In the early 2000s, after Napster had blown up the old model for buying and selling music, an idea took hold. It was repeated often enough that it became a cliché: people could be tempted away from piracy, the argument went, only if it became significantly easier to access music legally than to download it illegally. In the event, it is Spotify – with more than 500 million users, 200 million of whom pay for a monthly subscription – which has been primarily responsible for making this happen. It’s both surprising and unsurprising, then, that the company started up in a country where music piracy was more prevalent than anywhere else in Europe. In the mid-2000s, it was estimated that 1.2 million people in Sweden – out of a population of nine million – had shared pirated files. Three file-sharing services, Kazaa, μTorrent and the Pirate Bay, had Swedish roots. For at least some of the people involved, piracy went along with an ideological commitment to the free transmission of information: the Pirate Bay was set up in 2003 by Piratbyrån (the Bureau of Piracy), a Swedish collective critical of copyright law. Sweden’s Pirate Party was founded three years later, on a platform of providing ‘access to free communication, culture and knowledge’. It won 7 per cent of Swedish votes in the 2009 European elections.
When Daniel Ek and Martin Lorentzon founded Spotify in 2006, they were adamant that the company had to offer a free service. Neither of them had worked in music before. Ek was a 23-year-old programmer from Rågsved, a suburb of Stockholm. He had already built an online advertising firm, Advertigo, which he had sold to Tradedoubler, a digital marketing company co-founded by Lorentzon. Their background in advertising wasn’t irrelevant: if people were going to use Spotify without paying anything, its free service would have to be subsidised by ads. Making Spotify available for free was one of the ways the company tried to compete with illegal downloading; another was borrowing the technology that facilitated piracy. Within months of setting up Spotify, Ek and Lorentzon bought μTorrent and hired its founder, Ludvig Strigeus, as lead developer. The first wave of file-sharing networks, like Napster, had used peer-to-peer technology to connect someone who wanted to download a song with someone who already had the song on their computer. When you use torrenting software, by contrast, your computer takes bits of a song from lots of other computers on a network and downloads them all at the same time. This process is more reliable – and, if the parts are shared between enough people with fast internet connections, quicker – than downloading from a single source.
Speed was important to Ek. He wanted a song to play instantly, or what felt like instantly. That meant it had to come on within 285 milliseconds: any longer and, according to a study one of his engineers had read, the time lag would be noticeable. One of the ways Spotify achieved this was by downloading parts of a song to users’ computers when they played it for the first time. Another person who wanted to listen to the song would then be able to download it by torrenting: the torrents were divided in such a way that the song could start playing before it had been fully downloaded. The overlap with illegal file-sharing didn’t end there. Spotify was made available to thousands of invited users before the company had secured the licences to the music it was streaming. The songs available on the site early on came from employees’ own collections, and many of them had been obtained illegally. As the five co-authors of Spotify Teardown, a playfully critical study, put it, Spotify ‘began as a de facto pirate service’.It wasn’t until three years into the company’s existence that the last of the pirated songs were deleted; five years after that, Spotify abandoned torrents and began hosting all its music on central servers.
Spotify launched officially in Sweden in October 2008, and in the UK in February 2009. I installed the app a few weeks later. You were given the choice of a ‘premium’ subscription that allowed you to listen to music uninterrupted for £9.99 a month, or a free account, which played adverts every half hour or so. I was in my second year at university so, like everyone else I knew, I chose the free version. Record labels weren’t so keen on the free thing. It wasn’t just that royalties were smaller when someone listened on a free account. The labels were worried that, like illegal file-sharing, Spotify would make people think music was something they no longer needed to pay for – whether that meant paying for streaming or downloads, or buying CDs. That year, the global recorded music industry made $16 billion, down from a peak of $24 billion in 1999 (the year Napster launched). The situation would get even worse: by 2014, revenue had fallen to $14 billion. The music industry seemed to be dying, if it wasn’t dead already. But then things started to pick up. In 2021, revenue reached a record high of $25.9 billion (though in real terms this was only 60 per cent of what it had been in 1999), two-thirds of which came from streaming. This wasn’t all down to Spotify: it has never launched in China, where Tencent’s three streaming services – QQ Music, KuGou and Kuwo – have more than 500 million users between them. (Since 2017 Spotify has owned 9 per cent of Tencent Music, and Tencent 9 per cent of Spotify.) But as the picture started to look rosier, newspaper articles claimed that streaming, and Spotify in particular, had saved the music industry.
