On Drew Thurlow’s Generative Usage Royalty, and what we already know about how this ends
Drew Thurlow has a piece up at Alderbrook proposing a new royalty for AI. He calls it the Generative Usage Royalty: two tiers, a training-rights fee priced like sync and an output royalty distributed through a new collective body modelled on the MLC and SoundExchange. He grounds it in the long pattern of American music copyright — mechanical 1909, public performance in the 1930s, digital performance in the mid-1990s. Each time a new commercial use of music outran the existing system, a new royalty was built to catch up.
We need more pieces like this one, and more people writing them. The AI question is being settled right now, in private settlements and term sheets the rest of us will read about after the fact, and the number of working industry voices putting structured proposals on the public record is too small. Drew is doing what the rest of us should be doing. This is in his spirit, not against it. The fight is with the architecture, not with the writer thinking carefully about it.
The architecture has a problem. The pattern Drew invokes is real, but it has two beats, not one. A new use forces a new royalty. Then the rate is set by whoever has leverage at the table, and the mechanism for verifying payment is the first thing dropped under platform pressure. The 1909 mechanical did not arrive as a clean win for songwriters. The DPRA in 1995 delivered digital performance on terms hammered out by the parties that could walk away. SoundExchange has spent two decades explaining to letter-of-direction recipients why their money is sitting in a black box. The pattern arrives. The pattern also pays the long tail a fraction of what the press release implied.
We do not need to argue about this in the abstract. We have run this experiment twice in the last six years. Both times it failed.
Meta, 2020 to 2025
In 2020, Facebook signed blanket licensing deals with Universal, Warner, Sony/ATV and Merlin. The deals came with blank-check advance payments for a limited number of years. They did not come with any commitment from Facebook to report usage of music to labels and publishers. Tamara Hrivnak, then running music at Facebook, told Music Business Worldwide in late 2019 that the company had built “an awful lot of infrastructure to support our partners in music in a very short amount of time.” The infrastructure to verify what anyone was owed was not part of that build. Not in 2020, not in 2022, not in 2025.
In July 2022, Meta announced a 20 per cent revenue share to creators on eligible Facebook videos containing licensed music, with the remainder split between rights holders and Meta itself. The split has never been publicly detailed. By August 2025, Meta wound down the Reels bonus programs and routed full monetisation through its own royalty-free Sound Collection. Business accounts were pushed off licensed music entirely. In their 2024 Meta renewals, both Universal and Warner dropped premium music video licensing. Warner’s CFO Bryan Castellani put a number on what that cost his company: roughly ten million dollars per quarter. Sold publicly as an expansion.
The royalty got built. It paid the majors. It paid them less every year. It paid the indie sector through a Merlin blanket whose distributions, six years in, still arrive without the line-item usage data needed to audit them.
I have watched Meta payouts on indie statements since the deals began. The structure thins, the numbers fall, the column headers stay reassuring. There is no version of this story where the indies I work with got paid more in 2025 than they got paid in 2021. There is no version where they got paid with usage data attached. None of the architecture Drew proposes solves a problem that the platform did not already have a way to solve in 2020. The platform did not solve it because it did not have to.
TikTok and Merlin, October 2024
In October 2024, TikTok walked away from negotiations to renew its license with Merlin. The platform announced it would deal directly with individual Merlin members instead. A Variety source called it “classic divide and conquer.” IMPF called it a “thinly veiled attempt to divide independent labels and drive down the price of music.” The Worldwide Independent Network and IMPALA each issued statements. The American Association of Independent Music asked Congress to pass the Protect Working Musicians Act. Four indie bodies, four press releases, zero leverage.
The leak of the direct deal terms confirmed what everyone outside Merlin’s communications team already knew. The terms were worse. Per-view rather than per-creation, no MFN, no audit teeth, take it or leave it after the majors set the precedent in their own bilateral deals earlier that year. The few labels with the scale to push back got a slightly less bad version. The rest signed because the alternative was being delisted from the platform that had, between 2022 and 2024, become the single largest discovery surface in music.
Universal had pulled its catalogue from TikTok in early 2024 over royalty and AI concerns. They restored it in May. The party was big enough to walk got a better deal by walking. Nobody in the long tail can walk. That is the whole episode in one sentence.
The royalty got built. It is still being paid. It is being paid badly. The terms were set by the only counterparty with leverage at the table. The body created to give the indie sector its own leverage was disintermediated in a single October decision, because the platform realised the body had less power than the bluff suggested.
Who is the indie
Worth asking. Who is the indie sector any new royalty is supposed to protect?
