The Spotify Effect: Digital Distribution and Financial Growth

Conceptual Framework

To elaborate on this observation, I suggest approaching Spotify—and other so-called platforms—primarily through their relation to finance. This methodical choice comes with two caveats. First, talk of a “financialization” of media industries sometimes overstates the power of the financial sector over corporations or individuals, as if finance was entirely independent of a given economic system. Second, the term financialization tends to be used merely metaphorically, to evoke an abstract transformation of economy and society. Claims about finance’s power often lack empirical evidence. They should, however, “not be theorized apart from an analysis of the infrastructure that supports financial transactions” (Poovey 2015, 221).

As a consequence, I will analyze Spotify’s growth over the past decade in terms of how it literally relates to, and is modeled upon, finance’s market practices. This is not to say that music should be seen as a commodity on financial markets, or that music is a speculative object in a similar vein as fine art. Rather, seeing Spotify’s operations as financialized means acknowledging the degree to which it has adopted ideas, role models, procedures, organizational blueprints, and market devices from the world of finance. More specifically, financialization refers to the fact that Spotify in all of its facets has become an integral part of lending agreements that are widespread in finance. The background for this development is the breaking and making of music formats in the 1990s and early 2000s as dotcom start-ups tried to financially exploit the idea of free mass access to music. Recall the peer-to-peer file-sharing Internet service Napster, for instance, whose long-term purpose allegedly was “to ride the wave of interest in the new economy and secure an IPO, thereby leveraging money from the financial markets into the hands of its owners and venture capitalists” (Leyshon 2014, 69).

From the perspective of finance, a service like Napster or Spotify—which equally used peer-to-peer networks and unlicensed music in its early years—provided a source of income on which speculation could be built. As Andrew Leyshon and Nigel Thrift (2007, 98) have pointed out, financial capitalism “is dependent on the constant searching out, or the construction of, new asset streams, usually through a process of aggregation.” For example, rental payments can be used to generate bonds, raising huge sums from banks. Setting up such an income stream to raise loans means that a large part of the business consists of debt. The source of profit here is not the individual rent paid by a tenant but the automated “system for aggregating ground rents into a mass” (Leyshon and Thrift 2007, 105), which legitimizes the debt as a way to produce early returns. As a consequence, borrowers may mobilize capital externalizing the risk that their promised future income might never be realized. There are also social consequences, as financialization may lead to a “society of permanent restructuring where assets and ownership are endlessly churned” (Johal et al. 2007, 568). Aggregation, debt financing, and a dynamic of continuous restructuring are clearly observable in relation to Spotify, as shown below.

Another, more illustrative aspect of Spotify’s relation to finance is the way its growth story has been promulgated. Projective stories are constantly crafted to gain stakeholder support, establish venture legitimacy, and create models for other businesses to follow (Garud et al. 2014). Combined with what Melissa Gregg (2015) describes as the spectacle of “data work”—mobilizing big data graphs at showcases and tech demos, and betting on the affective properties of data visualization—such growth stories have become an integral part of both finance investments and Spotify. As anthropologist Anna Tsing (2005, 57) notes in a study on Internet investments, finance often conjures up scale: “Profit must be imagined before it can be extracted; the possibility of economic performance must be conjured like a spirit to draw an audience of potential investors.” Spotify regularly stages major media events to showcase the prospects of high future income, such as in New York City in May 2015, when automatically personalized playlists and “more targeted advertising” were hyped to compensate for the firm’s over-proportional $8.5 billion valuation. ²

To systematize the relationship between asset streams in finance and streaming as a delivery mechanism for cultural content, the remainder of this article is structured in two parts. The first part offers some historical background, briefly introducing the market practice of brokerage, which has been vital for turning Spotify into what it currently is. Given advertising’s key role in growing the service, the second part is devoted to a more detailed analysis of the advertising technology at work in Spotify’s free version. The research on which this analysis is based includes industry interviews, digital tools (to capture network data), and archival work. ³ As will be shown, so-called “programmatic advertising” techniques are modeled on financial practices. Such programmatic transactions of online display advertising work like stock-market trading; they involve personal data and algorithms in buying and selling ads through procedures of real-time bidding. The point in contextualizing and detailing these procedures is to demonstrate that Spotify’s growth is based on an automated online aggregation system—rather than “music streaming”—and that this system, while creating growth, also massively re-intermediates the relation between ad agencies, consumers, and the music industry.