Whether or not you share that view depends on what you understand the music industry to be. Two-thirds of all the music that’s streamed is controlled by the ‘Big Three’ record labels: Universal, Sony and Warner. They’ve done very well out of streaming: Universal made $4.8 billion from it in 2021. But as the authors of Spotify Teardown point out, Spotify’s ‘very existence remains dependent’ on the Big Three’s willingness to keep making their music available. This has allowed them to dictate their own terms: one of the things they got out of their early negotiations with Spotify was equity, their initial stake amounting to almost a fifth of the company. Smaller labels and distributors, however, aren’t in a position to make serious demands, either for themselves or for the musicians they represent. So it’s strangely difficult to answer the most important question about Spotify’s business model: how much does it pay artists? In The Spotify Play, their account of the company’s rise, Sven Carlsson and Jonas Leijonhufvud describe the way the company’s early deals with the record labels were ‘so complicated’ that it was impossible to provide ‘a neat figure’ for how much artists made per stream.
Another reason it’s difficult to come up with a figure is that royalty payments aren’t simply calculated according to how many times an artist’s music has been played. It’s a question, rather, of how many times it has been played relative to everyone else’s music: Spotify sets a sum to cover all the royalties paid each month, and each artist’s share of that pot is calculated according to their share of streams. The figure varies, but the average payment per stream is around £0.003, or 0.3p. That money doesn’t go straight to the artist. Most of it goes to whoever holds the right to the master, as the recording of a song is known; a smaller share goes to whoever has the rights to the song itself, which covers both the music and the lyrics. The rights to a master are usually held by the artist’s label, which will take a cut: the Big Three take between 70 and 80 per cent of the royalties for the master (the less established the artist, the more they take); an independent label typically takes 50 per cent. Royalties for song rights are more standardised: the publisher takes around a third of the amount and the songwriters get the rest. (The biggest publishers are the publishing wings of the Big Three: Universal Music Publishing Group, Sony Music Publishing and Warner Chappell Music.) All this means that an artist on a major label who writes their own music might actually receive as little as £0.001, or 0.1p, each time one of their songs is played. In other words, to earn one pound in royalties, one of their songs has to be played a thousand times.
Still, the numbers can add up. The two most played songs on Spotify, The Weeknd’s ‘Blinding Lights’ and Ed Sheeran’s ‘Shape of You’, have each been played more than 3.4 billion times; ten of Sheeran’s other songs have each been played at least a billion times. It was estimated last year that his total earnings from Spotify up to that point came to more than $80 million. Even Taylor Swift, once Spotify’s most high-profile critic, has made her peace with it. Swift removed her music from the platform in 2014 – ‘Valuable things should be paid for,’ she said – but relented in 2017. By last year, Spotify had paid her an estimated $70 million in royalties. For many less well-known musicians, however, Spotify royalty cheques are barely worth the paper they’re printed on. Che Chen, the guitarist in the New York experimental band 75 Dollar Bill, told the music website Pitchfork that one of their songs earned a monthly royalty of 20 cents. ‘We might make $100 a year from streaming,’ he said. By contrast, when the band put a live album on the platform Bandcamp – a kind of anti-Spotify, where artists and labels upload music and sell it directly to listeners – they made $4200 in just two days.
