Primary sources used in this piece include Spotify’s Form F-1 registration statement (SEC EDGAR, filed February 28, 2018) and the confidential draft registration statement that preceded it in 2017; “Spotify — Large Scale, Low Latency, P2P Music-on-Demand Streaming” (Kreitz & Niemelä, IEEE P2P 2010), written by Spotify’s own engineers; the leaked 2011 Sony Music–Spotify US licensing contract and the reporting built on it; the NBER working paper “Platform Power Struggle: Spotify and the Major Record Labels” (w33048); the academic book “Spotify Teardown” (MIT Press, 2019); and contemporaneous journalism from the period. Where a number comes from an audited filing, I use it as the filing states it. Where it comes from credible reporting but no primary document, the text says “reportedly.” Where accounts conflict — and on the label deals they conflict a lot — I show the conflict rather than picking a side.
Prologue: Seventy-Two Hours
On May 31, 2006, Swedish police raided The Pirate Bay and seized its servers. The site was, at that point, the most visited torrent directory in the world — a Stockholm-run institution founded by the anti-copyright group Piratbyrån in September 2003, and the clearest single symbol of the fact that an entire generation had stopped paying for music.
The site was back online on June 2. Roughly seventy-two hours.
It got bigger after the raid, not smaller. By Wikipedia’s accounting of the period, registered users grew from one million to 2.7 million within two years, and the site claimed over five million active users in 2007. When Swedish courts finally convicted its operators in April 2009 — a year in prison each and a 30 million kronor fine, raised on appeal — the site kept running anyway. The most heavily policed piracy operation on earth had absorbed a police raid, a criminal trial, and prison sentences, and lost approximately nothing.
A few kilometers away, in the weeks around that raid, two entrepreneurs were incorporating a company built on a different theory: that piracy could not be raided, sued, or legislated out of existence — it could only be out-engineered. The Swedish operating company, Spotify AB, was established in Stockholm that spring; the Luxembourg holding entity, Spotify Technology S.A., was incorporated on December 27, 2006, a date confirmed in the company’s SEC filing.
The state had tried force. Spotify was going to try a progress bar so fast you never saw it.
Part I: The World They Were Born Into
Two collapses made Spotify possible, and both were well underway before the company existed.
The first was the collapse of the recorded music business. Spotify’s own F-1 states it plainly: global recorded music revenues fell from $23.8 billion in 1999 to $16.9 billion in 2008 — a roughly 30% decline in the nine years before Spotify launched — as, in the filing’s words, “Growth in piracy and digital distribution were disrupting the industry.” The draft registration statement adds the longer arc: revenues kept falling to $14.3 billion by 2014, a 40% peak-to-trough collapse, and the trend did not reverse until 2015. The NBER paper on the Spotify–label relationship, using RIAA data in inflation-adjusted terms, puts the US decline at $23 billion in 1999 to $7.4 billion in 2014. Warner Music Group posted negative net income in six of the seven years between 2005 and 2011. This matters for everything that follows: every negotiation Spotify ever had with a label happened against the backdrop of an industry that had been shrinking for as long as anyone could remember.
The second collapse was of the assumption that Swedes would ever pay for music again. Sweden was not incidentally piratical; it was structurally piratical, and the state had — without intending to — built the infrastructure. Between 1998 and 2001, a government home-PC program subsidized computers through employer tax breaks; per the Swedish Internet Foundation’s history of the reform, roughly 850,000 machines were distributed, reaching close to a quarter of the country’s four million households, at a cost of about four billion kronor. Broadband followed the boxes. By one published count, Sweden had 28 broadband subscriptions per 100 people in 2005, against 17 in the US — where dial-up was still common — and a global average of 3.7. Klarna’s founder later argued that streaming “could only happen in a country where broadband was the standard much earlier,” and the timeline supports him: Spotify built a streaming product while Apple, serving a dial-up-and-DSL America, was still selling downloads.
Put the two together and you get the strange market Spotify was born into. Sweden had the best consumer internet in the world, the most normalized piracy culture in the world, and a music industry whose local revenue had already been hollowed out. For a label executive in Stockholm in 2007, the counterfactual to licensing some startup’s free streaming service was not lost CD sales. Those were already gone. The counterfactual was zero.
Kazaa — co-created by Niklas Zennström and Janus Friis, who will reappear in this story on the losing side — and then The Pirate Bay had trained an entire population in a specific behavior: search for a song, click, listen, pay nothing. Any legal product that wanted those users could not ask them to change that behavior. It had to perform the same behavior, better.
Part II: The Founders
The F-1’s management section is unusually useful here, because founder backstories are normally the most embellished part of a startup history, and this one is written under securities law. Per the filing: before founding Spotify in 2006, Daniel Ek founded Advertigo, an online advertising company acquired by TradeDoubler; held senior roles at Tradera, the Nordic auction company acquired by eBay; and served as chief technology officer of Stardoll, a fashion and entertainment community for pre-teens. He was in his early twenties and had already touched three of the businesses that would define Spotify’s mechanics: advertising (the free tier’s revenue), marketplaces (two-sided dynamics), and consumer internet products for people with no money (the free tier’s audience).
Martin Lorentzon founded TradeDoubler, a Stockholm internet-marketing company, in 1999 — the F-1 says “founded”; TradeDoubler’s own history credits a co-founder, Felix Hagnö. The two men met through the deal flow: TradeDoubler acquired Ek’s Advertigo in March 2006, reportedly for around SEK 10 million. Within weeks they had decided to start something together; accounts place the decision in April 2006 and the Swedish registration that June. Lorentzon had the money — he had reportedly sold TradeDoubler holdings for tens of millions of dollars after its 2005 IPO — and reportedly put in an initial million euros of his own, covering developer salaries, offices, and early licensing costs. The Wikipedia account of the period adds a human detail I can’t independently verify but which multiple profiles repeat: the two bonded partly over the aimlessness of having gotten rich young, and built Spotify in part to have something worth doing.
