The Library in the Air

On Friday, 29 September 2023, Netflix mailed its last red envelopes. The company that had sent its first DVD, a copy of Beetlejuice, in 1998 had shipped more than 5.2 billion discs in twenty-five years. In the final weeks it sent some loyal members up to ten extra films, and it told everyone who still had discs at home that they could keep them.¹ For a few thousand people, the last thing Netflix ever gave them was something they owned. Everything else the company offered, by then to nearly 250 million households, was a licence to watch for as long as they paid and for as long as the title stayed in the catalogue.

That exchange, owning a few things for access to almost everything, is now how most of the world gets its recorded culture. In 2025 streaming brought in more than $22 billion for the recorded music industry, 69.6 percent of its worldwide revenue, from 837 million paid subscription accounts and many more free listeners.² Spotify alone had 290 million paying subscribers and 751 million monthly users at the end of that year, and said it had paid more than $11 billion to the music industry.³ Netflix took in $45.2 billion in revenue in 2025, and the studios that once licensed it their films now run rival services or, in Warner Bros.’ case, have been sold.⁴

This essay follows how that happened, from music piped down telephone wires and radio stations that streamed a baseball game through a modem, to a library that seems to float in the air above every phone and television. It follows the decisions that gave streaming its shape: Napster’s index and the labels’ lawsuit, Apple’s 99-cent song, YouTube’s open door, Netflix’s free trial for a whole industry, Spotify’s free tier, the season released all at once, the studios’ retreat into their own services and the arrival of music no one played. Its deeper subject is what changes when a culture moves from things people keep to a stream they rent: what they value, how they spend and remember their time, and how the money they pay finds its way, or fails to find its way, to the people whose work they hear.

Before streaming: the schedule and the shelf

For most of the twentieth century, recorded culture reached people in one of two ways. The first was the schedule. Radio, and then television, sent programmes to everyone at once, and listeners took what was on when it was on. Someone else chose the sequence, and the audience shared it. The second was the shelf. Records, then cassettes, compact discs, videotapes and DVDs, were objects people bought, kept, lent and sold, and played whenever they wished.

Each form had its economics. Broadcasting was paid for by advertisers, by licence fees or by the state, because no one could charge each listener for a signal in the air. The shelf was paid for by sales: each copy had a price, and the price paid for the making of the next record. The schedule offered abundance without choice; the shelf offered choice within the limits of what a person could afford and store. A record shop might stock tens of thousands of titles, but a household owned a few hundred.

There was already a third, quieter form. In the 1920s an American Army general, George Owen Squier, patented ways of sending programmes along electrical wires, and a company built on his ideas, Wired Radio, piped music into businesses. In 1934 it took a new name, Muzak, reportedly a blend of “music” and “Kodak.”⁵ Muzak was music delivered as a service rather than sold as an object, and chosen for its effect rather than for its own sake: to make shoppers linger or workers keep pace. Few people thought of it as culture. It was the first subscription to sound.

In 1974 the Welsh critic Raymond Williams, watching British and American television for his book Television: Technology and Cultural Form, noticed something about the schedule that its makers rarely said aloud. Broadcasters did not really offer a series of separate programmes. They offered what he called planned flow: an evening arranged so that one thing led into the next, with trailers and announcements smoothing the joins, so that viewers kept watching the channel rather than choosing each programme.⁶ The unit of television, he argued, was the flow, not the show. Streaming would at first seem to dissolve the flow, giving each viewer a library to choose from. The history that follows suggests it rebuilt the flow in a more personal form.

The first streams: RealAudio and the narrow pipe

The word streaming describes a technical choice. A downloaded file must arrive in full before it plays. A stream begins playing as soon as enough of it has arrived, and plays while the rest keeps coming. The listener never holds a complete copy, which made streaming attractive to rights holders long before it was attractive to audiences.

In April 1995 Rob Glaser’s Seattle company, Progressive Networks, introduced RealAudio at the annual convention of the National Association of Broadcasters, and on 5 September that year it streamed a Seattle Mariners game against the New York Yankees live over the internet, one of the first major sporting events heard that way.⁷ Most listeners connected through dial-up modems. The sound was thin, and the stream stopped whenever the connection faltered while the player waited for more data, a pause that came to be called buffering. Progressive Networks, renamed RealNetworks, built a business selling servers to broadcasters and giving away players to listeners, and fought Microsoft for the format of internet audio and video through the late 1990s.

These early streams carried broadcasts, not libraries. They extended radio’s schedule to the computer and inherited its limits. The pipe was too narrow and the catalogue too thin for anyone to imagine abandoning the shelf. What changed that was not a streaming service at all, but a way of copying files.

Napster and the end of scarcity

By the late 1990s three things had come together. The MP3 format could compress a song to a tenth of its size on a CD with little audible loss. Hard drives had become large enough to hold thousands of songs. University networks and early broadband made sending a song a matter of minutes. Millions of people had libraries of music on their computers. The question was how to find what someone else had.

Napster, released in 1999, answered it with a design that became the essay’s first consequential decision. The songs stayed on users’ computers and passed directly from one user to another. But Napster ran central servers that kept an index of what every connected user had, so a search for any song returned an instant list of people offering it. The index made the service fast and simple. In February 2001, according to Media Metrix, about 26.4 million people in fourteen countries used it at home.⁸ For a generation of listeners it was the first experience of having almost all recorded music available at once, for nothing.

The record labels sued. On 12 February 2001 the US Court of Appeals for the Ninth Circuit, in A&M Records v. Napster, held that Napster was likely liable for its users’ infringement. Its users were not making fair use, the court found, and Napster had both actual knowledge of the infringement and the ability to supervise it. The central index was the crux: Napster knew what was being shared and could have stopped it.⁹ Under the injunction that followed, the free service could not continue.

There had been another path. On 31 October 2000 Bertelsmann, the German company that owned the BMG labels, announced an alliance with Napster. It would lend the company money to build a membership service that paid artists, labels and publishers, would make BMG’s catalogue available and would drop its lawsuit once the service worked. It urged the other labels to join.¹⁰ They did not. Universal, Sony, Warner and EMI continued to sue, and the paid Napster never launched in the form planned.

The labels’ decision is easy to judge harshly, and Hindsight and the Illusion of Inevitability explains why that judgement should be resisted. At the time they were enjoying the best years in the history of recorded music. US revenue peaked in 1999 at $14.6 billion, much of it from compact discs that cost little to make and sold at high prices.¹¹ Licensing a service built on unauthorised copying looked like rewarding the theft of their product, and the five majors had no agreed view of what a song online should cost. Their legal case was strong, and they won it.

But the problem they faced was not Napster. It was that listeners had discovered they could have any song, at once, and would not unlearn it. When Napster closed, sharing moved to services with no central index: Kazaa, LimeWire, Grokster and BitTorrent. The Supreme Court held unanimously in 2005, in MGM Studios v. Grokster, that a company that actively induces infringement can be liable even without a central server.¹² That closed some services and changed none of the behaviour. US recorded music revenue fell for fifteen years, to about $7 billion in 2014.¹¹ The lesson of Napster’s architecture was absorbed by the pirates, who built systems with no one to sue; the lesson of its popularity took the industry a decade to act on.

Ninety-nine cents: the song becomes the unit

The first legal answer that people adopted came from a company that sold computers. On 28 April 2003 Apple opened the iTunes Music Store, offering more than 200,000 songs from the five major labels at 99 cents each, with no subscription.¹³ It sold more than a million songs in its first week, although accounts differ on whether that took seven days or five. A Windows version followed in October.

The labels had already launched their own services, and they had failed: they were sold by subscription, restricted what users could do with the files and each offered only some labels’ music. Apple’s store was easy to use, covered all five majors and sold songs one at a time at a single price. The labels insisted on copy protection, and Apple supplied its own, FairPlay, which let songs play only in iTunes and on iPods. In 2007 Steve Jobs argued publicly that the protection should be abandoned altogether, a story told in this site’s commentary on his open letter, Thoughts on Music. Within two years the labels allowed unprotected sales.

The 99-cent song changed music’s unit of value. The album had been a bundle, a dozen songs sold together for the price of the one or two people wanted most. iTunes let people buy the one or two. For listeners this was a gain. For artists whose albums sold on the strength of a single, it was a loss, and it foreshadowed a later change in which the unit would shrink further, from the song bought to the song played.