The biggest artists have, of course, always earned more from their music than anyone else. But in the vinyl and CD eras, underground bands were effectively operating in a separate economy from pop stars. It didn’t matter that more people listened to pop: these bands had dedicated fans, who were more likely to spend money on an album than the typical music listener. What’s new about Spotify’s payment model is that it makes music a zero-sum game, putting an artist like 75 Dollar Bill in direct competition with Ed Sheeran. The more streams the big artists get, the less everyone else makes. Taylor Swift decided it wasn’t worth keeping her music off Spotify, but for others it isn’t worth being on there. When I searched for 75 Dollar Bill recently, I saw that the band had removed all its albums from the platform. The British classical label Hyperion never put its music on there in the first place. ‘The Spotify model does not work for classical,’ Simon Perry, the label’s former director, told the BBC, ‘because as a proportion of listeners, there is not enough traffic for it to generate the sort of income a label needs to invest in a performer and recording.’
The shift from paying artists according to how many copies they’ve sold to paying them per stream might seem fairer, a reward for making music that people enjoy. (It’s not a completely new system: radio stations have always paid artists based on how often their songs are played. It’s just that these payments used to be dwarfed by the royalties from CDs and vinyl.) But how often someone listens to an album or song isn’t the only marker of how much they like it. The music I love isn’t necessarily the music I play most. Some things are too difficult or too intense for background listening; some things mean so much to me that I deliberately ration them. I don’t think this music is worth less because I play it less frequently. But Spotify directly equates appreciation with streams: it tells you when you’re in the top 1 per cent of an artist’s listeners, as if enthusiasm can be quantified.
For a song to count as having been streamed – and for an artist to qualify for a royalty – it must play for at least thirty seconds. If an artist can’t hold someone’s attention for those first thirty seconds, they don’t get paid. Some have adapted the way they go about making music accordingly. In 2010, less than 20 per cent of number one songs in the US had choruses that started within the first fifteen seconds; by 2018, almost 40 per cent did. Meanwhile, hits have got shorter: between 2013 and 2018, the length of the average song on the Billboard Hot 100 chart in the US fell from 3 minutes 50 seconds to 3 minutes 30 seconds. (The rise of TikTok, where only a segment of a song tends to be played, has intensified this trend.) Since an artist is paid the same amount irrespective of a song’s length, it makes sense to keep it short, in the hope that a listener will get through more songs. So as pop songs got shorter, albums got longer: between 2013 and 2018, the average duration of the most streamed albums on Spotify increased by almost ten minutes, to sixty minutes. (There have been some endearing efforts to exploit the thirty-second rule: in 2014, the American funk band Vulfpeck released Sleepify, an album consisting of ten silent songs that each lasted 31 or 32 seconds, in order to earn enough money to fund a tour. They made almost $20,000 before Spotify took the album down.)
Rather than trying to make Spotify work for them on its own terms, some artists have decided instead to take it on. In 2020, the Union of Musicians and Allied Workers launched the campaign Justice at Spotify, approaching the inequities of the streaming economy as a kind of workplace dispute. More than 28,000 artists have signed up to its list of demands, which include an end to payments based on an artist’s share of overall streams and the introduction of royalties worth at least one cent per stream. The irony is that Spotify’s biggest critics used to be the biggest artists, because they thought they had the most to lose. Yet a recent poll of professional musicians in the UK found that 82 per cent earn less than £200 a year from streaming; only 7 per cent earn more than £1000. ‘Music lovers’ might think they’re getting a ‘good deal’ from streaming, a report by the House of Commons Digital, Culture, Media and Sport Select Committee said in 2021, but there’s a danger that ‘some of the music they love may not be being made in ten years’ time.’
Daniel Ek originally conceived of Spotify as ‘the future of the record store’. People already knew what they wanted to hear; Spotify’s job was to help them find it. But in the early 2010s, the authors of Spotify Teardown write, as the likes of Apple and Amazon launched streaming services with equally large libraries, Spotify ‘began to transform itself’. It would no longer be ‘a simple distributor of music’. It was now ‘the producer of a unique service’: recommending music. Ek had come to believe that ‘the biggest unsolved question for most users is, how can you help me figure out what I’m going to listen to?’ So much music was coming out that it was hard not to feel overwhelmed. In the early 1970s, five thousand albums were released in the US each year; in 2013, almost 130,000 were released. Today the equivalent of three million albums’ worth of music is added to Spotify every year. (There are more than a hundred million songs on Spotify, millions of which have never been played; a website called Forgotify lets you stream them randomly.)
