AI Made Music Easier to Create. Did It Make It Any Easier to Earn a Living?

Generative AI has demolished some of the oldest barriers to making a record. It has also created an almost unlimited supply of new music. For independent musicians, that may turn attention — not production — into the industry’s scarcest resource.

By Jason Fite

North Port, Florida · September 29, 2026

There has probably never been an easier time in history to make a song.

That sentence would have sounded ridiculous for most of the recording industry’s existence. Making a record once meant accumulating expensive things: instruments, microphones, tape machines, studio time, engineers, producers, session players — and enough money to keep all of them in the same room until something worth releasing came out.

The laptop dismantled much of that infrastructure. Digital distribution dismantled another piece of it. A musician could record in a bedroom and place the file on the same services carrying Taylor Swift, Bad Bunny, and the Beatles.

Generative AI is attacking what remains.

A user can open a service such as Suno, describe a piece of music, provide lyrics or other direction, and generate a polished recording in minutes. Suno says more than 100 million people had used the platform by early 2026. It reports more than 2 million paid subscribers. Its newest model suite, v6, launched on September 9 and was developed with industry partners including Warner Music Group, BMG, and Believe.1

For an independent musician, the implications are enormous. AI can be a sketchbook, an arranger, a demo studio, a production assistant, or — in its most extreme application — almost the entire production process.

I know that from the chair, not the press release. I studied audio production at Middle Tennessee State University. I write the lyrics. I write the melodies on bass and guitar. Sometimes I compose the demo in GarageBand. That demo — and a lyric sheet with cues — goes into Suno, where I write the prompts and builds for the arrangement I want around the music I already played. The vocal is generated from a model trained on my own voice, not a stranger’s and not a celebrity clone. When a build is close, I pull it into Logic Pro, mix it, and master it myself before it goes to the distributor. DistroKid will have the file on Spotify, Apple Music, Tidal, and Amazon within days if the rights are clean and the AI box is checked. Then I start another one. What used to be a four-figure week can be an evening plus the years it took to learn the instruments and the room.

There is a problem hidden inside that capacity.

Everybody else gets the same machine.

The catalog that nobody hears

In 2025, an average of roughly 106,000 new ISRCs — the identifying codes attached to individual recordings — were delivered to digital music services every day, according to Luminate. That was already up 7 percent from the previous year. The more revealing number is what happened next: 88 percent of tracks in Luminate’s global streaming data accumulated 1,000 streams or fewer during 2025. Eighty-six percent of the year’s net increase in tracks was concentrated among recordings with 100 streams or fewer. Most of the growth in the world’s recorded-music catalog occurred at the nearly unheard end of the market.2

Then AI hit the accelerator. In January 2025, Deezer said it was receiving roughly 10,000 fully AI-generated tracks per day. By September that figure had reached 30,000. In April 2026 it was about 75,000, or 44 percent of daily deliveries. By June, Deezer said, approximately 90,000 fully AI-generated tracks were arriving every day. At the peak they represented more than half of all new music delivered to the service. Apple Music’s Oliver Schusser said in April that more than a third of monthly deliveries to that store were 100 percent AI.3

It is difficult to imagine a cleaner demonstration of what happens when technology removes a production bottleneck. The music business has entered the age of abundance. The musician may be entering the age of attention scarcity.

That is the question this piece is built to answer, and it is a business question, not a metaphysical one. Is generative AI democratizing the music business — or only the ability to make music?

The $11 billion paradox

There is an obvious counterargument to the idea that streaming has become economically hopeless: the streaming business is generating more money than ever.