One more line on the team, hedged appropriately because the sourcing is secondary: after selling Advertigo, Ek reportedly served briefly as CEO of uTorrent — the world’s dominant BitTorrent client — until it was sold to BitTorrent Inc. in December 2006, and uTorrent’s creator, Ludvig Strigeus, reportedly then joined Spotify as a developer. If accurate, and it is widely reported, it means the company that beat file-sharing employed the engineer who had built file-sharing’s best client. That is not color. It is the whole thesis in personnel form.
Ek’s stated framing of that thesis, as reported by Music Business Worldwide, was that the goal was a service “better than piracy and at the same time compensates the music industry.” Not cheaper than piracy — nothing is cheaper than free. Better.
Part III: The Thesis and the Pitch
The thesis had three parts, and each was radioactive in 2006.
First: music would be free at the point of use, funded by advertising, because the competition was free and no paywall would survive contact with a population trained by Kazaa. Second: the free tier would eventually push people into a paid subscription — a model with essentially no successful precedent in music. Third: the labels, who had spent the previous half-decade suing services that looked exactly like this, would license their entire catalogs to it.
The third part took two years and is the most myth-encrusted stretch of the whole story, so here is what the record actually supports.
The NBER working paper states, citing the reporting record, that after “much cajoling” Ek convinced all of the major labels to license their content — and that the majors “took ownership stakes and advance payments” as part of the deals. That much I treat as solid: equity plus advances, in exchange for catalogs. The paper deliberately gives no equity percentage.
The percentage that circulates — that the majors and the indie consortium Merlin together got roughly 18% of the company — comes primarily from Music Business Worldwide, which in 2018 dug the original share issuances out of Luxembourg’s corporate registry and reported that in October 2008 the four majors plus Merlin were issued 352,176 shares for a nominal aggregate price of €8,804.40, which the outlet computed as an 18% combined holding: roughly 6% to Sony BMG, 5% to Universal, 4% to Warner, about 2% to EMI, and 1% to Merlin. I present that as reported-but-unconfirmed: it is one outlet’s reading of registry documents, the per-label percentages vary slightly across accounts, and neither Spotify nor the labels has confirmed the figure. What is not in dispute is the direction of the trade — the labels took equity essentially for free alongside their licensing terms, and it eventually paid: after the April 2018 listing, Sony reportedly sold half its stake for around $750 million and Warner sold three-quarters of its stake for roughly $400 million.
Why did companies that were suing college students hand their catalogs to a Swedish startup? Because the alternative in Sweden was nothing, and because the deal was built so the labels won either way. Advances guaranteed cash whether or not the product worked. Equity captured the upside if it did. And the underlying leverage never moved: the US Copyright Royalty Board later formalized what everyone in these negotiations already knew — that an interactive streaming service “must have” the repertoire of every single major to survive commercially. There were four suppliers, each holding a veto, and content licensed from Universal, Sony, Warner, and Merlin still accounted for approximately 88% of Spotify streams as late as 2016, per the draft registration statement. Spotify never had negotiating power. It had a desperate counterparty, which is a different thing, and the deals show it: the labels priced their desperation into advances, minimums, and equity, and Spotify paid.
The deals Spotify could actually close in 2008 were European. The product launched in October 2008 across a handful of European markets — the engineers’ own 2010 paper describes it as available in six European countries from that date — reportedly alongside a $21.6 million Series A that included Li Ka-shing’s Horizons Ventures, Creandum, and Northzone. The United States, the largest music market in the world, stayed closed for almost three more years. That gap is the single most important fact about Spotify’s early growth, and I’ll come back to it.
Part IV: The Growth Machine
This is the longest section because it is the actual subject. Spotify’s growth machine had four interlocking mechanics — scarcity, speed, a borrowed social graph, and a free tier that the company itself would later describe, in a securities filing, as a funnel. Each has a legend attached. Here is what the record supports.
The invite and the progress bar
Spotify did not launch open. The 2008 European launch paired paid accounts with an invite-gated free tier, and the invite stayed load-bearing for years — growth-marketing accounts of the period describe each loosening of access (open free signups in early markets, the eventual retirement of invites, mobile) as a deliberately pulled lever rather than a floodgate opening. I treat the lever-by-lever chronology as reportedly-true rather than verified; what the primary record does show is that the constraint was real and the demand behind it was real. By the engineers’ own account, the service had over seven million users and more than eight million tracks by mid-2010, from a standing start in six mid-sized European markets, with essentially no paid marketing story anyone has ever credibly documented.
The scarcity did two jobs. It made a free product feel like a privilege — in a piracy-saturated market, where the competing product was infinitely available, an invitation was the one thing The Pirate Bay could not offer. And it throttled the burn: every free user cost money (more on that arithmetic below), so the invite system was also, quietly, a cash-flow valve.
The early numbers were tiny. As of June 2009, per an academic analysis of the period, Spotify reportedly had about 170,000 registered users and monthly advertising income around £82,000 — a rounding error against royalty obligations that came with guaranteed minimums attached. Nobody inside or outside the company could claim, on the 2009 evidence, that advertising would ever pay for the royalties. The free tier survived those years as a bet, not as a business.
Faster than free
The pitch to users was never “legal.” Legality is a feature for lawyers. The pitch was speed — that clicking a song in Spotify would start playback faster than a downloaded pirate file could open, and so much faster than a torrent that the comparison became embarrassing.
This is the rare growth legend with a primary source attached, because Spotify’s engineers published the measurements. The Kreitz & Niemelä paper (IEEE P2P 2010) reports that in a March 2010 measurement week, “the median playback latency is only 265 ms” — including cached tracks — with a 75th percentile of 515 ms and under 1% of playbacks experiencing stutter. A quarter of a second. Human-perception territory. The paper also documents the constraint as engineering religion: the peer-to-peer layer was explicitly forbidden from degrading playback latency, with the latency-critical opening seconds of any randomly accessed track served from central servers while P2P handled the bulk.