Over the last decade, Spotify has developed a range of methods to help listeners decide what they want to hear. When you open the app, the home page recommends a number of things – artists, albums, playlists – based on what you’ve listened to in the past. Many of these playlists are collections of songs put together by Spotify employees. They can be immensely popular: one, RapCaviar, has 15 million followers, giving it more influence in determining which rap breaks through than any radio station or TV channel. But there are also Spotify’s ‘made for you’ algorithmic playlists, with each user getting their own customised version based on their listening history. One that’s promoted heavily is Discover Weekly, a selection of thirty songs that a user hasn’t listened to before. Some of the customised playlists are dedicated to a specific mood or activity: ‘Running Mix’, even ‘Grief Mix’. (My Grief Mix features a surprising number of rap songs.)
When Spotify began introducing these features, it compared its role to that of a ‘new best friend’, offering ‘hand-picked recommendations’. This implies agency: you’re given a recommendation, and it’s up to you how you act on it. But that’s not how Spotify’s Autoplay function works. Once you get to the end of an album or selection of songs, Autoplay kicks in; the songs that follow are meant to be similar in some way to whatever you’ve just been listening to. (Some songs are more likely to follow than others: Spotify recently launched a service called Discovery Mode, which allows artists and labels to increase the likelihood of a song coming up on Autoplay in exchange for a reduced royalty rate.) Autoplay has been switched on by default since the start of 2017. To disable it, you have to go into Spotify’s settings and actively opt out. Otherwise, the songs will keep playing. A new feature called DJ, so far available only in North America, comes closer to simulating the experience of listening to the radio: an audio commentary accompanies the songs selected for you, using AI to provide facts about what’s playing.
Recommendations now ‘drive close to half of all users’ streams’, according to Spotify’s co-president Gustav Söderström. In Computing Taste, an ethnography of the data scientists and product managers working in ‘the world of music recommendation’, Nick Seaver gives an account of the way this sort of technology operates. The job of his interviewees, who tend to work for private companies hired by streaming services, is to help their clients ‘answer an apparently simple question: what’s next?’ There’s a tendency, when talking about a platform like Spotify or Netflix, to refer to ‘the algorithm’, as if a single formula determines what each person is recommended. But at Willow, one of the companies Seaver writes about (he refers to the companies and their employees by pseudonyms), there are ‘dozens and dozens’ of algorithms, tracking many different things: ‘What does a song sound like? What device is a user listening on? What has a user listened to in the past?’ These different bits of information are then ‘orchestrated together’ by another algorithm, which establishes how a user goes about listening to music: whether they like being introduced to new things, say, or prefer sticking to what they already know. ‘Every recommended song is a little test, a probe meant to fill out a picture of what a given user likes.’
But a Spotify user will like different things depending on the time of day, their mood, whether they’re cooking or running or filing their taxes. As Tom, a product manager at Whisper, tells Seaver, ‘One listener is really many listeners.’ Someone might enjoy listening to techno, and jazz, and R&B. But that doesn’t mean they want to hear a song that’s a blend of all three: an effective music recommendation system will know when to play them R&B, and when to play techno. How does an algorithm – or a set of algorithms – distinguish R&B from other kinds of music, or one kind of R&B from another? ‘It turns out,’ Seaver writes, ‘that training a computer to identify musical genres from their sound alone is very hard to do.’ Take Christian rock: ignore the lyrics and you might easily mistake it for non-Christian rock. Since music recommendation software can’t tell the difference between the two based on how they sound, it has to be able to recognise that Christian rock attracts a particular set of listeners. It can then recommend songs appropriate for a Christian rock fan, based on what like-minded users have been playing.