Background: A History of Brokerage

Market interfaces providing access to videos or songs have often been criticized for devaluing their offerings to maximize corporate gains. If “value depends on scarcity,” growing a service means devaluing audiences, content, and ad rates (Wolff 2015, 34). While not owning, producing, or servicing anything, the argument goes, Spotify, YouTube, Airbnb, and GrubHub are just “indescribably thin layers that sit on top of vast supply systems (where the costs are) and interface with a huge number of people (where the money is)” (Goodwin 2015). Growth, then, seems to depend both on the allegedly infinite “shelf space” of the Internet, and on a type of intermediary that has the capacity, or is ruthless enough, to clear that shelf. In short, a broker.

As a theoretical figure and sociological type, the broker has a long history; the figure’s importance notably re-emerges in contexts of state crisis (Lindquist 2017). Brokerage relates to an increase in transnational flows, and to entrepreneurs with a special expertise in manipulating boundaries: between legal and illegal, commercial and noncommercial, formal and informal, and with the talent to relocate transactions from one side to the other. The broker is a specific type of middleman: an actor who gains from the mediation of valued resources that he or she does not control. According to a well-known distinction by Bruno Latour (2005, 39), what distinguishes the broker as a mediator from an intermediary in the broader sense is that while the latter transports meaning without transformation, mediators “transform, translate, distort, and modify the meaning of the elements they are supposed to carry.” Brokerage has always been morally ambiguous, especially where inequality forms the basis for brokerage opportunities, and where the broker is seen to exploit an exclusive, central location in a given network and becomes identified with a structural function of regulating the circulation of values.

Spotify was founded with the stated aim to mediate between the interests of two conflicting economic actors, the music industry on the one hand and nonauthorized file-sharers on the other. Spotting an opportunity in the structural hole between these two disconnected groups, the company’s brokerage role developed as that of a market maker. ⁴ Although co-founder Daniel Ek and his engineering colleagues did not have anything of value to offer themselves—neither a cultural good nor valuable contacts in the field of cultural production—they successfully began promoting trade by reducing the physical, social, and temporal distance between an industry (largely American) and its economic free riders (mostly in Sweden and Europe during the early years). To do so, they had first to transform the meaning of online music listening, shifting focus “from access to context” so that a business model based on advertising revenue could take hold (Burkart 2013; Wikström 2013, 12). While this business model failed to generate sufficient revenue, necessitating the constant remodeling of the free version, Spotify has come to shape the notion of streaming far beyond its own platform. “In order to spotify something,” Jonas Andersson Schwarz (2013, 54) summarizes polemically,

A commercial service provider has to take something which is free (that which previously constituted some sort of common), chop off its tail (that is, curtailing the supply and minimizing the technical quality), encode it (that is, setting up rules for how it can be used) and, lastly, impose a fee for using it (either through an advertising model or through a subscription model).

