The Stack On Software and Sovereignty

My argument then is not another prophecy of the declining state withering away into the realm of pure network, but to the contrary, that the state ’ s own pressing redefinition takes place in relation to network geographies that it can neither contain nor be contained by. As cloud-based computing platforms of various scales and complexities come to absorb more and more social and economic media, and do so on a planetary scale, the threads linking one data object to one jurisdiction bound to one geographic location become that much more unraveled. It ’ s not that the state cannot follow those threads, rather that when it does it takes leave and becomes something else. Other exemplary exceptional territories, such as tax-free Special Economic Zones, also mark the visible effects of network globalization in general, but the underlying function of emergent Cloud systems to enable these demands special consideration, especially at the level of the individual User. The rights and conditions of citizenship that were to whatever degree guaranteed by the linking of information, jurisdiction, and physical location, all within the interior view of the state, now give way perhaps to the riskier prospects of a Google Grossraum, in which and for which the terms of ultimate political constitution are anything but understood.

For example, we know that the modern state is also bound to a particular set of protocological interests and generative legal codes and that the membranes of their jurisdictions are defined by them, but the transposition of juridical media from legislation to computation produces a rather different set of ground rules. Certain political positions are built already into the hardware, not allegorically but literally. For example, for many processor chips from the last few decades, “ core user ” is a sovereign figure who can generate subordinate administrative subjects — who in turn can control the calculative access of other users. ²⁰ What law, passed where, could undo this polity, based not on formal constitution or really even the perfunctory end user agreement? As we have seen, states respond forcefully, sometimes by attempting to directly compete with these kinds of sovereignty and sometimes by absorbing them as their own new norms.

114

Cloud Layer

At the scale of agonistic geopolitics, the variance in strategies for governing sovereign space, between states, platforms, and their combinations, seems to do more to multiply the variables than to reduce them toward consensual maps. For the emblematic Sino-Google conflict, the productive friction is also the result of a superimposition of two incommensurate logics of territory and governance: one is a cognitive capitalist proto- polis predicated on universal information rationalization, and the other is a geographically circumscribed central command that sees the Cloud as an extension of the body and charge of the national state. In this case, the two are like unmerged layers of an image pasted onto the same canvas, each a different shape and scale, but each demanding the recognition and acquiescence of the other. In the messy practices of billion- User scale technopolitics, even this arch opposition cannot itself hold. What if instead of simply staying put or leaving China, the Google Grossraum instead evolves over time and blankets or pours down on Chinese sovereign territory in such a way that the state would be as powerless to enforce policy on it as it is on the rain? ²¹ And if we imagine this circumnavigation to be well subscribed by Chinese publics, then how divided are their loyalties and how will they identify with dual citizenships such as these? What do states become in relation to this perforation when the circumscription of land guarantees that much less? For party planners, nightmares do such things make. ²² Others, looking back from that virtual future in which Google (or some similar Cloud platform) more fully connects China ’ s continent of Users with or without the formal acknowledgment of the Chinese government, may have more fondness for a Cloud Polis in which multiple User identities compete, conflict, and overlay without also congealing into a tidy individual User citizen-subject. That is, the superimposition of dissensual sovereignties applies to the people as well. Toward a design of such systems, what are the real limit conditions of the Cloud layer as an actual machine?

1. Cloud Infrastructure

What is the Cloud, exactly, if understood as a vast discontiguous apparatus? Like The Stack as a whole, the Cloud layer is itself a terraforming project, covering the globe in subterranean wires and switches and overhead satellite arrays, simultaneously centralizing and decentralizing computing and data storage and the social relations that depend on them (and vice versa). Cloud infrastructure ’ s energy and water appetite demands specific landscape accommodations, including putting data centers in unexpected locations (inside office towers, ice caves, underground bunkers, coal mines, carved-out mountains, dead malls, inhospitable islands, old churches) all in the name of strategies both rational and strange (cheap energy, cheap space, proximity to oceanic passage, lax regulation on data storage, earthquake and flood avoidance, perimeter security, ideal temperature control). ²³ Above ground, Cloud infrastructure is serviced by torqued logistics networks of warehouses, roads, factories, containers, ports of call, airports, and package routing hubs that together allow it to route physical objects in the way it routes data packets. In doing so, the layer absorbs metadata about each thing and each relation of exchange, all translated by algorithmic militations of cognition, data semantics, manufacture, demand, and response optimization as it links these across continents, compressing and expanding economic rhythms and cycles.

