Even with the transformation of AI, we must rewrite US Copyright Law. It is a mess. Photo Credit by Michael Whalen
AI music is not going away. It is separating into a transient synthetic economy and an artist-centered business built around identity, ownership, relationships and fans.
Something fundamental is happening in music, and I think we have been asking the wrong question about generative AI. We keep debating whether AI music will replace human musicians, whether Suno will survive its lawsuits, whether AI songs deserve royalties and whether listeners can tell the difference between something created by a person and something created largely by a machine. Those questions matter, but they may not describe where the business is actually going.
The more interesting possibility is that we are not heading toward one music business transformed by AI. We are heading toward two different music industries occupying some of the same platforms but operating under different economics, expectations and definitions of value.
One will be built around synthetic music that is generated quickly, consumed quickly and often replaced just as quickly. The other will remain artist-centered, where identity, ownership, history, audience and human connection are still the foundation of the business.
Nicholas Gunn, founder of Blue Dot Music and a British-born multi-instrumentalist, producer and songwriter, described the shift to me this way:
âPersonally, Iâm quite clear that we are entering into a time of two types of business in music. One that is generative AI driven that is highly transient represented on the DSPs, under AI labeling, and on generative AI sites such as Udio. The other being an artist centric model being represented on the DSPs without labeling and via DTC.â
I think Gunn is right, and the implications are considerably bigger than whether a particular Suno song stays on Spotify. We may be watching the music business divide into two parallel economies, each capable of becoming substantial but built around almost opposite ideas of what music is worth.
The First Industry Is Music as Content
Generative AI has made something historically difficult extraordinarily easy: producing something that sounds like a finished recording. A person can now generate songs, alternate versions, vocals, instrumentals and entire albums at a scale that would have been impossible only a few years ago. The technical accomplishment is extraordinary, but the economic consequence is even more important because the marginal cost of creating another piece of recorded music is collapsing.
That does not mean the music is necessarily bad. It means the economics surrounding its creation have fundamentally changed. When another convincing recording can be created in seconds, enormous quantities of music inevitably appear, and that is exactly what is happening across the streaming ecosystem.
The problem is that AI solved a scarcity problem the music industry did not actually have. There was never a shortage of music available to listeners. Before Suno, Udio or any of the current generation of models existed, there was already more excellent recorded music in the world than any person could hear in a lifetime.
What remained scarce was attention. Interest was scarce. Cultural relevance was scarce. Trust was scarce. The ability to create a relationship between an artist and an audience was scarce, and none of those things became abundant merely because another million audio files could suddenly be generated.
AI dramatically increased inventory. It did not automatically increase demand.
The Synthetic Music Economy
That new inventory can still become a substantial business, and I think it will. The mistake is assuming that it will look like the artist business we already understand.
There will be music generated for playlists, moods, games, fitness applications, meditation, retail environments, social media, personalized entertainment and applications we have not invented yet. A listener may eventually request ninety minutes of instrumental music generated specifically for a rainy morning in Boston, listen to it once and never encounter those recordings again.
That is still music, but it is not necessarily an artist career. The value comes from satisfying a particular moment rather than building a catalog that somebody wants to revisit twenty years later.
This is what Gunn means when he describes the generative side as highly transient. A generative platform may create the music because somebody needs it now, and the track may fulfill its purpose before disappearing into an effectively infinite pool of alternatives. There is no requirement that anyone remember the title, identify with the artist, buy a ticket or care what the creator produces next.
There is nothing inherently wrong with that model. We already consume enormous quantities of media in similarly functional ways, but we should stop pretending that functional synthetic music and artist development are simply two versions of the same business.
The Second Industry Is Music as Identity
The other business looks almost like the opposite. It revolves around artists whose identity matters to the listener, where the audience is not simply requesting a certain kind of sound but is interested in what a particular person does next.
That distinction is enormous because an artist accumulates meaning over time. An artist has a history, makes choices, changes direction, succeeds, fails, surprises people, develops a catalog and creates relationships that exist independently of any individual recording.
A generative system can make another song. It cannot automatically manufacture that history.
This is why direct-to-consumer becomes increasingly important in the artist-centered side of the business. Streaming will still matter because it remains the dominant way most people encounter recorded music, but the strongest relationship will increasingly be the one that exists directly between artist and listener.