And that P2P layer is the most under-told part of the machine. During that same measurement week, only 8.8% of played music data came from Spotify’s own servers. 35.8% came from other users via the peer-to-peer network; 55.4% came from the local client cache. Spotify ran the fastest music service in the world while paying for less than a tenth of its own bytes — its users’ bandwidth and disk space carried the rest. The peer-discovery design, per the paper, borrowed directly from the file-sharing world: a BitTorrent-style tracker plus Gnutella-style search. Spotify beat piracy with piracy’s own toolkit, and the toolkit doubled as an infrastructure subsidy during precisely the years the company could not afford infrastructure. The P2P network was retired in April 2014, once — per the company’s statement at the time — its own server fleet could carry the load. The scaffolding came down after the building stood.
The two-year wall
Between the October 2008 European launch and July 2011, the United States was closed to Spotify. The blocker was licensing: the US majors would not grant the free tier. The delay is usually narrated as label stubbornness, but the leaked Sony contract lets us see what “yes” eventually cost, and it reframes the whole episode.
The Sony Music US agreement — a 42-page contract signed January 18, 2011, about six months before the US launch, leaked in 2015 and widely authenticated after The Verge’s copy was forced down and Digital Music News reported receiving over two dozen independent copies — obligated Spotify to pay Sony $25 million in advances over the first two years plus $17.5 million for an optional third, with per-stream minimums reported at roughly $0.00225–$0.0025 and a most-favored-nation clause guaranteeing Sony whatever better terms anyone else negotiated. Rolling Stone’s analysis of the contract structure put roughly 60% of revenue flowing out to rightsholders under it — before publishing royalties. One label. Multiply by four majors, all of whom knew about the MFN dynamics, and the two-year delay stops looking like foot-dragging and starts looking like the price of admission being negotiated line by line: the US launch happened when Spotify agreed to guarantee tens of millions of dollars against a free tier that had never once covered its own costs. Reporting on the leak also noted the advances likely never reached Sony’s artists — a detail that detonates three years later in the Taylor Swift section.
Sean Parker enters the record here, and I hedge his role to what the sourcing supports: he reportedly invested about $15 million through Founders Fund, and is reported to have personally worked the Warner and Universal relationships in the run-up to the US deal, along with the Facebook connection. The mythologized version makes Parker the door-opener; the documented version makes him one accelerant among several. I can’t adjudicate further on the record I trust.
July 2011: the borrowed graph
Spotify launched in the US in July 2011 — reportedly July 14 — with paid accounts and an invite-gated free tier, replaying the European scarcity playbook in the one market where American tech press coverage could amplify it. Within a year, per growth-marketing accounts of the period, it had reportedly passed three million US users, a fifth of them paying.
The distribution weapon of that launch was Facebook. The partnership wired Spotify into Facebook’s social graph in the autoshare era — friends’ listening surfacing in feeds, the product broadcasting itself with every play. I keep the description qualitative because the dossier of documents I trust doesn’t quantify the channel’s contribution, but the strategic shape is clear and multiply reported: Spotify entered the US not as an unknown Swedish utility but as a socially visible object inside the largest network on earth, at the exact moment that network was pushing “frictionless sharing” as its platform strategy. In Rogers’ diffusion terms — more on this below — Facebook manufactured observability for a product whose usage is otherwise private.
By March 2011, just before the US launch, Spotify reportedly had 6.67 million users and one million paying subscribers — a 15% conversion rate that was already, quietly, the most important number in the company.
December 2013: the funnel finds its mouth
Mobile was Premium-only for years — the app reportedly arrived in fall 2009 as a subscriber perk. That made the free tier a desktop product in a world going mobile, which capped the top of the funnel exactly where the next billion listeners were arriving.
In December 2013, Spotify launched free mobile listening (shuffle-based), paying for it with whatever licensing concessions that required. The share of listening on mobile reportedly tripled between 2013 and 2014. And the company clearly understood what it had built, because starting in February 2014 it began formally tracking where its paying subscribers came from — instrumentation that would later produce the single most important sentence in its F-1.
The funnel, quantified
Here is the sentence. Spotify’s F-1 describes the free tier as a funnel “driving more than 60% of our total gross added Premium Subscribers” since it began tracking the data in February 2014. The draft registration statement adds the kicker: “Only a small percentage of new Premium subscribers join via paid acquisition channels.”
Sit with what that means. The largest music service in the world acquired the substantial majority of its paying customers from its own free product, at essentially zero marginal marketing cost, and said so under securities law. The free tier’s costs lived in cost of revenue — royalties on free streams — rather than in a sales-and-marketing line, which means Spotify’s “customer acquisition cost” was structurally disguised as a gross-margin problem. That’s why the company looked so alarming on a P&L and so unstoppable on a cohort chart, simultaneously, for a decade.
The filings let us watch the machine improve:
- Users: 91 million monthly active users and 28 million Premium subscribers at the end of 2015 (a ~31% conversion rate); 125 million and 48 million at the end of 2016 (~38%); 159 million and 71 million at the end of 2017 (~45%), across 61 countries and territories — a scale the F-1 asserts was “nearly double the scale of our closest competitor, Apple Music.”
- Churn: quarterly Premium churn fell from 7.5% in Q4 2015 to 6.0% in Q4 2016 to 5.1% in Q4 2017 (the annual-average series runs 7.7% → 6.6% → 5.5%).
- Growth rates at the end of the period: Premium subscribers up 46% year over year; MAUs up 29%.
The conversion rate roughly tripling from 2011’s reported 15% to 2017’s 45% is the growth story, compressed to one ratio. Two caveats the filings themselves force: these are company-reported operating metrics, not independently audited ones, and a March 2018 amendment trimmed the year-end MAU figure from 159 million to 157 million after excluding roughly two million accounts that had suppressed ads with unauthorized apps — a rounding error at that scale, and an oddly fitting one: even Spotify’s fraud problem was people pirating the free tier.
