Of dog kennels,magnets,and hard drives: Dealing with Big Data peripheries

Abstract

How did the 3.5-inch Winchester hard disk drive become the fundamental building block of the modern data center? In attempting to answer this question, I theorize the concept of "data peripheries" to attend to the awkward, uneven, and unintended outsides of data infrastructures. I explore the concept of data peripheries by first situating Big Data in one of its many unintended outsides—an unassuming dog kennel in Indiana housed in a former permanent magnet manufacturing plant. From the perspective of this dog kennel, I then build a history of the 3.5-inch Winchester hard disk drive, and weave this hard drive history through the industrial histories of rare earth mining and permanent magnet manufacturing, focusing principally on Magnequench, a former General Motors subsidiary, and its sale and movement of operations from Indiana to China in the mid-1990s and early 2000s. I then discuss how mobilities of rare earths, both as materials and political discourse, shape Big Data futures, and conclude by speculating on how using the situated lenses of data peripheries (such as this Indiana dog kennel) can open up new methods for studying the material entanglements of Big Data writ large.

From data center to data periphery

The hard drive. It is here, among us, sorting us, storing us. Where is it? How did it get here? How do we know it? For an object so central to the maintenance of digital living, the hard drive remains a fickle and obscure thing, always tucked away, always hiding. A mere 3.5-inch box. A black box? Not exactly. The hard drive, and specifically its 3.5-inch incarnation, is spectacularly knowable, if only we learn where and how to know it. Through the rise of Critical Infrastructure Studies, we have come to know the data center as a dominant signpost of digital materiality (Hogan, 2015a, 2015b; Holt and Vonderau, 2015; Hu, 2015; Johnson, 2019a; Vonderau, 2019), but what of the hard drive—that crucial magnetic brick populating all those centers? A tremendous amount of material and labor must be mobilized before data centers can get on with the business of being data centers, and in order to understand the complex flows of these materials and labor, we must first understand the hard drive—where it comes from, how it’s made, and who makes it. Contrary to popular belief about the prevalence of solid-state storage solutions (Metz, 2012; Vincent, 2016), the 3.5-inch hard disk drive—virtually identical in form to the one that came with your or your parents’ Gateway PC in 1996—still overwhelmingly populates most data centers today (Brewer et al., 2016). Having stubbornly endured as a standard storage solution for personal computers since the late-1980s—and for data centers since the late 1990s—the 3.5-inch Winchester ¹ hard disk drive remains deeply embedded in global data infrastructure.

In this article, I explore the hard drive not as an object, but as an infrastructural standard linking a series of diverse socio-material processes (Carse and Lewis, 2017)—processes bound up with histories of labor, colonialism, neoliberalism, and commodity flows. Before loaded into server racks, hard drives are mined, smelted, smoothed, flattened, and packed into 3.5-inch-wide aluminum boxes. As the data center’s foundational functional unit, the hard drive, and particularly the 3.5-inch Winchester form factor (see Figure 1), ² has been and continues to be a dominant influence on the form and function of global data storage infrastructure (see Adrian, 2017; Brewer et al., 2016). How did the 3.5-inch Winchester hard disk drive become the building block of the data center? What follows is a history of how The Cloud came to be built, 3.5 inches at a time. This is not a business, or technical history of the disk drive industry (e.g., Christensen, 1993), but rather a history that situates the 3.5-inch hard drive as window into the diverse industrial and post-industrial worlds in which Big Data is bound up.

Figure 1.

Connor Peripherals CP-340, the first NdFeB-powered 3.5-inch Winchester hard disk drive, introduced in 1986 (Carlson, n.d., courtesy of the Computer History Museum).

Figure 2.

Coco's Canine Cabana at 405 Elm St. Valparaiso, Indiana (Google Maps).

To this end, I introduce the concept of data peripheries to think about how Big Data, and its requisite centers, leverage and depend on a multitude of social, political, and material conditions that proliferate well outside the standard analytical purview of data center studies. I use the term “peripheries” not to reinforce tired imperial concepts of core/periphery binaries, but rather as an attempt to denaturalize the data center as the primary place where the materiality of Big Data storage is analytically accessed. In this regard, the concept of data peripheries builds upon recent scholarship on the uneven material flows, and idiosyncratic regional politics of global data architectures, such as Chan’s (2014) concept of “networking peripheries”, and Johnson’s (2019a) illustration of data centers in Iceland as “infrastructural in-betweens”, which both highlight how cultures of data move through, occupy, and transform spaces. Data peripheries, intentionally juxtaposed directly against the now rhetorically naturalized idea of data centers, move these conversations into the undertheorized world of the material supply chains undergirding Big Data, attending more deeply to the industrial mobilities of materials and labor that precede the data center, much of which come out of the Global South. In foregrounding how logistical mobilities weaving in and out of the Global South shape the material forms of Big Data infrastructure, the concept of data peripheries attempts to think beyond current relations of datafication (Milan and Treré, 2019) by bringing questions of supply relations, and their associated frictions and mobilities, deeper into conversations about the socio-material impacts of what Hogan (forthcoming) calls the “data center industrial complex”. In this regard, building on Thylstrup’s (2019) notion of “data toxicity” or “data-out-of-place,” I bring Big Data’s peripheries into relief through the industrial “traces” and imprints left in the wake of its supply chain. In exploring these “traces”, I chart Big Data’s tentacular industrial relations, questioning how they might be reimagined from the bottom-up, and from the outside-in.