FUGA is Universal. The Orchard is Sony. ADA is Warner. The “independent” distribution layer is, in significant part, a wholly-owned arm of the three majors. The genuinely independent operators — labels without major distribution, publishers running their own administration, artists on DistroKid plans — lost their collective negotiating leverage in October 2024 when TikTok broke the Merlin block.
The November 2025 class action filed in the Northern District of California, Nguyen v. Suno, is the receipt. The lead plaintiffs are independent artists distributed through DistroKid, TuneCore, CD Baby and similar aggregators. Their recordings were used to train Suno alongside major-label recordings. The major-label settlements announced in October and November 2025 cover the major-label catalogues. The indie catalogues had no seat at the table. They filed a class action because there was no other mechanism available to them. The same settlements that suggest the industry is figuring AI out — the Udio and Suno licensing deals — were the trigger for the indie suit because the deals explicitly excluded them.
The ECAD paragraph
Drew proposes that PROs are the natural administrators of his Generative Usage Royalty in their respective territories. In Brazil, the PRO is ECAD, which in practice means a federation of seven member associations. Of the twelve per cent of streaming revenue allocated to composition rights in the Brazilian split, ECAD administers three per cent. The remaining nine per cent — the mechanical — is administered outside the PRO entirely, primarily by Backoffice Music Services, an operator without the regulatory standing or transparency obligations of ECAD itself. Industry estimates put the share of the Brazilian independent market that receives nothing from the mechanical pool at around seventy per cent. The credits expire in three years.
The PRO Drew proposes as the natural administrator, but it has already lost the analogous fight on the previous platform shift. It ceded the larger slice to a back-office operator. It is not at the table with any platform that matters. It will not be at the table with OpenAI. The collective body Drew describes already exists here. It is the structural condition we have been working on for a decade.
The equity table
The royalty is not the deal.
In June 2025, Bloomberg’s Lucas Shaw reported that Universal, Sony and Warner were in licensing talks with Suno and Udio that would include both license fees and equity. Music Ally framed it as “licensing fees and equity stakes” in plain language. By the end of November 2025, the framework had moved from rumour to settlement. UMG settled with Udio on October 26, taking equity in the company’s pivoted 2026 platform. Warner settled with Udio on November 19 on similar terms. Warner settled with Suno on November 25, sold Songkick to Suno as part of the deal, and a $250 million Series C closed at a $2.45 billion post-money valuation in the same week. Menlo Ventures led. NVIDIA participated. So did Hallwood Media. Sony is still suing. UMG is still suing Suno. They are still suing on the way to the same deal. Habitual line steppers.
This is the Spotify playbook on a compressed timeline.
In 2008, Sony BMG took a six per cent equity stake in Spotify. Universal took five. Warner took four. EMI took two. Merlin, representing the entire independent sector globally, took one. That one per cent is the rate at which the collective body created to give indies major-label parity was permitted to participate in the equity upside the last time this exact deal got done. By 2018, Spotify went public at a valuation of thirty billion. Sony sold half its stake in 2018 for around 768 million. Warner sold its full stake for 504 million. Universal held, and as of late 2024, its remaining stake was worth approximately three billion dollars.
The press release version of that story is mostly Warner: $126 million credited to artist accounts on their June 2018 statements. Read the fine print, and most of that money still ran against unrecouped balances. For artists in the red on Warner’s books, it disappeared into the debt.
Sony went further. In summer 2018, Rob Stringer announced that Sony would bypass unrecouped balances entirely on the Spotify equity payout — over $250 million paid through to artists who, on any normal accounting, would have seen none of it. UMG followed in November. Warner did not. The estimated annual cost to Sony was $12.5 to $25 million.
That is nothing. That is, in fact, the strongest argument for the case Drew wants to make — that the system can be made to pay. I want to take it seriously because it is true.
It is also one CEO’s bet, not a Sony policy and not an industry pattern. The Stringer playbook — 2018 bypass, 2021 expansion of the unrecouped program, real-time royalty data, monthly cash withdrawals — is the work of one leader at one major, spending reputational capital on the way through a once-in-a-decade liquidity event. Pre-Stringer Sony was Sony BMG with the rootkit and the Mottola years. Post-Stringer is whoever inherits the chair. What really moved him in that summer is the kind of thing we will have to wait for a book to find out.