Anyone can upload: YouTube and the safe harbour

Video took a different route. YouTube, founded in 2005, let anyone upload a video and share it. Much of what people uploaded was their own: home videos, commentary, early vloggers. Much of it was not: clips of television programmes, music videos and films. The 1998 Digital Millennium Copyright Act gave platforms a safe harbour from liability for what users posted, provided they removed infringing material promptly when the owner sent a notice. YouTube relied on that harbour. It removed videos when asked and did not screen them in advance.

The growth was immense. On 9 October 2006 Google agreed to buy YouTube for $1.65 billion in stock, although the company had fewer than seventy employees, no profit and serious copyright claims pending.¹⁴ In March 2007 Viacom, owner of MTV and Comedy Central, sued YouTube and Google for $1 billion, alleging what it called brazen infringement.¹⁵

In October 2007 YouTube introduced a system it called Video Identification, later known as Content ID. Owners supplied reference copies of their work. The system compared each upload against them and let the owner choose to block a match, track it or run advertising against it and take a share of the revenue.¹⁶ This turned a legal dispute into a market. A studio that found its clip on YouTube no longer had to choose between suing and ignoring it; it could be paid. Viacom pursued its case for seven years, lost twice at trial level and settled in March 2014 with no reported payment.¹⁵

The decision to stay open and build automated matching shaped the web’s video culture. It allowed remix, commentary and amateur performance to flourish on a scale no licensing system could have arranged in advance. It also put the judgement of what counted as fair use largely in the hands of an algorithm tuned to owners’ claims, and taught a generation of creators that the platform, not the law, was the effective referee. YouTube’s later role as a music service, and its recommendation system, which What You Might Also Like follows in detail, both rest on the bargain struck in 2007.

Netflix: from envelopes to the stream

Netflix began in 1998 by renting DVDs by mail, and soon by monthly subscription. By the mid-2000s it had a large, profitable business and a problem it could see coming: discs would decline as broadband spread. On 16 January 2007 it began to roll out a feature called Watch Now. Subscribers could stream films and series on their computers at no extra charge, with monthly allowances of six to forty-eight hours depending on their plan. The catalogue was tiny, about 1,000 titles against some 70,000 on disc, and Disney was absent.¹⁷

The decision to give streaming away, rather than sell it separately, deserves attention, because it is the essay’s clearest example of a company cannibalising itself without hurting itself, the trap described in Why Good Management Fails avoided by a small, free experiment. A separate streaming product with 1,000 titles would have been a poor bargain and would have failed. As a free extra, it let millions of subscribers try streaming with no risk and no decision, and taught them a habit. Netflix then licensed more catalogue, brought its player to televisions and other devices, and by the fall of 2011, according to the network equipment company Sandvine, accounted for 32.7 percent of peak downstream internet traffic in North America.¹⁸

The second half of the transition was harder. In July 2011 Netflix separated its DVD and streaming plans, raising the price of the combined service by up to 60 percent, and in September announced that the DVD business would be renamed Qwikster. Customers revolted. Netflix reversed the split within weeks and lost about a million more subscribers than it had expected, a sequence examined in this site’s note on Reed Hastings’s Qwikster letter. The DVD service continued, shrinking, until the last envelopes in 2023.

Spotify: the free tier as an argument

Spotify was founded in Stockholm in April 2006 by Daniel Ek and Martin Lorentzon.¹⁹ File sharing was widespread in Sweden, the home of The Pirate Bay. Spotify’s founding argument was that piracy could be beaten only by something better than piracy: faster, with the whole catalogue, and free. Gabe Newell, co-founder of the game company Valve, made the same point about games in 2011, saying that piracy was a problem of service rather than price.²⁰

Free was the hard part. A service could not offer every song without licences from the major labels, and the labels had spent a decade fighting free music. Spotify’s answer, when it launched in October 2008 in several European countries, was two tiers. Listeners could use a free version with advertisements and some restrictions, or pay a monthly fee for an advertising-free premium version. The free tier was a funnel: its purpose was to make people want to pay.

The price of the labels’ consent was ownership and guaranteed money. In 2009 a report said the four major labels had together acquired about 18 percent of Spotify’s shares, which the company had previously denied; later estimates put the stake at 15 to 20 percent.²¹ In 2015 The Verge published a leaked contract between Spotify and Sony Music, signed around the time of Spotify’s US launch. It promised Sony up to $42.5 million in advances over two years with an optional third, and included a most-favoured-nation clause that let Sony demand any better terms another label received.²² The labels, in other words, were paid in advance, paid as shareholders and paid per stream, and the structure protected the largest catalogues first.

Did the free tier work? The most careful early study, by the economists Luis Aguiar and Joel Waldfogel, examined music consumption between 2013 and 2015. They found that streaming displaced music sales and also displaced piracy, and that for the industry as a whole the losses in sales were roughly offset by the new streaming revenue: streaming was close to revenue-neutral in those years.²³ That was not the triumph Spotify’s founders described, but it was an end to a decade of decline. As subscriptions accumulated the effect turned positive. Global recorded music revenue grew every year from 2015 to 2025.²

How the stream learned to bend

Behind the catalogues lay an engineering change that made streaming reliable enough to replace the shelf. The early streams used specialised protocols and servers that held a connection open to each listener and sent a single quality of stream. When the network slowed, the player stalled. In the late 2000s the industry moved to adaptive streaming over ordinary web connections. A video was cut into short segments, each encoded at several qualities, and the player requested one segment at a time, choosing the quality according to how fast the last segments had arrived. Apple’s HTTP Live Streaming came with the iPhone’s 2009 software, and the international standard known as MPEG-DASH followed around 2012.²⁴ The stream now bent to the network instead of breaking. A viewer on a slow connection saw a blurrier picture rather than a frozen one.

Using ordinary web traffic meant the segments could be stored in caches near the viewer, like any web page. Netflix took that to its conclusion. In June 2012 it announced Open Connect, its own delivery network: internet providers could connect directly to Netflix at no cost, or install Netflix’s own storage appliances inside their networks, filled overnight with the titles their customers were likely to watch.²⁵ The library was not really in the air. It was on servers in thousands of buildings, placed as close to viewers as the providers would allow.

Not every provider agreed. In late 2013 Netflix streams to Comcast customers slowed, as the two companies disputed who should pay to carry the traffic between their networks. In January 2014 a federal appeals court struck down the US net neutrality rules then in force. On 23 February 2014 Netflix and Comcast announced a multi-year agreement under which Netflix connected directly to Comcast’s network and, by most accounts, paid for the connection.²⁶ Comcast and others described it as an ordinary commercial interconnection deal. Critics saw a large provider charging a content company for access to its own customers. The dispute showed that a library that seems weightless rests on cables whose owners can set terms.

All thirteen at once

By 2011 the studios that licensed films and series to Netflix had noticed that they were building a rival’s audience, and began to raise their prices and hold back their work. Netflix’s response was to make its own. In March 2011 the trade press reported that it had outbid HBO and AMC for House of Cards, a political drama with David Fincher directing and Kevin Spacey starring, committing to two seasons of twenty-six episodes in all without a pilot, for a sum reported at around $100 million. The figure was never confirmed, and a later report suggested it was considerably less.²⁷

The second decision was how to release it. Television had always released series weekly, to build anticipation and to sell advertising across months. Netflix had no schedule and no advertisers. On 1 February 2013 it released all thirteen episodes of the first season at once.²⁸ Viewers could watch as fast as they liked, and many watched in a weekend. Critics at the time worried that the shared suspense of waiting would be lost. The practice, already familiar from DVD box sets, got a name, binge-watching, and became the default for Netflix’s originals.

Binge-watching has been studied. In 2017 Liese Exelmans and Jan Van den Bulck reported a survey of 423 adults aged eighteen to twenty-five. About 80 percent described themselves as binge viewers. Those who binged more often reported poorer sleep, more fatigue and more symptoms of insomnia, while ordinary television viewing showed no such link, and the association was explained by what the researchers called cognitive pre-sleep arousal: viewers lay in bed still thinking about what they had watched.²⁹ The study was a survey, and it could not prove that binge-watching caused the poor sleep. But it gave substance to a remark Reed Hastings made on Netflix’s earnings call in April 2017, when asked about competitors such as Amazon and HBO: the company’s real rival, he suggested, was sleep.³⁰ It was a joke, and also a description of the business. A service paid by subscription wins when its members watch enough to keep paying, and the scarce resource it competes for is their time.

On 6 January 2016 Hastings switched on Netflix in 130 more countries during a keynote speech, bringing the service to about 190, all except China and a few countries under US sanctions.³¹ He said Netflix would lead in the new markets with its originals. The company had become a global television network without a broadcast licence in any of them.