Global recorded-music revenue reached $31.7 billion in 2025, according to IFPI, an eleventh consecutive year of growth. Streaming generated more than $22 billion and represented 69.6 percent of recorded-music revenue worldwide. Paid subscription streaming grew 8.8 percent, with 837 million users of subscription accounts globally.4

Spotify says it paid more than $11 billion to the music industry in 2025, up more than 10 percent from the previous year. Roughly half of those royalties, the company says, were generated by independent artists and labels. More than 81,000 artists generated at least $10,000 from the service in 2025. More than 13,800 generated at least $100,000. More than 1,500 crossed $1 million. More than a third of the artists generating at least $10,000 were DIY musicians, or had begun their careers that way.5

Those are royalties generated on Spotify, not money deposited into an artist’s account. Payments pass through labels, distributors, publishers, and collecting societies. The trend is still real. So is the contradiction.

The pool of money is getting larger. The number of people trying to reach it is getting larger faster. AI supercharges the second number.

Independent-statement studies in 2026 put Spotify’s master-side median somewhere around $0.0033 to $0.0036 per stream globally, with some U.S. books nearer $0.004. At those illustrative rates, a million streams is about $3,300 to $4,000 in master royalties before a distributor’s cut and before any split. Two hundred fifty thousand streams is about a thousand dollars. Since April 2024 a track needs at least 1,000 streams in the previous twelve months, plus an undisclosed minimum of unique listeners, before it is eligible for Spotify’s recorded-music royalty pool at all.6

Ditto Music’s Independent Music Report 2026, published September 29 from a survey of 5,039 artists using Ditto in 112 countries, puts the income side in one line: 74 percent of independent artists earned less than $1,000 from music last year. Only 9 percent earned $5,000 or more. Twenty-two percent call music their full-time job. Sixty-four percent hold another job. Fourteen percent lean on family or savings. Twenty-nine percent believe their music income is sustainable for the next three years. Eighty-two percent cannot afford to tour — and live is still the second-largest income line after streaming.7

The pool grew. The typical independent check did not become a living. Creating more music and creating more income are different jobs that happen to share a file format.

What it costs, what ships, what pays

Follow the work at three points on the same continuum and the contradiction stops being abstract.

A traditional independent session — room, engineer, players, mix, master — still costs hundreds to several thousand dollars if people are paid. The output is one or two songs. The songs enter the same store as everything else. They have a story a publicist can tell. They do not have a structural advantage over a file made on Tuesday night.

A hybrid workflow, the one I use, collapses the middle of that invoice without handing the song to the machine. The writing is mine: lyrics, bass and guitar melodies, often a GarageBand demo. Suno is asked to build the production around that demo from a prompt and a cued lyric sheet, and to sing it in a model of my voice. The last mile is conventional again — Logic Pro mix, self-master, distributor. DistroKid’s 2026 rules allow the result if the uploader owns the rights, does not impersonate a living artist, discloses AI at ingest, and does not mass-upload spam. Since spring that disclosure travels as credits onto Spotify, Apple Music, and YouTube. Suno’s own terms assign commercial rights only for audio generated on a paid plan after a permitted download. Free-tier output is personal use. Upgrading later does not launder a free file. In September Suno capped exports — seven lifetime downloads on free, twenty a month on Pro, sixty on Premier — a spam brake that also admits the old pipe was being used as a factory.

A fully generative workflow can ship a catalog the way a content farm ships listicles. Academic work on AI uploaders finds a small slice of accounts producing a disproportionate share of AI tracks, at lower quality, with hit rates a fraction of other AI projects. Platforms have learned the signature: duplicate metadata, bulk instrumentals, SEO titles, purchased streams. That is not a music career. It is an attempt to mint claims on a royalty pool.

Release volume is the tell. Human working musicians in recent samples average on the order of one new track a month. AI-tagged accounts average several. Ten songs nobody hears are not ten times the opportunity of one song nobody hears. A thousand are not a career. They are inventory.

The million-song problem

Streaming is not a conventional store shelf. Spotify does not run out of space because somebody uploads another 40 million songs. Human beings run out of time. There are 24 hours in a day. A listener cannot consume the 90,000 AI songs Deezer was receiving every day in June, much less the human catalog arriving beside them and the century of recorded music already on the shelf.

Generative AI can increase supply almost without limit. It cannot manufacture another hour of human attention.

