$198 billion, 57%, and the number nobody models
Music in the Air stops at the society's door. Goldman Sachs forecasts the music industry to $198.1 billion by 2035 and models the share that reaches rights holders. What happens after the society is paid has a published size and, from April 2027, a clock.
A forecast of rights-holder revenue is not a forecast of what rights holders receive.
Goldman Sachs's Music in the Air 2026 takes the global music business from $112.5 billion in 2025 to $198.1 billion in 2035 (p.12), and models how much of that retail money reaches rights holders: about 57%, held flat for ten years (pp.15–16). Then it stops. The money has reached the label, the publisher or the collecting society, and the model treats that as the end of the pipe.
It is not the end of the pipe. After the society is paid, whether the owner is paid depends on three things the report never mentions: whether the work is registered, whether the usage is matched to it, and whether the share is claimed. In 93 pages the report does not name a single collecting society. The words unmatched, unclaimed and black box do not appear. Its one nod to collection is an AI product (pp.9, 57, 61).
This is not a complaint about the model. Its job is to forecast the industry's revenue, and it does that with care; every Goldman figure below is Goldman's, cited to its page. The complaint is about everyone who will quote the $198 billion as if it were a forecast of their own income. For a catalogue owner, the number that matters is produced by the step the report does not model, and that step has a published size.
Two steps modelled, one not
Goldman's model has two steps between a listener and a rights holder. There is a third, and it is where the owner gets paid or does not.
The first step is retail revenue: what listeners and advertisers pay. Goldman grows it at 5.8% a year to 2035, with recorded music adding $45.8 billion, live music $27.7 billion and music publishing $12.1 billion at 4.8% a year (p.12, Exhibit 13).
The second step is the share of that retail money which reaches rights holders. Goldman holds it at about 57% for the whole forecast, because "the industry market structure and negotiating leverage between rights holders and DSP remain largely the same" (pp.15–16). On that basis recorded-music rights-holder revenue goes from $31.7 billion to $56.1 billion (p.16).
The third step is the one the report does not draw. A streaming service pays a mechanical or performing-right society a pool of money for a month's usage. The society pays out the part it can match to a registered work with a claimed share, and holds the rest. The owner receives a statement of what was paid. The report notes that rights holders "derive value from i) mechanical rights (reproduction) and ii) performance rights" (p.60); it does not model how either reaches them.
Here is the uncomfortable part. Nobody in the chain is paid to check that the owner was paid. The service has discharged its licence when the pool is paid. The society has discharged its rules when it distributes what it can identify and holds the rest for the period its rules allow. The owner reads a statement that looks complete, because it is a complete list of what matched. A royalty statement is a record of what matched, not of what was earned. Nothing in Goldman's 57%, and nothing on anyone's statement, says how far apart those two numbers are.
How big is the step Music in the Air leaves out?
For one right, in one country, the answer is in an annual report filed every year. It is $630 million and it is not shrinking.
The MLC (the US collective that administers mechanical royalties from interactive streaming and downloads) is unusual among societies in publishing, by usage year and by reason, how much it is holding. Its 2025 Annual Report, with figures as of the March 2026 distribution, shows $630.5 million of unmatched and unclaimed blanket royalties held across 2021 to 2025 usage, plus $27.5 million in dispute. "Unmatched" means The MLC cannot tell which work a recording embodies. "Unclaimed" means it can, but less than 100% of the work's shares has been claimed by a member.
