97% Can't Tell the Difference

I'm not a music journalist, not a critic, not a music industry professional. I'm someone who loves music. A DJ at my own scale within the Kombava collective in Lisbon, passionate about it my whole life. If I'm writing about artificial intelligence in music today, it's because Wednesday evening something made me stop and think.

TMC. Quotidien. Yann Barthès welcomes Antoine Monin, CEO of Spotify France and Benelux, for the platform's 20th anniversary. The conversation quickly drifts toward AI. And Monin says something that stays with you long after the closing credits: "Who are we to judge?"

His position is clear. AI in music is a technological revolution like the dozens music has already lived through. The synthesizer, the sampler, the vocoder, DAW software. Each time, the same outcry. Each time, music continued. He offers a figure: 26% of tracks indexed on Spotify are now linked to AI in some way. And he adds: "If you have no talent, you can use every tool you want - you still won't have talent."

That deserves a closer look at the data.

AI Uploads: What the Numbers Show - and What They Don't

Spotify doesn't publish precise data on the proportion of AI tracks in its uploads. But Deezer does, and it's the only streaming platform in the world to do so publicly since January 2025. What they show is vertiginous.

In ten months, the share of 100% AI tracks in daily uploads went from 10% to 34%. In raw volume: from 10,000 to 50,000 tracks per day. A 400% increase. And Spotify acknowledges that all platforms receive "essentially the same catalogue."

But here's the paradox - and this might be the most important number in this article: those 34% of uploads represent just 0.5% of actual listening. AI music is a tsunami of production, not consumption. Not yet.

(Look at the chart above. That tiny green bar? That's 0.5%. We almost left it out, it's so invisible. We kept it. That's the whole story.)

Meanwhile, Spotify removed 75 million spam tracks in twelve months. To put that in perspective: Spotify's active catalogue contains roughly 100 million titles. They cleaned out the equivalent of three-quarters of their entire library.

Suno: 7 Million AI Tracks Per Day

To understand why the curve is climbing so fast, you need to look at the machine feeding it. Suno is four former Kensho AI engineers in Cambridge, an open-source model published on GitHub in 2023, and 18 months later: 100 million users, 2 million paid subscribers, $300 million in annual recurring revenue. Seven million tracks generated every day. Seven million.

The CEO of the Recording Academy - the people who run the Grammys - recently said that "every" songwriter and producer he knows is now using Suno. To unblock ideas, test melodies, generate starting points. The key word: every.

And there's something worth acknowledging here: AI enables musical crossovers that simply wouldn't have been attempted before. Country blended with boom-bap hip hop. Combinations no producer would have tried, that sometimes yield something genuinely interesting. It's not all spam. There's real creative value possible.

In November 2025, Suno settled a $500 million dispute with Warner Music Group. The agreement authorizes training its models on WMG's catalogue and includes the acquisition of Songkick. Universal and Sony are still pursuing their own proceedings. The underlying generative AI music market is estimated at around $570 million in 2024, projected to reach $2.8 billion by 2030.

I've been using Suno for a few weeks. Not for this article - for fun. I type a prompt in the evening, listen to what comes out, iterate, start again. And I'm taking a pleasure I hadn't anticipated in producing tracks without knowing how to play a single instrument. I made one, called "Miradouro" - we come back to it at the end.

But there's an angle that neither Monin nor Quotidien touched, and that Alexis - a friend from Kombava and the only actual music producer in our collective - pointed out in a recent conversation. The question isn't just philosophical, it's economic. An AI track doesn't negotiate its royalties. An AI artist doesn't ask for an advance, doesn't demand promotion, doesn't have a tour to coordinate. For a platform like Spotify, AI is the ideal content supplier: infinitely scalable, always available, and cheap.

Alexis goes further. 95% of people aren't music obsessives. They just want something in their ears. The first Spotify playlist that loads, done. Some lofi hip hop to work to. Background sound for the commute. For that 95%, AI changes nothing about the experience. It's us - the 5% of genuine music lovers - who see the problem. And maybe that's also why Spotify isn't in any rush to solve it.

Booba, AI, and the Question Nobody Asked

Let's talk about Booba. A fan from the beginning, less so recently. But his recent appearance on "Carré d'As" alongside Oxmo Puccino, Pit Baccardi and Lino was a rare moment in French rap. Four legends gathered to talk about the future of their art.