The website matters. The mailing list matters. Concerts matter. Physical products matter. Memberships matter. Limited editions matter. Community matters, and direct communication matters because all of those things create an economic relationship that cannot be reduced to an anonymous stream.
The irony of the AI revolution may be that it pushes serious musicians back toward one of the oldest ideas in the business: knowing who actually cares about your work.
The DSPs Will Contain Both Industries
Spotify, Apple Music, Amazon, YouTube and the other DSPs are unlikely to become exclusively human or exclusively synthetic. They will probably contain both worlds, but those worlds may increasingly be identified and treated differently.
We are already seeing the beginning of AI disclosure systems. Platforms are developing ways to detect synthetic recordings, identify AI involvement, suppress spam, remove fraudulent activity and distinguish between AI-assisted production and fully generated material.
That creates the possibility Gunn described, where part of the DSP ecosystem becomes clearly identified as generative. Those recordings may still receive streams and may still have listeners, but they become a recognizable category rather than pretending to be indistinguishable from artist-driven music.
That is psychologically important. Once the difference is disclosed, the listener is no longer merely choosing between two pieces of audio. The listener is potentially choosing between two kinds of cultural product with different expectations attached to each.
For years, the assumption was that AIâs ultimate victory would come when nobody could tell the difference. I increasingly think the opposite is more likely:Â the difference will become part of the metadata.
Labeling Changes the Product
An AI label sounds like a small technical adjustment, but it could fundamentally change consumer behavior and the economics around the music. Imagine a streaming platform identifying a recording as wholly AI-generated, partially AI-generated or artist-created with limited AI assistance. Suddenly the presence or absence of that designation begins to communicate something about the nature of the product.
That does not mean listeners will always choose the human artist. Plenty of people will happily listen to synthetic music when it serves their purpose, particularly in background, functional or personalized settings.
The distinction matters because the expectations are different. Nobody needs to know the biography of the system creating meditation music for a hotel lobby, and nobody expects to buy a concert ticket to see the model perform its greatest hits.
Artist music asks for something else. It asks the listener to care not only about the output but about the source, the history behind it and the person who made it.
Which Brings Us Back to Suno
The Suno litigation matters because it sits directly at the point where Music AI 1.0 is colliding with Music AI 2.0. The first phase of the business essentially argued that enormous quantities of existing music could be used to train commercial models without traditional licensing because the training process was legally protected or transformative.
The record companies disagree, and the litigation remains unresolved in important respects. What matters for the future of the market, however, is that the commercial behavior surrounding the lawsuit is already changing.
Warner reached an agreement with Suno built around licensed next-generation models. BMG has now entered a broad arrangement involving both recordings and publishing. Udio has pursued similar relationships with major rightsholders, and other generative companies are building or announcing models based on authorized catalogs.
Most significantly, Suno itself has said that its earlier models will eventually be retired when its next generation of industry-partnered technology arrives. That does not resolve the lawsuits, but it tells us something important about where the company believes the sustainable business is going.
The future of commercial generative music may therefore be decided partly before the courts finish deciding the first generation. Licensing is becoming a market requirement even while the legal argument over whether it was always required continues.
So What Happens to Everything Already Made?
This may become one of the strangest catalog problems the music industry has ever encountered. Millions of recordings have already been created using early generative models whose training data remains legally contested, and some of those recordings have already been distributed commercially.
Those recordings do not automatically become illegal because the model that produced them was trained on disputed material or later retired. The legality of training and the legality of a specific output are not the same question, and collapsing those two issues would be a mistake.
That is why I do not expect some massive overnight deletion of everything created with early Suno or Udio models. The more likely outcome is much stranger and potentially more important for musicians.
The old AI catalog becomes stranded.
Stranded Music
A stranded recording still exists. The problem is that its ability to move through the professional music ecosystem becomes increasingly restricted as the rules around it change.
Imagine an artist who generated an album using an early Suno model. The artist paid for the appropriate subscription, obtained whatever commercial-use rights Suno offered at the time, downloaded the recordings and distributed the album through an aggregator.
The album stays online and nothing dramatic happens. Then, over the next several years, the surrounding business changes.
The original model is retired. New Suno models are licensed. Distributors begin asking which AI systems were involved in production. DSPs require standardized AI metadata. Music supervisors begin requesting provenance information before approving tracks for television or advertising, while catalog buyers begin asking how AI was used before acquiring masters.
