Two Million Subscribers, Two Missing Majors: Inside Suno's Licensed AI Music Models, September 10, 2026

0:00Welcome to the DX Today podcast. Your daily deep dive into the AI ecosystem. I'm Chris, and joining me is always Laura. Thanks, Chris. Today we are digging into a story that has been building for two full years and finally landed yesterday. And I think it genuinely changes the shape of the argument about AI and creative work. Set the scene for me, because a lot of people have heard the name and have a vague sense that there is a controversy without really knowing where any of it actually stands right now. The company is Suno. Cambridge, Massachusetts founded in 2022 by Mikey Schulman, Martin Kamacho, and George Kutsko. You type a description of a song and it hands you back a finish track with vocals on it. That description undersells how much friction it removed. Because for years, the confident take was that music generation would never get past the uncanny valley of sounding almost right, but not quite convincing. It got past it startlingly fast. More than 100 million people have used the product. Two million of them are paying subscribers. The company crossed $300 million in annual recurring revenue

1:01back in February. And those are not vanity numbers. Those are the kind of numbers that make an entire industry stop treating you as a novelty and start treating you as either a partner or a defendant. There's no third option once you are that large. Both, as it turns out, Suno raised around in June of 2026 that valued it at $5.4 billion. And it has taken in close to a billion dollars in venture capital overall. So walk me through what actually shipped yesterday. Because I gather this is not just a routine version bump where the model gets marginally better at guitar tone and everyone moves on. Three models shipped. There is version six, the flagship for paying subscribers. There is version six wild, which is deliberately less predictable and more exploratory. And there is version six mini, which is the free tier. Fine, that is a normal product ladder. What makes this different from every other model release? We cover on this show, weekend and week out all year long. This one was trained from scratch on licensed music, Warner Music Group, BMG, and believe all handed over catalog

2:03under commercial agreements. The chief product officer, Jack Brody, said it was trained entirely from scratch from the ground up. Okay, that is the sentence that matters. Because the entire legal and moral fight over generative music has been about training data that was taken rather than bought. And this flips that premise. It flips it partially and I want to be precise here. Because the partial part is where all of the interesting tension actually lives. And most of the coverage I read yesterday glided right over it without pausing. Then let's not glide over it. What is missing from that licensed pile? And how much does the absence actually matter to the quality of what comes out the other end? Universal Music Group and Sony are not in it. Soono stated flatly that there is no music from Universal or Sony that goes into training its version six models. Those two remain plaintiffs in the Boston copyright case. So the two largest record companies in the world are still suing. While the third and fourth largest have taken the deal and are now commercial partners in the very same product that is an extraordinary split inside one industry.

3:06It is the single most important fact in this story. Warner settled in November of 2025 and promised license models within the year. Yesterday was Warner collecting on that promise right on schedule. I want to push on something. If the two biggest catalogs in recorded music are excluded is the resulting model actually worse because that would be the obvious argument for holding out and refusing to license. Brody addressed that almost preemptively and his answer was pointed. He said a lot of the things that make these models better are actually not the data it is trained on. That is a remarkable thing to say out loud while standing next to your licensing partners because it is essentially telling every rights holder in the room that their catalog is a nice to have. It cuts both ways though. It is also the argument that makes licensing affordable in the first place. If catalog were the whole ballgame, no startup could ever pay what the majors would demand for it. Let's talk about the money because revenue sharing is one of those phrases that sounds meaningful and frequently turns out to mean almost nothing

4:08once you read the actual arithmetic underneath it. The structure is that a portion of Suno revenue goes to the licensing partners starting from day one. Not after some future milestone, the rights holders then distribute that internally to their artists and songwriters. And there it is. That second sentence is where I get nervous because the history of the recorded music businesses, largely a history of money stopping somewhere between the label and the artist. You're not wrong to flag it. Robert Kinkall, who runs Warner, framed it as artists and songwriters sharing and revenue generated by more than two million paying subscribers. But no percentage was published anywhere. No percentage published means we cannot evaluate it. We are being asked to accept a structural principle as if it were a settlement. And those two things are genuinely not the same thing at all. Agreed. And I would add that the subscription prices give you a ceiling to reason about. Plans run from roughly $8 a month up to about $30 a month at the top tier. So do that arithmetic for me out loud because I suspect the total pool here is smaller than the language around it