I elaborate the concept of data peripheries through the story of a single, key ingredient in the manufacture of the 3.5-inch hard disk drive—the neodymium–iron–boron (NdFeB) magnet. The neodymium magnet was the first cobalt-free rare earth permanent magnet, and one of the central components in hard disk construction that helped cement the 3.5-inch hard disk drive as both a technical standard, and one of the most ubiquitous forms of information storage in human history. Since the mid-1980s, the proliferation of the NdFeB magnet has played a central role in the miniaturization of electronics and has helped make the hard disk drive the most popular data storage solution of the PC age—as opposed to solid state, bubble memory, floppy disks, or other competing storage solutions of the time (see The Computer Chronicles, 1985). The history of the rare earth mining and permanent magnet manufacturing industries are hard drive histories—histories that begin and persist outside the box, in the data peripheries, both intended and unintended. I chart how the evolution of the rare earth mining and permanent magnet manufacturing industries propelled the 3.5-inch hard drive to prominence in the late-1980s and have helped it retain its outsized relevance as a technical standard within otherwise constantly transforming computing infrastructures. Standards are reflective of and reproduce embedded value systems (Busch, 2011; Star, 1999). The analytic of data peripheries offers a lens through which to investigate the embedded forms, processes, and values within Big Data supply chains. Following labyrinthine resource flows through Africa, the United States, China, and even an Indiana dog kennel, I illustrate how rare earths, magnets, hard drives, and dogs are all brought together into the always-already precarious friction (Tsing, 2005) we have come to call Big Data. I explore the concept of data peripheries by first situating Big Data in one of its many unintended outsides, an unassuming dog kennel in Indiana. From the perspective of this dog kennel, I then build a history of the 3.5-inch Winchester hard disk drive, weaving this hard drive history through the industrial histories of rare earth mining and permanent magnet manufacturing. In the penultimate section, I discuss how mobilities of rare earths, both as materials and political discourse, shape Big Data futures. To conclude, I speculate on how using situated lenses of data peripheries (such as this Indiana dog kennel) can open up new methods for studying the material entanglements of Big Data writ large.

Unintended outsides

“The practices and actors in dog worlds, human and nonhuman alike, ought to be central concerns of technoscience studies” (Haraway, 2016: 95). To know the hard drive and its history we first need to get our bearings. Rather than reading its “grammatology” (Kirschenbaum, 2008), or traveling through the object in a “media descent” (Allen-Robertson, 2017), we will instead travel volumetrically inward, simultaneously from all altitudes, from the fringes of data peripheries, all the way to the data center, and back again. To begin, we will situate ourselves in the fray of one (of many) of the hard drive’s unintended outsides—a place where the hard drive catalyzed and was itself caught up in the wake of something altogether larger and more complex than itself. We open, of course, in a dog kennel in Valparaiso, Indiana.

In 2009, at 405 Elm St. in Valparaiso, Indiana, Kathy DeFries, owner of Excel Machine Technologies, opened Coco’s Canine Cabana inside the other half of her 15,000 square-foot machine shop. With no shortage of tropical paraphernalia, dogs have over 10,000 square feet of play room, and can lounge freely under the latticed palm frond roof of the centrally situated cabana (Wieland, 2009). Indeed, a pooch’s paradise! But these wily canines are part of a much larger, more complex history than they realize. This unassuming dog kennel, and its associated machine shop, resides in the same sprawling complex that once housed the headquarters of one of the world’s largest manufacturers of rare earth permanent magnets, a company founded by General Motors in 1982 called Magnequench. This company, and he carefully constructed narrative about its titanic rise and swift abandonment of Valparaiso in the mid-1990s, is central to how the United States has come to understand and legislate rare earths as strategic minerals. In 2008, during her first campaign for President of the United States, Hillary Clinton made a strategic stop in Valparaiso to talk about Magnequench. She lamented how George W Bush and his administration had let the jobs there slip through their fingers into Chinese hands, and how China now held a crucial stake in America’s defense technology. In addition to use in hard drives, optical drives, and speakers, NdFeB magnets made by Magnequench are essential for the production of precision-guided missiles, and other heavy weaponry (Burrington, 2018). Regardless of Bush’s aggressively neoliberal outsourcing policies, the long process of trans-pacific technology transfer in Valparaiso began in the 1990s, well before he entered office. By 2003, Magnequench had packed up the remainder of their operations, leaving their vast industrial campus largely empty, and Kathy with a struggling machine shop, bereft of the lucrative contracts the Magnequench empire attracted (Robison and Ratnam, 2010). After years of declining revenue, she transformed a portion of her machine shop into Coco’s Canine Cabana, where happy dogs and industrial machining continue to exist in precarious collaboration.

About a year after Kathy launched this business, China instituted a temporary ban on rare earth mineral exports—first to Japan, and then to the United States—which immediately incited both nations, as well as the technology and automotive industries, to frantically reexamine their strategic and economic dependence on Chinese rare earth oxides (Aston, 2010; Thibodeau, 2010). Suddenly, as if overnight, rare earths were everywhere—in our defense technology, our hard disk drives, our cell phones—and they were in danger! Rare earth minerals have passed in and out of the public imaginary for over 100 years, remaining largely obscure—an almost mythic group of metals that we are told at strategic times we all depend on, somehow. This was just such a strategic time. Shortly after the export ban, Seagate, the world’s largest manufacturer of hard disk drives, warned consumers that prices of their drives could be subject to fluctuation, depending on the outcome of the ban—one of the first public statements from a major corporation noting the geopolitical complexities of obtaining rare earth minerals for digital technology (Hatch, 2014). This rare earth “panic”, and the flood of hyperbolic journalism it produced, was the first moment in the 21st century when rare earth minerals entered the public imaginary as key material constituents of modern living (Klinger, 2017). However, rare earths have a long, dynamic, complex history, and have been active agents in the co-construction of modern living for quite some time (see Mann, 1950). This article works to bring into relief a piece of this thick, complex history, illustrating how peripheral material conditions, like rare earth mining, magnet manufacturing, and Coco’s Canine Cabana constitute crucial relations within data infrastructures, and how these relations might illuminate new ways of understanding the socio-material impacts of Big Data. The unique circumstances surrounding the opening of Coco’s Canine Cabana in Valparaiso, Indiana, provide a critical lens through which to better understand the complex peripheries and processes with which the building of Big Data is caught up.