The studios take their films home

The studios drew the obvious conclusion. If a library was the product, they owned the best libraries. In August 2017 Disney announced that it would end its distribution deal with Netflix, beginning with its 2019 films, and launch its own services. Its chief executive, Bob Iger, told analysts the move would give Disney much greater control over its own destiny, and later recounted the decision at length in his memoir, The Ride of a Lifetime.³² Disney+ launched in the United States, Canada and the Netherlands on 12 November 2019 at about $7 a month, and Disney said 10 million people had signed up on the first day, some of them on free trials.³³ Other studios and technology companies launched their own services, and what had been one library became a dozen, each with its own subscription.

The low price bought subscribers at the cost of years of losses. Disney’s streaming businesses reported their first quarterly operating profits in 2024, after losses that one estimate put at around $11 billion since launch.³⁴ Other services found the economics harder. In 2022, after the merger that created Warner Bros. Discovery, the company removed a set of series and films from HBO Max, including the science-fiction series Westworld, and wrote off hundreds of millions of dollars in content as part of what it called a strategic review. The comedian John Oliver, whose own show ran on HBO, called the service “a series of tax write-offs.”³⁵ For viewers, the removals were a lesson in what access means. A work that had been available everywhere could become available nowhere, because the owner found it more valuable as a tax deduction than as a title in a catalogue.

The reckoning: advertising, passwords and the last disc

In April 2022 Netflix reported that it had lost about 200,000 subscribers in the first quarter, its first decline since 2011, partly because it had suspended service in Russia. Its share price fell sharply. It lost about 970,000 more in the second quarter.³⁶ The company that had grown by being cheaper and simpler than cable now faced a crowded market, higher costs and saturation in its richest countries.

It reversed two of its oldest positions. Netflix had carried no advertising in its whole history as a streaming service; in late 2022 it launched a cheaper plan with advertisements. It had long tolerated, even joked about, password sharing; it now estimated that more than 100 million households were using someone else’s account. In 2023 it began charging for members outside the account holder’s household in more than 100 countries. In the second quarter of that year it added 5.9 million members, and said that revenue had risen in every region where paid sharing was introduced.³⁷ By 2025 its advertising revenue had passed $1.5 billion, its operating margin was 29.5 percent, and it reported more than 325 million paid memberships.⁴ The service had become a habit strong enough that its terms could change.

Who gets paid

The money that streaming collects has to be divided, and the division became the field’s most persistent quarrel.

In television, the quarrel became a strike. In the broadcast era, writers and actors were paid residuals when their work was rerun or sold to new outlets, and those payments grew with a show’s success. Streaming services paid fixed residuals that did not depend on how many people watched, and kept their viewing figures private. The Writers Guild of America struck for 148 days in 2023, ending on 27 September with an agreement that, among other things, created a bonus residual for high-budget streaming series and films watched by at least 20 percent of a service’s domestic subscribers in their first ninety days, raised foreign streaming residuals by 76 percent, and gave the guild confidential access to viewing data.³⁸ For the first time, a streamed work’s success was tied, modestly, to the pay of those who wrote it.

In music, the quarrel is about the pool. Most streaming services divide royalties pro rata: all the subscription and advertising money in a market goes into one pool, and each rights holder receives a share equal to its share of all streams. A subscriber who listens only to a small jazz label does not pay that label; their fee is divided according to what everyone listens to. The alternative, called user-centric, would divide each subscriber’s fee among the artists that subscriber actually played. Advocates argue that user-centric payment would be fairer to niche artists and less vulnerable to fraud; the services and the largest labels have so far preferred to keep the pool.

The services and labels chose to adjust the pool rather than replace it. In September 2023 Deezer and Universal Music Group announced what they called an artist-centric model, launched in France the following month, which gave a double weighting to “professional artists,” defined as those with at least 1,000 streams a month from at least 500 listeners.³⁹ In 2024 Spotify stopped paying recording royalties on tracks with fewer than 1,000 streams in the previous twelve months. It said 99.5 percent of streams were of tracks above that threshold and estimated that about $40 million a year would move from the smallest tracks to the rest; it also imposed longer minimum lengths for noise and other functional audio and charged labels and distributors for fraudulent streams.⁴⁰ Critics noted that for many emerging musicians the threshold removed small but real income. The same year Spotify reported its first full year of operating profit, €1.4 billion.⁴¹ At the start of 2026 Daniel Ek became executive chairman and two long-time executives, Gustav Söderström and Alex Norström, became co-chief executives.⁴²

A different worry concerned not how the pool was divided but who was in it. In early 2025 the journalist Liz Pelly, in Harper’s Magazine and her book Mood Machine, reported that Spotify had for years commissioned inexpensive recordings, credited to invented artists, to fill playlists for moods such as relaxation and focus, under an internal programme known as Perfect Fit Content.⁴³ The music was meant to be heard but not listened to, much as Muzak had been. Its presence in popular playlists meant less money in the pool for musicians whose work it displaced.

Music no one played

Generative AI arrived in this world of abundant, pooled, mood-driven listening with an obvious use: making more of it. Services such as Suno and Udio generated complete songs from text prompts. On 24 June 2024 Universal, Sony and Warner sued both, alleging that they had been trained on copyrighted recordings without permission. The cases moved quickly from courtroom to contract. Universal settled with Udio on 29 October 2025, and the two agreed to build a licensed music-creation service for 2026, with artists choosing whether to take part. Warner settled with Suno on 25 November 2025 and sold it the concert-listing service Songkick.⁴⁴ Sony and Universal continued to pursue Suno, filing further claims in September 2026.⁴⁵

Meanwhile the machine-made tracks reached the catalogues. Deezer, which built a detector for fully AI-generated music, reported that such tracks were about 10 percent of daily uploads in January 2025, 34 percent by November, and more than half at their peak in June 2026. They accounted for only 1 to 3 percent of streams, and Deezer said that up to 85 percent of the streams those tracks received in 2025 were fraudulent, generated by bots to collect royalties from the pool.⁴⁶ The thresholds and fraud penalties of 2024 now had a new purpose. A royalty system designed to pay musicians for being heard was being mined by software that made sounds no one wanted to hear, played by software that no one was.

The library changes hands

The last turn of the history so far was a fight over ownership of a library. On 5 December 2025 Netflix agreed to buy Warner Bros.’ studios and its HBO Max service for $27.75 a share, an enterprise value of $82.7 billion, leaving Warner Bros. Discovery’s cable networks behind.⁴⁷ The company that had begun by renting other studios’ films in envelopes would have owned one of the oldest film studios in Hollywood. Paramount Skydance responded with a hostile bid for the whole company. On 26 February 2026, after Paramount raised its offer to $31 a share in cash, Netflix said that at the price required to match, the deal was no longer financially attractive, and walked away with a $2.8 billion break fee.⁴⁷ Paramount’s purchase was cleared by the US Justice Department in June 2026 and by the European Commission in July, survived a challenge by state attorneys general and closed on 6 October 2026.⁴⁸

The episode compressed the whole history. The streamer had become rich enough to buy a studio, and the studio’s library had become the prize because, in a world of rented access, the one thing that cannot be rented from someone else is the work you own.

What streaming is today

Streaming is now the main way people hear recorded music and the fastest-growing way they watch film and television. In music it is a small number of large services, Spotify, Apple Music, YouTube, Amazon and, in some countries, regional rivals such as Tencent’s, offering nearly the same catalogue of more than a hundred million tracks, and competing on price, recommendation, bundled audiobooks and podcasts and their place on phones. Paid subscriptions are the industry’s largest source of revenue, and the labels’ share of that money depends on contracts negotiated at the services’ founding and renegotiated since.²

In video it is several large libraries, each mixing licensed and owned work, subscriptions and advertising, scheduled live events and on-demand catalogues. Netflix is the largest by revenue and has become the industry’s benchmark for profitability. The studios have consolidated, and the biggest consolidation of all, Paramount’s purchase of Warner Bros. Discovery, closed weeks before this essay was published.

The underlying technology is the adaptive stream delivered from caches near the viewer, so ordinary that it is noticed only when it fails. The underlying economics is access by subscription or advertising. The underlying legal form is the licence, which lets the listener play, not keep, and lets the service remove what it no longer wants to pay for. And the underlying design, increasingly, is the queue: the next song, the next episode, the next video, chosen by a recommendation system and started automatically unless the viewer intervenes.