None of those conditions reproduces in the AI settlements. The deals are private. The valuations are private. The cap tables are private. The training data is sealed in settlement filings that UMG and Sony are still trying to pry open in discovery in their own ongoing cases against Suno. There is no IPO-equivalent visibility moment forcing the question into the open. There is no CEO at any major currently positioning themselves the way Stringer did in 2018. And the Sony of the 2018 bypass — the one major that actually paid through — is the major still suing Suno in 2026. Not settled. Not paying. Not making the gesture. The leader bet has a face. The face is not in the room.
Who cashes the check?
The majors are not being paid for music. They are being paid for waiting.
They sue, they settle, they take license fees and equity, and they wait for the exit. Suno told its Series C investors that current revenue — two million paying subscribers and three hundred million ARR as of February 2026 — is the warmup. Quote from the deck: a billion in revenue by 2028, “before we consider monetising consumption. The next step is a $500 billion company. The phrase is in the pitch deck, not in a press release.
Suno is hiring a former Warner executive as Chief Music Officer and a former Patreon and Spotify executive as Senior Director of Artist Partnerships. They are not staffing a tool. They are staffing a platform. Seven million songs generated per day, full Spotify catalogue equivalent every two weeks, a hundred million total users, in-platform listening, a TikTok-style social feed in the product roadmap, and a concert discovery service that Warner handed over in the settlement. This is Spotify 2014 with a generative training corpus underneath it. Mikey Shulman says publicly that the “AI Spotify” framing is “obviously wrong.” He says this while building it.
Udio took the deal and gave up the business. Walled garden, no downloads, fan engagement platform, no open generation, the 2026 model is a remix tool with licensed inputs. Udio is the Deezer of AI. Lower share, friendlier to the industry, willing to be domesticated. Suno took the deal, kept the platform, kept the open generation, kept the consumption surface, and added Songkick to it. Suno is the actual Spotify analogue. The majors are positioned across both. They will cash the check the way they cashed it in 2018. The escrow is already drawn up.
ECAD cannot hold equity in a US-domiciled AI company. Neither can ABRAMUS, UBC, SOCINPRO, the MLC, or SoundExchange. The collective bodies Drew describes as natural administrators are structurally locked out of the payout structure that the majors are actually using. They will administer a royalty pool. The pool will be a fraction of what the equity pays. The class action filed by independent artists in November 2025 is not asking for a seat at the equity table. It is asking, for now, to be told what their recordings were used to train and what the training was worth. The answer they get will determine whether the next class action does ask for equity, and how loud the answer is allowed to be.
Will the check arrive
Will the check arrive at the underlying rights holder? Is the metadata good enough to find them?
After Meta in 2020, with no usage reporting committed at signing and none added since, the answer is no. After TikTok in October 2024, with direct deals stripped of MFN protection and audit teeth, the answer is no. After ECAD’s mechanical surrender to a back-office operator that pays seventy per cent of the indie market nothing, the answer is no. After years of MLC and SoundExchange black boxes paying the long tail in opaque pool distributions while the registered titles sit in dispute queues, the answer is no.
Spotify, with all its problems, with the pro-rata formula that has held the long tail to a third of a cent per stream for a decade, with the discovery mode royalty cut and the noise threshold and the ghost-artist playlists, looks like a temple by comparison. At least Spotify pays according to a public formula. At least Spotify reports streams per track in territories I can audit. At least Spotify’s 2018 IPO was the event where Warner credited 126 million dollars to artist accounts, and Stringer chose to bypass unrecoupment to pay artists their share.
The AI royalty, as currently sketched, is darker on every axis. No public formula. Undisclosed weights in the output pool. No reporting infrastructure built. Administered by a body that does not yet exist. Distributed through a metadata infrastructure that is already broken at scale and that no party at the negotiating table has any incentive to fix. All while the majors who would design the framework have already cashed out at the equity table running alongside it.
A new royalty is not a safeguard. Leverage is the safeguard. The long tail has none. The “indies” the framework is supposed to protect are mostly majors in a different jacket, and the few that remain genuinely independent lost their collective leverage in October 2024. The class action is the new instrument because the old instrument was disintermediated.
The royalty will be built. Something will pass, called the Generative Usage Royalty or close enough. It will be administered by something resembling a collective body. The collective body will deliver a pool distribution to participating rights holders, weighted by formulas the rights holders cannot inspect, against attribution that the architects of the framework already concede they cannot solve. The majors will already have closed their equity rounds and the first IPOs by then.
We need more people writing about this in public. Drew is doing the work. The architecture needs pressure. The reporting will not build itself, and the safeguards will not arrive if the only people putting frameworks on paper are the ones the majors are already positioned to cash out around. The system eats training data and excretes equity. The check for the long tail is the part that does not come out.
Plan accordingly.