In her “Companion Species Manifesto” Haraway (2016) asserts, “the world is a knot in motion” (p. 98), a shoelace, if you will, perpetually tied and retied in various ways. For Haraway, attending to this knot, and its ongoing knotting, is a feminist intervention, one that attempts to understand “how things work, who is in the action, what might be possible, and how worldly actors might somehow be accountable to and love each other less violently” (p. 99). At its core, the “Companion Species Manifesto” is a “kinship claim” (p. 101) about how we get on with and relate with the world and its overwhelming complexity. Haraway argues that individuals, never only themselves, are always already companions, and as such do not and cannot pre-exist their relations; rather they are constituted through them. The same can be said about Big Data, and the industrial traces left by its myriad supply peripheries. One cannot exist without the other. This article, too, is a kind of kinship claim. In Valparaiso, Big Data histories are tied up with dog histories, and threading this knot is the 3.5-inch hard drive, and the wayward journey of political economic relations that has kept it lodged firmly in the bowels of Big Data infrastructure. Through this lens, Coco’s Canine Cabana emerges as a tool to help us reconceptualize, territorialize, and reconfigure Big Data’s social, material, and canine footprint, allowing us to situate both the intended and unintended relations in these processes. What can we see when we look at Big Data from this unassuming dog kennel in Indiana? What narratives emerge?

In mapping these narratives, I draw on a litany of critical infrastructure and feminist technoscience scholars who have charted the industrial underbelly of media technologies (Bozak, 2012; Cubitt, 2017; Mattern, 2017; Maxwell and Miller, 2012), the cultural geographies of global media infrastructures (Hu, 2015; Parks, 2015; Starosielski, 2015), as well as the nuanced environmental and ecological imaginaries of media technologies (Mukherjee, 2020; Murphy, 2016; Peters, 2015). Data centers, as the most immediately visible and voluminous incarnations of The Cloud, have become sites of romantic speculation, and an increasing amount of critical scholarship (Burrington, 2014; Carruth, 2014; Hogan, 2015a; Holt and Vondreau, 2015; Johnson, 2019b; Vonderau, 2019) seeks to render new maps and classifications of the cloud’s “materialities, geographies, and logics” (Mattern, 2016). In this spirit, I ask, what is the data center’s foundational, functional unit? What, above all things, must it obtain in order to do its job? Even after all these years, the hard disk drive (NOT the solid-state drive) remains the fundamental building block of the data center, due in large part to the early popularization of the 3.5-inch Winchester form factor (which previously proliferated through most desktop tower PCs). The 3.5-inch hard disk held and continues to hold an immense manufacturing volume over competing drive sizes. As such, the 3.5-inch hard disk drive has been a central driving force in the generative material form of Big Data infrastructure because, in the simplest terms, it is the cheapest and most ubiquitous form of digital storage. But why has this narrow 3.5” path dependency remained so difficult for even the world’s largest tech companies to move beyond? And what can the stubborn ubiquity of the 3.5-inch Winchester hard disk drive illustrate about the material pasts, presents, and futures of Big Data? Answering Parikka’s (2016) call for an even harder hardware studies, this tale is replete with minerals, rocks, constructions, and destructions across local and global scales. Through the lens of data peripheries, media objects like hard drives, iPhones, and their requisite data networks, are not seen as gravitationally dense institutions around which resources accrete like planetary dust around a nascent star system—but rather as brittle, unstable, and temporary organizations of material conditions, deeply dependent on the interminable precarity of global supply chains. The social, political, and ecological conditions undergirding the availability of these technologies are perpetually unstable, and tech companies must navigate and negotiate with this ongoing precarity to hold those conditions in place. Big Data appears globally distributed, but the determinants of its material form are propped up by old nationalisms, colonial violence, and geopolitical frictions, all of which leave deep traces in our data infrastructures (Thylstrup, 2019), if only we know how and where to look.

The 3.5-Inch Winchester hard disk drive

There is nothing terribly special about 3.5 inches as a particular size of disk drive, nor does a disk drive being 3.5-inches wide make it technically superior to drives of other sizes. The 3.5-inch Winchester inherited its size from PC floppy disk drives (Brewer et al., 2016), and as such was built to fit seamlessly with PC motherboards, which, in the mid-1980s, were standardized at 3.5 inches for most mass-market desktop PCs (The Computer Chronicles, 1985). By the late 1990s, the 3.5-inch Winchester hard disk had become the most dominant and commercially available size of hard disk, outselling all others by a factor of over 100 (Disk/Trend, Inc., 1994). As a result, this standard, an old holdover from the early days of the PC industry, became repurposed by early builders of what would one day be termed Big Data. It was simply a matter of availability. Google’s first “data center” in 1999 was a 7 × 4’ closet down the hall from now-defunct search engine Alta Vista. This closet housed about 30 servers (Moscaritolo, 2014), and based on hard disk manufacturing volume data from the Disk/Trend Report (1994), all these servers were likely full of 3.5-inch Winchester hard disks. The hard disk industry grew swiftly and exponentially after 1999 as the dot com era, and the subsequent rise of social media manufactured a massive global need for data storage. In 1999, the global hard disk industry moved a total of 174.4 million units; in 2010, yearly sales had almost quadrupled to over 651 million units, almost all of them 3.5-inch drives (StorageNewsletter, 2018).