This is not a society failing. The MLC's average match rate for its first five years is 91.9%, it reprocesses the remainder every month, and it publishes more about its held money than most societies do. The point is what the system settles to after years of that effort. For 2021 usage, the oldest year, $81.9 million was still unpaid in March 2026: 14.3% of the $571.9 million collected for that year, four to five years after the streams happened. A year earlier it was $90.9 million; a year before that, $103.0 million. The residue falls by about a tenth a year, which means most of it will never be matched by effort alone.
| Usage year | Unmatched | Unclaimed | In dispute | Total held | Collected for the year |
|---|---|---|---|---|---|
| 2021 | $40.8m | $39.5m | $1.6m | $81.9m | $571.9m |
| 2022 | $48.7m | $42.3m | $3.5m | $94.5m | $697.7m |
| 2023 | $76.6m | $62.7m | $7.8m | $147.0m | $1,019.2m |
| 2024 | $87.2m | $59.3m | $7.3m | $153.8m | $967.7m |
| 2025 | $108.1m | $65.3m | $7.4m | $180.8m | $914.2m |
| Total | $361.4m | $269.1m | $27.5m | $658.0m | $4,170.5m |
Source: The MLC 2025 Annual Report, appendix section F, figures as of the March 2026 distribution. "Collected" is net of royalties identified as subject to direct voluntary licences after reprocessing. The 2025 column is young: its initial distributions ran to March 2026 and reprocessing had barely begun.
Two more lines from the same report. $52.7 million had been matched, found eligible and still not paid, because the amount was below the payment threshold, the payee's details were wrong or incomplete, or the member had not finished registering. That is money with a known owner who has not been paid for a paperwork reason. And of the $397.3 million of pre-2021 "historical unmatched" royalties the streaming services handed over in 2021, $240.0 million (60.4%) had been distributed by October 2026. The rest is still waiting for someone to register and claim it.
The MLC administers US mechanical royalties from interactive streaming and downloads. The US performing right, and both rights everywhere else, sit with societies that report their held balances differently or not at all. CISAC's 228 member societies collected €13.97 billion for creators in 2024 (Global Collections Report 2025). The MLC's $630 million is the published residue of one pipe, not a total for the system. Nobody publishes the total, which is the point.
The MLC's unclaimed royalties now have a clock
In April 2027 The MLC starts paying unclaimed royalties to other people, by rule, on a schedule, inside Goldman's forecast window.
The Music Modernization Act requires The MLC to hold unclaimed royalties for at least three years and then distribute them to copyright owners by relative market share. The MLC has held them for twice that. On 30 September 2026 it announced the start date: April 2027, beginning with the remaining royalties from January 2021 usage, then one usage month in each monthly distribution after that.
The rules are in its Policy for the Distribution of Unclaimed Accrued Royalties (September 2026). The April 2027 distribution uses the database snapshot taken on or around 10 February 2027. Registrations, matches and claims count only if they are processed before that snapshot. The unclaimed pool for each month, licensee and offering is allocated pro rata to the shares that were claimed in that snapshot. A completed distribution is final. A share claimed afterwards is paid for later periods, "excluding only royalties for periods that were already subject to a market share distribution".
Say what that is. It is a transfer from the unregistered to the registered, by rule, on a schedule. The money does not vanish into a black box; it is paid, in proportion to market share, to whoever had their paperwork in order on the day of the snapshot. The statute requires a publisher that receives it to pass at least 50% of the part attributable to each songwriter back to that songwriter. Nothing requires anyone to pass any of it to the owner who never registered.
The MLC has committed to the pace for 2021 usage; the closing dates for 2022 onward assume it continues at one usage month per month. Each distribution is preceded by at least 90 days' notice, and the notice, not this chart, fixes the date.
Put the two charts together. Goldman's window runs from 2025 to 2035. At the announced pace, every usage year from 2021 to 2028 is fully redistributed inside it. For a catalogue owner, the first deadline in the system that is set by rule rather than by effort is 10 February 2027, and the publishing growth in the forecast arrives, year by year, through a door that closes behind it.
For a buyer, this changes what an unclaimed balance is. Before the snapshot it is recoverable with statutory interest. After it, it is someone else's income, and the policy says The MLC corrects only its own errors. For a lender, it is the first date on which part of the collateral's history becomes unrecoverable by rule. For a publisher, it is the date by which "we will tidy the registrations later" stops being a plan.
Where the growth is, the register is thinnest
Goldman puts most of the next decade's new subscribers in emerging markets. The authors' societies there collect a rounding error.