And the cost of it all, per the F-1: through December 31, 2017, more than €8 billion in cumulative royalties paid to artists, labels, and publishers since launch (the earlier draft statement had put it at more than €7 billion through September 30, 2017 — the two figures are consistent, one quarter apart). Revenue of €1,940 million, €2,952 million, and €4,090 million for 2015, 2016, and 2017 — a 45% compound growth rate — against net losses of €230 million, €539 million, and €1,235 million in the same years. Gross margin was 12% in 2015 and 14% in 2016 per the filings; royalties and other cost of revenue consumed roughly 86–88% of every euro. The free tier specifically ran underwater for years: ad-supported cost of revenue was 116% of ad-supported revenue in 2015 and 112% in 2016 — Spotify paid out more in royalties on free listening than free listening earned — and only crossed into positive territory in 2017, at 90%, when the renegotiated 2017 label deals (lower rates, contingent on growth targets, per the F-1) finally took hold. One fairness note on the scariest number: the €1,235 million net loss in 2017 was mostly non-operating finance costs from convertible notes; the operating loss was €378 million.
That is the machine, whole: a free product that cost more than it earned for nearly a decade, deliberately, because it was not a product. It was the top of a funnel that converted at rates no advertising budget could buy.
Part V: The Wars
The incumbent that wasn’t fighting
Spotify’s most important competitor spent the entire war selling something else. Apple’s iTunes was the legal-music incumbent — and it was a download store, built around ownership, purchase decisions, and syncing, in an era when Apple’s home market was still climbing onto broadband. Pandora, the biggest American streaming name of the period, was lean-back internet radio: no on-demand choice, no catalog access, a structurally different product with a structurally different (and cheaper) license. For roughly seven years, the largest technology company in music and the largest streaming company in America both declined to build the thing Spotify was building. That gap in the record is not a footnote; it is the single largest gift the company ever received, and I’ll price it in the luck audit.
Rdio: the control group
If you want to know whether Spotify’s model was obvious, the market ran the experiment. Rdio was founded in 2008 by Carter Adamson with Janus Friis and Niklas Zennström — the creators of Kazaa, the very software that trained Spotify’s home market to expect free music — and launched on-demand streaming in the US before Spotify got there. It reportedly raised about $125 million. It was, by many contemporary accounts, a beautifully designed product.
It died in 2015, its assets reportedly sold to Pandora for around $75 million. The autopsy, per the post-mortem record: Rdio required payment from day one — no ad-supported free tier at launch — and ran an almost nonexistent marketing and distribution operation, attempting the freemium pivot too late to matter. The men who built the piracy generation’s favorite tool would not subsidize a free tier, and the company they built lost to the one that would. Freemium was not the obvious move; it was the move everyone else was too fiscally sane to make.
Taylor Swift and the price of free
On November 3, 2014, Taylor Swift pulled her entire catalog from Spotify — then claiming over 50 million users, more than 10 million paying — and withheld her new album, 1989, entirely. Spotify had refused to window the album behind the paywall; Swift refused, in her words, to contribute her life’s work to “an experiment that I don’t feel fairly compensates the writers, producers, artists.”
The numbers fight that followed is the clearest documented window into streaming economics ever opened in public, and the two sides’ figures — both plausibly true — measured different things. Big Machine’s Scott Borchetta told TIME the label had received $496,044 for domestic streaming of Swift’s music in the prior twelve months, less than 50,000 albums’ worth, and less than her Vevo videos earned. Spotify countered that her global payout across the same period was $2 million, that it had paid roughly half a million dollars in the single month before the pull — without 1989 even on the service — and Daniel Ek claimed an artist of her scale was on track to exceed $6 million a year as the audience grew. Ek also used the moment to state that Spotify had paid $2 billion to labels and publishers cumulatively — the company’s own claim, contemporaneous with the dispute.
Note what nobody in the fight disputed: the money was flowing to the labels. Spotify was already paying out the large majority of its revenue — a company representative had told researchers in 2013 that “80 to 85 percent” of the money coming in “goes out again to labels and artists” — but what reached artists depended on label contracts Spotify didn’t control. A 2015 French study reported that major labels kept 73% of Premium payouts, publishers and writers took 16%, and artists received 11%. And the artist-payout grievance long predated Swift: the record shows a Swedish artist pulling his catalog in 2009 and a Norwegian indie label reporting roughly $3 in earnings on more than 55,000 streams that same era. Swift’s protest was the loudest, not the first — and the structural target of it, arguably, should have been the parties who held the equity and kept the advances.
1989 sold nearly 1.7 million US copies in its first two weeks, per Nielsen SoundScan. Spotify’s user base grew more than 50 percent that year anyway. Both facts are true; that’s the point. The biggest artist in the world at the peak of her leverage could not dent the funnel.
June 2015: Apple arrives
Apple Music launched in June 2015 — the platform owner, with the App Store, the default install, and the deepest balance sheet in the industry, finally building the on-demand product. Ek’s reported response was a tweet reading “Oh ok,” quickly deleted.
Here is what the filings say happened next: Spotify’s Premium base went from 28 million (end-2015) to 48 million (end-2016) to 71 million (end-2017), while quarterly churn fell every year — 7.5% to 6.0% to 5.1%. Growth accelerated and retention improved after the most dangerous possible competitor entered. By the F-1, Spotify believed it stood at nearly double Apple Music’s scale, and contemporary reporting put Apple at roughly 36 million subscribers against Spotify’s 71 million. (One qualifier from the period’s coverage: Apple was reportedly adding US subscribers at a faster percentage rate by early 2018 — the war wasn’t over, and in the US it still isn’t.)
Why did the funnel survive Apple? Because Apple didn’t copy it. Apple Music launched paid-only with a trial, not with a perpetual free tier — the same choice Rdio made, made this time by a company that could have afforded otherwise. Spotify remained the only place on earth where all the music was free forever, legally. As long as that stayed true, the top of the funnel belonged to Spotify, and 60% of its new subscribers kept walking in through it.
The story told vs. the record
Three legends, checked against the documents:
“Spotify killed piracy.” Partially true, and it deserves the measured version. What the record supports: the industry’s revenue decline reversed in 2015 for the first time in a decade and a half, Spotify had paid over €8 billion into the industry by the end of 2017, and its founding premise — out-convenience the pirates — visibly moved a generation of listeners. What the record also shows: The Pirate Bay remained the most visited torrent site in the world until late 2014, six years into Spotify’s rise. Piracy wasn’t killed. Its reason was.