Hard drive sales peaked in 2010, and have somewhat declined since (StorageNewsletter, 2018). However, data centers continue to build around the 3.5-inch hard drive, even considering significant investments in lighter and faster solid-state (SSD) technology. While the rise of solid-state technology has cut into hard disk sales, most of the decrease is due to the replacement of hard disks by SSDs in personal computing devices like laptops. The future of SSDs in global data infrastructure, however, remains unclear. The uneven adoption of technically superior SSDs in data centers is key to understanding how the geopolitical peripheries of the 3.5-inch hard drive continue to influence data center design decisions.

In May of 2016, Western Digital closed a $19 billion acquisition of SanDisk, one of the world’s largest producers of solid-state storage solutions (Vincent, 2016). The SSD fervor that resulted in this consolidation began in 2011, when numerous companies began experimenting with SSDs to help bear the brunt of potential hard disk price hikes anticipated during the rare earth panic. In 2012, Wired reported that prominent cloud storage companies like Amazon, Dropbox, and Facebook had embarked on the transition to SSD, but in reality, this transition was rather limited. These companies began transitioning to SSDs primarily in their operational servers (servers that actively retrieve data), while cold storage servers (servers that store inactive data) continued to use 3.5-inch hard disks (Metz, 2012). As of 2020, SSD adoption in data centers still remains slow, and new server designs continue to accommodate the 3.5-inch architecture.

Facebook has purchase contracts with both Seagate and Western Digital (the two largest hard drive companies by market share), and stocks its cold storage racks with millions of 3.5-inch drives (Morgan, 2014). Solidifying their commitment to the 3.5-inch standard, in 2017 Facebook announced its next-generation storage platform called Bryce Canyon, a high-density solution built entirely around the 3.5-inch enterprise ³ hard disk. It can pack seventy-two of them into a single storage enclosure, and, says Facebook, it will increase average storage density over twenty percent (Adrian, 2017). Furthermore, in 2016, Google released a white paper encouraging disk drive manufacturers to continue prioritizing Winchester hard disks over solid state, stating that at this juncture, an industry-wide change in form factor would constitute a long and complex process (Brewer et al., 2016). Decisions like these indicate that data center architectures are deeply embedded in a palimpsest of older, entangled standards (Star, 1999). It is easier for Google and Facebook to invest in older, more standardized technology, than commit to new, more objectively efficient, less mechanical storage solutions. Thus, while solid state technology continues to make inroads, the 3.5-inch hard drive, and it’s significantly low cost-per-gigabyte, still drive a majority of the decisions around innovation, expansion, and scale of cloud storage. The data center industry continues to design and build solutions to house and accommodate the 3.5-inch form factor, despite the inefficiencies of its outdated, heavily mechanical design. As such, the hard disk drive still remains, after all this time, one of the most critical infrastructures for data storage worldwide. Yet, this has more to do with the peculiarities of its material supply chain than with the affordances of the object itself.

Rare earth metals, and specifically the NdFeB magnet, have made possible the miniaturization of electronic devices, including speakers, headphones, cell phones, televisions, computers, tablets, and hard disk drives. They consist of about 17 elements, whose unique optical, conductive, and magnetic properties make possible a generous amount of our modern technological world. They help power most of the world’s electric motors, make LED lighting possible, produce the brilliant reds on our digital screens, power solar panels and wind turbines, and have been singularly instrumental in the expansion of digital computing. As the scalability of the NdFeB magnet increased in the 1980s, hard disk drives became more versatile and quickly supplanted other forms of digital storage. The NdFeB magnet first found its way into personal computer hard disk drives in 1986, powering the Conner Peripherals CP-340 (Porter, 2005). This little magnet, and the political economy of its production and distribution, helped make hard disks (like the CP-340 and its descendants) small and cheap such that, by 1990, hard disk drives represented a multibillion-dollar market, in which the “industry [was] clearly moving toward smaller, higher capacity hard drives, particularly those made for the notebook-size computers.” (Harrington, 1990: 131).

As mentioned earlier, in the mid-1980s, the 3.5-inch Winchester hard disk drive became a popular peripheral for personal computers largely because it was the same size as standard PC floppy disk drives, which made it technically simple to swap out as an internal storage solution (see The Computer Chronicles, 1985). However, during this time, the future of hard disk storage was by no means a foregone conclusion. In 1985, Alan Shugart, founder and former CEO of Seagate (currently the world’s largest hard drive manufacturer), shrugged off hard drives as a fad that would be eclipsed by the utility and security of floppy disks (The Computer Chronicles, 1985). Shugart’s position was a popular one in 1985, as PC hard disks had not yet proven themselves as particularly fast or reliable. In the early 1980s, most saw hard disks primarily as appendages to large mainframe computers, and unnecessary for PCs. Industry insiders derided early PC hard disks, calling them “cheap and slow,” and a step backward in the evolution of magnetic storage (Disk/Trend, Inc., 1980: DT7-8). Thus, going into 1986, it was not immediately clear that hard drives would succeed as ubiquitous storage solutions.

