‍

The licensing deals are being signed. The settlements are being announced. The trade press covers each one as evidence that a new market is opening up for rights owners.

Our read is different. The value that AI companies needed from creative IP has already been extracted. What's happening now is the payment of fines, the settling of court cases, the management of legal exposure. That's not a licensing market opening. That's trickle-down economics being used to describe something that isn't trickling down.

And if you look at who is actually sitting at the table, the trickle-down framing becomes hard to sustain.

‍

Who the architecture was built for

‍

Look at the licensing arrangements that have been struck around future model training. Then look at whose names appear on them. The big three. The largest publishers. The major rights owners who can negotiate at the scale that makes a counterparty relationship worth having for an AI company.

Now look for a smaller label catalogue in those arrangements. An independent fund's portfolio. An artist who owns their masters outright. You won't find one. Not one structured deal that passes money through to the smaller rights owners whose work is also in the training set.

We're not saying anyone deliberately excluded them. This isn't a motive argument. It's an architecture argument. The settlement structure being built around future AI models is shaped so that money flows to a handful of large counterparties and stops there. Whatever those counterparties collect, there is no mechanism carrying it further down. The architecture excludes independent and smaller rights owners by design, not by intent.

So as a catalogue owner outside the major label system, the picture looks like this. Your IP contributed to the training data. The value the models needed from it has been extracted. And the licensing structure being built to compensate for that extraction was not designed with you in the room.

We think this is the story. The biggest companies cream the profits. Everyone else gets the supply shock with none of the licensing income.

And the supply shock is the part that most of the conversation is ignoring.

‍

The liquidity event nobody seems to be pricing

‍

Set the revenue side aside for a moment and look at supply.

Deezer publishes its numbers, and they represent the clearest single piece of evidence we've seen on this. In January 2025, roughly 10,000 AI-generated tracks were being uploaded to the platform per day. By January 2026, that figure was around 60,000 per day. By 2026, AI tracks accounted for 44% of all daily uploads. Of the streams those AI tracks generate, Deezer reports that 85% are fraudulent, which is its own problem, but set that aside.

Spotify removed roughly 75 million tracks it described as spammy in a twelve-month window. The platform's total catalogue sits somewhere around 250 million. Sit with that ratio for a second.

That is not a trickle of novelty content at the edges of the market. That is the supply of recorded music roughly doubling, in months, with production cost falling toward zero. Human creativity being replaced at scale, and cost of production coming down at scale, simultaneously. There could not be more potential disruption ahead, and the market is treating it as a footnote.

Here is why that matters for the thing you actually own. Streaming royalties come out of a pool. The pool divides by share of streams. If the denominator, the total volume of music competing for those streams, roughly doubles, then every existing track's share of the pool gets diluted before you account for any change in listener behaviour. More supply into a fixed revenue pool means less per stream for everything already in it. That is not a controversial claim. That is how the pool works.

The same supply-and-demand logic plays out across the rate dispersion between platforms. People point to high-rate DSPs as a structural haven, somewhere the per-stream economics still look defensible. But a small premium platform pays a higher rate partly because it is small. If it scaled into a mass-market platform, wouldn't it face the same pressure on per-stream rates that the large platforms already face? The rate dispersion is not a structural escape. It is the same economics observed at different scales.

‍

What a market due for a correction looks like

‍

Our background is in pricing markets for a living. Sports, equities, anything with liquidity and a number attached. You develop a feel for what a market looks like when it's about to reprice. Not certainty. A feel. And music rights is showing a lot of the symptoms at once.

Catalogue multiples have sat roughly flat for about three years. Flat, through a period where the underlying supply picture changed more than it has in the entire history of recorded music. A market that doesn't move its prices while its fundamentals move underneath it is either seeing something we haven't, or it isn't looking.

Meanwhile the survey data says 86% of institutional investors plan to increase their allocations to music rights, and around 21% describe themselves as unconcerned about AI's impact. We are not going to quote individuals or put words in anyone's mouth. The price behaviour says it plainly enough on its own. Multiples flat for three years against a doubling of supply is the complacency, expressed as a number.

We are also watching what the smart money is doing rather than what it is saying. Selling to wealth funds. Selling to GIC. ABS instruments appearing in the market, complex derivatives wrapped around catalogue assets. These are not the portfolio moves of investors with a high-conviction long position. They are the moves of investors managing their exit.

We will be honest about the other side of this. We might be early. Markets can stay mispriced for a long time, longer than seems reasonable, and "due a correction" is not the same as "correcting next quarter." Plenty of people who called a top were right about the direction and wrong about the timing by years. So we hold this loosely on timing and tightly on direction.

But if you are modelling a catalogue's discounted cash flows on assumptions built from a time when maybe 10,000 AI tracks were being uploaded per day instead of 60,000, you are pricing an asset that no longer exists. The numbers underneath the model came from a world that is likely to be irrelevant to what lies ahead.

That is our read. The value has already been extracted and it is permanent. The licensing architecture that's meant to compensate for it structurally skips most of the people who own rights. The supply of music is roughly doubling while the revenue pool is not. And the market is priced as if none of those three things is happening.