4 Lessons about AI, Formats, and Content Economics
Something interesting found its way into my feed recently. “You have got to see this!” my wife, a classically trained violinist and composer, noted. Amy Slaton, of 1000-Lb Sisters fame, has recently released a song called Pounds Hollow on Spotify. She even engaged in a tongue-in-cheek “Thank you for your consideration” Grammy campaign. The one hitch is that she didn’t really write that song, at the very least, not the music, and quite obviously can’t sing to save her life (you’ll have to watch the clip to see for yourselves). Instead, she used Suno’s popular AI tool to create what I can generously call the majority of the song. If you read the credits, though, they list her as lyricist, songwriter and composer. No mention of AI. Spotify acted promptly(ish) and removed the song from the platform (at time of writing).
Traditionally, this would be the point to turn into a philosophical debate about creativity in the age of AI. After all, an AI-made track made it to No. 58 on the Billboard Hot 100 earlier this year. But philosophy doesn’t show up on a P&L. We will also not be discussing whether AI is coming for creative jobs, the argument that drove Hollywood’s actors and writers to the picket lines. As interesting and important as those discussions are, this piece has a much narrower focus.
If you own IP assets and engage in their monetization for a living, the more pressing question for you is: what will AI do to the value of my catalog?
About 25 years ago, music was the first content industry to be hit by a technology (read: the internet) that changed how people consume it. Film, TV, publishing and everything else built on IP came next. When that transition took place, I was working for Sony BMG, one of the four major global music labels which had to confront that transition head on. My boss, the president of the Global Digital Business Group, needed to balance the growth of digital formats against the rapid decline of physical ones (namely, CDs). It was there that I cut my teeth in harnessing technology to drive revenue growth and learned important business lessons about technology and content economics. Allow me to share the four most important ones.
Lesson 1: No format lasts forever but demand endures.
In 2008, I was invited to speak at my alma mater about the ongoing transition in the music industry. I shared with the eager MBA students one of my favorite charts, an updated version of which is shown below.
The lesson could not be clearer: in the US, vinyl was king until it was replaced by cassette tapes and then the Compact Disc. The CD reached its zenith in 2002 and was replaced by downloads (for a while). Downloads, the “own your music” model, peaked at 43% of revenue in 2012 and were 3% by 2023. No format that ever made the bulk of the industry’s money still does.
And yet the industry is still here. Focusing on the overall revenue line (wholesale, in 2025 dollars), US revenue fell from $16.0 billion in 1999 to $6.5 billion in 2014 but has climbed back to $11.5 billion. Not quite all the way, still about 28% below the peak, but, and this is the important point, people never stopped listening to music. They stopped listening to music on one format moving on to another.
What new format will AI bring? We are still in the early innings, and it is hard to gauge which format (if any) will replace streaming, the current king of the music castle. In an age of infinite creative possibilities, could stems (pre-mixed groups of combined tracks that make up the components of a song) become a much bigger revenue stream than they are today? Time will tell.
Lesson 2: The shiny new object is unlikely to last.
I remember a distinct conversation I had with a colleague in 2007 about the growth of ringtones. At the time, Sony had its own ringtone store on Verizon’s phones and lucrative deals with the other carriers. My colleague, who was keenly interested in expanding the universe of ringtone buyers, asked for my opinion on how to achieve that. Having read the focus group reports and seen the first iPhone in action, I posited that ringtones might have reached their peak and that he might want to try to sell more ringtones to the people who were already buying them.
Today, ringtones are all but extinct. They were a product built on a temporary interface, and when the interface changed (read: the iPhone’s introduction in 2007), the revenue went with it.
The ringtone-equivalent of the AI era might be the licensing deals for model training. Universal settled with Udio in October 2025, and Warner with Suno in November. In June, the National Music Publishers’ Association (NMPA) struck an industry-wide deal with Udio that values songs and recordings equally. Disney put $1 billion into OpenAI and licensed more than 200 characters to its Sora video tool.
While financial terms have not been publicly disclosed, there is reason to wonder how long the leverage will last: Suno’s newest model was trained on licensed catalogs plus its own users’ data. If user data (or synthetic data) becomes the moat, each additional catalog could be worth less.
Lesson 3: What sticks often starts out looking small.
In 2007, subscriptions services were barely a blip on the US recorded music revenue chart. Napster had about 830,000 subscribers. Steve Jobs was telling Rolling Stone that people want to “own their music” and music industry heads (naturally) agreed with him. I certainly wasn’t about to part ways with my impressive CD collection. That same year, YouTube launched Content ID, which fingerprints uploads and lets the rights owner block them or share the ad revenue. Until that point, the industry (represented by the RIAA) had been suing its fans. Now it had an option to monetize them. By 2016, Google said it had paid out over $2 billion, with half the music industry’s YouTube revenue coming from fan content (although labels dispute this figure). More impressively, by 2025, streaming (including all subscription types) was 82% of the US market, and I had long given up my CD collection.