Goldman expects emerging markets to supply up to 75% of net subscriber additions by 2035, growing from 438 million to 956 million paid subscribers at 8.1% a year (p.22). Each pays rights holders about $12 a year against $30 in developed markets (p.34). Latin America is the fastest-growing region, with recorded-music revenue up 17.1% in 2025 (p.23).
Recorded revenue reaches labels and distributors directly. The writer's and publisher's money takes the long way round: a local society collects it, matches it against its own repertoire, and sends it to the writer's home society under a reciprocal agreement, or to a sub-publisher. Every hop is a registration that has to exist and a match that has to be made. The third step, done abroad, is the third step done several times.
Africa, authors' collections 2024
Latin America, authors' collections 2024
Emerging-market rights-holder revenue per subscriber
Read those three together. The region adding the most listeners pays the least per listener, and its authors' societies either collect almost nothing or went backwards in a year the recorded market grew by a fifth. Goldman's model is right that the retail money will arrive. Whether the writer's share of it gets home depends on registrations in societies most Western publishers have never filed with. For a publisher counting on emerging-market growth, the forecast is the easy part.
AI pays through the same register
Every licensing template in the report pays by identified work and share. An opt-in is a registration. A per-generation royalty is a matching problem with a new input.
Goldman's base case puts generative AI at a $3.3 billion uplift to subscription streaming rights-holder revenue by 2035, about 10%, with a 50% rights-holder share of retail against roughly two thirds in streaming today (pp.65–66). The report lists how that money would be paid: subscription and licensing fees under an opt-in structure, revenue shares, and "per-generation royalties" based on usage of licensed catalogues in user outputs (p.62). Its table of licence fee types includes an "Output Royalty", a "per-generation royalty triggered when AI model produces content influenced by licensed catalog" (p.54, Exhibit 77). It records a society building "infrastructure that identifies when songwriters' and composers' work is used by AI" (p.53).
Every one of those pays a specific work and a specific share. The opt-in list is a register. The attribution system is a matching engine. The output royalty is a usage line that has to find its owner, like every usage line today. Whatever share of the $3.3 billion is real arrives through the same door as the mechanical royalties above, with a harder matching problem, and the report's own risk table already names "market share dilution of royalty payouts" as the AI-era risk for labels (p.9). The work that is registered, identifiable and claimed gets paid from this pool. The rest joins the residue.
Goldman counts to the society's door. Nobody counts past it. From April 2027, the money nobody counted is paid to the people who did.
What a buyer, lender or publisher should check
None of this requires a view on the discount rate. It requires asking for evidence from the society side, not the statement side.
Treat the forecast as the top of the pipe
Goldman's $198.1 billion and 57% describe money reaching rights holders as a class. They say nothing about whether a given work is registered, matched and claimed. Price the forecast on the works whose registrations you have seen.
Ask for registrations, not statements
A royalty statement lists what matched. The evidence that the owner is being paid in full is the registration at each society, for each work, with each share and each identifier, and the absence of the work from that society's unmatched or unclaimed lists.
Look the catalogue up in the held pools
The MLC publishes its unmatched and unclaimed data for members to search. Works that appear there are earning US mechanicals that are not being paid to anyone. Equivalent lists exist at other societies with less publicity and shorter windows.
Date every finding against the clock
January 2021 usage is fixed on or around 10 February 2027, then one usage month a month. A registration or claim that is submitted but not processed before the snapshot misses it. Allow at least one monthly cycle of headroom.
Follow the writer's share abroad
For income from emerging markets, establish which local society collects, under which reciprocal agreement or sub-publishing deal the money comes home, and whether the works are registered there at all. If no one can answer, the forecast for that territory is unsecured.
Treat AI opt-ins as registrations
Whatever the licence template, AI money is paid by identified work and share. A catalogue with incomplete identifiers and unclaimed shares will be as hard to pay from an output royalty as it is from a stream.
