Generative AI may not be heading toward one transformed music business. A growing view in the industry is that it is producing two: a synthetic-music economy built on speed and volume, and an artist-centered economy built on identity, ownership and direct audience relationships. Both can share the same streaming platforms while running on very different economics.
Nicholas Gunn, founder of Blue Dot Music and a British-born multi-instrumentalist, producer and songwriter, describes the split this way:
“Personally, I’m 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.”
Abundant recordings, scarce attention
Generative tools have made it easy to produce something that sounds like a finished recording. Songs, alternate versions, vocals, instrumentals and whole albums can now be generated at a scale that was impossible a few years ago, and the marginal cost of another recording is collapsing.
That solves a scarcity problem the industry never had. Listeners already had more recorded music than anyone could hear in a lifetime. What stayed scarce was attention, cultural relevance, trust and the relationship between an artist and an audience. AI increased inventory, but it did not automatically increase demand.
The synthetic side can still become a large business, built around music for playlists, moods, games, fitness, meditation, retail spaces and personalized listening. A listener might request ninety minutes of instrumental music for a rainy morning, use it once and never hear it again. That is functional music rather than an artist career, and its value lies in meeting a moment, not in building a catalog.
The artist economy and direct relationships
The artist-centered business revolves around people whose identity matters to the listener. An artist accumulates history, makes choices, changes direction and builds a catalog and relationships that outlast any single recording. A generative system can make another song but cannot manufacture that history.
Direct-to-consumer channels become more important on this side. Streaming remains the main way most people find recorded music, but websites, mailing lists, concerts, physical products, memberships, limited editions and community all create an economic relationship that cannot be reduced to an anonymous stream.
Labels in the metadata
Platforms are already building ways to detect synthetic recordings, identify AI involvement, suppress spam, remove fraudulent activity and separate AI-assisted production from fully generated material. If streaming services label recordings as wholly AI-generated, partly AI-generated or artist-created with limited AI help, the label itself starts to signal what kind of product a listener is getting.
That will not stop people choosing synthetic music when it serves a purpose, particularly for background or functional listening. It does mean the two categories carry different expectations.
Licensing and the stranded recording
The Suno litigation sits where the first generation of AI music meets the second. Early commercial models were built on the argument that training on existing music did not require traditional licensing. The record companies dispute that, and the cases remain unresolved in important respects, but commercial behavior is shifting regardless.
- Warner has reached an agreement with Suno built around licensed next-generation models.
- BMG has entered a broad arrangement covering recordings and publishing.
- Udio has pursued similar relationships with major rightsholders.
- Suno has said its earlier models will eventually be retired when its industry-partnered technology arrives.
That raises the prospect of “stranded” recordings: tracks made with early models, distributed through an aggregator, that stay online but become harder to move as the rules change. Distributors may ask which AI systems were involved, DSPs may require standard AI metadata, and music supervisors and catalog buyers may want provenance before approving a track. Nothing happens to the audio itself, but the commercial environment around it shifts. The legality of training and the legality of a specific output are separate questions, so a mass deletion of early-model music is considered unlikely.
Provenance is an old concern in music, covering who wrote the song, who owns the master and whether samples were cleared. Generative AI adds a new one: what made this recording. A buyer does not need to dislike AI to care, only to dislike uncertainty. Music from licensed, well-documented systems moves more easily between distributors and into film, television, advertising, games and catalog sales.
Distributors take different lines
Distributors and aggregators sit in the middle of the transition, and their policies are not yet consistent. 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 as long as rights are respected, impersonation is avoided and the platform is not abused with mass-generated spam.
Over time the key question is likely to shift from whether someone used AI to which system they used, what it was allowed to train on and what role it played. Because no one can audit the training history of every model, the industry may need a model-level trust system, with some providers known for licensing arrangements and provenance controls and others treated as opaque. Unlicensed models would not need to be outlawed. Commercial institutions can marginalize them simply by declining the risk.
Why artists may matter more
Synthetic music prizes efficiency, customization and low cost. The artist business prizes scarcity, identity, reputation, context and relationships. Historically a recording has been an asset that holds value for decades. Much generative music may behave more like a service, consumed once with near-zero catalog value because an equivalent version can be generated instantly.
If so, an artist who knows five thousand people who care about the work may have a more durable business than an anonymous project that can generate millions of tracks. Production was never the main obstacle to a music career. Building an audience was, and AI does not solve that. Creation became nearly limitless while human attention stayed finite.
The second generation of commercial AI music has to prove the business works, which points to licensing, provenance, disclosure and better metadata. Much of the music made in the first era may stay technically available while almost nobody listens.