Nothing necessarily happened to the recording itself. The environment around the recording changed, and that can be enough to alter its commercial value.
An album may continue streaming through its existing distributor but become difficult to transfer somewhere else. It may remain usable in some contexts but become unattractive in others, particularly where sophisticated buyers require a clean chain of rights.
The track is not necessarily banned. It is simply less portable than it once was, and that is what makes it stranded.
The Value of Provenance
The music industry has always been obsessed with provenance, even if we rarely use that word. Every serious transaction eventually asks who wrote the song, who owns the master, who controls the publishing, whether samples were cleared, whether featured performers signed releases and whether the seller can actually prove ownership.
Generative AI adds another question to that chain:Â What made this recording?
That question will matter because an AI system is not merely another microphone, compressor or synthesizer. Depending on the model and how it was used, the technology may introduce questions involving training rights, performer likeness, copyrighted material, authorship and ownership.
A professional buyer does not have to dislike AI to care about those questions. The buyer simply has to dislike uncertainty.
That is where licensed models acquire their greatest commercial advantage. Their value is not merely that they appear legally cleaner; their value is that music created through them is easier to move through the professional ecosystem.
The Music That Can Move
A recording created with a properly licensed and well-documented system has a clearer path through the business. It can move from one distributor to another, into film and television, into advertising, into games and potentially into a catalog acquisition without carrying the same degree of unresolved provenance risk.
That does not guarantee that every licensed-model recording will be valuable, protectable or creatively worthwhile. It simply means the rights conversation begins from a cleaner position.
This is why I believe licensed models will gradually push unlicensed commercial models toward the margins. The unrestricted model may be capable of producing equally impressive audio, and in some cases it may produce better audio, but professional value is not determined exclusively by what sounds best in a blind listening test.
A clean chain of rights matters. Documentation matters. Licenseability matters. Indemnification matters, and the ability to explain where the asset came from matters whenever meaningful money changes hands.
The music that can travel has more economic value than the music that cannot.
The Distributors Become the Border
The distributors and aggregators are going to sit directly in the middle of this transition. For years, digital distribution became almost frictionless: upload the audio, provide the metadata, attach the artwork and send the release into the streaming ecosystem.
Generative AI changes the risk calculation because the distributor is now potentially handling not just music but unresolved questions about how that music was created. Different companies are already responding differently.
TuneCore has moved toward requiring generative systems based on licensed datasets. CD Baby has taken a much more restrictive approach to AI-generated content. DistroKid currently permits AI music provided rights are respected, impersonation is avoided and users do not abuse the platform through mass-generated spam.
Those policies are not consistent yet, but the inconsistency is less important than the underlying change. The distributor is no longer treating every finished WAV file as though it presents the same rights profile.
The distributor is becoming a checkpoint.
Over time, I expect the important question to shift away from the vague question of whether someone âused AI.â The more important question will become which AI system they used, what that system was permitted to use and what role it played in creating the recording.
The Approved-Model Era
The music business cannot realistically investigate the full training history of every model involved in every release. That would be impossible at scale, which means the industry will eventually need some kind of model-level trust system.
Certain providers will be known to have licensing arrangements, provenance controls and contractual frameworks that distributors understand. Music generated or assisted by those systems becomes relatively easy to accept.
Other models will remain opaque. Those models become harder to accept because every submission carries more uncertainty and requires more diligence than a distributor or DSP wants to perform.
This does not require governments to outlaw unlicensed models. Commercial institutions can marginalize them simply by declining to accept the risk.
That is how an unlicensed system can continue to exist technologically while becoming increasingly irrelevant professionally.
The Two Industries Will Value Different Things
The synthetic music economy and the artist economy will not merely produce music differently. They will value different things.
The generative side will prize efficiency, customization, speed, enormous scale, infinite variation and extremely low production costs. Its strength will be that it can create the right piece of audio for a particular situation almost instantly.
The artist business will increasingly value scarcity, identity, reputation, context, ownership and relationships. Its strength will be that a listener wants something specifically from that artist and not simply another acceptable version of a genre.
One business is optimized around availability and utility. The other is optimized around meaning and attachment, and those differences will produce completely different economics.
The generative business can always create another song. The artist business depends on whether somebody actually wants the next song from you.