“The labels were dragged kicking and screaming.” The filings and registry documents tell a different story: the labels took equity (reportedly ~18% combined, for a reportedly nominal €8,804), took tens of millions in advances (per the leaked Sony contract), held per-stream minimums and MFN protection, kept — per the 2015 study — roughly 73% of premium payouts, and reportedly sold listing-era stakes for over a billion dollars combined. They weren’t dragged. They were paid at every stage, in every state of the world, including the ones where Spotify died.
“Freemium was obviously going to work.” The free tier lost money on its own revenue — cost of revenue above 100% — every year on record until 2017. The best-funded competitor refused to copy it; Apple refused to copy it; the biggest artist in the world publicly seceded from it; and the company reached its 2018 listing having lost over €2 billion across just its three filed years (2015–17), with the free tier’s economics turning positive only months before the filing. “Obvious” is not the word. It was a bet that stayed open for nine years.
Part VI: Structural Analysis — the Thinking Tools
Before the scorecard renders a verdict, two frameworks earn their keep on this story as ways of seeing it. As with every piece in this series: these organize the situation, they don’t judge it.
Porter’s Five Forces, in one paragraph, because one force is the story. Supplier power in recorded music streaming is close to the theoretical maximum, and it never softened: four suppliers controlled catalogs that accounted for roughly 88% of Spotify’s streams as late as 2016, the CRB itself concluded an interactive service “must have” each major’s repertoire to be commercially viable, and MFN clauses meant no supplier could be played against another. Every other force reads normal — buyers fragmented and powerless individually (but frictionless to churn), substitutes brutal (the substitute was free), entry barriers low for a product and crushing for a catalog, rivalry fierce but thinned by the bust of undercapitalized clones. This is why Spotify’s gross margin was 12–14% while software companies its size ran 70%: the industry structure routed the economics to the suppliers, permanently. Spotify’s genius was never escaping Porter’s diagnosis. It was building the only company that could survive inside it.
Rogers’ Diffusion of Innovations explains the growth curve. Run the five adoption factors: Relative advantage — a quarter-second to any song ever recorded, versus torrent queues, fake files, and malware; the advantage was felt in the first ten seconds of use. Compatibility — this is the subtle one: Spotify demanded almost no behavior change from a pirate. Search, click, listen, pay nothing. It was piracy with better engineering. The behavior change it did demand — paying — was deferred, optional, and arrived only after the habit was locked. Complexity — near zero. Trialability — total; free forever, invite as the only gate (which converted trialability into desire). Observability — engineered, first through invite envy, then industrially via the Facebook autoshare partnership. Products that score like this spread at compounding rates regardless of their income statements. Spotify’s did.
The Evidence Scorecard: Would You Have Funded Spotify?
Now the judging instrument — the weighted, gated scorecard described on the series hub, scored the way a disciplined investor would have had to score it at the time, without the answer key. Two snapshots: the 2008–09 launch era, when the company was an invite-only free-music service in a handful of small markets with negative-gross-margin licensing deals; and end-2014 into 2015 — mobile-free launched, the funnel instrumented, Swift gone, Apple incoming.
The gates first
Four kill questions, answered with period evidence:
- Was the pain real — frequent, expensive, urgent? Split verdict, and the split matters. For the industry, the pain was existential and quantified: a 30% revenue collapse by 2008, on its way to 40%. For the consumer — the party who’d eventually need to pay — the “pain” was the friction of free: malware, fake files, no mobile, moral unease for some. Frequent, yes (music is a daily behavior). Expensive, no. Spotify’s real gate answer was that it monetized the industry’s desperation and the consumer’s convenience gap, not a consumer’s burning problem. Passes, narrowly, with that asterisk.
- Was there a nameable buyer? In 2008: advertisers (nameable, small — reportedly £82k a month against 170k users) and a hypothetical subscriber who had never existed at scale in music anywhere on earth. Marginal pass at best; this gate was answered by faith until roughly 2011.
- Could the market be venture-scale? The top-down market was shrinking — a $16.9B industry in decline. The bottom-up case was the honest one: hundreds of millions of people already exhibiting the listening behavior daily at price zero; converting even a modest fraction to ~€60–120 a year rebuilds the industry with a new landlord. Pass — but note it required believing the bottom-up case against a falling top-down number.
- How much behavior change did adoption demand? With a paywall: enormous (pay for access, own nothing — a model with no successful precedent). With the free tier: almost none. The free tier existed, structurally, to defeat this gate. Pass — conditional on freemium.
No gate fails outright. Two pass only because of the free tier — which is exactly where the money bled. Proceed to scoring with that tension named.
The weighted score
| Category | Weight | Snapshot 1: 2008–09 | Snapshot 2: end-2014/2015 |
|---|---|---|---|
| Product-market fit | 30 | 16 | 26 |
| Distribution | 25 | 13 | 23 |
| Unit economics | 20 | 3 | 8 |
| Market quality | 10 | 6 | 9 |
| Team / founder-market fit | 10 | 8 | 8 |
| Moat / defensibility | 5 | 1 | 2 |
| Total | 100 | 47 | 76 |
Product-market fit — 16, then 26. In 2009 the behavioral evidence was real but thin: coveted invites, fast organic growth to a reported 170k registered users, and a product whose measured speed beat the alternative. No retention data was public; willingness to pay was entirely unproven. Only behavior counts, and the paying behavior didn’t exist yet — 16. By 2015 the behavior was overwhelming: 28 million people paying roughly €7 a month (91M MAU, per the filings), annual churn falling through 7.7%, usage surviving the catalog loss of the world’s biggest artist. Not higher than 26 because churn in the 6–8%-a-month range is still a leaky product, and ARPU was already sliding (€6.84 in 2015, headed toward a reported €5.32 by 2017) as discounted family and student plans — reportedly at least a fifth of new subscribers over the following two years — did more and more of the growing. Demand was proven; demand at full price less so.