The NdFeB-powered 3.5-inch hard disk changed things. As it became more popular throughout the late 1980s and early 1990s, the market for larger drives shrunk substantially, eventually forcing most hard disk manufacturers to abandon the manufacture of other form factors, like the previously dominant 5.25-inch hard disk (Martin, 1987: 106). The flood of 3.5-inch drives into the market eventually, over the years, drove the cost-per-gigabyte of hard disk storage down to just a few cents (Komo, 2014). Today, the 3.5-inch Winchester hard disk drive remains the most affordable mass storage option for data centers, and Seagate predicts that hard disks will retain price advantage over SSDs until at least the mid-2030s (Mellor, 2019). While the low cost-per-gigabyte can in part be attributed to successive improvements in platter density (Christensen, 1993), the enduring ubiquity of the 3.5-inch form factor has just as much to do with the complex political economy of rare earth mining and permanent magnet manufacturing. The volatile political economies of these industries profoundly affected the conditions out of which 3.5-inch hard drives emerged as the basic material building blocks for global data storage infrastructure. Thus, in order to understand the continued dominance 3.5-inch hard drive, and its relationship to Big Data, we must attend to the peripheral preconditions of rare earth mining and permanent magnet manufacturing, the trail of toxic traces left in their wake, as well as new relations sprouting from the rubble. Coco’s Canine Cabana emerged much like Tsing’s (2015) Matsusake mushroom—from the soil of capitalist ruins. The magnet industry in Valparaiso, Indiana, played an early role in the history of digital data storage, and its decay has cultivated the conditions for dogs to gather in its wake.

The neodymium–iron–boron magnet

Rare earth elements, despite their name, are not rare, as reserves exist on nearly every continent. However, their extraction and separation are labor & capital intensive, and their applications are often so hyper-specific that no functional replacements exist. The iPhone, for example, contains roughly 62 separate elements, 12 of which cannot be substituted for any other known material. As such, objects like the iPhone are some of the most comprehensive representations of the Earth’s geologic makeup ever constructed, and millions of people carry them daily in their pockets, never once considering their profound material influence. We are regularly in contact with dozens of elements that, before the 1970s, had barely left the earth’s crust, much less entered consumer life.

While rare earths have been embedded in media technologies since the late-19th century (Klinger, 2015), they gained pronounced influence in the 1960s with the development of the europium-ytterbium red phosphor, which helped popularize and standardize color television broadcasting (Walsh, 2015), and the samarium cobalt permanent magnet (SmCo), which quickly became instrumental in heavy industry and military technologies (Herbst, 1993). Shortly thereafter, the SmCo magnet found other applications in media technologies. It was used in the diaphragms of the headphones of the Sony Walkman in the late 1970s—one of the first commercial applications of rare earth permanent magnets (Popular Science, 1982). Since the 1980s, the development of rare earth applications has marched in tandem with the rapid proliferation of digital and energy technology. They are responsible for LED lights, LCD screens, touch-screen technology, modern wind power, electric car batteries, as well as the miniaturization of personal electronics, including hard disk drives (Humphries, 2013).

The NdFeB magnet emerged violently, frantically invented as a response to a spike in cobalt prices in 1978, brought on by pronounced civil unrest in Zaire (now Democratic Republic of Congo; Robinson, 1984), still the world’s primary supplier of cobalt (Frankel, 2016). That summer, communist forces from neighboring Angola invaded Zaire’s Shaba province, challenging Zairian President Mobutu Seko’s control over the region’s mining operations (Cowell, 1981). Seko wanted independence for Zaire, but the state relied heavily on the influx of western capital and skilled labor from Europe, specifically in its mining industry (Odom, 1992: 9). Once the invasion halted mining operations, the price of cobalt skyrocketed by nearly a “factor of 8” (Robinson, 1984)—yet for only a short time. By 1983, the market had once again stabilized, and so had Zaire’s mining industry. However, during that five years, government agencies and private industries worldwide scrambled to offset the monumental costs incurred from cobalt’s sudden rise. Magnequench released its NdFeB magnet to market in 1986 as a potential stable replacement for its volatile SmCo progenitor, and one that could secure a future for US magnet manufacturing which, at the time, looked bright because, until the early-1990s, the US was the world’s foremost supplier of rare earths.

From the 1960s to the late 1980s, most of the rare earth elements used worldwide were extracted from the Mountain Pass Mine, situated on the California-Nevada border, just outside of Primm (United States Geological Survey, 2002), which conveniently provided an already-accessible standing reserve of raw resources for the ramping up of rare earth production after the turmoil in Africa in the late 1970s. Yet the fortunes of the Mountain Pass Mine turned in the 1990s due to a web of economic, political, and environmental factors that coincided with the shifting forms of both the permanent magnet and personal computer industries. In late June of 1995, GM announced the sale of its Magnequench operation for $70,000,000 to the Sextant Group, an American investment firm, and two Chinese materials companies: San Huan New Material High Technology Inc., and the China National Non-Ferrous Metals Import and Export Corporation (Greising, 2005; Koenig, 1995). This buyout, supported by both the Clinton administration in the US, and the Chinese government, allowed China to move up the rare earth value-added chain, affording them the capability to not only mine their own rare earth materials, but also produce NdFeB magnets, as opposed to simply exporting material to Japanese manufacturers (Kiggins, 2015: 9). By 1995, principally driven by hard disk drive sales, rare earth permanent magnets had grown to a nearly a billion-dollar industry, 77% of which was dominated by Japanese manufacturers (Fastenau and Loenen, 1996). The movement of Magnequench to China significantly altered this balance.