The AI era version of this is still taking shape. Spotify’s AI covers and remixes tool is licensed by several rightsholders, including Universal. Disney plans to stream curated fan videos made with Sora on Disney+. Warner’s CEO recently called it “creation-based revenue”. None of these revenue streams currently appears to be meaningful. In the not-too-distant future, one of them might look a lot bigger.
Lesson 4: Forecasts extrapolate the present.
According to a persistent urban legend, in the early 80s, AT&T hired McKinsey to run a study evaluating the viability of mobile phones as a new market opportunity. McKinsey famously came back projecting that the US would have 900,000 cell phone subscribers by 2000. Fast forward 20 years and you’ll find 109 million subscribers. McKinsey was off by a factor of about 120. This is not meant to disrespect McKinsey but rather to illustrate that forecasts fail because they measure consumer behavior today and assume it stays the same. As we have shown already, that is rarely the case.
Let’s run a quick thought experiment. Suno has reportedly passed 2 million paying subscribers and $300 million ARR in February 2026. Last month, its CEO said it blasted through these metrics. If music catalog licensors kept 20 cents of every dollar paid to the platform (generous assumption), we’d be talking about $60 million a year, minuscule in comparison to almost all other revenue streams. Could Suno’s subscribers grow like the mobile carriers’ did? At an implied $150 ARPU, if Suno gets to 100 million subscribers it would generate $15 billion. At 20% revenue share, catalog owners could capture $3 billion of that. For context, it took Spotify 11 years to get to 100 million paid subscribers. Suno’s investors seem to think that’s doable. Its latest funding round valued it at $5.4 billion.
The key lesson is not to assume that consumer behavior today would stay the same forever.
What is a GP to do?
The genesis of catalog-backed securities might be traced back to one of my favorite artists. In 1997, David Bowie sold bonds against record sales of his early catalog. They started with a stellar credit rating from Moody’s. Then the CD market cracked, and in 2004 that same agency cut them to one notch above junk, citing weak recorded-music sales. The bonds never missed a payment, though, and were paid off in 2007. Bowie Bonds, as they became known, could be viewed as the grandaddy of today’s catalog-backed securities, and they neatly encapsulate an important lesson about collateral that turned out to be a bet on a format.
Not everyone was as fortunate with their bets. In 2007, Terra Firma, Guy Hands’ PE shop, bought EMI at a £3.2 billion EV. My colleagues and I at Sony BMG raised an eyebrow. It was quickly evident to us that the Terra Firma crew hadn’t fully understood what they’d bought, that costs they found after closing surprised them, and that their thesis about digital music wasn’t evolving as fast as they needed it to. By 2011, Citigroup, the lender, took control and wrote down EMI’s debt by 65%. Guy Hands and his team were very smart people and the folks they brought to run EMI (some of whom I’ve personally met) were top notch operators. The asset wasn’t the problem either. Sony Music, which ultimately ended up with the crown jewel of EMI Group – EMI Music Publishing – in the ensuing asset sale, is still minting money off it.
Will AI push catalog values up or down?
The bear case: Fitch warned in October 2025 that AI music could weaken demand for artist-made content and the royalties behind music bonds.
The bull case: So far, the numbers haven’t validated Fitch’s warning. Catalog is driving over 72% of US streams. In fact, US music publishing revenues alone grew from $2.95 billion in 2014 to $7.3 billion in 2025 (based on NMPA figures in 2025 dollars). That’s a 2.5x in 11 years. In March of this year, Francisco Partners sold Kobalt to Brookfield-backed Primary Wave for about $1.5 billion, double what it paid for it four years earlier.
All this seems to suggest that Guy Hands’ mistake was a hold period and a debt load mismatching a transition that took fifteen years to play out.
Which leaves us with three questions that matter more than any forecast:
(1) Do you have the time? The music industry took fifteen years to reach bottom and eleven more to climb back. A typical fund hold is 5-7 years. If the transition outlasts your fund, the question isn’t whether the asset is good. It’s whether your capital structure can wait.
(2) Which side of the behavior change are you on? For every new AI revenue line, ask: is it more like a ringtone, a feature on someone else’s interface, or like a subscription, a recurring relationship you own? Whose customer is it? And can you meter it, the way Content ID did?
(3) Are you ready to monetize? Are your rights and consent data licensing-ready? Obtaining the necessary artist permissions is a complex and labor-intensive process. Know what you have and what’s missing. It is the difference between collecting money now or a year from now.
What the last twenty-five years suggest is that the songs themselves – as an asset class – came through just fine. Asset value tended to go up and to the right, though not always in real terms and not for every owner. So, find the sentence in your model that sounds like “people will always want to own their music.” Otherwise, you might end up with a Pounds Hollow ringtone on your hands.