The industry's best-read forecast counts to the society's door, and everyone who quotes it will quote the number at the door. The number that matters to an owner is the one on the far side, and for the first time it has both a published size and a published deadline. Check it before February 2027, or watch it become part of someone else's market share.
Goldman Sachs: Goldman Sachs Global Investment Research, Music in the Air: 10 Years On, 15 October 2026, 93 pages. Page numbers in this article are the PDF's, which run one ahead of the printed footer. Global retail revenue $112.5bn (2025) to $198.1bn (2035E), 5.8% CAGR; recorded music +$45.8bn, publishing +$12.1bn at 4.8%, live +$27.7bn: p.12, Exhibits 12 and 13. Retail-to-rights-holder share of about 57% held flat: pp.15–16. Recorded-music rights-holder revenue $31.7bn to $56.1bn: p.16. Mechanical and performance rights as the sources of rights-holder value: p.60. "AI forensic royalty recovery tools" and "improvements in royalty collection": pp.9, 57, 61. Emerging markets up to 75% of net adds, 438m to 956m subscribers, 10% penetration against 42%: pp.22, 34. Rights-holder ARPU $12 against $30: p.34. Latin America recorded music +17.1% in 2025: p.23. AI licence fee types including the Output Royalty: p.54, Exhibit 77. Opt-in structures and per-generation royalties: p.62. Attribution infrastructure: p.53, Exhibit 76. Generative AI base case $3.3bn, about 10%, at a 50% rights-holder split: pp.65–66, Exhibit 86. "Market share dilution of royalty payouts": p.9, Exhibit 7. The report is distributed to the firm's clients; a public summary is at goldmansachs.com. The counts of zero mentions of The MLC, CISAC, "unmatched", "unclaimed" and "black box" are TrackForge's word search of the full text.
The MLC, held royalties: The Mechanical Licensing Collective, 2025 Annual Report, appendix section F, figures as of the March 2026 distribution: royalties held by usage year and category; royalties collected by usage year after reprocessing (2021: $571,859,897.20); $52,690,895.03 matched, eligible and undistributable; average match rate 91.9% for 2021–2025 usage; historical unmatched royalties of $397,286,419.31 collected and $229,029,821.03 distributed as of March 2026. The year-earlier figures for 2021 usage ($90.9m in March 2025, $103.0m in March 2024) are from the 2024 and 2023 annual reports. Historical royalties distributed by October 2026, $239,955,887.57 of $397,286,419.01 (60.40%): HUR report, October 2026.
The MLC, market share: Policy for the Distribution of Unclaimed Accrued Royalties, September 2026: same snapshot as the month's blanket distribution (§3.8); claims count only if processed before it (§6.4); pro-rata allocation to claimed shares (§5.1); finality (§6.5); correction of The MLC's own errors only (§6.6); 90 days' notice (§7). Market Share page, read 23 October 2026: distributions to begin in April 2027 with remaining January 2021 usage, one usage month each month, snapshot on or around 10 February 2027, 2021 royalties held for at least six years, just over $6 million remaining for January 2021 after the September 2026 distribution, later claims paid "excluding only royalties for periods that were already subject to a market share distribution". Press release, 30 September 2026: the announcement of the April 2027 date and the one-month-at-a-time pace. The songwriter minimum of 50% is 17 U.S.C. §115(d)(3)(J)(iv).
CISAC and IFPI: CISAC Global Collections Report 2025, 6 November 2025: €13.97 billion collected for creators in 2024 (+6.6%) by 228 societies in 111 countries; Africa €90 million (+14.2%); Latin America and the Caribbean €786 million (−0.6%). IFPI Global Music Report 2025: recorded-music revenue in 2024 grew 22.5% in Latin America and 22.6% in Sub-Saharan Africa.
Disclosure: TrackForge checks whether works are registered, matched and claimed at collecting societies, and recovers royalties that are not. No TrackForge client or catalogue figure appears in this article; every number is from the published source named beside it.