That distinction explains one of the stranger facts in the boom. AI music is uploaded at staggering volume. People are not listening to it in proportion. Fully AI-generated tracks reached more than half of Deezer’s new deliveries at the June peak and account for only 1 to 3 percent of streams. Deezer excludes detected AI music from editorial playlists and algorithmic recommendations. Apple puts AI listening “much lower” than 0.5 percent. Spotify, which does not publish an AI upload share, told ABC News in July that “well under 1 percent” of consumption goes to fully AI-generated tracks, and that it removed more than 75 million “spammy” tracks — AI and otherwise — in the year to September 2025.8

Deezer says as much as 85 percent of streams associated with fully AI-generated tracks in 2025 were detected as fraudulent and demonetized. The industry’s AI problem is therefore partly a music problem and partly a spam problem. A recording is a potential claim on a royalty pool. If generating a recording becomes instantaneous, bad actors can create catalogs at machine scale and pair them with artificial listening. Spotify’s September 2025 spam filter was written for that pattern: mass uploads, duplicates, search-engine manipulation, artificially short tracks. The company said out loud that AI makes those catalogs easier to generate at scale.9

For a legitimate independent, the bind is specific. AI can raise productivity. Productivity is useful only if somebody wants the product. The algorithm rations attention. It does not mint it.

When nobody can tell

There is another possibility: AI music may not remain identifiable as a category for long. In October 2025, Ipsos ran a blind listening test for Deezer with 9,000 people in eight countries. Participants heard three tracks — two generated entirely by AI, one made by humans — and were asked to identify them. Ninety-seven percent failed to classify all three correctly. Failing one of the three was enough to count as failing the test; the figure does not mean 97 percent of individual guesses were wrong. It does mean synthetic music has crossed a perceptual threshold.10

The same survey found the split that will probably define the next phase. Fifty-one percent believed AI would contribute to more low-quality or generic music. Sixty-four percent thought AI could lead to a loss of creativity. Forty-six percent thought AI could help them discover more music they liked. Listeners may not reject synthetic music because it sounds synthetic. They may still care how it was made.

That is why the distinction between AI-generated and AI-assisted music is no longer academic. In July 2026 a group of organizations including IFPI, the RIAA, A2IM, WIN, IMPALA, the Recording Academy, SAG-AFTRA, and the Human Artistry Campaign announced a voluntary labeling framework separating the two. YouTube now lets music partners designate deliveries as “Fully Gen AI,” “Partly Gen AI,” or “No Gen AI.”11

A producer might use AI to sketch an arrangement, replace an element later, manipulate a vocal, extend a passage, brainstorm a lyric, or build a demo that is then rebuilt with instruments. At what point did AI make the song? There is no universally accepted answer. The law does not provide a clean one either.

Who owns the song

In January 2025 the U.S. Copyright Office concluded that material produced by generative AI can receive copyright protection when a human author has contributed sufficient expressive elements. Human-authored material that remains perceptible in an AI-assisted work can be protected, as can sufficiently creative human arrangements or modifications. Merely providing prompts is not enough by itself. Using AI as an assistive tool does not automatically disqualify the work.12

The workflow matters. The legal position of someone who accepts an unchanged generation is different from that of a writer who authors the lyric, composes the melody on guitar and bass, records a demo, directs the generation around that demo, and then mixes and masters the result. Under the Copyright Office’s 2025 standard, prompts alone are not enough; perceptible human authorship can be. That is why the order of operations in a hybrid session is not a trivia detail. It is the difference between a file and a claim. Copyright in the output is still only half the fight. The other half is what went into the models.