Cloud Layer

115

Those Cloud economics are predicated on efficiencies of centralization made possible by simultaneous and exponential price drops in data computation, storage, and transmission, which are driven in turn by increases in the amount of data produced by the billions of devices that continuously pump information into these platform services. ²⁴ This virtuous cycle of devices in the field uploading data to cheap central storage, for processing in datacenters and redistribution to other users and devices, fuels the rapid centripetal centralization of information toward a small number of global Cloud platforms capable of supporting and leveraging service infrastructures at such a massive scale. At the edges of this network, the merger of data processing, storage, and distribution into a composite platform makes possible the generic provision of computational capacity into an on-tap service utility for large organizations and individual Users (e.g., people, sensors, devices). In many cases, individual applications are provided free of immediate charge and monetized through the capitalization of User cognition by advertising. In other cases, Cloud application interfaces charge based on per unit of use as defined by different combinations of number of Users and total amount of data processed, stored, and transmitted. The ultimate effect of the centralization of Cloud services, paired with faster wired and wireless bandwidth to end devices, is to link almost any person, thing, animal, building, car, plant, or smart mote into supercomputing platforms. This allows each such end User to perform or contribute to otherwise impossible calculations, all of which further centralize the platform ’ s economy of uses, effects, and profits. Access to live data and peer-to-peer communication streams means that aggregate responses to individual interactions can be calculated in remote data centers and relayed back to the User almost instantaneously, as if the computation and all the data it is based on were happening in the palm of her hand. Dumb devices are infused with seemingly cosmic omnipotence. Servicing generic supercomputation draws any thing and any place into a formal interrelation and interdependency, flattening both hierarchies and nuances. Today ’ s end user Cloud apps will grow in complexity and capability only as exponentially larger pools of data are aggregated, analyzed, optimized, and made available through networks of application programming interfaces (APIs) with which developers can link oceanic reservoirs of information into smart services that learn as they are put to greater use. Cloud services will include greater artificial intelligence streamed into graphical, voice, or object interfaces with which we interact in naturalistic ways. Most likely, however, they won ’ t do this primarily over the open Internet as we know it, but through more narrowly designed and owned networks in which competitive advantages (drawn from physical infrastructure to data optimization to energy efficiency to format lock-in) will drive and delimit everyday computational economics.

116

Cloud Layer

Another reason the open Internet may provide a diminished fraction of the load for next-generation Cloud services is that it is getting full. There is a finite amount of information that can be pumped through the network as it is currently constructed, and that limit may be appearing on the near horizon. In order to ensure that future services can stay online and continue to scale, Cloud platforms will likely continue to build their own secondary and tertiary versions on top of or in place of the “ old Internet, ” using both open and closed protocols to consolidate proprietary publics and networks. How full are the physical channels of the Cloud layer already? While it is extraordinarily difficult to exactly map the scope and shape of the Internet, as so many avenues are closed off from scrutiny, ²⁵ it is nevertheless generally agreed that the total amount of data produced and transmitted doubles every one or two years (perhaps faster). A recent study estimated that in 2018, 6.5 zettabytes of data (trillions of gigabytes) will pass through the Cloud, with 31% in the public Cloud and 69% in the private Cloud. ²⁶ The increase is seen as a function of faster processors in end user devices, which can produce more data and feed then into common platforms, but the growth rate of shared data is increasing faster than the spectral efficiency of the optical fiber networks over which it passes can accommodate it. Even techniques like multiplex polarization will probably not suffice in the long term. ²⁷ We may like to think of fiber optics as providing functionally limitless bandwidth, but at a zettabyte order of magnitude, for a planetary Cloud ultimately linking trillions of devices, this is not true. Unless new technologies, such as widespread multicore fiber, are widely deployed in the next decade, the physical limits of how much information can actually pass through a given channel may introduce new economies of bandwidth scarcity, prioritization, and pricing. ²⁸ At the same time, newer networks with greater carrying capacity, as well as faster signal throughput, may provide new kinds of Cloud services previously only imagined, such as holodeck-quality virtual environments. Such accomplishments may steer primate evolution toward shared intersubjective experience, introducing fantastic new genres of narrative, design, architecture, poetry, medicine, and music at a planetary scale, or it may allow a select few to watch 8K LOLcat videos from 10 angles at once.