How streaming changed what we own, watch and hear

From owning a copy to holding a key

A record or a disc was property. Its owner could play it, lend it, sell it or leave it to someone else, under the legal principle that a lawful copy, once sold, is the buyer’s to dispose of. A stream is a service. The subscriber holds a key to a library they do not own and cannot keep, and the library’s contents can change without their consent. Most people accepted the exchange willingly, because the key opened more doors than any shelf could hold. The cost appears only when something they cared about disappears, or when they stop paying and find they have nothing.

From the schedule to the queue

Streaming promised to abolish the schedule, and in one sense it did: anything could be watched at any time. But services soon rebuilt sequence. The next episode starts in a few seconds, the next song follows the last, a recommended video begins when the current one ends. The schedule was public and shared; the queue is private and personal. It answers to a model of each viewer’s preferences rather than to a broadcaster’s view of the evening, and it is tuned to keep the session going.

From the album to the playlist and the mood

The album was a work with an order, made to be heard from start to finish. iTunes split it into songs, and streaming arranged songs into playlists, many of them organised not by artist or genre but by activity and feeling: focus, sleep, workout, rainy day. Music increasingly reaches listeners as a background for something else. For some artists this is a path to an audience they would never otherwise have found. For others it is pressure to make music that fits a mood without demanding attention, the kind of music Muzak was built to supply.

From the copy sold to the play counted

The record business sold copies, and a copy paid the same whether it was played once or a thousand times. Streaming pays per play, from a shared pool. That changes what is rewarded: songs that are played often, in the background, by many people, earn more than songs that are loved intensely by a few. It changes what is measured, since every play is recorded. And it invites fraud, because plays, unlike sales, can be manufactured.

From the scarce shelf to the flooded catalogue

A record shop or a video store had limited space and chose what to stock. Streaming catalogues have no shelf, and any musician can upload through a distributor for a small fee. Deezer alone now receives well over a hundred thousand new tracks a day, by its own figures, and a growing share are made by machines. Abundance was the promise. The new problem is that when everything is available, being available is worth very little, and the power shifts to whoever decides what is noticed.

What the history suggests

Streaming repeats several patterns that run through the site’s essays. The labels’ fight with Napster is a case of a defining strength becoming the constraint: the compact disc’s profits made it hard to imagine a business without it. YouTube’s open door and Napster’s index are both cases of permissive defaults as borrowed time, with different endings because one found a way to share the gains before the courts closed the door. Netflix’s free streaming in 2007 is a small working thing beating a grand design. The studios’ retreat into their own services and Netflix’s turn to owning its shows are cases of incumbents absorbing the challenger’s best idea, in both directions. And the persistent quarrel over royalties is an instance of who pays shaping what gets built: the pro-rata pool rewards scale and background listening, and the music made for it reflects that.

What is particular to streaming is the replacement of the copy by the licence across a whole culture, almost without anyone deciding it. No law abolished the record collection, and no one was forced to give up their discs. The convenience of access simply outcompeted the security of ownership for most people most of the time. Streaming is also unusual in how directly its economics shape its art. Because services are paid by time spent, they favour works that hold attention for many hours; because royalties are paid per play, musicians are paid for repetition; because catalogues are pooled, the most valuable thing a company can own is a catalogue no one else has. These are not neutral facts about delivery. They are choices about what culture is for and who it serves, which is where the philosophy begins.

Understanding the philosophy

The rest of this essay examines eight ideas beneath streaming, one at a time. Each is explained in plain terms, placed in the tradition of thought that has examined it, traced in the history above, tested against its strongest objection and related to how people and organisations decide today. Two of them are explicitly psychological, and a third draws on the psychology of law. The connections to critics, philosophers, psychologists, economists and sociologists are interpretations offered to make the ideas clearer. The engineers, executives and lawyers who built streaming were solving practical problems, and did not, as a rule, set out to answer these thinkers.

1. The aura and the endless copy

The idea in plain terms

A painting hangs in one room. To see it you must go there, and seeing it is an event in your life. A recording can be played anywhere, by anyone, as often as they like. Something is gained, because millions can hear what only a few once could. Something may be lost, because a work that is always available is never an occasion. Streaming takes this to its limit: not one copy per household but one library for everyone, present wherever there is a signal.

The tradition

In 1935 the German critic Walter Benjamin wrote an essay, first published in French the following year in the journal of the Frankfurt Institute for Social Research, now usually known in English as “The Work of Art in the Age of Mechanical Reproduction.” Benjamin argued that a work of art had traditionally possessed what he called an aura: its unique existence in a particular place and time, the history of who had owned and seen it, and the distance it kept from the viewer, rooted originally in ritual and worship. Photography and film, he argued, destroyed the aura by multiplying the work and bringing it to the viewer instead of the viewer to the work.

Benjamin did not mourn this simply. He saw in reproduction a chance for art to leave the sphere of cult and tradition and enter politics and everyday life, to be judged by the many rather than revered by the few. His essay is ambivalent: the aura’s decay is a loss of a certain kind of experience, and also a liberation from the authority that experience carried.

Where it shows up

Streaming is the furthest point of Benjamin’s process. A recording on a disc was still an object with a history, bought on a particular day, worn by particular hands. A stream has no history at all for the listener; it is identical for everyone and leaves nothing behind. The industry’s own figures show a counter-movement. In 2025, as streaming passed 69 percent of recorded music revenue, physical formats grew by 8 percent, driven largely by vinyl, a format bought by many people who also stream.² Live concerts, which cannot be streamed without losing what makes them live, became the main source of income for many musicians. Netflix’s move into live events, sport among them, offers what its library cannot: something that happens once, at a time everyone shares.

The strongest objection

The aura may be a myth that served the people who controlled access to art. The sense that a work is precious because it is rare is partly a sense of social distinction, and Benjamin’s own sympathies lay with the crowd. Listeners today form intense attachments to songs they have only ever streamed; the attachment lives in the listener’s history with the work, not in an object. And the vinyl revival could be read as another kind of consumption, a premium product sold to people who want a souvenir of something they hear elsewhere.

There is force in this. But it does not dissolve the observation that people seek out the unrepeatable precisely when everything else is repeatable. The aura migrated from the object to the occasion.

What it means today

For anyone who makes or sells culture, the question is what in their work cannot be copied. In a world of perfect reproduction, scarcity moves to presence, timing, physical objects and relationships between artist and audience. Services that understand this sell events and memberships as well as libraries. Artists who understand it treat the stream as a way to be found and the occasion as the way to be valued. For listeners, Benjamin’s essay suggests a small discipline: to make an occasion of some of what they hear, because the medium will not do it for them.

2. Planned flow and the private queue

The idea in plain terms

Television never really sold programmes. It sold an evening, arranged so that one thing flowed into the next and viewers stayed with the channel. Streaming seemed to end that: every viewer could choose. But the next episode starts automatically, the next song follows the last, and the recommended video is already loading. The flow returned, built for one viewer at a time, and it is still designed to keep people watching.

The tradition

Raymond Williams, a Welsh critic and one of the founders of cultural studies, argued in Television: Technology and Cultural Form (1974) that broadcasting’s defining characteristic was planned flow. Studying the output of British and American channels, he found that the published schedule of discrete programmes concealed a continuous sequence, with trailers, announcements and advertisements smoothing every join, so that the real experience was watching television, not watching a programme. Williams also insisted that technologies do not determine their uses: television’s form came from the social and commercial choices of those who built it.

A decade later the American critic Neil Postman, in Amusing Ourselves to Death (1985), argued that television’s form shaped what could be said through it. A medium built on images and constant change, he wrote, turned news, politics and religion into entertainment, whatever the intentions of the people speaking. Postman, like Williams, located the medium’s effect in its structure rather than its content.

Where it shows up

The history above shows the flow rebuilt piece by piece. Netflix released whole seasons at once in 2013, which favoured continuous viewing. Its player begins the next episode after a short countdown. Music services fill silence after an album ends with similar songs. YouTube’s autoplay, discussed in What You Might Also Like, starts a recommended video when one ends. The difference from Williams’s flow is that the sequence is no longer chosen by a scheduler for a nation but by a model for a person, and it is tuned by measuring what keeps each person watching.

The strongest objection

Flow may be what viewers want. Many people use television and music as company, and choosing every item would be work. Autoplay can be turned off. A library with a queue is still more under the viewer’s control than a broadcast schedule ever was, and people who want to choose deliberately can. Williams’s evidence came from a period of a few channels and no remote controls; his concept may describe a passing arrangement rather than something essential.