In its attempt to span the chasms between industry and piracy, Spotify’s role as a cultural broker could be explained as situational (and metaphorical), but it subsequently took on a structural (and literal) brokerage function in line with the company’s intensifying relations to finance. In this respect, Spotify’s prevalent classification as a “tech company” is somewhat obscure, since the company was from its very start in the business of providing content to audiences, while selling those audiences to advertisers. These are defining characteristics of the media sector that put Spotify in line with cable television and satellite industries. The firm’s original articles of association list “Internet-based services within digital media such as music, games, and television, and related activities,” and its first subsidiaries—Spotify Sweden AB and Spotify Service AB ⁵ —were established to expand advertising sales and media buying, and public relations and communication, respectively (Bolagsverket 2009–2012). The most obvious reason why Spotify still came to be identified with technological innovation rather than with the practices of a traditionally operating media firm is “that being thought of as a tech company brings with it the potential for much higher valuations from the investment community” (Napoli and Caplan 2016). Venture investing requires an industrial self-categorization such as Spotify’s—“data processing, hosting, web portals” (Statistics Sweden 2016), according to the Swedish Tax Agency’s industrial classification standard (SNI)—since investors see greater potential in the technology sector than the media sector. And they especially did throughout Spotify’s two initial series of venture capital funding granted during the global financial crisis of 2007–2009, a period when the availability of such capital had decreased, except for Internet start-ups (Block and Sandner 2009).

At the end of a decade-long process of financialization, Spotify is neither particularly Swedish nor just about music. While invocations of the company’s “Swedishness” are needed to sustain the venture capital vision of so-called “European unicorns” (start-ups valued over $1 billion), and to position Spotify at the sexy, cool end of digital innovation, the company acts now as a digital broker whose history of equity rounds, market and debt capitalization, and board of directors firmly ties brokerage strategies to U.S.-based financial interests. Between October 2008 and June 2015, Spotify raised $1.6 billion in seven rounds of investment from twenty-six investors, including Coca-Cola, Goldman Sachs, and Palo Alto–based Technology Crossover Ventures. By 2012, Spotify had begun trading securities on the U.S. financial market; in 2015, it raised $500 million in the form of a loan convertible into Spotify shares, and in 2016, another $1 billion in convertible debt that will enforce a rapid joint stock exchange listing in Sweden and the United States (Constine 2016). Further strengthening ties to both finance and other media, Technology Crossover Venture partner and former Netflix board member Barry McCarthy was appointed Spotify’s Chief Financial Officer in 2015, replacing Daniel Ek in key leadership functions at subsidiaries such as Spotify LTD, Spotify Service AB, Spotify Europe AB, and Spotify Sweden AB. As illustrated by the recent addition of television content, Spotify now operates increasingly like a traditional American media company, while retaining the benefits of financial and regulatory loopholes granted to European tech firms. ⁶

An Analysis of Spotify’s Advertising Technology

Neither Daniel Ek nor Martin Lorentzon had worked in the music industry before founding Spotify. Rather, both of them came straight from advertising technology backgrounds. Lorentzon had made himself a fortune with his company TradeDoubler, which he established in 1999. Having survived the first dotcom crash, TradeDoubler expanded to become a leading affiliate-marketing network in Europe and went public in 2005. In March 2006, TradeDoubler acquired Advertigo, a startup that had developed a technology for “contextual advertising” from its founder, Daniel Ek. Soon after their initial meeting in summer of 2006, the two Swedish multi-millionaires started a business together, which in July that same year was named Spotify. The name did not hold any particular meaning, in part because the business did initially not focus on any particular digital content, but instead aimed to develop a more general “media distribution platform” based on peer-to-peer technology. ⁷

While Spotify has been testing ads since 2007 to monetize the service, the use of advertising technology has proliferated with increasing pressure from investors to deliver on the company’s promise of vast future income. To grow into its $8.5 billion valuation, in 2015, Spotify implemented programmatic ad sales, a new transactional procedure of automated media buying modeled on finance’s stock exchanges. The automation of online ad buys via interconnected online “trading desks” allows inventory to be auctioned off within milliseconds. This procedure yields several advantages. It allows liquidation of overabundant inventory that human sales teams have failed to sell. Automation thus attempts to establish a functioning market where there arguably was none before. Programmatic advertising also scales efficiently, by allowing advertisers to aggregate interactions with audiences across the web, without having to cherry-pick those audiences manually across thousands of publisher sites. For ad agencies and especially media planners, establishing automated trading desks was a defensive measure against Google and other players, ensuring their own relevancy in a field where brands were enabled to buy directly from ad networks (Turow 2011, 82). Low barriers of entry and high profitability additionally incentivized market growth. Importantly, programmatic advertising also appeared the only way to stop the downward spiral of digital advertising prices, provided that it was capable of improving performance and reducing “waste,” or unwanted audience segments (Simmons 2016).