Cloud Layer

117

The pressure of that physical limit also pushes against the geoeconomics of the Internet backbone and private Cloud platforms, forcing how their development impacts on political geography as a whole. Instead of thinking of the Cloud as a bunch of individual private computers connected to data centers by big public pipe, it is perhaps better to think of any (at least partially) gated Cloud network as a single vast discontiguous computer, linking servers to browsers and back again, such that functions that may have once happened “ inside ” any one device now happens on the device ’ s “ out side, ” now on and in the network itself. Information input, storage, processing, display, and so on are handled by different components in the wider computer architecture, but when very fast optical links connect devices, supercomputers, dense populations, server farms, and enterprise clusters, they can delink and distribute that machine archi tecture, spreading it even across oceans. ²⁹ But in practice, the Cloud (and any one Cloud platform) is much more than one shared network computer. Different actors (e.g., tel cos, states, standards bodies, hardware original equipment manufacturers, and cloud software platforms) all play different roles and control hardware and software appli cations in different ways and toward different ends. Internet backbone is generally provided and shared by tier 1 bandwidth providers (such as telcos), but one key trend is for very large platforms, such as Google, to bypass other actors and architect complete end-to-end networks, from browser, to fiber, to data center, such that information retrieval, composition, and analysis are consolidated and optimized on private loops. Consider that if Google ’ s own networks, both internal and external, were compared to others, they would represent one of the largest Internet service providers in the world, and by the time this sentence is published, they may very well be the largest. Google indexes the public Internet and mirrors as much of it as possible on its own servers so that it can serve search results and popular pages quickly to Users, regardless of where the original page may originally be coded, sourced, and hosted. It also maintains largescale networks for the distribution of its own User-facing content, such as YouTube videos, Gmail, and Search, and for the shuttling of the incredibly large “ copies ” of the Internet from one Google data center to another around the world. We could also then think of Google, the integrated platform architecture, as a kind of metacomputer that absorbs and processes data so as to maximize its own growth. When I log in to Gmail, my laptop becomes, for a moment, a peripheral device of that larger Google metamachine spread across the world. This ubiquity affects how data traffic itself is composed and managed. For example, software-defined networking (SDN) allows a network to prioritize, plan, track, and optimize the ideal routes of sets of packets in a way that is more sophisticated, composable, and responsive but less neutral and less dependent on political decisions made elsewhere. ³⁰ In the long run, the investment and expertise in developing the infrastructure necessary to maintain the world ’ s largest consumer Cloud platforms may trend away from networks on which packets are sorted and routed in neutral, but not always reliable, pathways across shared pipes, and toward very large data providers and Cloud platforms, such as Google, or toll road developers, such as Comcast, controlling traffic over their own networks for their particular purposes. The future maturation of global Cloud platforms as quasi-independent geopolitical domains may be determined to a significant degree by their ability to deploy their own competitively effective, end-to-end cyberinfrastructures. ³¹ This will redraw the practical boundaries between open and curated computational services, public and private data, state and corporate geography, User and service, and Interface and infrastructure. The end result supports more ubiquitous, finer-grained supercomputation, but deals a severe blow to the social ideal of general-purpose computing, as hardware devices would also likely be further tuned to suit specific platform strategies, perhaps locking out others (see the discussion of the Apple Cloud Polis below). As we already see for financial networks, general Cloud-based geopolitical enclaves may appear as indirect functions of exclusive networks, monetized by providing premium services to those that can pay, including many that are traditionally managed by states, leaving other publics to wrestle over thirty-one flavors of net neutrality. As a global trend, this mode of computational production also bears the potential for a much more stratified mesh of enclaved Cloud experiences and a multitiered, more socially asymmetrical Cloud that provides premium services for some (including privacy , the ultimate Cloud service) and bare-bones services for others (their identities and interactions more exposed to the elements and to exploitation.) Still, the most pressing geopolitical design problems are those posed by what Cloud platforms already are and do. By examining some of our dominant platforms more closely — reading and interpreting them — we can perhaps predict what they will become, but, more important, perhaps steer their becoming as well.