Distribution — 13, then 23. The 2009 machine was genuinely novel — invite scarcity plus a free product in piracy’s home market, near-zero CAC — but it was licensed market-by-market, meaning Spotify’s distribution was gated by the consent of four suppliers, and the largest market on earth was closed. A channel you cannot legally enter is not yet a repeatable channel — 13. By 2015 the funnel was instrumented and quantified in what became the F-1: more than 60% of all gross Premium adds from the free tier, “only a small percentage” from paid acquisition, with the Facebook graph and the 2013 mobile unlock feeding the top. One repeatable channel, owned, measured, compounding — 23, docking two points because the channel’s cost sat in cost of revenue and its legal existence still depended on license renewals.
Unit economics — 3, then 8. This category is the story, so I’m going to spend a paragraph on it. In 2008–09, every part of the model lost money on its own terms: royalty advances and per-stream minimums due regardless of revenue, ad income around £82k a month reportedly, and a free tier whose marginal listener generated cost with no offsetting revenue to speak of. There was no evidence — none — that any customer would ever become profitable. That is a 3, and the non-compensatory rule says no other category can rescue it. By the 2015 snapshot it improves to an 8, not more: conversion was proven and marginal acquisition was nearly free, but the filings show gross margin of 12% (2015) and 14% (2016), the ad-supported tier still paying out 116% and then 112% of its own revenue in royalties, net losses of €230M and then €539M, and ARPU falling. The honest 2015 statement was: the funnel works, the suppliers take the surplus, and profitability depends on renegotiating with a cartel that has never once needed you. (That renegotiation happened — 2017’s lower contingent rates took the free tier to 90% cost of revenue and the company to a 21%-ish gross margin per analyses of the filing — but a 2015 investor couldn’t underwrite it, only hope for it.)
Market quality — 6, then 9. A shrinking top-down market with an impeccable why-now (broadband, smartphones, supplier desperation) and a gigantic un-monetized behavior pool — 6. By 2015 the industry’s global revenue had inflected positive for the first time since 1999, with streaming as the engine: the “why now” had become “now” — 9.
Team / founder-market fit — 8, then 8, capped per the series rule. Documented, not narrativized: two founders with exits (Advertigo; TradeDoubler co-founded), reportedly self-funding the earliest years, with a skill map that matched the problem almost perfectly — ad-tech, marketplaces, and (reportedly) the literal engineering leadership of the world’s dominant BitTorrent client. Under pressure the documented decisions were consistently right: Sweden-first, speed-as-religion, paying for mobile-free, refusing to window 1989. Held at 8 both times because retrospective team scores are the most hindsight-polluted number on any card, and this series caps the category.
Moat — 1, then 2. In 2009: non-exclusive licenses, a copyable client, a P2P cost edge that was temporary by design — 1. By 2015, scale (double Apple), data and playlists, and the sheer capital required to replicate the free tier’s losses gave real but shallow protection; run VRIO and the free tier itself was Valuable and (still) Rare — the labels had granted it to nobody else at scale and competitors refused its costs — but not Inimitable by a platform owner willing to burn money. Meanwhile 88% of the product’s raw material belonged to four suppliers with MFN clauses. A moat that your suppliers can drain on renewal is a 2.
Reading the gap
47, then 76. On this series’ bands, the 2008–09 Spotify scores as a trap — below 55, carried by team and a novel channel, sunk by unit economics that weren’t merely unproven but structurally negative and owned by counterparties. The honest reading of the early snapshot is that a rational institutional investor passes, and the record suggests most did: the funding that mattered came from the founders’ own pockets and a reportedly unconventional syndicate (a Hong Kong billionaire’s fund, Nordic seed firms) rather than from the Sand Hill Road consensus.
The 76 lands in promising, with one major risk — and the card names the risk without ambiguity: unit economics under absolute supplier power. That risk never fully retired; it was still the bear case in the F-1 and it is still the bear case today.
What changed between 47 and 76 is precisely enumerable, which is the point of scoring twice: (1) paying behavior materialized — 15% conversion by early 2011, 31% by 2015; (2) the distribution channel became owned and measured — the funnel instrumentation of February 2014; (3) mobile-free removed the cap on the funnel’s mouth; (4) the market’s why-now arrived on schedule; (5) the suppliers, holding equity, kept renewing. Nothing about the team changed. The evidence changed.
Kill criteria — stated as of 2009, checked against history
The five pieces of evidence that would have made a rational investor walk away, and what actually happened:
- If free-tier economics never close, the company bleeds out before conversion matters. FIRED, for years — ad-supported cost of revenue exceeded its revenue every filed year until 2017. The company outran it with capital and renegotiation; this criterion nearly killed the company and merely didn’t finish.
- If any major pulls or refuses renewal, the product breaks overnight (four suppliers, ~88% of streams, “must have”). Partially fired — the US majors withheld the market entirely for almost three years; an artist-level pull (Swift, 2014) tested the mechanism and the funnel held. A label-level pull never came; equity and advances are the likeliest reason why.
- If free-to-paid conversion stalls in the low single digits, the free tier is a charity for the labels. Never fired — reportedly 15% by March 2011, ~31% by end-2015, ~45% by end-2017.
- If a platform owner launches before Spotify reaches scale, defaults beat products. Fired late and survivably — Apple moved in June 2015, seven years after launch, against a company that finished that year with 28 million subscribers and falling churn.
- If the labels convert their leverage into terms that cap gross margin permanently, there is no business here even at scale. The live one. Partially fired continuously — 12–14% gross margins through 2016 — and only loosened (not resolved) by the 2017 growth-contingent renegotiation.
Every crisis in Part V appears on this list, stated in advance from 2009 information. That is what the scorecard is for.
Hidden Forces Nobody Talks About
The Swedish state built the launchpad, twice. A tax-subsidized PC in a quarter of households, then world-leading broadband — and, as the unintended second-order effect, the world’s most sophisticated piracy culture, which functioned as free market education. Sweden’s government spent four billion kronor teaching Swedes to expect music through a wire; Spotify collected the graduates. The reform was still on the books when Spotify was founded in 2006. No US startup had this environment, which is part of why no US startup built this company.