A key promise of programmatic advertising is that of a competitive edge when it comes to demonstrating advertising’s new “relevance to the consumer” (Internet Advertising Bureau [IAB] 2016). Spotify promotes its platform to advertisers as an innovative setting for programmatic targeting techniques. According to Spotify, these techniques include both demographic targeting and content targeting. Playlists, tailored to specific urban activities (such as “Morning Commute”) and moods (such as “Life Sucks”) are combined with data on genre preferences, age and gender, geography, language, and streaming habits alongside broader interests, lifestyle, and shopping behaviors, fueled by third-party data providers. It is, in short, a business model based on technology and process: music is promoted as merely functional for defining and microtargeting divisions of audiences.

Despite Spotify’s efforts to match ad content to listening behaviors, however, users of the free version have questioned the efficacy of this model. As recently as August 2016, the Spotify Community website listed numerous complaints that document the lack of relevance ads still have for consumers. Listeners complain about the frequency, repetitiveness, and loudness of audio ads, and about the lack of proper targeting, noting that some ads did not match basic data sets including age and gender, location of user IP and user language, genre preferences, and listening context: “With all the information you have on every user, I still have to listen to publicity on music genres I hate,” one user commented, while another found it to be “incredibly off-putting, for example, to be listening to a nice soothing string of classical or folk music and suddenly have raucous, obscene hip-hop blaring at you” (Spotify Community 2016). However, such criticism blames publishers such as Spotify for what are in effect decisions made by marketers, most of whom do not micro-target their ads but instead opt for broad media reach. Marketers, in other words, have not generally adopted Spotify’s high value, “premium environment for premium brands” vision at scale so far and are not regularly using the service’s targeting features.

Spotify itself has pointed out that ads serve a dual purpose, generating a revenue stream for the company but also prompting advertising-averse users to pay for Spotify Premium (Blattberg 2015), a claim that is substantiated by the fact that as of August 2016, the most frequent audio and display ads promoted Spotify’s own ad-free service. Furthermore, the apparent inefficiency in avoiding negative responses to ads also has been related to the late roll-out of programmatic media buys. Having introduced programmatic ad sales only in February 2015, Spotify was still “building the market” in November 2015, concentrating its efforts on the United States where by March of 2016, automated trading was used for 30 percent of all ad formats, while adoption was already higher in Europe and Australia (Bieber 2016).

Since issues of user targeting or the ad tech networks involved in placing ads are not reliably addressed via trade journals or interviews, a way to gain additional insights about the function, scope, and effects of Spotify’s advertising technology is to conduct experiments with the help of digital tools. Programmer Roger Mähler and I therefore conducted such tests using freely available tools between August and October 2016. A first experiment established an artificial Facebook account, which was used to log in to Spotify’s free desktop web player (https://play.spotify.com) to test the hypothesis that user data can be easily forged and thus are not suited for valuing Spotify’s ad inventory on desktop, refuting the company’s claims of “100 percent authenticated user data” (Spotify for Brands 2015). A second experiment tested the hypothesis that programmatic advertising might lead to a proliferation of new and hidden intermediaries in the online distribution value chain. This experiment used an ad tracking plugin called Ghostery (ghostery.com) to map stakeholders involved in placing ads and was combined with Fiddler, a tool for capturing network data (www.telerik.com/fiddler).