118

Cloud Layer

1. Cloud Polis

In tracing the outlines of our current Cloud-based polities, we see that platforms have assumed and absorbed several core political functions of the modern state and innovated new modes of governance besides. Cartography , for example, has long been a core competence of states, particularly when they are busy expanding their empires, naming locations, planting flags in them, and rendering these on their own maps for the first time. Throughout the colonial era of globalization, the basic description of a “ new ” territory over which a state might wish to internalize into its jurisdiction, and over which it contests control, may be legally precedent to whatever private strategies are subsequently put in motion, or they may come later as decorative legitimation of earlier speculative adventures. ³² Either way, sovereigns have long claimed and enforced the right to name and objectively represent the territories they govern. Today contested lines on maps still symbolize geopolitical tensions over sovereign influence, as the many intrigues over islands in the South China Sea attest. Consider then the curious episode in 2010 when Google Maps slightly shifted the line marking the border between Nicaragua and Costa Rica. Troops were summoned and war over the ambiguous territory seemed possible. The naming and measuring of the ground over which and into which politics might maneuver was, however unintentionally, remade not by either of these states but by a Californian software company. The situation was defused, but a new precedent was made. In the US context, compare this to Lewis and Clark ’ s westward excursion, or the cartographic corps sent out during the New Deal to expand and reinforce federal territorial vision and the right of final say-so over what parcel is where. More recently, Google made news by formally recognizing Palestine as a delineated nation, well ahead of its eventual full status at the United Nations, ³³ and between the two institutions, it ’ s not altogether obvious which recognition will prove with the benefit of historical hindsight to be the more efficacious. These episodes hint at a potential swap in the relative positions between the cloud and the states in the definition of a last-instance geography that is now also described to states by Cloud platforms instead of the other way around. The ambiguities of the rotation in priority, and perhaps of which institution finally underwrites which, persist in differing models of Cloud Polis. For one, the state shape-shifts into a Cloud platform, and for the other, Cloud platforms functionally displace states by assuming their functions.

Cloud Layer

119

States move into the Cloud. We see this in guises from the benign to the sinister, from progressive gestures of information democratization like data.gov to the troubling surveillance armatures associated with the National Security Agency, the Patriot Act, and superjurisdiction over international Cloud data. ³⁴ From the prosaic gathering and reportage of information to its citizens, on the one hand, to the clandestine hoarding of massive troves of data about its citizens (and other countries ’ citizens), on the other, states in the cloud transform what states can see, and what “ seeing like a state ” (and listening and sensing like a state) actually entails. ³⁵ With this shift in perceptual techniques, habits, and dispositions, states are inspired to look for some things and not others. The work of contemporary governance is transformed toward the management of multiple, irregular, asymmetric layers of Cloud platforms, and as the “ eyes ” of the state evolve, its bones and blood will follow. ³⁶

Much has been made in recent years about a reinvention of government through the more intensive and intelligent strategizing of its institutions as public data-generating, -sorting, and -disseminating systems, and their optimization along the lines of any other large peer-to-peer information service. The sunny ethos of “ government as a platform ” imagines Users accessing democratic states in the same way that a private Cloud platform allows Users to retrieve and program structured data for their own secondary purposes. ³⁷ It sees platforms spawning new clients, developing economies and microtargeted utilities, and it asks then, “ If states are conceived as platforms, wouldn ’ t they work the same? ” One version of this presumes a general inversion of government ’ s data-capturing mission now outward toward the exterior-facing world of active constituents. This might lead to a greater modularization and even outsourcing of governmental services to constituents interested in providing and innovating these. ³⁸ However, the PRISM/Snowden affair and the general revelation of the core role that consultants like Booz-Allen Hamilton and Palantir play in the federal information complex demonstrate that this disintermediation of the work of the Cloud-based state is not only well underway, and that the more individualistic-entrepreneurial story of “ government as a platform, ” as told by publisher, Tim O ’ Reilly and others, is at most a supplemental augmentation of a well-established apparatus. In fact, one shudders at the thought of what a more individually distributed subcontracting of state Cloud surveillance might look like. At the same time, the most far-reaching visions of the open government movement might still represent a transformation of the state itself into something unknown — perhaps an apps-based queryable megamachine like Planetary Skin or Google Earth, but with obvious and significant differences. ³⁹ The optical positions of a state — how it sees the world and its constituents and how its citizens see themselves reflected through the ambient quantitative commons — might bear all the benefits and bankruptcies of earlier forms of communicative reason. As these try to reform technocratic media into humanist channels, they are not necessarily predicated on an overcentralization of expertise; rather, we see how some core mechanisms of governance on the ground become less like a central command machine and more like an ambient, generalized utility at hand for anyone interested in parsing the databases and spreadsheets and deploying them toward new designs. Here the maximal state and