The reply is that defaults decide most behaviour, a finding set out in The Architecture of Choice, and that a default designed to continue serves the business’s measure of success, time spent, whether or not it serves the viewer. That the flow can be switched off does not make it neutral.

What it means today

Product teams that design sequences should ask whose evening they are planning. The choice between stopping by default and continuing by default is a choice about whose interest the product puts first. A service could offer natural stopping points, let viewers set limits that hold, or end a session gracefully. Leaders who measure success by hours watched should recognise that the measure rewards flow, and that Williams’s observation from 1974 describes what their systems are likely to optimise.

3. Why people obey: deterrence, legitimacy and the convenient path

The idea in plain terms

People in the 2000s could take almost any song for free. The industry tried to stop them by suing, and largely failed. A decade later most stopped of their own accord, because paying for a stream became easier than taking a file. That history raises an old question: do people obey rules because they fear punishment, because they think the rules are fair, or because obeying is the easiest thing to do?

The tradition

In 1968 the economist Gary Becker published “Crime and Punishment: An Economic Approach,” which treated offenders as rational people who weigh the gains of breaking a law against the probability of being caught and the severity of the penalty. On this view, enforcement works by raising the expected cost of offending. The labels’ campaign against Napster and, later, against file sharers fits Becker’s model: raise the risk, and behaviour will change.

In Why People Obey the Law (1990), the psychologist Tom Tyler reported studies of residents of Chicago that pointed elsewhere. Their compliance with the law depended less on the risk of punishment than on whether they regarded the authorities as legitimate, and legitimacy depended heavily on whether they felt they had been treated fairly. This became the research tradition of procedural justice. In Code and Other Laws of Cyberspace (1999), the legal scholar Lawrence Lessig added a third view: behaviour online is regulated by four forces, law, social norms, markets and architecture, and the architecture of software often regulates more effectively than law, because it makes some actions easy and others impossible.

Where it shows up

Every part of this history shows all three forces at work. The courts closed Napster, but not file sharing, because the architecture moved to networks with no centre. The labels’ lawsuits, by many accounts, damaged their standing with the listeners they needed, which in Tyler’s terms eroded legitimacy. What finally changed behaviour was a new architecture and a new market: Apple made buying a song easier than finding a file, and Spotify made streaming easier than either. Aguiar and Waldfogel’s finding that Spotify displaced piracy is evidence that convenience did what deterrence could not.²³ Gabe Newell’s remark that piracy is a problem of service makes the same point from the seller’s side.²⁰

The strongest objection

Convenience may be enough to explain everything, with no need for legitimacy. Streaming is cheaper per song than buying, and free with advertisements; people choose what is cheap and easy. Becker’s model accommodates this if the time and effort of piracy are counted as costs. Legitimacy is hard to measure and easy to invoke after the fact.

Yet the persistence of the royalty disputes suggests legitimacy matters for the future of the arrangement. Musicians who believe the system is unfair are more willing to withhold work, support regulation or encourage listeners to pay them directly, and listeners who believe the system cheats artists are more tolerant of piracy when services fragment and prices rise.

What it means today

For companies that depend on people respecting rules they cannot fully enforce, the lesson is to make compliance the easiest path and to keep the rules visibly fair. Enforcement can remove a target, as it removed Napster, but it rarely removes a demand. For leaders in industries facing a new way of getting their product, the question is whether they are designing a better legal path or only defending the old one. And for streaming services, the legitimacy of the payment system is not a public-relations matter; it is part of what keeps listeners paying rather than taking.

4. Imagined communities and the lost appointment

The idea in plain terms

When millions of people watched the same programme at the same hour, they shared more than the programme. Each knew that others were watching, and that knowledge made them a public: people who could talk about it the next day and who felt part of something larger. Streaming lets everyone watch at their own time. It gives each viewer freedom and takes away the appointment that made a nation, or a generation, feel it was watching together.

The tradition

In Imagined Communities (1983) the political scientist Benedict Anderson argued that modern nations became imaginable through print. A reader of a morning newspaper, he observed, performs a private ritual while knowing that thousands of strangers are performing it at the same moment, and this awareness of simultaneity, in what Anderson, borrowing from Benjamin, called homogeneous, empty time, lets people imagine themselves as members of a community most of whom they will never meet. The French sociologist Émile Durkheim, in The Elementary Forms of Religious Life (1912), had described what happens when people gather physically for a shared ritual: a collective effervescence, an intensity of feeling that binds them and renews their sense of belonging. Broadcasting extended both: a shared ritual of simultaneous attention without physical gathering.

Where it shows up

Streaming’s release practices are choices about this. Netflix’s decision to release whole seasons at once in 2013 fitted its nature as a library and broke with the weekly appointment. The conversation about a show became scattered over weeks, and spoilers became a hazard of social life. Several services have since released some series weekly or in parts, recovering a little of the shared rhythm. Live sport, the one category streaming cannot release on demand, has become the most valuable form of television precisely because it must be watched together, at once.

The strongest objection

The national audience of broadcasting was never as unified as nostalgia suggests, and its unity often meant the exclusion of minorities and tastes the schedule ignored. Cable television had already fragmented it. Streaming has enabled new communities around works that broadcasters would never have aired, connected through social media rather than simultaneity. A fan community discussing a series across continents is no less real for not watching at the same hour.

This is right, and the change is better described as a shift from one large imagined community to many smaller ones. The cost is that fewer works are shared by everyone, and that common reference points, once supplied by a few channels, are now supplied by whatever the recommendation systems make popular at once.

What it means today

For anyone releasing a product, timing is a decision about community, not only about convenience. A weekly release builds a shared rhythm and a longer conversation; a full release maximises individual freedom and immediate viewing. Organisations that want their people to share an experience, a strategy, a crisis, a celebration, should notice that shared attention requires an appointment. The library makes everything available to everyone and nothing common to all.

5. The binge and the two selves

The idea in plain terms

Watching six episodes in a night can feel wonderful while it happens and leave little behind: a blur of plot, a late night, a vague regret. Waiting a week between episodes can feel frustrating and leave clearer memories. Psychology suggests the difference is real, because the self that lives through an experience and the self that remembers it judge by different rules, and streaming is designed to serve the first.

The tradition

The psychologist Daniel Kahneman, summarising decades of work in Thinking, Fast and Slow (2011), distinguished the experiencing self, which lives moment by moment, from the remembering self, which keeps score and makes choices. In a 1993 experiment, Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier asked people to hold a hand in painfully cold water for sixty seconds, and on another trial for sixty seconds followed by thirty more at a slightly warmer temperature. Most chose to repeat the longer trial, because it ended better. Fredrickson and Kahneman described two regularities: memories of experiences are dominated by their peak and their end, and the duration of an experience is largely neglected. These findings have been replicated in many settings, though their size varies with the kind of experience.

A second strand concerns choice over time. The psychiatrist George Ainslie, in a 1975 paper on “specious reward,” showed that people and animals discount future rewards in a way that makes their preferences reverse as a reward approaches: at ten in the evening a viewer prefers sleep at midnight to another episode, and at midnight, with the next episode seconds away, prefers the episode.

Popular accounts of cliffhangers often cite the Zeigarnik effect, the claim from the 1920s that people remember interrupted tasks better than completed ones. A 2025 meta-analysis by Romain Ghibellini and Beat Meier found no reliable memory advantage for unfinished tasks, but did find a robust tendency to resume them, a separate finding associated with Zeigarnik’s colleague Maria Ovsiankina.⁴⁹ The cliffhanger works not because viewers remember it better but because an unfinished thing pulls them to continue.

Where it shows up

Exelmans and Van den Bulck’s study found that young adults who binged more often slept worse, through the arousal of thinking about what they had just watched.²⁹ Reed Hastings’s remark that Netflix’s competitor was sleep acknowledged the trade.³⁰ The autoplay countdown arrives at the moment Ainslie’s analysis predicts preferences will reverse, and the cliffhanger supplies the pull Ovsiankina described. Duration neglect means that six hours of viewing may be remembered as hardly longer than one.

The strongest objection

The remembering self is not obviously the one that should rule. A binge enjoyed is enjoyment; why privilege the memory of it? Many people binge books and always have. Viewers know what they are doing and choose it, and it is paternalistic to call their choices mistakes. The research on binge-watching relies on self-reports and correlations.

The objection is serious, and the distinction it blurs is the one in Wanting What We Want to Want. The right response is not to condemn the binge but to notice who sets the default. Viewers choose within a design that makes continuing effortless and stopping a decision. That design favours the experiencing self because the business is measured in time spent, and the remembering self, which might have chosen differently at ten o’clock, is not consulted at midnight.