Access to cheaper rare earth materials from China coaxed American magnet firms away from the more expensive product flowing out of Mountain Pass. These firms followed in the footsteps of microchip and hard drive manufacturers, who were some of the first industries to outsource production to Asia in the 1970s and 1980s. Between 1988 and 1998, China’s market share of NdFeB production increased from 14.4% to 40% (Goldman, 2014: 152), and kept rising. While China currently controls about 95% of the rare earth mining, processing, and value-added manufacturing industries, this control has never been traditionally hierarchical. The rare earth market in China in the 1990s and early 2000s was very lightly regulated, and these loose regulations kept prices low, which increased global demand. Tsing (2005) would term this a resource frontier, a boundary condition at which awkward collisions manifest in the proliferation of “human subjects and natural objects” (p. 30). Yet, while the concept of frontier often connotes a sort of beginning, a time and space prior to standardization and ordering, frontier conditions and imaginaries continue to persist in the unintended outsides of rare earth spaces. While China currently maintains state control of its rare earth mines, unsanctioned mining and mineral trading flourishes, taking advantage of export controls and domestic discounts to keep prices low for various international customers. So common are “illegally” traded rare earths, that industry journals sometimes index their prices (Abraham, 2015: 104). At the same time, speculation on potential rare earth projects outside of China (which rarely materialize) has become its own micro-industry. Less than a year after the panic in 2010, there were over 400 proposed projects, a vast majority of which raised capital, and subsequently vanished (Klinger, 2017).

Throughout the 1990s and early 2000s, China had unofficially allowed unsanctioned mining in order to drive growth, which decentralized most of the industry, making more recent attempts to increase oversight difficult. Smuggling of rare earth material became a common and very lucrative practice (Wübbeke, 2015: 23–24). Because of this, Magnequench spent the 1990s and early 2000s battling manufacturers, importers, and third-party buyers in court, attempting to retain control of its product. Through a series of lawsuits and trade complaints filed in the late-1990s and early-2000s, it becomes clear how profoundly the peripheral resource frontiers of rare earth mining and permanent magnet manufacturing affected the early computer industry, as a number of prominent tech manufacturers and retailers weave in and out of these cases, revealing deep, inter-industry dependencies.

The extent and complexity of these dependencies are often obscured by what Tsing (2009) calls niche-segregation—a practice of responsibility abdication in supply chain capitalism. In this practice, supply chains function as linkages of borderline unrelated niches, where each niche operates at the “edge of economic sustainability … and … legitimacy” (p. 172). These niches exist alone, as their own centers, but also as peripheral operators in a larger, semi-linked process we have come to call a supply chain. Supply chain histories are difficult to tell because often, the various segregated niches are not immediately visible to one another, nor to researchers, and additionally, tech companies tend not to archive purchase records. However, in the case of the rare earth magnet industry in the 1990s and early 2000s, patent infringement accusations were incredibly common and, as a result, through the litany of court documents attached to these accusations, a more cohesive narrative of the deep connections between hard drives, magnets, and rare earths can be drawn out. These cases also highlight just how unsettled supply chains often are, and how infrastructural power is constantly contested.

Shortly after the Magnequench sale in 1995, the recipe for its patented magnetic powder leaked throughout China. “Counterfeit” NdFeB magnets then began flooding into the United States, and importers from all over the country took advantage of the surplus. In response, in 1996 and 1997, Magnequench filed patent infringement suits against three separate companies (YBM Magnex, Inc., Pacific Century Enterprises, and Polymag, Inc.), accusing them of stealing the formula for their bonded magnetic powder (United States International Trade Commission, 1998). Then, in 1998, Magnequench, along with Sumitomo, the world’s second-largest rare earth magnet manufacturer, filed a joint complaint to the International Trade Commission, stating that both of their proprietary NdFeB recipes had been infringed upon, and that US importers were knowingly benefitting from this infraction (United States International Trade Commission, 1998). It is difficult to determine exactly how many importers dealt in NdFeB magnets in the late 1990s, but the complaint to the International Trade Commission lists seven US companies and one Taiwanese importer. The complaint sought damages for violations of six individual patents, illustrating the volatility of the rare earth value-added manufacturing chain, as well as the speed with which this industry was growing.

It is crucial to note, in the mid-1990s, the rapidly expanding hard disk drive industry was by far the largest recipient of these counterfeit materials, as, at that time, hard drives were the number one application of NdFeB magnets, accounting for 55% of the market (Bloomberg News, 2004; Fastenau and Loenen, 1996). Because of this, Magnequench did not merely go after the direct importers. In 2001, Magnequench filed lawsuits in district courts in New York and Indianapolis against a slew of electronics and computer companies who had allegedly obtained magnets from at least one of the importers listed in the ITC complaint, including Acer America Corp., Circuit City Stores, CompUSA, Compaq, Philips, Toshiba, Sony, and Hewlett Packard (Bloomberg News, 2004). In a quintessential evocation of the ideology of niche-segregation (Tsing, 2009), a representative from Compaq proclaimed that, while magnetic technology “is deeply embedded in components that we purchase from other companies … we strongly believe[s] that the responsibility lies with the component suppliers, and not with the end-user equipment makers” (Quan, 2001). Despite this appeal, Magnequench won the ITC complaint in 1999, which theoretically should have heralded a halting of “counterfeit” imports into the United States.

