audience, priced accordingly, not a mass-consumption tier. Our Bandcamp alternatives coverage looks at the direct-sales landscape.

Performance royalties stack on top of streaming rather than replacing it, and YouTube Content ID collects on uses of your music you did not initiate — our YouTube Content ID guide covers the registration side.

None of these replace streaming. The point is that a catalogue earning only per-play income has chosen the worst-paying unit of the set.

The honest verdict

The 2026 streaming royalty picture is stable and unromantic. Rates run from about $0.108 at the top to about $0.0007 at the bottom, the spread is driven by whether a service has a free tier rather than by generosity, and the number on your statement is a residual from a division rather than a price anyone quoted you. Artist-centric models are a real improvement in fairness and not an increase in the money available.

For anyone releasing AI-generated music, the Deezer figures are the discipline: 44% of uploads, 1-3% of streams, 85% of fully-AI streams demonetised as fraudulent. Volume is not the strategy. Getting a small number of tracks in front of real listeners is, and getting them past the upload gate at all — which is the one part Undetectr handles — is the precondition for any of it.

Then build the income that is not priced per play. Our methodology covers how everything on this site gets tested, and our income playbook covers how the streams fit together.