And it's in the middle of that conversation that the Duc reveals he used AI to create "DD" - his tribute to Désiré Doué that entered the top 10 - and "Seychelles." Then he adds, with a knowing smile: "I've got features with Shakira - without Shakira."

The Duc says it openly. How many artists are doing the same without saying so?

This is exactly the blind spot in the debate. We talk about "AI music" as a monolithic block. But the reality is a spectrum. Booba writes his lyrics and generates the beat - is that 50% human? 70%? The Beatles used AI to restore Lennon's voice. David Guetta created a fake Eminem feature. The Velvet Sundown, an entirely AI fictional band, accumulated millions of streams before anyone noticed. And iHeartRadio launched its "Guaranteed Human" programme, promising to broadcast only human voices.

Who draws the line? And more importantly - is it really the public deciding what they listen to?

The Spotify Algorithm: Are You Really Choosing Your Music?

Think about your listening over the past week. How many tracks did you actively search for yourself - by artist, by title, by a specific desire? And how many arrived in your ears via Discover Weekly, Release Radar, Daily Mix, or the autoplay that kicks in after your last chosen track?

I listen to a lot of playlists, like everyone. But being a DJ forces me to break out of the loop - to search, to dig, to listen to things the algorithm would never surface for me. Laurent Garnier says the power of DJing is sharing the unexpected. He's right. But the unexpected has to be found first. And the algorithm, by definition, will never take you there. It takes you toward the predictable.

The recommendation algorithm doesn't just distribute music. It selects it. It composes your menu. And if tomorrow the machine also composes the music itself, the audience isn't choosing anything at all. They're consuming what the algorithm both produces and recommends. A closed loop.

Push the logic further. Spotify has already launched "Prompted Playlists" - AI-generated playlists from a text prompt. The next step is predictable: music generated on the fly, calibrated to your mood, your history, your habits. No shared catalogue. No artist. Just a continuous stream, built for you. Like the Instagram feed, like targeted advertising, but as a soundtrack.

This isn't science fiction. It's the logical extrapolation of what already exists.

Push it one more step: imagine a morning where Spotify knows your sleep data, your agenda, your prompts from the night before. It generates a track - not from a catalogue but from scratch - calibrated to your exact mood at 7:43am on a rainy Thursday. Music that exists only for you, at that moment. Ultra-personalised. Ultra-disposable. The absolute opposite of an album you'll still be listening to ten years later.

It's technically possible. It's probably profitable. And it might be exactly the future nobody actually wants to live in.

There's an inverse signal worth noting. Vinyl has never sold as much as it does today. I've been buying my favourite albums at a feverish pace for the past two years myself. Not out of nostalgia. Because a vinyl record is a work. You choose it, you put it on the turntable, you listen to both sides. It's the exact opposite of the stream. Before, you had to go out and buy a record. You consumed music differently. And apparently, more and more people want that back.

The Return of Live: What AI Will Never Replace

I just came back from Hans Zimmer at the MEO Arena in Lisbon. And what struck me was watching real musicians - the best in their field - playing together in front of thousands of people. No algorithm produces that. No prompt replaces it.

And people want it. Look at the records Céline Dion is breaking. Look at the queues. There's a massive hunger for music experienced together, for shared performances in the flesh.

AI makes music production infinite. Costs tend toward zero. Volume tends toward infinity. In a world where anyone can produce a track in 25 seconds, value migrates mechanically toward what can't be replicated: presence, the moment, the body.

The movement is already underway, and it started before AI. Streaming revenue per artist stagnates or falls. For musicians, money is increasingly made in live performance. AI didn't create this trend. It accelerates it.

But audiences now expect something different from live shows. Not just a concert. An experience. Look at the Sphere in Las Vegas, reinventing what it means to "see a performance." Look at what Kendrick Lamar builds on stage. Look at Tomorrowland, where visual immersion has become as essential as the music itself. Tomorrow's live show isn't yesterday's. It's bigger, more immersive, more multisensory.

And then there are moments that give you hope, simply. Thomas Bangalter with Fred Again. A former Daft Punk and a producer from the new generation, together, on stage, live. No AI. No filter. Just two artists and the audience between them. Proof that the people who matter in music are returning to the living.

This might be the best news in this article: AI isn't killing music. It's putting the human body back at the centre. The performance, the direct emotion, the risk of the moment. Everything the closed loop will never be able to replicate.