Users were the data center. For its capital-starved first six years, roughly nine of every ten bytes Spotify served came from somewhere other than Spotify’s servers — 55.4% from listeners’ own disk caches, 35.8% from other listeners’ machines, 8.8% from the company, per its engineers’ published measurements. The user base involuntarily subsidized the infrastructure of the product that was converting them. The P2P layer came down in April 2014, quietly, exactly when the company could afford to replace its users’ bandwidth with its own.
The equity made the suppliers into shareholders of the disruptor. The confirmed fact is that the majors took ownership stakes and advances in the original deals; the reported detail is ~18% for a nominal sum in October 2008. Either way, the strategic consequence is under-discussed: every time a label weighed killing the free tier, it was weighing an asset on its own balance sheet. The parties with the contractual power to kill Spotify had a financial interest in not using it. Contrast the one famous catalog pull: it came from an artist — the one class of participant holding no equity and receiving, per the 2015 French study, 11% of the payout pool.
There was never a market for the raw material — and the regulator said so. The CRB’s “must have” finding formalizes the structure: an interactive service needs all four majors, so no competitive supplier dynamics ever emerge, so streaming margins are set by negotiation with a de facto cartel, forever. This single structural fact explains the advances, the MFN clauses, the 116%-cost free tier, the 12% gross margins, the seven-year profitless scale-up, and why moat-building before scale was arithmetically impossible. Most Spotify commentary treats thin margins as a bug in the company. Porter would call them a feature of the industry that the company simply chose to inhabit anyway.
The piracy DNA was literal, and the company later fought the people who documented it. The academic Spotify Teardown team — whose publisher’s description calls the company’s origin “a partly illicit enterprise that grew out of the Swedish file-sharing community,” and whose research reported early beta versions serving unlicensed MP3 files — found Spotify representatives openly acknowledging the file-sharing links in 2013, and then found Spotify contacting the Swedish Research Council in an attempt to threaten their funding when the research methods became known. The Council’s own lawyer reportedly called the pressure unprecedented. A company built on out-engineering pirates, staffed (reportedly) by uTorrent’s creator, using BitTorrent-style peer discovery, spent its maturity policing the story of where it came from. The origin wasn’t shameful. It was the edge.
The Luck Audit
Specific breaks, and the skill that made each exploitable — no “right place, right time.”
Lucky: the suppliers were desperate on schedule. Spotify negotiated its founding deals with an industry nine years into a ~30% global revenue collapse, against counterparties like Warner losing money in six of seven years. That desperation — cash scarce, hope scarcer — is why catalogs plus equity could be had by a Swedish startup at all, and reportedly for a nominal share price. Five years earlier, the labels were suing Napster into the ground and feeling fine; five years later, iTunes-era digital revenue might have restored just enough confidence to say no. The skill: Ek spent two years of documented “cajoling” and — crucially — structured the deals so the labels won in every scenario: advances if Spotify failed, equity if it succeeded. He didn’t overcome their incentives. He rebuilt them.
Lucky: Sweden was a free laboratory. Spotify’s home market was one where the labels’ downside from a free-music experiment was approximately zero — piracy had already taken the revenue — and where broadband and PC penetration were, by published counts, years ahead of the US. The counterfactual was zero, so the experiment was cheap to permit. The skill: launching where the argument was unbeatable, proving conversion there, and then spending that proof (plus up to $42.5 million in Sony advances and their MFN cousins) to buy entry into markets where the argument was weak.
Lucky: the giants stood still for seven years. Apple kept selling downloads until June 2015. Pandora stayed radio. Rdio — built by the Kazaa founders, funded with a reported nine figures — refused the free tier until too late. None of this was Spotify’s doing; a 2010 Apple streaming service with a free tier probably ends this story. The skill: using the window at maximum aggression — US entry 2011, mobile-free 2013, funnel instrumentation 2014 — so that when Apple finally moved, the F-1 could claim nearly double its scale, and churn was already falling.
Lucky: the smartphone wave crested exactly under the model. Streaming access beats download ownership only in a world of always-connected pocket computers; that world arrived, on someone else’s capex, in precisely Spotify’s growth years. The skill: recognizing that mobile was the funnel’s ceiling and paying the licensing price to make free mobile happen in December 2013 — after which listening reportedly tripled on mobile and the tracked funnel began filling.
The skill that needed no luck: the 265-millisecond religion. The engineering constraint that no cost optimization could touch latency is the one fully endogenous piece of the story — measured, published, and the entire consumer-facing basis of “better than piracy.” Luck opened every window above. A quarter-second product is what climbed through.
What This Actually Means
I’ll resist the founder-lessons ending; most of this configuration is not replicable on demand. But four patterns are worth holding.
Freemium here was not a pricing strategy; it was a distribution machine purchased at negative gross margin. Spotify’s free tier lost money on its own revenue for nearly a decade — and delivered more than 60% of all paying customers at almost no marketing cost. The P&L booked the funnel as a royalty problem; the cohort data revealed it as the cheapest acquisition channel in consumer subscription history. If you evaluate businesses only on current-period margins, you would have killed this company every single year until 2017. That is a genuine limitation of margin-first analysis, and Spotify is its canonical exhibit.
You can build a very large company inside someone else’s market power — but the card will show it forever. The scorecard’s unit-economics category never healed past an 8 in this piece, and that’s not a scoring quirk; it is the CRB’s “must have” finding expressed as a number. Spotify proved that absolute supplier power caps your margin, not necessarily your scale. Whether that trade was worth the reported ~$28.5 billion of listing-day market value is a question the labels — holding both the royalties and, reportedly, the stakes — never had to ask.
Convenience beat free — but measure the claim. Spotify out-competed a price of zero on speed, reliability, and zero behavior change, which is the strongest evidence anywhere that convenience is a price consumers pay. And yet the world’s biggest torrent site kept its crown until 2014, and the industry recovered to a fraction of its 1999 peak. Spotify didn’t kill piracy. It made piracy pointless for people whose time was worth more than $10 a month — and built the machine that converted exactly those people.
