AI and Spotify
Will GenAI come for Spotify and other DSPs like it has for SaaS?
Spotify’s stock has come down a decent amount over the past six months and I’ve received a few questions about this and the connection to AI, so I figured I’d share my thoughts on how GenAI plays into the music value chain.
Music meets AI.
AI helps generate music. Lots of music. Decent music. Music that sounds like any genre you like.
Much like other AI tools this has three large impacts:
Professionals can become power users to streamline their process, lower costs, brainstorm.
Lay people can become creators! Yay! We can all create music. Some people will do this and now become ‘traditional artists’.
New forms of music/sound will be created. Much like the selfie didn’t exist until phone cameras made it easy and possible, which then gave rise to whole new apps, like Snapchat. ‘Vibe coding’ music will give rise to new formats - music remixes, audio GIFs? funny sound effects etc.
Music however has copyright, which is quite different from the open web and the written word. Licensing is a thing, and the entire industry is built around enforcing it. Much like the NYT, Reddit and StackOverflow are working on ensuring they get their fair share from LLM model makers, so to are UMG, Sony and Warner Music.
What’s this look like in the music value chain? Probably not all that different from what it looks like today. It complicates payouts and rights for mashups and collaborations that are AI generated, but covers, mashups and collaborations aren’t new in the industry.
New GenAI music companies
Suno.ai (or other similar platforms) license rights from music labels. Users pay to use these platforms to generate/create music. Suno then pays the labels based on whatever contract they’ve inked.
Why do users pay Suno? For some it’ll be fun and fulfilling to create music. For others it will help them create music to monetize. Where will they monetize this music? Probably DSPs. Sure, they could do a Patreon model, but DSPs are the natural aggregators in the music supply chain. Artists want to be where most of their listeners are - that’s Spotify, YouTube Music etc.
There’s always a chance that a Suno.ai could generate enough traction to break out and become a major hub of its own for listening - i.e become it’s own DSP or an ‘instagram of music’ (come for the creation, stay for the network). But this is very small. The market is already owned by the largest companies in tech. Your regular user simply won’t change their listening app. This thesis was tried and failed in 2022 with NFT based streaming music companies. They all failed on distribution, despite having decent traction with artists.
Existing DSPs
Existing DSPs (Spotify for example) could develop or license GenAI models to incorporate these features within their existing offering. Always wanted to hear what Taylor Swift x Snoop Dog sounds like? Always dreamt of an epic guitar solo in Coldplay’s Fix You (I have)? Here you go! You can also save that song and add it to your playlist, and share it with friends. This kind of mashup would need to pay royalties to Snoop Dog, Taylor Swift, and Coldplay via their record labels as well as to the AI model company (who would also pay a licensing fee to those same labels for the data).
Perhaps their is a gold mine out there in new GenAI related music features. Perhaps it’s related to new social features, or media types. Perhaps not.
Music is a weird asset. Unlike books, blog posts or pictures, the catalogue of older music is vastly more valuable than newly created ones. While there’s an infinite number of times you can remix Back in Black, you’re still utilizing Back in Black. There’s a limit to the new use cases and paradigms of GenAI in music. They revolve around the core song building block. I don’t think that changes.
AI Spam?
A third option is that AI generates music and distributes it, at massive scale. Call it a music “agent” or simply someone who wants to game the current DSP payout system (similar to frauds today) by uploading hundreds of thousands of music tracks.
Except that these aren’t frauds. These are “real” songs. “AI slop” even though it might be quite good music. This kind of behavior would be frowned upon by the industry I’d Imagine.
I’d assume this gets fought back against, either by banning, treating it like spam or giving a lower royalty rate for clearly identifiable AI music, making it less profitable to begin with.
Distribution wins
This leads me to the conclusion that it’s easier for the popular DSPs to integrate GenaI music features than it is for a GenAI company to win share from popular DSPs.
GenAI also offers the ability for Spotify to vertically integrate into commoditized music genres (sleep, focus, white noise), perhaps increasing margins as the payout to the real world artist gets replaced with a lower payout to a licensed AI model.