5:10makes it sound to a working musician reading the headline. Take two million subscribers at a blended rate somewhere in the middle of that range. You get a few hundred million dollars a year in gross revenue. A portion of that gets shared across three major rights holders, which is real money at the corporate level and can still be almost invisible at the individual level. Once it is divided across catalogs containing millions of recordings and hundreds of thousands of writers. That is the honest read, though I would argue the interesting part is not this year's check. It is that a creation-based revenue stream now exists at all. Alongside the streaming stream. Say more about that distinction because I think that is the part most people will miss while they are arguing about whether the percentages fair or insulting to working artists. Streaming pays when someone listens to an existing recording. This pays when someone makes something new using the model. Those are structurally different economics. And the second one has never had a revenue line before. The Chief Product Officer made a related claim that I want you to stress test,

6:11which is that this new revenue does not cannibalize what artists already earn from conventional streaming platforms. I am skeptical of that claim is stated. Attention is finite. If a listener spends an hour generating and playing their own tracks, that is an hour not spent streaming somebody else's catalog. Right, and the counter argument would be that the person making custom background music for a short video was never realistically going to license a Warner recording for that purpose in the first place at any price. That is the strongest version of the industry's case, and it is genuinely plausible for a large share of use. The uncertainty is what happens as quality rises and the substitution starts biting into real listening. Let's turn to the part that did not get resolved yesterday because I noticed the litigation column is still remarkably crowded even after these deals were signed and announced with a great deal of crowded is an understatement in Germany, the collecting society Gemma won its case against Suno in the Munich Regional Court on July 31st. That is a decided loss not appending complaint.

7:11And elsewhere because a single adverse ruling in one European jurisdiction is survivable, but a pattern of them across multiple territories starts to look like a structural problem for the business model. Coda and Denmark sued back in November of 2025. So Canon Canada filed a claim on September 2nd just over a week ago and Round Hill music is seeking more than a billion dollars. A billion dollars from a company valued at 5.4 billion is not a nuisance suit. That is an existential number. And I gather Round Hill has not signaled any interest in settling. None at all, publicly. And then there is a class action filed on September 1st by Jason Isbel and three other artists. And that one is not about recordings. It is about names and likenesses. Explain why that distinction matters because I think most listeners would assume a copyright case and a likeness case are basically the same complaint wearing slightly different legal clothing. They are not licensing a catalog does not license an identity. You can lawfully train on a recording and still be accused

8:13of letting users conjure something that sounds like a specific named performer. So the licensing deals, however large they turn out to be, do not actually close that exposure at all. Which means soon-o-bought piece with three companies and left an entirely separate category of legal claim completely untouched. Exactly. And there is one more piece of history that keeps resurfacing. Which is that soon-o-admitted in court that it used a common downloading tool to pull audio from a major video platform. That admission is going to follow them for years because it converts an abstract argument about fair use into a very concrete factual record about how the earlier material was actually obtained. It does. Although I would note the counter argument the company is effectively making now, which is that version six is a clean room build and the old models were retired the same day it launched. Retired the same day is a detail I want to dwell on because that is not just a legal cleanup gesture. That is also a product decision imposed on two million paying customers overnight. And they noticed there was immediate criticism from existing

9:16subscribers. Some of whom said the new models sounded different from what they had built workflows around and threatened to cancel outright. Which is the recurring tension in this whole industry, isn't it? Legal safety and user preference are pulling in opposite directions and the company chose legal safety without giving anyone a migration window. It chose survival, I would say. And there is a second phase coming that may matter more than the models themselves. An artist opt in program where individual artists choose to participate and get compensated when activated. Now that is interesting because it moves the unit of consent from the label down to the individual artist, which is the thing critics have been demanding this entire time. It is and believe in TuneCore announced artist participation the day before launch. TuneCore is notable because it had previously refused to distribute this kind of material and has now reversed that position entirely. A distributor reversing course is a genuine signal about where the independent side of the market thinks this is heading because those companies live or die on the trust of the artists who use them. Right. And Suno is not operating alone here. UDO has taken a walled garden approach with universal,

10:21Warner, Merlin and Cobalt and Spotify struck its own deal with universal back in May. So the entire category is converging on licensing at roughly the same moment, which suggests the holdouts are negotiating rather than objecting on principle. Whatever the public statements happen to say. That is my read. The artist Ali Gatye put it in a way I keep thinking about saying it is our job as artists to help ensure this becomes a net positive for music. If you had to name the one thing our listeners should actually watch over the next few months. What is the signal that tells us whether this model worked or quietly failed? Watch whether an individual artist publishes what they were paid. Not the label, not the aggregate, an individual. That single disclosure will tell us more than every press release combined. That is a good test and it is falsifiable, which is more than most of what gets promised in the space. Laura, thank you. This was a genuinely useful hour of untangling. Always a pleasure, Chris. And I would just add that whatever anyone happens to think about the technology itself, the era of training quietly first and apologizing to rights holders later is

11:26clearly and permanently over now. That's all for today's episode of the DX Today podcast. Thanks for listening and we'll see you next time.