Major record companies sued Suno and Udio in 2024, alleging the companies copied copyrighted recordings to train their models. Both companies argued fair use. The industry has since split between licensing and litigation. Warner Music Group reached agreements with both companies in 2025; the Suno deal settled the prior case and set plans for licensed models and opt-in uses of participating artists’ names, likenesses, voices, and compositions. Universal settled with Udio. BMG and Believe signed with Suno in 2026; Believe had blocked unlicensed-generator deliveries only months earlier. Universal- and Sony-affiliated labels filed a second suit against Suno on September 18, 2026, in Boston federal court, alleging copying of 60,202 recordings and arguing that v6 — trained in part on user “creations” from earlier models — is “model laundering.” The allegations are unproven. Suno disputes them.13

The lawsuits are about copyright. The business question underneath them is about value. If recorded music is valuable enough to train a system that produces new music, who participates in the value that system creates? Independent musicians live in the gap between those two sentences.

One industry, four answers

The philosophical debate is already a set of incompatible store rules.

DistroKid allows music made with AI tools if the artist controls the rights, does not impersonate, does not infringe, and does not mass-generate to manipulate platforms. LANDR caps AI songs at twelve a month. TuneCore has blocked some named generators at the door. CD Baby, under Downtown’s content rules, is among the strictest. Bandcamp announced in January 2026 that music generated wholly or substantially by AI is not permitted, framing the decision around human creativity and the direct artist–fan relationship. Tidal has declined to attribute royalties to wholly AI-generated recordings. YouTube is building disclosure infrastructure. Spotify is building spam and impersonation infrastructure. Deezer detects, tags, excludes from recommendations, demonetizes fraudulent listening, and now removes certain synthetic tracks that go unplayed for six months.14

There is no single AI music economy. There are several being invented at once. Each is a bet about what listeners ultimately value.

Sync offices are running a parallel experiment. A brief that once went to three composers now goes to three composers and a generator. The generator is cheaper and faster. The composer still wins when the client needs a warranty — a cue that will not blow up in a brand-safety review over training data, voice likeness, or a missing split sheet. Speed was democratized. Clearance was not. The cue that pays is the one that clears.

The part AI cannot generate

The easiest conclusion is that AI will flood the stores until human musicians cannot compete. The listening data does not yet support replacement. Uploads exploded. On Deezer they remain a small percentage of plays. Global recorded-music revenue is growing. Spotify reports more artists crossing $10,000, $100,000, and $1 million thresholds.

There is evidence for abundance. There is not yet evidence that abundance automatically means substitution. One industry-backed study commissioned by CISAC projected that generative AI could put 24 percent of music creators’ revenues at risk by 2028, particularly through substitution and competition. That figure is a forecast under a set of assumptions — not an observed 24 percent decline.15

The future may be messier, and more like the thing musicians are already living. AI can make creation more accessible and careers more difficult at the same time. Those ideas are not contradictory. The ability to make a record was once scarce. Distribution was once scarce. Both became abundant. What remains scarce is the thing musicians have always needed: somebody who cares.

That is why direct relationships may become more valuable in an AI-saturated market, not less. If the store contains effectively unlimited music, identity, story, community, a stage, a shirt, a mailing list, a sync relationship, and a reason to come back are the assets that cannot be duplicated by adding tracks. Bandcamp’s ban is a bet on that kind of scarcity. Whether the bet holds is empirical. It is also the closest thing the independent sector has to a moat that does not depend on an algorithm’s mood.

Music is early, not unique. Any field where the output is a file and the bottleneck used to be production is running the same trial. Illustration, stock photography, voiceover, code, beat licenses, video cues: cost per unit falls, unit count explodes, the buyer’s time does not. Productivity per person goes up. Productivity as an advantage goes down, because the other people got the same update. When supply becomes elastic and demand does not, per-unit revenue takes the adjustment. Platforms can protect a royalty pool by filtering fraud and raising thresholds. They cannot protect a given artist from a shelf that got longer overnight.

I can close the production ledger at midnight. The other ledger still opens in the morning with the question it opened with when the barrier was a studio lockout and a reel of tape: will anyone spend the next three minutes on this, and if they do, does the three minutes turn into a life.