Transient Music Versus Catalog
For most of recorded-music history, the recording has been treated as an asset. Albums remain available for decades, catalogs develop value, old recordings are rediscovered and master rights can be sold long after the original release because music can continue generating revenue over time.
The generative economy may create a different kind of musical object. It may create recordings that are consumed rather than collected.
A person asks for something, the system generates it and the listener uses it. There may be no reason to remember the title, identify the supposed artist, follow the project or hear the same recording again.
In that environment, the recording behaves more like a service than an asset. Its usefulness may be immediate while its long-term catalog value approaches zero because another equally useful version can be generated instantly.
That is an extraordinary shift because the traditional recording business was built around the idea that successful music could accumulate value. Much synthetic music may behave in exactly the opposite way and depreciate almost immediately.
The Artist Business Moves the Other Way
The artist-centered side of the industry may become more valuable precisely because synthetic music becomes unlimited. When recordings become abundant, the things that cannot be generated instantly become more important.
A real artist has a history with an audience. The artist possesses a point of view that develops over time, a body of work, a reputation, a live presence, relationships and a community of people who want to know what happens next.
This is where direct-to-consumer becomes crucial. An artist who knows five thousand people who genuinely care about the work may have a healthier and more durable business than an anonymous synthetic project capable of generating five million tracks.
The numbers can make that comparison look absurd until you examine the economics. Attention is more valuable than inventory, and a relationship is more durable than a stream.
The Great Irony of AI Music
AI was supposed to make artists less important because it could separate the finished recording from the traditional process required to make one. Instead, it may make being an artist more important because the recording itself is becoming less scarce.
When anyone can generate something that sounds finished, technical completion is no longer sufficient evidence of value. When millions of songs can appear every day, abundance ceases to be impressive.
Something else has to carry the value, and that value begins moving toward identity, judgment, reputation, taste, performance, trust and the ability to build an audience that wants to remain connected over time.
The machine can generate another record. It cannot automatically generate the relationship surrounding that record.
This Is Where Vanity AI Runs Into a Wall
The first generation of AI music encouraged an understandable fantasy. If making records was the obstacle preventing someone from having a music career, then suddenly the obstacle appeared to be gone.
The problem is that making recordings was never the principal obstacle. Building an audience was.
AI can solve production at extraordinary scale. It cannot force another human being to become interested in the result, and it cannot manufacture sustained attention simply because the output sounds competent.
That is why the enormous volume of synthetic uploads does not automatically translate into an equally enormous new artist economy. A million additional tracks are not necessarily a million additional competitors because competition requires listeners.
Most generated music appears to have almost none.
What we are beginning to understand is the difference between creation and demand. Technology made creation almost limitless while human attention remained exactly as finite as it was before.
Music AI 2.0
The next generation of commercial AI music will increasingly be licensed because professional music requires assets that can move safely through a rights system. The first generation could treat permission as an argument to be settled later because it was still proving that the technology worked.
The second generation has a harder test. It has to prove that the business works.
That means licensing, provenance, disclosure and better metadata. It means distinguishing AI assistance from fully generated recordings, while distributors decide which systems they are willing to accept and DSPs decide how synthetic material is labeled, recommended and monetized.
It also means acknowledging that the enormous amount of music already created during the first era will not necessarily retain the same commercial status forever. Some of it will remain viable. Some will become difficult to transfer, while some will lose professional opportunities because the origin of the recording is too uncertain for the next transaction.
Much of it will probably experience the most brutal outcome the music business has ever offered: it will remain technically available while almost nobody listens.
The more important development, however, is what emerges beside it. Two music industries begin to take shape.
One is a massive synthetic-content economy that is licensed, scalable, customizable, increasingly labeled and often transient. It may eventually generate more music than human beings ever could, and much of that music will be consumed without listeners caring who supposedly made it.
The other is an artist economy built around identity, relationships, audience, ownership and direct connection. Those artists will still use streaming services and many will use AI somewhere in the creative process, but their strongest businesses will extend beyond the DSP into relationships with people who care specifically about them.
Both industries can become large. Both can use AI, and both can produce excellent music, but they are not the same business because they are not creating the same kind of value.
The central mistake of Music AI 1.0 was assuming that once machines could make convincing recordings, they had recreated the music industry. What they actually recreated was one piece of it: production.