Environment compounds before companies do. State PC subsidies → broadband → piracy culture → engineering talent that had built the pirate tools → a market where labels had nothing left to lose. Every layer preceded the company; the founders’ contribution was recognizing the stack and building the one product that could only be built on top of it. The honest accounting holds both: nobody else could have gotten this lucky, and almost nobody else could have used it.
Sources and Notes
Primary sources:
- Spotify Technology S.A., Form F-1 registration statement, SEC EDGAR, filed February 28, 2018. Source of: 159M MAU / 71M Premium / 61 countries and territories (Dec 31, 2017); 46% and 29% YoY growth; the “funnel” disclosure (>60% of gross added Premium subscribers since February 2014); quarterly Premium churn 7.5%/6.0%/5.1%; revenue €1,940M/€2,952M/€4,090M and net losses €230M/€539M/€1,235M (2015–17); >€8B cumulative royalties through Dec 31, 2017; the $23.8B (1999) → $16.9B (2008) industry decline; founder biographies; the December 27, 2006 Luxembourg incorporation; ad-supported cost of revenue 116% (2015) → 112% (2016) → 90% (2017); the 2017 label renegotiations.
- Spotify Technology S.A., confidential draft registration statement (2017), SEC EDGAR. Source of: 91M/28M (2015) and 125M/48M (2016) user metrics; annual churn 7.7%/6.6%; ARPU €6.84/€6.20; >€7B royalties through September 30, 2017; ~88% of 2016 streams from Universal, Sony, Warner, and Merlin; “only a small percentage” of Premium adds via paid acquisition; nine consecutive years of industry decline and the $14.3B (2014) trough.
- Gunnar Kreitz & Fredrik Niemelä, “Spotify — Large Scale, Low Latency, P2P Music-on-Demand Streaming,” IEEE P2P 2010. Source of: 265ms median latency, <1% stutter, the 8.8%/35.8%/55.4% delivery split, the P2P design constraints, and the 7M-user/8M-track mid-2010 anchor.
- NBER Working Paper w33048, “Platform Power Struggle: Spotify and the Major Record Labels.” Source of: the confirmed equity-plus-advances deal structure, the CRB “must have” finding, Warner’s 2005–11 losses, and the RIAA-based US revenue collapse figures.
- The leaked 2011 Sony Music–Spotify US contract (surfaced by The Verge, May 2015; republished by Digital Music News), with analysis by Rolling Stone. Treated as authentic per the multiple independent copies reported after the takedown.
- Eriksson, Fleischer, Johansson, Snickars & Vonderau, Spotify Teardown: Inside the Black Box of Streaming Music (MIT Press, 2019), plus Rolling Stone’s reporting on the book and the research-funding pressure episode.
- Contemporaneous journalism: TIME (November 2014, the Swift dispute figures); TechCrunch (April 2014, the P2P shutdown); Billboard (June 2015, the Apple Music response); Music Business Worldwide (May 2018, the Luxembourg registry equity reporting); Taipei Times and the Swedish Internet Foundation’s Internetmuseum (the home-PC reform).
Disputed or single-source details (hedged in text):
- The ~18% combined label/Merlin equity stake and the €8,804.40 nominal price: Music Business Worldwide’s reading of Luxembourg registry filings; per-label percentages vary slightly across accounts, and no principal has confirmed the total. Presented throughout as reported, not established. The confirmed floor (per NBER) is that majors took equity and advances.
- Ek’s uTorrent CEO stint and Ludvig Strigeus joining Spotify: secondary profiles only; hedged as “reportedly.”
- Lorentzon’s ~$70M TradeDoubler proceeds and €1M initial funding; Advertigo’s ~SEK 10M sale price: single-sourced (Wikipedia/MBW); hedged.
- The $21.6M Series A (Li Ka-shing, Creandum, Northzone, Horizons) and the October 7 launch date: blog-tier sourcing; hedged.
- June 2009 figures (~170,000 registered users; ~£82,000 monthly ad income): one academic-secondary source; hedged.
- Sean Parker’s ~$15M Founders Fund investment and personal role in Warner/Universal talks: secondary accounts; hedged.
- March 2011 (6.67M users / 1M subscribers), July 14 US launch date, 3M US users in year one, fall-2009 mobile app, mobile listening tripling 2013–14, ~80% of subscribers starting as free users: growth-marketing accounts; all hedged as “reportedly.”
- Rdio’s ~$125.7M raised and ~$75M Pandora asset sale: blog-tier post-mortems; hedged.
- Sweden’s 2005 broadband comparison (28 vs 17 per 100): single published count; hedged.
- The Swift payout figures: Big Machine’s $496,044 (domestic, label receipts) and Spotify’s $2M (global, all rights) measure different scopes and may both be accurate; Ek’s $6M forward projection and $2B cumulative-payout statement are the company’s own contemporaneous claims.
- Pirate Bay user counts and the post-raid growth figures: Wikipedia’s aggregation of period reporting; hedged.
- “Early betas served unlicensed MP3s”: the Spotify Teardown researchers’ account as characterized in press coverage of the book; attributed, not independently verified here.
- Two filing footnotes for precision: a March 2018 F-1 amendment revised end-2017 MAUs from 159M to 157M after excluding ~2M ad-suppressing accounts; and the €1,235M 2017 net loss was mostly non-operating finance costs on convertible notes (operating loss: €378M).
Analytical frameworks:
- The weighted, gated evidence scorecard (gates → weighted categories /100 → two snapshots → non-compensatory rule → kill criteria) is this series’ standard judging instrument, described in full on the series hub; team is capped at 10 per the hindsight-pollution rationale there (after Gompers, Gornall, Kaplan & Strebulaev, JFE 2020).
- Porter’s Five Forces and Rogers’ Diffusion of Innovations (1962) are used as thinking tools — the first for supplier power, the second for adoption speed — and explicitly not as verdict-renderers.
- VRIO (Barney, Journal of Management, 1991) appears in the moat scoring.
Characterizations based on inference rather than direct evidence are flagged in the text with “reportedly,” “accounts place,” or similar qualifications.