What it means today

Designers can serve both selves: ending episodes on resolutions as well as cliffhangers, asking whether the viewer wants to continue, letting people set limits in advance that the product honours. Netflix’s own prompt asking whether a viewer is still watching is a small concession to the remembering self. Leaders who measure engagement should ask what their customers would say about their product in memory, not only what they do in the moment, because the remembering self decides whether to renew.

6. Music as mood and function

The idea in plain terms

People have always used music to do things: to work, dance, mourn, sleep. Streaming organised music around those uses. The most popular playlists are moods and activities, and a growing share of what plays is chosen because it fits a feeling without demanding attention. The question is what happens to music, and to listening, when its main job is to accompany something else.

The tradition

In 1938 the philosopher and musicologist Theodor Adorno published “On the Fetish-Character in Music and the Regression of Listening.” He argued that commercial music and radio were producing a deconcentrated way of listening, in which people recognised and liked what was familiar, attended to fragments rather than to the whole, and treated music as a commodity for producing pleasant states. With Max Horkheimer, in Dialectic of Enlightenment (1944), he extended the argument to what they called the culture industry, which manufactured standardised products while presenting them as individual choices. Adorno’s tone was contemptuous, notably towards jazz, and his elitism is widely criticised.

Psychology studies the same phenomenon without the contempt. Robert Zajonc showed in 1968 that mere exposure to a stimulus tends to increase liking for it, which helps explain why familiar songs please. Patrik Juslin and Daniel Västfjäll, in a 2008 paper in Behavioral and Brain Sciences, proposed several distinct mechanisms by which music induces emotion, among them reflexes to sudden sounds, associations learned through conditioning, emotional contagion from the music’s expressive character, memories it evokes and the play of expectation. The sociologist Tia DeNora, in Music in Everyday Life (2000), documented how people use music deliberately to regulate their moods and organise their days.

Where it shows up

Muzak was the first business built entirely on music as function, chosen to make workers productive and shoppers linger. Streaming made the same use personal and universal. Mood and activity playlists fill the services’ front pages. Liz Pelly’s reporting on Spotify’s Perfect Fit Content described inexpensive recordings, credited to invented artists, commissioned to fill such playlists, which suggests the service recognised that for these uses the identity of the musician hardly mattered.⁴³ Spotify’s 2024 rule requiring longer minimum play times for noise and functional audio before they earn royalties was a sign that such tracks had been taking a share of a pool meant for music.⁴⁰ The flood of AI-generated tracks has entered largely through the same door: music that fits a mood can be made without anyone in particular making it.

The strongest objection

Functional music is as old as music. Work songs, lullabies, marches and dances were all made to do something, and many great works began as accompaniment. People who use music to concentrate or relax are not regressing; they are using an art for one of its oldest purposes. Adorno’s critique reflects a narrow taste for concentrated listening to complex works. And streaming has also made concentrated listening easier than ever, since any album can be heard in full.

All of this is true. The concern that survives is economic and practical rather than aesthetic. When music is paid by the play and chosen for mood, the rewards go to what can be played most often without being noticed. Musicians whose work demands attention compete for the same pool with tracks designed not to demand it.

What it means today

For services, the honest course is to distinguish music listened to from sound used, and to pay and label them differently, so that listeners know what they are hearing and musicians are not competing with ambient filler. For musicians and labels, the question is which uses of their work they want to serve. For listeners, DeNora’s research suggests that using music for mood is legitimate and valuable, and Adorno’s warning suggests that a life in which music is always accompaniment may lose the experience of giving it full attention.

7. Superstars, the long tail and the pool

The idea in plain terms

Streaming was supposed to help the obscure. With unlimited shelves, every niche work could find its audience, and the long tail of small artists would together rival the hits. In practice, attention and money remained highly concentrated. Underneath that lies a question of fairness: when a subscriber pays a monthly fee, whose work should the money support?

The tradition

In 1981 the economist Sherwin Rosen published “The Economics of Superstars.” He showed that when technology lets one performer serve a vast audience at little extra cost, small differences in talent or reputation produce very large differences in income, because everyone can have the best at nearly the same price as the second best. Recording, broadcasting and streaming each made the effect stronger.

In 2004, in Wired, and in his 2006 book The Long Tail, Chris Anderson argued the opposite force: with unlimited shelf space and good recommendations, demand would spread along a long tail of niche products whose combined sales could rival the hits. In a 2008 article in the Harvard Business Review, Anita Elberse examined sales data from music and video and found that the tail was indeed growing longer, but it was also flattening, with most niche titles selling very little, and hits retaining or increasing their share. Carl Shapiro and Hal Varian, in Information Rules (1999), had explained why: information goods are expensive to make and nearly free to copy, so their economics reward scale and bundling.

The fairness question is old. Aristotle, in Book V of the Nicomachean Ethics, described distributive justice as allocating shares in proportion to merit, while leaving open what merit is. Robert Nozick, in Anarchy, State, and Utopia (1974), offered a different principle: a distribution is just if it arises from free transfers, and he illustrated it with fans who each pay a quarter to watch a basketball player they admire. On Nozick’s view, each payment belongs to the person it was given for.

Where it shows up

The pro-rata pool, used by most streaming services, does not follow Nozick’s fans. A subscriber’s fee is divided according to everyone’s listening, so a person who listens only to jazz partly pays the most-streamed pop stars. The user-centric alternative would follow Nozick, sending each subscriber’s fee to what they heard. Deezer’s artist-centric model and Spotify’s 1,000-stream threshold kept the pool and changed its weights, rewarding artists with established audiences and excluding the smallest tracks, partly to fight the fraud that a pooled system invites.³⁹⁴⁰ Rosen’s superstar effect and Elberse’s flat tail describe the result: the money concentrates at the top, and the many small artists Anderson imagined share a thin slice.

The strongest objection

The pool may be fairer than it looks. Subscribers pay for access to everything, not for particular artists; the value they receive includes the availability of the whole catalogue, which the pool supports. User-centric payment is more complex and might change the distribution less than its advocates expect, because heavy listeners would pay the same fee for more plays. Every system of distribution creates incentives for fraud; the pool is not uniquely vulnerable.

These are reasonable points, and they show that the choice is between principles rather than between fair and unfair. The pool treats a subscription as membership in a shared library; the user-centric model treats it as patronage of particular artists. Most listeners, asked, probably assume the second.

What it means today

Any platform that pools payments and divides them must decide what principle it follows and should say so plainly to those who pay and those who are paid. Thresholds and weights are not technical details; they are judgements about who deserves a share. Leaders designing revenue sharing, whether for creators, sellers or partners, should test their rules against the question Nozick’s fans pose: does the money go where the payer thinks it goes?

8. Who keeps the culture

The idea in plain terms

For a century, culture was kept by the people who owned copies: households with records, libraries with discs, collectors and archives. In a streaming world, the copies belong to the services and the rights holders, who keep what pays and remove what does not. A society’s memory of its own culture comes to depend on the commercial judgement of a few companies.

The tradition

The French sociologist Maurice Halbwachs argued in Les cadres sociaux de la mémoire (1925) that memory is social: individuals remember within frameworks supplied by the groups they belong to, and what a group stops rehearsing, it forgets. The German scholar of cultural memory Aleida Assmann, in an essay of 2008, distinguished two forms. The canon is the active memory of a culture, the works it keeps in circulation, teaches and performs. The archive is its passive memory, the store of what is no longer in use but is kept, so that later generations can rediscover it. A healthy culture needs both: the canon to share and the archive to draw on.

Copyright law long supported the archive indirectly. Under what American law calls the first-sale doctrine, the owner of a lawfully made copy may lend, sell or give it away. Libraries could build collections, second-hand shops could keep old records in circulation, and a work out of print remained somewhere on someone’s shelf.

Where it shows up

Streaming catalogues are canons with no archive behind them. When Warner Bros. Discovery removed Westworld and other series from HBO Max in 2022, viewers who had never owned a copy had nowhere to turn.³⁵ Licences between studios and services expire, and works move or vanish. Netflix’s last DVDs in 2023 marked the end of one of the largest circulating collections of film on physical media.¹ The consolidation of studio libraries, culminating in Paramount’s purchase of Warner Bros. Discovery in 2026, puts more of the twentieth century’s film and television in fewer hands.⁴⁸ Libraries generally cannot lend streams as they lend books, because streams are licensed rather than sold.