By analysing how something sounds and who listens to it, companies like Whisper divide music into different ‘clusters’. A cluster will often overlap with an existing genre, but sometimes it will refer to something new, a group of artists whose music shares certain characteristics without having emerged from a specific scene. The person at Spotify who gives these new clusters names is Glenn McDonald, the company’s ‘data alchemist’. McDonald has created a website, Every Noise at Once, that maps all of Spotify’s ‘genre-shaped distinctions’ – six thousand in total – on a single chart. Scrolling through it is like staring at a galaxy, with McDonald the astronomer responsible for naming each new star. Genres to the left are said to be denser and more atmospheric (cryptic black metal, epic black metal, Greek black metal); those to the right are spikier and bouncier (rave funk, hard minimal techno). Some of the genres listed are purely functional: ‘sleep’, ‘Pilates’, ‘pet calming’. Others are baffling: Spotify distinguishes between ‘small room’, ‘big room’, ‘deep big room’ and ‘escape room’. It’s fun to imagine what these might sound like, but there’s no need: click on any of the names and an example of the genre plays.
Browsing Every Noise at Once is a reminder of how narrow our experience of Spotify is. For me, the most exciting part of listening to music is falling in love with a song or album that sounds like nothing I’ve heard before. But people have complained to me that Spotify’s recommendations are overly on the nose: the problem isn’t that they don’t like what Spotify plays them, but that it’s too obviously the sort of music they might like. For all Spotify’s talk of ‘discovery’, the thing it really cares about is what Mike, the chief scientist at Willow, calls ‘the hang-around factor’. If someone skips a song, or stops listening altogether, then something has gone wrong. You mustn’t bore the listener, but you mustn’t startle them either.
If you search on Spotify for Pavement, the 1990s alternative rock band, the first song that comes up is ‘Harness Your Hopes’. It’s been played close to a hundred million times, almost three times more than any of the band’s other songs. For a long time it was an obscure B-side – so obscure that when Stephen Malkmus, Pavement’s frontman, heard it playing in a bakery a few years ago, he didn’t even recognise it as one of his own. According to a fascinating article by the journalist Nate Rogers, ‘Harness Your Hopes’ started to become popular in 2017, the year Autoplay was enabled on every Spotify account. There seems to be something about the song that made Spotify’s algorithms single it out. The American musician Damon Krukowski had a similar experience. ‘Strange’, by his former band Galaxie 500, wasn’t particularly popular when it was released in 1989, but it has become their most played song on Spotify. Writing on his blog, Krukowski pointed out that ‘Strange’ has ‘a more predictable song structure’ than most of Galaxie 500’s work, and asked whether Autoplay might be ‘separating out and rewarding … the most “normal” songs in each band’s catalogue’. What if, he worried, asking Spotify to play Galaxie 500 comes to mean ‘play the song by Galaxie 500 that most resembles songs by others’?
My own experiences of Autoplay have been underwhelming. The music that comes on isn’t dissimilar to whatever I’ve just been playing, but it does often seem a little blander. It’s as if Spotify has decided that the best way to hold my attention is by not demanding too much of it. Sometimes this is what listeners want. At one point, Willow noticed that users were switching off a particular playlist it had designed ‘at an unusually high rate’. The culprit was identified: a song with vocals, on an otherwise instrumental playlist. ‘What had happened, Mike guessed, was that the voice had brought the music into listeners’ conscious attention, spurring them to turn it off.’ Removing the track solved the problem. Now, once again, the playlist ‘faded back below the threshold of listener awareness’.
The terms we use for listening online – stream, torrent – seem to have anticipated this effect: if music is a tap that’s constantly flowing, sometimes we have to let it wash over us. This is what Spotify wants, or even expects. Spotify Wrapped, a slideshow generated for each user at the end of the year, breaks down what they have been listening to at different times: what they played in the morning, how they ‘seized the day’ and ‘embraced the night’. The assumption is that the app serves as a constant companion, soundtracking our different moods and experiences (listening to music doesn’t seem to count as an experience in itself). Gustav Söderström has said that he’d like users to ‘start Spotify in the morning and not really pause it until they go to sleep’.