Both experiments appealed to public concern regarding consumer surveillance, which seems to increase with mounting financial investments and corporate growth. Internet critics and journalists have long documented that social media companies such as Facebook use tools to micro-target their users based on emotional states. Similar claims have been made about Spotify, for example, by Stanford psychologist Michal Kosinski who developed a model for behavioral prediction now used by Cambridge Analytica (cambridgeanalytica.org), a firm notorious for “psy-ops” electoral manipulation in support of Brexit and the Trump campaign. As Kosinski and others argue in a paper titled, “The Song Is You,” industries should abandon the traditional order of knowledge that organizes music according to genres and styles and strategically exploit the link between music choices and personality traits (Greenberg et al. 2016). Yet while concerns over digital advertising’s social profiling are obviously not unfounded, such criticism often misunderstands the implications of programmatic targeting techniques.

One misunderstood issue here is that although programmatic advertising certainly indicates a shift away from larger statistical aggregates toward particular individual targets, such a “target” is not to be equated with an individuated human being but with an inferred one. Rather than being “you,” targets are like you: sets of demographic, psychographic, and other data points aggregated via various online sites, made commensurable or “bundled” together like mortgages in the early years of automated finance, then sold. If programmatic advertising’s promise rests on scaling ads across thousands of publishers, then this means a higher volume of ad traffic, but also that advertisers buy interactions with an audience wherever this bundle of aggregated data points spends time—that is, decoupled from content such as music, from publisher brands such as Spotify, and from actual listeners such as you (cf. Turow 2011, 85; Wolff 2015, 77).

Our first experiment unsurprisingly confirmed that Facebook user accounts can be easily forged, even when they—as in our case—are deliberately designed to appear forged in terms of conspicuous user features, behavior, and social network structures. The digital proxy we used, for instance, was based entirely on Google “Most Popular” searches for name, location, or images. Accepting all incoming requests, it quickly garnered more than four hundred “Friends,” many of which appeared equally forged. The account was then linked to Spotify, and Facebook friends were imported. Web browser authentification of user data therefore appears insufficient, as opposed to Spotify’s mobile app’s device ID. We also learned that browser authentification rests on User ID and geographical location, meaning that IP address-based tracking determines ad placements. This is because Spotify’s system seems geared toward a national advertising market that serves both international brands and niched or local ad inventory. Actual user language, gender, age, moods, and genres or listening context thus had little discernible effect on targeting users across ad formats, because Spotify appeared to push geotargeted ad content, transferring the user’s agency to the associated milieu or environment (cf. Barreneche and Wilken 2015, 506).

Such answers are partial at best, however, because they confine their analysis to the consumer-facing part of Spotify’s platform. A fallacy of today’s widespread “platform critique” (Gerlitz and Helmond 2013) is that it abstracts from what are sprawling, ephemeral networks of interaction that reach beyond any platform itself. Represented through the limited, momentous view of the critical analyst, a complex supply chain is reduced to one operational mechanism. In Spotify’s case, such a view notably ignores the role of marketers (while overemphasizing the publisher), and overstates technology (while ignoring socio-organizational issues). In response to this fallacy, we conducted a second experiment to map the larger advertising supply chain. While the findings are neither representative nor conclusive, they certainly offer material for further hypothesis testing.

Opening the Ghostery plugin, and browsing through the automated playlists presented by the service to our digital proxy, we got a list of more than thirty companies involved in Spotify’s networked interaction. Some of these companies provide trackers (such as ScoreCard), widgets (e.g., Facebook Social Graph, Taboola), and analytics (e.g., Google Analytics), others measure performance (New Relic), yet the largest part is related to advertising. Based on this list, it is possible to develop a preliminary typology of industrial actors to be found in programmatic ad sales:

Ad supply is based on a continuous auction of audience segments, an auction that includes real-time bidding managed via so-called ad exchanges, or digital marketplaces. In this brokerage setting, a publisher (Spotify) partners with one or several supply-side or “yield-optimization” platforms (in short, SSPs). This partnership is designed to maximize the prices for Spotify’s impressions, and to aggregate and manage Spotify’s relationships with ad buyers, as well as with ad exchanges and ad networks. Through the SSP, the publisher may determine who it wants its inventory bought by, the type of ads it wants displayed, and the range of pricing: preferences that are set before trading starts. Once a publisher’s sales partner or SSP (e.g., Rubicon) begins inputting available ad impressions into automated ad exchanges (e.g., Admeta), demand-side platforms (e.g., AdRiver) analyze and purchase them on behalf of marketers (brands, for example, Heineken) depending on attributes such as location or specific targets, with the intent to buy ad impressions as cheaply and efficiently as possible. In addition, publishers also may pre-sell part of their inventory at a fixed price to one or several ad networks (e.g., Adkontekst) for a particular targeted audience.

From a marketer or brand viewpoint, the exchange may start differently: by simply buying a so-called “whitelisted” inventory, which would also include Spotify; by accessing Spotify’s inventory through a DSP for open-auction bidding; or by directly contacting Spotify in case “something more than just inventory” is wanted: when first-party data really are required to target a brand in relation to content or demographic data, as in so-called private (pre-negotiated) marketplaces. ⁸ Trading is fueled by consumer data that are extracted by various data suppliers and aggregated by one or several data management platforms (e.g., BlueKai). As soon as the inventory has been sold, several media- and format-specializing ad servers (e.g., ADTECH) will deliver the ad units to the website, while the viewability and effectiveness of the ads are monitored (e.g., by Moat and Comscore). While the basic currency of the exchange is CPM, or cost-per-mille, a standard measure for calculating an ad’s cost per thousand advertising impressions, buyers and sellers measure differently: a marketer may optimize toward viewability or clicks through settings in the DSP platform, and a publisher toward number of buyers or revenues per thousand impressions. The entire exchange process from the client’s initial linking up to the publisher content server to the final placing of the ads takes less than 30 milliseconds.

Somewhat complicating the picture, several of the above-mentioned firms perform multiple roles in the supply chain and may be directed toward either publishers or marketers, or both (e.g., AppNexus), while none of them is exclusively working with Spotify. Dozens of firms, clustered in digital space but operating from often remote countries, offer highly specialized, competing or complementary services that are partly tailored to regional ad markets, subject to a constant transformation of technology and industry, and always actualized in their relations depending on a given brand and target. Automation thus entails a superintermediation of media buying—it leads to an “exploding number of ad technology products and services, new cost models, and an evolving patchwork of bundled buy-side offerings and solutions” (Dick 2016). As Jana Jakovlevic, Spotify’s Head of Programmatic Advertising, admits in an interview (2016), it is a “lot of middle-men . . ., definitely confusing, even more so for the buy side.” ⁹

For Spotify, programmatic advertising has two implications. First, there is an attempt to maximize ad sales in a short time. To sell the idea of “music streaming behavior as new currency for advertisers” to the venture capital investment community, Spotify aims to increase ad revenue by multiplying programmatic supply chain vendors. As the example above illustrates, the company cooperates not with one or two firms in each category of service, but with several ad exchanges, SSPs, DSPs, and so on. ¹⁰ In what resembles not so much a stock exchange as incrementally stacked exchanges, superintermediation manifests itself as the breeding ground for growing ad revenue. In 2015 alone, advertising nearly doubled, up 98 percent to $219 million (Ingham 2016). Second, automation may deliberately prompt negative responses to ads by pushing increasing volumes of ad content toward listeners who clearly do not seek to have their music experience interrupted. Although audio spots are one of the oldest ad formats, and although Spotify only displays one ad at a time (as compared with many other platforms), advertising-adverse user experiences seem to have increased. As a consequence, since 2015, Spotify has managed to convert more free listeners into paying subscribers than ever before. In the short historical moment of “Spotify before IPO” (and potential Facebook acquisition), then, automation has enabled a double “conversion” of free listening into scale and value.