The strongest objection

Digital preservation is easier than ever. The rights holders keep master copies; they have every incentive to make old works available when there is demand; streaming has made vast back catalogues more accessible than any shop ever did. Physical copies decayed, were lost and were never universal. Archives such as national film institutes and sound libraries keep what commerce does not.

The objection is right about storage and wrong about access. A master copy in a vault is not part of the public’s memory unless someone can see or hear it, and the decision to make it available rests with an owner who may prefer to write it off. Halbwachs’s point is that what is not rehearsed is forgotten.

What it means today

For services and studios, retirement policies are decisions about cultural memory and deserve the care given to product launches: notice, alternatives, and perhaps an obligation to deposit or license what is withdrawn. For public bodies, the question is whether the archive function that private ownership once supplied now needs to be secured by law, through legal deposit, library licences or rules for orphaned works. For individuals, the lesson of the last DVDs is modest: own a few things that matter, because a key can be taken back.

Bringing the threads together

The eight ideas connect at a few points. Benjamin’s aura and Halbwachs’s memory are both about what happens to a work when it stops being an object: it gains reach and loses the history and permanence that objects carry. Williams’s flow and Kahneman’s two selves are about time: the queue is designed for the experiencing self, while the shared appointment and the remembered season belonged to the self that keeps score. Becker, Tyler and Lessig explain how streaming won, by making the legal path easiest, and Rosen, Anderson, Nozick and Adorno explain what that path rewards: scale, repetition and music that fits a mood. Anderson’s imagined communities sit between them, because the appointment was both a way of spending time and a way of sharing it.

Underneath all of them lies the same transfer. Decisions that individuals once made, what to keep, when to watch, what to hear next, whom to pay, are now made by services, through defaults, catalogues and contracts. Most people accepted the transfer because it gave them more than it took. The ideas above describe what it took.

The library in the air

The library seems to float above every phone, available everywhere and owned by no one who uses it. In fact it sits on servers inside internet providers’ buildings, under contracts that expire, in catalogues whose owners can remove what they no longer wish to pay for. The history of streaming is the history of people choosing access over ownership, one convenient step at a time, and of companies deciding, on their behalf, what the access would include, how it would be sequenced and how the money would be shared.

A few working principles follow for people who build, run or rely on such services.

A default is a decision made for everyone who does not decide. Autoplay, the full-season release, the pro-rata pool and the removal of an unprofitable title were each defaults that most users never chose and few could change. Whoever designs one should be able to say whose interest it serves.

The convenient path is the one people take, and it is also the one they can be led along. Streaming beat piracy by being easier, and the same ease carries viewers past midnight and listeners into playlists of music no one in particular made. The power to make something effortless is the power to decide what people do.

A payment system is a theory of who deserves what. Every threshold, weight and pool encodes a judgement about merit, and those who pay and are paid have a claim to understand it. A system that cannot explain itself to the people inside it will lose their trust long before it loses their money.

Memory needs an archive as well as a canon. A culture whose works are kept only while they pay will forget what does not, and the forgetting will be invisible because nothing marks the absence. Someone, a company, a library or a state, has to be responsible for keeping what is not currently wanted.

And the occasion still matters. In a world where everything can be played at any time, the things people value most are those that happen once, or that they choose to make happen once: the concert, the shared premiere, the album heard from start to finish with nothing else to do. The library in the air holds almost everything. What it cannot hold is the attention people decide to give, which remains, as it always was, theirs to spend.

Selected bibliography

• Durkheim, Émile. Les formes élémentaires de la vie religieuse. Paris: Alcan, 1912. English translation: The Elementary Forms of Religious Life.
• Halbwachs, Maurice. Les cadres sociaux de la mémoire. Paris: Alcan, 1925.
• Benjamin, Walter. “L’œuvre d’art à l’époque de sa reproduction mécanisée.” Zeitschrift für Sozialforschung 5 (1936). English translation: “The Work of Art in the Age of Mechanical Reproduction.”
• Adorno, Theodor W. “Über den Fetischcharakter in der Musik und die Regression des Hörens.” Zeitschrift für Sozialforschung 7 (1938).
• Horkheimer, Max, and Theodor W. Adorno. Dialectic of Enlightenment. 1944; published 1947.
• Becker, Gary S. “Crime and Punishment: An Economic Approach.” Journal of Political Economy 76, no. 2 (1968): 169–217.
• Zajonc, Robert B. “Attitudinal Effects of Mere Exposure.” Journal of Personality and Social Psychology 9, no. 2, part 2 (1968): 1–27.
• Williams, Raymond. Television: Technology and Cultural Form. London: Fontana, 1974.
• Nozick, Robert. Anarchy, State, and Utopia. New York: Basic Books, 1974.
• Ainslie, George. “Specious Reward: A Behavioral Theory of Impulsiveness and Impulse Control.” Psychological Bulletin 82, no. 4 (1975): 463–496.
• Rosen, Sherwin. “The Economics of Superstars.” American Economic Review 71, no. 5 (1981): 845–858.
• Anderson, Benedict. Imagined Communities: Reflections on the Origin and Spread of Nationalism. London: Verso, 1983.
• Postman, Neil. Amusing Ourselves to Death: Public Discourse in the Age of Show Business. New York: Viking, 1985.
• Tyler, Tom R. Why People Obey the Law. New Haven: Yale University Press, 1990.
• Fredrickson, Barbara L., and Daniel Kahneman. “Duration Neglect in Retrospective Evaluations of Affective Episodes.” Journal of Personality and Social Psychology 65, no. 1 (1993): 45–55.
• Kahneman, Daniel, Barbara L. Fredrickson, Charles A. Schreiber, and Donald A. Redelmeier. “When More Pain Is Preferred to Less: Adding a Better End.” Psychological Science 4, no. 6 (1993): 401–405.
• Lessig, Lawrence. Code and Other Laws of Cyberspace. New York: Basic Books, 1999.
• Shapiro, Carl, and Hal R. Varian. Information Rules: A Strategic Guide to the Network Economy. Boston: Harvard Business School Press, 1999.
• DeNora, Tia. Music in Everyday Life. Cambridge: Cambridge University Press, 2000.
• A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001).
• MGM Studios, Inc. v. Grokster, Ltd., 545 U.S. 913 (2005).
• Anderson, Chris. The Long Tail: Why the Future of Business Is Selling Less of More. New York: Hyperion, 2006.
• Assmann, Aleida. “Canon and Archive.” In Cultural Memory Studies, edited by Astrid Erll and Ansgar Nünning. Berlin: de Gruyter, 2008.
• Elberse, Anita. “Should You Invest in the Long Tail?” Harvard Business Review, July–August 2008.
• Juslin, Patrik N., and Daniel Västfjäll. “Emotional Responses to Music: The Need to Consider Underlying Mechanisms.” Behavioral and Brain Sciences 31, no. 5 (2008): 559–575.
• Kahneman, Daniel. Thinking, Fast and Slow. New York: Farrar, Straus and Giroux, 2011.
• Exelmans, Liese, and Jan Van den Bulck. “Binge Viewing, Sleep, and the Role of Pre-Sleep Arousal.” Journal of Clinical Sleep Medicine 13, no. 8 (2017): 1001–1008.
• Aguiar, Luis, and Joel Waldfogel. “As Streaming Reaches Flood Stage, Does It Stimulate or Depress Music Sales?” International Journal of Industrial Organization 57 (2018): 278–307.
• Iger, Robert. The Ride of a Lifetime: Lessons Learned from 15 Years as CEO of the Walt Disney Company. New York: Random House, 2019.
• Ghibellini, Romain, and Beat Meier. “Interruption, Recall and Resumption: A Meta-Analysis of the Zeigarnik and Ovsiankina Effects.” Humanities and Social Sciences Communications 12 (2025).
• Pelly, Liz. Mood Machine: The Rise of Spotify and the Costs of the Perfect Playlist. 2025.
• IFPI. Global Music Report 2026. London: IFPI, 2026.

