Wall Street Turned Songs Into Bonds. Then AI Started Writing Them.

In March 2026, a North Carolina man named Michael Smith pleaded guilty in a Manhattan federal court to a scheme that sounds absurd until you see the number attached to it. Using AI music generators, Smith created hundreds of thousands of songs, then used armies of bots to stream them billions of times. The payout for this synthetic catalog of nobody-music: more than $8 million in royalties siphoned out of the same pool that pays Drake, Taylor Swift, and the estate of every legacy artist you can name.

Smith got caught. What should worry investors is everyone who didn't — and the fact that the machine he exploited is now industrial-scale. In April 2026, the streaming service Deezer disclosed that fully AI-generated tracks now make up roughly 44% of the music uploaded to its platform every day — on the order of 75,000 new synthetic songs every twenty-four hours. The company also estimated that up to 85% of streams on those AI tracks in 2025 were fraudulent.

Now here is why this belongs in a markets briefing and not just a culture column. Over the last five years, Wall Street quietly turned song catalogs into one of its favorite new fixed-income products — and it did so on the explicit promise that streaming royalties are a stable, inflation-resistant, bond-like cash flow. That promise is now colliding with an infinite supply of machine-made competitors for the exact same pool of money.

How songs became bonds

The pitch was seductive, and for a while it was right. A catalog of durable hits — the kind people stream every single day, decade after decade — throws off a remarkably predictable annuity. Unlike an oil well or a patent, a great song doesn't deplete and doesn't expire for the better part of a century. So the smart money went shopping.

Blackstone took Hipgnosis Songs Fund private in a roughly $1.6 billion deal in 2024, absorbing some 45,000 songs, then led a landmark $1.47 billion music-royalty asset-backed securitization (ABS) — literally packaging the royalty streams into bonds and selling them to institutions. KKR built a private-credit book against catalogs. Primary Wave closed its fourth fund at $2.225 billion in April 2026 and struck a deal to acquire Kobalt. By the ratings agency KBRA's count, more than $12.9 billion in music-royalty-backed bonds have been rated since 2020, with senior tranches priced like the high-grade credit they were marketed as — roughly 3.8% to 4.2% yields — and equity tranches reaching for 14% to 18%.

Every one of those structures rests on the same load-bearing assumption: that the royalties flowing to a given catalog will hold up, or grow, as global streaming grows. That is the assumption AI is quietly rewriting.

Briefings like this land in members' inboxes before the market prices them in. Join free →

The pool is fixed. The claimants are not.

Here is the mechanic almost nobody outside the industry understands, and it is the entire thesis. Spotify, Apple Music, and the other major platforms pay on a pro-rata model. Every month, they take all subscription and ad revenue, remove their cut (around 30%), and drop the rest into one giant pool. That pool is then split among rightsholders according to each one's share of total streams on the platform.

Read that again, because the word "total" is the whole problem. The pie is a fixed size — it's just the subscription revenue. It does not grow because more songs exist. So when 75,000 machine-generated tracks arrive every day, and bot farms inflate their stream counts, they are not baking a bigger pie. They are cutting the existing pie into more slices. Every fraudulent or synthetic stream is a claim on the same fixed pool that the ABS bonds and the nine-figure catalog acquisitions are being paid out of.

For a decade, catalog buyers underwrote their purchases on rising per-stream economics and relentless subscriber growth. AI-generated supply attacks that from the denominator: even if the pool keeps growing, an exploding number of claimants means the per-stream rate — the thing every catalog valuation is discounted off of — comes under pressure that no spreadsheet from 2023 modeled.

So the question that decides whether this is a genuine crack in a $12.9 billion bond market or a manageable nuisance is not "is AI music real." It clearly is. The question is who actually eats the dilution — because it is not spread evenly, and the answer determines exactly which of these assets are mispriced and which are quietly protected.