However, Magnequench had trouble maintaining a hold on its patent. In 2004, the company filed yet another patent infringement suit—this time against Microsoft and Phillips. This suit alleged that Microsoft used counterfeit magnets in the hard drives of its XBOX, and Philips in its CD-RW optical drives, implying that “counterfeit” Magnequench magnets continued to proliferate through the United States well after the 1999 ITC verdict (Magnequench v. Microsoft, 2004). Magnequench understood that data storage applications represented the core of their market and attacked electronics companies to protect market share against the increasing demand of the computer industry. The demand for hard disk storage had expanded exponentially between 1992 and 1999, from 10¹⁶–10¹⁸ bytes per year (Coughlin, 2012). These hard disks required NdFeB magnets, and Magnequench’s sale in 1995 allowed for the scaling-up of operations due to the constant availability of under-market rare earth minerals, immediately benefitting hard drive manufacturers. These counterfeiting accusations illustrate how political economic relationships come to be defined by their unintended outsides. These are data peripheries. All of these entanglements may seem peripheral to the hard drive as a technical object, but they are absolutely central to the hard drive as a process.

Rare earth mobilities

An estimate in 2013 placed China’s share of the rare earth market (including the refining and processing market) at 97% (vs. 46% in 1994) and cites rare earth production in the US as effectively zero (Mancheri, 2013: 9). Many have challenged this monopoly, but the path dependencies have become so deep that, even with significant disruption, most have failed. During the rare earth panic, the price of neodymium skyrocketed from $19/kg to $244/kg, an almost 13-fold price increase (Hruska, 2012)—an echo, in some ways, of the 1978 Cobalt crisis in Zaire. This sent shockwaves through the international business community, and led to hyperbolic, and largely false speculation that China was hoarding supply, and seeking to manipulate the market (Reuters, 2012). This was also the first time in recent history that rare earth prices directly impacted end-user products, and in early 2011, both Seagate and Western Digital began to disclose rare earth precarity in their quarterly SEC reports (Hatch, 2014), which were some of the first investor-facing statements in the electronics industry dealing with rare earth supply problems.

The 2011 rare earth price shock did not, after all that, dramatically affect end-user hard disk prices (Wubbeke, 2015: 35), but it did produce a terrific amount of political anxiety, which served to build and cement binary narratives about China’s alleged control over the rare earth industry. These narratives, in turn, galvanized multiple efforts in the US to regain supply security. Some undertook new magnet research. The University of Minnesota began working to develop iron nitride magnets—permanent magnets with reportedly twice the maximum energy product of NdFeB—citing China’s export restriction as the reason for developing them (ARPA-E, 2011). These magnets have yet to reach market. The export restriction also prompted the staggered resurgence of US rare earth mining. Molycorp sought to reopen the Mountain Pass Mine, and officially began new operations in 2012. Molycorp also purchased Magnequench in a further attempt to consolidate a US rare earth supply chain, even though all of Magnequench’s operations remained in China (Howe, 2012). In 2012, many lauded the future of rare earths in the United States, and the tech industry shared this optimism. In Wired, mining analyst John Kaiser said, “In five years there will be rare earths produced all over the world and China will lose its edge” (Venton, 2012). The possibility for a dramatic reorientation of the rare earth market was heralded in 2012, and Molycorp’s purchase of Magnequench significantly raised expectations of the US gaining ground in the rare earth mining and value-added manufacturing industries. At least in the media narrative, Molycorp appeared poised to rebuild the profitable Mountain Pass/Magnequench relationship it had enjoyed so briefly in the early 1990s.

However, very little of this came to pass. As with cobalt shortly after the crisis in 1978, rare earth prices eventually returned to pre-restriction levels. But unlike with the cobalt crisis, the material supply chains for permanent magnets have not shifted all that much. The sudden price deflation dramatically impacted Molycorp’s efforts to revitalize Mountain Pass, eliminating almost all prospects of future profits. In March 2015, Molycorp announced that it likely could not keep Mountain Pass in operation, given these changing conditions; In June of that year, the company filed for Chapter 11 bankruptcy (Jamasmie, 2015) and shuttered Mountain Pass. Molycorp emerged from bankruptcy in 2016 as Neo Performance Materials and continues to operate Magnequench. Yet, despite the executive offices in Canada, all processing operations remain in China. The company’s hope to reinstate the US as a worthy competitor in the rare earth trade vanished.

Undeterred, the US continues to manufacture politically expedient narratives about rare earth security that continue to inform supply and manufacturing conditions for data storage today. These largely binary narratives have long remained central tools for industry leaders and policy makers, often rendering rare earth speculation far more profitable than material production. The U.S. Department of Energy currently has multiple rare earth projects in development, but a majority of the processing and value-added manufacturing capabilities remain firmly in China (Johnson and Gramer, 2020). Additionally, instead of focusing on developing proven reserves, a considerable amount of energy and funding is being directed at efforts to extract rare earth minerals from coal and coal byproducts (Jackson, 2019; Simeone et al., 2018)—all under the banner of US rare earth security. Former Secretary of Energy Rick Perry even made a special trip to Hazelton, Pennsylvania in September of 2017 to announce that coal country will soon supply critical rare earth minerals for digital and defense technology. However, nearly three years later, these efforts remain speculative, and illustrate how much of rare earth scarcity is structural, rather than actual (Klinger, 2017). Indeed, rare earths, and their alleged scarcity, are less a material reality than a story told over and over again, adapted ever so slightly to reify and reinforce existing fears, anxieties, and political desires. Over the years, this story has reverberated through the electronics industry in ways both intended and unintended, entangling mineral, magnet, data, and dog futures tightly together.