As part of this project to monopolise our ears, Spotify has moved beyond music. Since 2019 it has made a series of expensive investments in podcasting, buying production companies such as Gimlet and the Ringer, striking deals to produce shows with the Obamas and Prince Harry and Meghan, and paying more than $200 million for the exclusive rights to the Joe Rogan Experience, sometimes described as the most popular podcast in the world. (It’s almost certainly the most controversial: Neil Young and Joni Mitchell have removed their music from Spotify in protest at Rogan’s statements about Covid.) Since this spending spree began, the company has started referring less to ‘music’ and more to ‘audio’. Its ‘key strategy’, Ek said in 2020, is now ‘around owning background moments, around owning, really, audio’.
It can sometimes seem that this strategy has succeeded. TikTok has superseded Spotify in certain respects, namely the ability to make a song a hit, but it’s Spotify, more than any other platform, which has become synonymous with the experience of listening. Nobody signs up to a video streaming service like Netflix expecting to be able to watch whatever they want – you know that you’re paying for a limited selection. But the premise of Spotify is that you don’t need anything else. The idea of abandoning it might therefore seem tantamount to abandoning music itself. But there is a case for doing so. Even by the standards of the streaming economy, Spotify gives artists a bad deal: Apple Music pays three times as much per stream, Tidal four times. Yet Spotify hasn’t had to cope with a backlash of the kind faced by Amazon or Uber – mostly, I think, because we tend not to recognise the labour that goes into making music. People who have quit Spotify speak of other, more immediate benefits. It’s easy to identify with the people interviewed by Liz Pelly, for a Guardian article last year, who had become unhappy with their relationship to music: the way they played it in order to have something on in the background, or prioritised listening to individual tracks at the expense of albums, or skipped a song if they didn’t immediately like it. They admit that swapping streaming for records and CDs and MP3s has taken effort. Yet they all say that they now hear music differently. The experience is more focused, more engaged. As one interviewee puts it, it’s ‘something I’m doing, rather than something I’m receiving’.
Considered in basic terms, the rise of Spotify makes for an impressive story: a Swedish upstart has managed to outdo some of the wealthiest companies in history. But the more it has grown, the less Swedish it has become. In 2016, Ek and Lorentzon wrote an open letter to the Swedish government calling for a cut in taxes on employee stock options and the abolition of rent controls in Stockholm. ‘If no changes are made,’ the letter read, ‘we will have to consider focusing our expansion on countries other than Sweden.’ The following year Spotify opened a new office in New York, which soon employed more people than its Stockholm headquarters. When it listed on the New York Stock Exchange in 2018, it was through a holding company based in Luxembourg.
On the day of the listing, Spotify shares were worth $149. As I write, they’re going for $133. The company is worth $23 billion, but it has never made a profit. (It doesn’t help that the cost of a subscription has remained frozen at £9.99; if it had risen in line with inflation, it would now be £15.) Spotify wants us to keep listening, but the more we listen the more it has to pay out in royalties. It’s not clear that it can ever find a way to reconcile these two things. Unlike Amazon and Apple it can’t subsidise music streaming with money from more profitable parts of its business, or use it to sell phones and smart speakers. It’s not the only tech company struggling: many of them have been cutting jobs, and in January, Spotify announced that it was laying off 6 per cent of its employees. But Ek’s ambitions aren’t limited to streaming. In 2020, he committed to investing €1 billion in European startups seeking to ‘tackle the infinitely complex problems that our society is dealing with’. The most recent of these investments was €100 million in Helsing, ‘a new type of defence and artificial intelligence company’. Its goal? Bringing ‘software-based capabilities to the armed forces’.
Send Letters To:
London Review of Books,
28 Little Russell Street
London, WC1A 2HN
Please include name, address, and a telephone number.