120

Cloud Layer

Clouds become de facto states. “ Platform as governance, ” the inverse of “ government as of platform, ” is suggested just as strongly by today ’ s Cloud Polis. The structures and limits of a global Cloud platform cannot not reorient the contexts in which the fragility of state sovereignty is contested. Like charter cities, we see specific Cloud platforms absorbing forms of sovereign differentiation between people and places that used to be the exclusive domain of the state and can only imagine what kinds of “ sovereignty services ” this may lead to. How will the Cloud-based delineation (or absence of delineation) of land, identity, energy, value, territorial interiority and exteriority, and so on, come to mutate the overlapping layers of everyday law and life? It is less that Cloud platforms ultimately replace states than that two domains become dramatically less distinct from one another, interlacing and folding up in new ways, producing emergent institutional forms not reducible to the direct combination of the two. The ability of the Cloud to achieve this is based not only on the networking of information between locations, but on its interdependent position in the larger Stack, linking energy economies, Cities, Interfaces, and Users into a global but wholly uneven platform. That governing logic, privileging the Cloud layer as timekeeper and space maker, is built into the technical architecture of The Stack and imposes itself on states and traditional geopolitics as much as the inverse. ⁴¹ That is, Cloud platforms not only have geopolitical ramifications and implications; they are a geopolitical condition and constitution in their own right. Cloud Polis is populated by hybrid new geographies, new governmentalities, awkward jurisdictions, new regimes of interfaciality, and so new (and old) imaginary communities, group allegiances, ad hoc patriotisms, and inviolable brand loyalties will inevitably follow. Some of these may be archaic fundamentalisms now given a new mission; others are more novel and progressive, or soupy mixtures both futuristic and atavistic at once. The assignment claimed by planetary-scale computation is then not only to challenge the state ’ s monopoly on legitimate violence (the force of material as well as of material force) but also its monopoly on legitimate citizenships. As discussed, the fate of every location, and person and User, is overlayered with multiple, asymmetric, and irreconcilable platform allegiances, rights bequeathed and values extracted. That layered patchwork is the result of both the global circulations of “ citizens, ” understood as a mixture of people and things, as well as of the fractured ability of any one imagined community to incorporate the lives of its adherents when it must share space (and Users) with multiple simultaneous alternatives. ⁴²

Cloud Layer

121

What is the proper role and ambition for this mode of Cloud Polis? How should we imagine the role of a geopolitical actor with the future scale, program, and capacity of, say, Google or Apple or Alibaba, as well as the smaller players that will fill out the space cleared by similar enterprises? First, the space over which Google ’ s (limited) capacity to generate or mediate proto-citizenship and economic sovereignty is neither comprehensive nor properly pluralistic. It is derived not from some classical constitutional claims, but from the mission statement to maintain the everyday nuts and bolts of planetary computation ( “ to organize the world ’ s information, and to make it universally accessible and useful ” ) and from this the presumption of an historical mission. In the administration of these global input and output interfaces, a kind of governance over the networks and territories mediated by them does emerge, Cloud Polis taking shape not through master plan but through accumulated microtechniques of regularization. By contrast, if the modern capitol building was an architecture that symbolized a legal, jurisdictional, and geographic center from which the terms and limits of sovereign citizenship might radiate concentrically, then the data center located out of site or even in the extraterritorial, un-national plane of the open sea, draws instead a technical and symbolic mesh of uneven threads flossed through multiple territories at once. From it, smaller assemblages of economic activity emanate, and practical norms of privacy and participation are worked out by both everyday habit and deliberate predation.