Coco’s canine cabana and big data futures

Since 2010, rare earths (and the security paranoia induced by their supposed scarcity) have remained a steady constant in the public imaginary. In addition to anxieties about critical materials security, rare earths are now often associated with profound ecological destruction—the Baotou region of Inner Mongolia acting as the poisonous rare earth poster child. This region houses the world’s largest open pit rare earth mine, and due to its presence, nearby villages have been poisoned and depopulated, and former farmers have been reduced to selling radioactive waste they collect on their property to nearby reprocessing plants (Bontron, 2012; Maughm, 2015). Baotou’s toxic lakes have become some of the most recognizable, and most cited visual markers of the ecological consequences of Big Data’s peripheries (see Davies & Young, 2014; Maughm, 2015). These recent optics have caused some tech companies to consider reshuffling their supply chains to avoid association with ecological horrors. Last year, in response to both these environmental revelations, and the supply concerns exacerbated by the rare earth panic of 2010, Apple announced intentions to strike rare earth mineral mining from its product supply chain, and instead rely entirely on recycled sources (Schlanger, 2017). The company’s decision to eliminate mined rare earths gives credence to the rising collective consciousness of mining’s socio-environmental damages, as well as to the geopolitical uncertainty produced by China’s veritable monopoly on rare earth mining and processing. While this theoretically means that Apple’s billions of iPhones, iPads, and Macbooks will use only recycled rare earth, this decision actually has nothing to do with its data centers or hard drives. Furthermore, Apple, as well as Google, apparently now power all their data centers using 100% renewable energy (Marcacci, 2018). But this, also, has nothing to do with hard drives. These claims of sustainable supply chains and dedication to renewable energy solutions are largely constructed imaginaries that rely on the relative opacity of the hard disk drive industry. Neither Apple nor Google manufacture hard drives. Only three manufacturers currently service the world’s cloud storage needs (Seagate, Western Digital, and to a far lesser extent, Toshiba), and, per the latest SEC reports, all of these companies continue to rely on mined rare earth minerals to build their magnetic motors. In order to justly imagine digital and sustainable futures, we must break apart these supposed black boxes, and attend to their wayward data peripheries.

Coco’s Canine Cabana, and its owner, Kathy Defries, garnered minor attention in 2010 during China’s rare earth export restriction, playing the part of yet another rust-belt manufacturing casualty to China. An exposé in Bloomberg News (Robison and Ratnam, 2010) used Coco’s as a lens for creeping anxieties over both US outsourcing and materials security. According to this article, the Pentagon had been “asleep at the wheel” as China swooped in and scooped up the US rare earth magnet industry, leaving nothing but industrial ruin and doggy daycares in its wake, framing Coco’s as symptomatic of a sort of failure of US ambition and foresight. But is Coco’s Canine Cabana merely a local curiosity randomly caught up in global processes, or is it something more, something altogether essential? If we look beyond the snowballing economic and political anxieties, Coco’s can open new ways of thinking about the relationship between us and our data, as one not of security and fear, but one that might point toward new epistemologies and imaginaries of what it takes, and what it should take, to exist justly in a datafied world (Milan and Treré, 2019). Dogs, after all, as Haraway (2016) asserts, should be central concerns to those in technoscience studies, and, I argue, should also be central concerns for those looking to understand the rhizomatic creep of data infrastructures. Holding a puppy in her arm, Kathy DeFries notes, “When things got slow for manufacturing, we had this big empty shop floor … It’s a great stress reliever” (Robison and Ratnam, 2010). For workers at 405 Elm St., dogs and heavy industry exist side by side, informing each other, growing each other. In fact, in 2014, the Open Arms Animal Rescue opened next door to Coco’s (O’Leary, 2016), helping to further transform this former manufacturing campus into a haven for animals. Where once magnets made their way into smart bombs and hard drives, now dogs are bathed, loved, and cared for. These dogs, products of magnet histories, weave, too, through hard drive histories. They may frolic on the peripheries, but these peripheries are where the conditions for magnets and hard drives were and continue to be realized.

Instead of a business or technical history, this article has attempted to tell an alternative material history of the 3.5-inch hard disk drive—a history from the fringes, the peripheries, where processes are rarely settled, and forms rarely concretized. It has leveraged a vast amount of information that neither fits neatly together, nor forms a cohesive whole, in large part because nothing about our relationship to digital technology is stable, cohesive, or whole. Much more work is needed to tease out the tangled mess of relations touched on in this article. The field needs more ethnographic work on tech manufacturing supply chains, as well as more feminist and decolonial scholarship (like Nakamura’s (2014) work on microchip manufacturing by indigenous women) of how diverse manufacturing regimes within and without the Global South (e.g., rare earth mining, magnet manufacturing, and hard drive construction) influence, build, and maintain Big Data imaginaries. The 3.5-inch hard disk drive—this little aluminum box, stuffed with magnets, steeped in politics—has exerted tremendous structural force over the form and function of Big Data infrastructures. Seeing Big Data through the hard drive situates data centers as temporally bounded structures, maintained through the ubiquitous availability of 3.5-inch boxes. Data centers are boxes too, boxes through which smaller boxes constantly move, bound for elsewhere. Data centers, then, are only temporary concrescences of data peripheries. To better understand the material conditions of our digital world, we need to explore the awkward mobilities of these peripheries—and not only in landfills, mines, and factories, as so much of the supply chain exists outside of itself. We, too, must pay heed to the unintended outsides, like Coco’s Canine Cabana. How can we start thinking about Big Data infrastructures in terms of their peripheral dog kennels? What does Big Data look like when seen from these spaces? What new relations become visible? Beyond just the supply chain for the data center, we must fashion detailed maps of its data peripheries, especially the ones that never quite make it back. They are out there, with us, always, these unintended outsides. We depend on these relations—the faded ones, the ragged ones, the peripheries with no centers. To that end, to really know our data and its material futures, we must also know Coco’s Canine Cabana—not only that it is a product of Big Data, of hard drives and their magnetic histories—but also that it is there for us, as a safe space for our furry companions, should we ever pass through Indiana, in need of a dog sitter.