Research limitations

Several figures rest on single or secondary sources. Napster’s user count comes from Media Metrix’s sample of home users in fourteen countries, not a global census. The labels’ combined equity stake in Spotify was reported in 2009 and later estimated at 15 to 20 percent; the company has not confirmed it, and the leaked Sony contract was published by the press, not by the parties. The reported $100 million commitment for House of Cards was never confirmed, and a later report suggested a lower figure. Accounts differ on whether iTunes sold a million songs in its first five days or its first week, and on the date of Video Identification’s launch, which some sources place in early 2008. The figure of about $11 billion in losses for Disney’s streaming businesses before profitability comes from one news report, and reports differ on which quarter of 2024 counted as the first profitable one. Warner Bros. Discovery’s 2022 write-off is reported at $825 million in a regulatory filing and at up to $3.5 billion in press accounts; the company did not say that tax considerations alone drove the removals. Spotify’s 2025 operating profit is reported in euros and dollars at different figures, and Netflix’s membership count for 2025 comes from press reports of its results after it stopped reporting subscribers quarterly. Liz Pelly’s account of Perfect Fit Content rests on her reporting and interviews; the date of her Harper’s article is given variously as December 2024 and January 2025. Deezer’s figures on AI-generated uploads and fraudulent streams are the company’s own and have not been independently verified, and the estimate of Deezer’s total daily uploads is derived from them. The legal status of the remaining suits against Suno was unresolved at the time of writing. Accounts of Paramount’s completed purchase of Warner Bros. Discovery give different headline values, depending on whether debt is included. The research on binge-watching and sleep is correlational. The philosophical connections to Benjamin, Williams, Postman, Becker, Tyler, Lessig, Anderson, Durkheim, Kahneman, Fredrickson, Ainslie, Zeigarnik, Ovsiankina, Ghibellini and Meier, Adorno, Horkheimer, Zajonc, Juslin and Västfjäll, DeNora, Rosen, Chris Anderson, Elberse, Shapiro and Varian, Aristotle, Nozick, Halbwachs and Assmann are the author’s interpretation; their works are paraphrased apart from short, attributed phrases. The decisions in the decisions box are judged by the author, separating what could be known at the time (process) from what happened (call).

Footnotes

1. “Netflix officially terminates DVD rental service with final mailings,” Fox Business, 29 September 2023; on the number of discs shipped, iClarified; on discs kept by members, Popular Science. ↩ ↩²

2. IFPI, “Global Music Report 2026: global recorded music revenues grow 6.4%”; Music Business Worldwide. ↩ ↩² ↩³ ↩⁴

3. “Spotify hits 290m paid subscribers in Q4,” Music Business Worldwide; Billboard. ↩

4. “Netflix Q4 earnings exceed expectations as subs top 325 million,” TV Technology; TelecomLead. ↩ ↩²

5. “Before Muzak, this general brought innovation to the Army,” History.com; Wikipedia, “Muzak”. ↩

6. Museum of Broadcast Communications, “Flow”; Wikipedia, “Flow (television)”. ↩

7. “25 Years of Internet Radio,” Streaming Media; Wikipedia, “RealAudio”; Streaming Media Blog, “The early history of the streaming media industry”. ↩

8. “Napster users fan out in search of file-swapping apps,” ClickZ, reporting Media Metrix data. ↩

9. A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001), Justia; US Copyright Office, fair use summary. ↩

10. “Bertelsmann and Napster form strategic alliance,” 31 October 2000, press release archive; Stereophile. ↩

11. “After Napster, the music industry winds up humming,” CBS News, citing RIAA data; Hypebot. ↩ ↩²

12. MGM Studios, Inc. v. Grokster, Ltd., 545 U.S. 913 (2005), Justia. ↩

13. Apple, “Apple Launches the iTunes Music Store”, 28 April 2003; Sound & Vision. ↩

14. “Google to acquire YouTube in stock transaction,” Search Engine Watch, 9 October 2006. ↩

15. “Viacom and Google settle YouTube lawsuit,” CBS News, March 2014; Fierce Video. ↩ ↩²

16. “Google unveils YouTube video identification,” The Register, 16 October 2007; eWeek. ↩

17. “Ten years ago, Netflix launched streaming video,” Quartz; CinemaTech, 16 January 2007. ↩

18. “Netflix represents 32.7% of US peak downstream traffic,” Broadband TV News, reporting Sandvine. ↩

19. Wikipedia, “Spotify”; Britannica, “Daniel Ek”. ↩

20. “Experiments in video game economics: Valve’s Gabe Newell,” GeekWire, 2011. ↩ ↩²

21. “Spotify truthiness,” The Register, August 2009; “Labels to share Spotify IPO money with artists,” The Register, February 2016. ↩

22. “Sony’s contract with Spotify leaks,” FACT, May 2015; Hypebot. ↩

23. Luis Aguiar and Joel Waldfogel, “As streaming reaches flood stage, does it stimulate or depress music sales?”, International Journal of Industrial Organization 57 (2018), RePEc; working paper at NBER. ↩ ↩²

24. Wikipedia, “Adaptive bitrate streaming”; Wowza, “MPEG-DASH explained”. ↩

25. “Netflix rolls out its own CDN: Open Connect,” TechCrunch, 4 June 2012; Netflix, “How Netflix works with ISPs around the globe”. ↩

26. “Inside the Netflix–Comcast deal,” Fortune, 24 February 2014; Engadget. ↩

27. “Netflix rumored to spend $100 million or more outbidding cable networks for one new TV show,” Engadget, March 2011; Consumerist. ↩

28. “Netflix releases all 13 episodes of its own TV show House of Cards at once,” Techdirt, 1 February 2013. ↩

29. Liese Exelmans and Jan Van den Bulck, “Binge viewing, sleep, and the role of pre-sleep arousal,” Journal of Clinical Sleep Medicine 13, no. 8 (2017), PubMed Central. ↩ ↩²

30. “Netflix: our biggest competitor is sleep,” Entrepreneur, April 2017. ↩ ↩²

31. Netflix, “Netflix is now available around the world”, 6 January 2016. ↩

32. “Disney will end Netflix deal and offer its own streaming services,” NPR, 9 August 2017; CBS News. ↩

33. “More than 10 million sign up for Disney+ in first day,” Axios, 13 November 2019; The Hollywood Reporter. ↩

34. “Disney celebrates its first quarter of streaming profits,” eMarketer; “Disney streaming turns profit for first time since launch,” City AM. ↩

35. “HBO Max to remove 20 originals and 36 titles overall,” TheWrap; “Warner Bros. Discovery tax write-down,” ComicBook.com; on John Oliver’s remark, Yahoo News. ↩ ↩²

36. “Netflix shares plummet 25% amid loss of 200K subs,” Light Reading; “Netflix only loses 970,000 subscribers in second quarter,” CNBC, 19 July 2022. ↩

37. “Netflix credits password-sharing crackdown with massive 5.9 million global subscriber additions in Q2,” Next TV; Engadget. ↩

38. “WGA ends strike, releases details on tentative deal with studios,” CBS News; ABC7. ↩

39. “Universal Music Group and Deezer to launch new ‘artist-centric’ streaming payment model,” Variety, September 2023; Digital Music News. ↩ ↩²

40. “Confirmed: from next year, tracks on Spotify will have to be played 1,000 times before they start earning money,” Music Business Worldwide; MusicTech. ↩ ↩² ↩³

41. “Spotify posts $1.5bn annual operating profit for 2024,” Music Business Worldwide; Axios. ↩

42. “Daniel Ek to become Spotify executive chairman,” Music Week, 30 September 2025. ↩

43. Liz Pelly, Mood Machine (2025), discussed in “Liz Pelly of Spotify exposé Mood Machine”, Hypebot; Wheeler Centre. ↩ ↩²

44. “Major record companies sue AI music generators Suno, Udio for ‘mass infringement’ of copyright,” Music Business Worldwide, June 2024; “Universal Music settles Udio lawsuit,” Music Business Worldwide, October 2025; “Warner Music signs deal with AI music startup Suno, settles lawsuit,” TechCrunch, 25 November 2025. ↩

45. Wikipedia, “Suno (platform)”; The Suno and Udio lawsuits and label settlements, Dubspot. ↩

46. “50,000 AI-music tracks are now uploaded to Deezer every day,” Music Ally, November 2025; “Deezer says over 50% of daily song uploads are now fully AI-generated,” Billboard. ↩

47. “Paramount pays Netflix $2.8 billion breakup fee after Warner Bros. deal win,” TheWrap; “Netflix bows out of WBD bid,” NewscastStudio, 27 February 2026. ↩ ↩²

48. “Paramount’s hard-fought takeover of Warner Bros. Discovery closed Tuesday,” CNBC, 6 October 2026; TechCrunch. ↩ ↩²

49. Romain Ghibellini and Beat Meier, “Interruption, recall and resumption: a meta-analysis of the Zeigarnik and Ovsiankina effects,” Humanities and Social Sciences Communications 12 (2025), RePEc. ↩