While a lot of general concerns have been raised by automation, its more direct consequences tend to be overlooked. These include, again, not only a higher volume of ads and growing collections of user data (Terdiman 2015) but also a decoupling of the music from the buying and selling of ads: it is the target, not the song, that creates the context for any given ad. Somewhat ironically, automation thus even puts wealthy publishers like Spotify in a “precarious position,” because it coerces them to follow the demands of database-driven mediators (Turow 2011, 86). Spotify has become a broker depending on other brokers for realizing its claims on future income—a middleman on top of middlemen. In addition, the new ad tech infrastructure has increased costs for publishers and marketers, instead of reducing them. Programmatic media buying is more expensive (especially regarding costs per exposure) than insertion-order, adding an “ad tech tax” to the budget, often still without transparent pricing (Morrissey 2016). Finally, automation causes friction within the advertising supply chain, as it disrupts long-established relations between ad agencies, media planners, and publishers. ¹¹

Conclusion

As this article has demonstrated, Spotify today should be regarded less as a Swedish music streaming service than as a U.S.-based media company operating at the intersection of technology, advertising, finance, and music. The source of its profit is, to put it simply, an automated online aggregation system. This system aggregates and “bundles” data input from consumers, advertisers, and cultural producers, and allows for efficient, low-cost billing and collecting. While aggregation is a general facilitating principle of all streaming services (Vonderau 2015), its most important function for Spotify over the past decade has been to establish the prospect of future income streams that would provide venture legitimacy in the present. Spotify’s current growth strategy is to create value before launching an IPO in 2017 or 2018. “Growth” here relates to the attempt to accumulate fictitious capital, in the sense of capital only indirectly related to the growth of real production. It is a strategy that does not primarily aim to turn songs (or audiences) into commodities but to treat them as a form of collateral that can be mobilized to secure loans. This is an investment in something yet to come, built on a “bit of fake-it-till-we-make-it hopefulness” where the hope is that,

“at some tipping point, a different kind of advertising, one not based on immediate response but on investing in shifts of mood, opinion, and desire, of creating the grand illusions and stories that propel consumer life—and big media margins” will emerge. (Wolff 2015, 87)

Given that markets do not exist in isolation but are embedded into another, the core or leading market governing Spotify’s distribution of songs and listener data thus far clearly has been the financial market. It is in the financial sector, not the music or advertising industries, where the largest transactions of capital have been documented, and from which models, procedures, and market devices have been adopted. This has lead to a co-dependence between constant growth imperatives and debt finance, and to an ongoing restructuring of the social worlds involved in making Spotify’s “ad-support” operational. “Growth” here refers both to the seemingly boundless horizontal expansion of the music streaming market—more content, more users, more data, more ads—and to the vertical scaling up across different sectors, where losses in one market may create assets in another, so that constantly shrinking revenues for independent musicians, for instance, caused by the effects of scale on Spotify’s proportional revenue model, are projected to go along with increasing asset trading elsewhere. Rather than just being two- or “multi-sided” (Evans and Schmalensee 2012), digital markets thus resemble stacks—where trading sites are stacked into or on top of each other, in often opaque, unaccountable, and unsustainable ways. ¹² While scaling connotes transparency and a single plane of development, stacking connotes multiplying, opaque markets, and a form of embedding markets in other markets that makes Spotify’s behavior dependent on actors “elsewhere.”

Highlighting these co-dependencies (White 2002), this article has demonstrated the need to move beyond the misleading notion of platform when explaining the operations of digital media companies (cf. Helmond 2015). It also substantializes Anna Tsing’s (2005, 57) ethnographic observation that financial capital crucially depends on the effects of spectacle, for “finance can only spread as far as its own magic.” I have shown that Spotify’s “magic” indeed exists: it invokes acts of trespassing, border-crossing, and the manipulation of boundaries in ways that connect the world of finance to our own imagination.