AI democratized access to something musicians spent generations trying to obtain: the means of production. A person with a laptop can now experiment at a scale that would have been financially absurd not long ago. Democratizing production is not the same as democratizing success. If everyone can make more music, making more music stops being an advantage. The advantage becomes making something people deliberately choose from an ocean of alternatives — and giving them a reason to come back when the next hundred thousand songs arrive tomorrow.

AI may have solved one of the oldest problems in music. It made creation abundant. Now musicians have to survive what comes next.

Source notes

1. Suno user and subscriber figures: company statements via Music Business Worldwide and Axios, 2026 (100 million+ users; 2 million+ paid subscribers; ~$300 million ARR as of February 2026). v6 launch September 9, 2026, developed with WMG, BMG, and Believe (Axios, MBW). Suno v6 launch reporting

2. Luminate 2025 year-end reporting: ~106,000 new ISRCs/day, up 7%; 88% of tracks under 1,000 streams; 86% of net catalog growth among tracks with 100 streams or fewer.

3. Deezer AI-delivery series, January 2025–June 2026, as reported by Music Business Worldwide, TechCrunch, and Music Ally. Apple Music: Oliver Schusser, April 2026.

4. IFPI Global Music Report 2026 (released March 18, 2026): $31.7 billion recorded-music revenue in 2025, +6.4%; streaming >$22 billion / 69.6%; paid-subscription streaming +8.8%; 837 million paid-account users. IFPI Global Music Report 2026

5. Spotify public reporting on 2025 payouts and artist-earning thresholds (Loud & Clear / 2026 company statements): >$11 billion paid out; independent share described as roughly half; 81,000+ artists at $10,000; 13,800+ at $100,000; 1,500+ at $1 million. Spotify Loud & Clear

6. OG Records independent-statement study, April–July 2026 usage months: Spotify median $0.0036/stream master-side before distribution fee. Spotify 1,000-stream threshold in force since April 1, 2024.

7. Ditto Music, Independent Music Report 2026, September 29, 2026: n = 5,039 artists, 112 countries. Ditto Independent Music Report 2026

8. Deezer stream share 1–3%; Apple AI listening “much lower” than 0.5%; Spotify Sam Duboff to ABC News, July 16, 2026; spam-removal figure via MBW, September 25, 2025.

9. Deezer: up to 85% of 2025 streams on fully AI tracks classified fraudulent and excluded. Spotify September 2025 spam-filter announcement.

10. Ipsos / Deezer blind test, October 2025: 9,000 listeners, eight countries. Deezer / Ipsos survey

11. Voluntary labeling framework announced July 2026 by IFPI, RIAA, A2IM, WIN, IMPALA, Recording Academy, SAG-AFTRA, Human Artistry Campaign. YouTube delivery designations for music partners.

12. U.S. Copyright Office, January 2025 report on copyright and artificial intelligence, human-authorship standard. U.S. Copyright Office AI reports

13. RIAA-coordinated suits against Suno and Udio, June 2024. WMG–Suno settlement November 2025; UMG–Udio October 2025; BMG–Suno August 2026; Believe–Suno September 8, 2026. UMG and Sony second complaint against Suno, D. Mass., September 18, 2026, 60,202 recordings; allegations disputed.

14. Distributor and store policies as of 2026: DistroKid help center; LANDR monthly cap; TuneCore generator restrictions; CD Baby / Downtown; Bandcamp announcement January 13, 2026; Tidal royalty rules for wholly AI recordings; Deezer six-month unplayed-takedown policy announced with the June 2026 upload figures.

15. CISAC-commissioned study on generative AI and creator revenues; 24% by 2028 is a modeled projection, not an observed decline. CISAC study

Method. First-person production detail is the author’s working practice as an independent songwriter and producer. Industry figures are from named public reports and platform disclosures. On-the-record interviews with distributors and licensing offices would be added in a commissioned revision.

Explore the music behind the essay

Meet Jason Fite and listen to The Fite and Saint June. If you would like to help sustain the work, shop official merch or support independent music.