Again, this is not to suggest that even this form of Cloud Polis somehow formally secedes from national interest and control, but rather that as national interest and control pivot to the Cloud, the Cloud absorbs assignments of sovereign identification. Data centers and large switches also continue to operate under some normal authority, such that each Westphalian state claims (contested) rights over data within its geographic loop. The data that any one data center might house and distribute might be globally accessible in principle but also may be filtered or unfiltered by both national and international authorities (including mass interception of all traffic, national firewalling and keyword sorting, targeted deep packet inspection, or, in the recent examples of Egypt and Syria, complete blocking of a national top-level domain). International and transnational jurisdiction are even less settled, as evidenced by US superjurisdictional enforcement episodes and by controversies at the UN International Telecommunications Union over the role of member states to control, filter, and tax entry points of data entry into and out of their recognized territories (or even to adopt locally idiosyncratic Internet addressing schemes). ⁴³ The lines are unclear, dotted, smudgy, knotty, and self-contradictory, but the lines do exist and are multiplying in number. As the cat-and-mouse games of perforating and mending the membranes of national virtual borders continue, a kind of accelerated evolutionary episode is playing out, and the mutual constraint between state and extrastate actors brings strange moments of symbiosis. Instead of the simplistic schema of centers versus networks, centralization versus decentralization, elites versus the people, we observe instead states becoming networks and networks becoming states, nodes becoming more like edges and edges becoming more like nodes. The intrapenetration of Cloud geography and state geography results in another spatial-institutional model based on the logics of platforms and on platform sovereignties that don ’ t fit into those frameworks. Something like Cloud Polis is still an awkward embryonic form of governance, both for and through planetary computation, and likely will be for some time.

122

Cloud Layer

That said, as the Cloud is planetary is scope, state control of its systems is guaranteed only to the extent that private providers continue to respect (by consent, force, joint venture, outright merger) the practical sovereignty of the national jurisdictions in which their servers are installed, where their offices are headquartered, how exactly their data structure economic exchanges, and how their monetization of those data is or is not taxed. That arrangement is both resilient and unstable in various measures and tracks the successes and failures of globalization itself. Today most Cloud service providers have constraining jurisdictions built into service plans. For example, Amazon Web Services segregates the serving of hosted data according to several geographic “ availability zones, ” allowing developers to deploy specific versions of their application to specific users in specific countries according to local laws and priorities, regardless of where Amazon might be hosting or mirroring their data. However, the sorting of state space and data space is not always so neat, and the unintended effects of innovative interconnections between Cloud publics can be calamitous. Recall the Bank of Iceland ’ s ill-fated online banking venture in the United Kingdom, where easy access to “ offshore ” digital deposits and inopportune loans help to crash both nations ’ economies and contributed to a long-term destabilization of Iceland ’ s (and Europe ’ s) solvency. Ultimately, as a design problem, the potential political economic architecture of the Cloud Polis is at best a blurry composition, and like any other form of governance, its emergence is unevenly and asymmetrically violent. ⁴⁴ While it standardizes and flattens some symbols, geographies, and economies, it simultaneously frays and multiplies others. There is no way to guarantee either outcome in advance, and it is precisely the unforeseeable reversibility of these effects on any one location that makes the Cloud Polis central to the forms of sovereignty it mediates and which mediate it. ⁴⁵ In other words, the increasing reliance on Cloud infrastructure produces unintended geopolitical effects precisely because its power to govern through the observation of geographies, and the calculation of strategic information in the competition over those spaces, is always in the end reversible. Government data willingly opened to the public for its edification and employment, reinforcing the bonds and identification with a particular state apparatus generating this bounty, are also a potential basis for countergovernance, resistance, revolution, and the multiplication of unpredictable microsovereignties. The opposite — oversecuritization making systems brittle — is equally risky. That any innovation may accomplish one and the other, or how it may be deployed for one agenda and also its opposite, depends to a degree on the Interfaces we use to engage them and on what they as Interfaces connect and disconnect for us and from us. This is how The Stack connects Users and Clouds, Cities and Interfaces. But even then, any one Interface (that is, any one instrumental image of the platform as it is decoding its functions to us) is enrolled by more than one imagination at the same time. Interfaces are dream worlds, however restricted. They are also the real techniques by which power in inscribed by and for the imagined communities of geopolitical intrigue. But because interfacial images, like the machines they mediate, are tools for more than one geography at once, even the semiotic specificity of any one interface is a poor guarantee against the ultimate reversibility of the Cloud Polis. Just ask Costa Rica.

Cloud Layer

123