The cornerstone of the digital technical system – the data centre – is not only a warehouse lost in rural zones, certainly very well-described by Rem Koolhas in his article ‘Brave World: People, Places, Technology’. This is only one part of the phenomenon, far from representing the complexity of the system that comprises a globalized territorial continuum, each piece of which is strongly intertwined with the others. If these storage buildings’ strategy of visibility, form and programmatic evolution is trying to find its way, it is also the network dimension and the multiplication of these infrastructures that raise questions. They function in a complex interconnection system that links the small data centres to the large ones, those of colocation to enterprise data centre. The digital storage industry shows three major trends: growth, interdependence and ultra-centralization.

The first trend is the growth of data centres in every territory. The large American and Chinese Big Tech companies are continuing the development and construction of data centres on a very large scale, preferably in rural and semi-urban territories where land and energy are inexpensive. The second major trend is centralization. The analysis of the industrial mutations of globalization has largely showed relocations of flows in territories where labour and raw materials are less expensive, while keeping company headquarters in the large metropolises that comprise the network of world cities. With the digital, relocation is limited and is concentrated on the national territories with dedicated hubs.

If the technologies described above as alternatives sketch a frugal world in which the digital permits more freedom, communication, more creativity than surveillance, centralization and individualized tracking, the possible futures are being woven in a more complex fashion than that binary face-off, encountering growing climate, environmental, migratory and democratic crises.

What will the Internet of tomorrow be? Reflections, debates and international regulations on the digital lead to a rather centralizing perspective whose trajectory the first scenario proposes to extrapolate. The second scenario asks us to follow another trend: that of the structuring of a decentralized digital public service coexisting with a small-scale infrastructural diversity. Lastly, the third scenario marks a radical break, the end of data centres and the digital world as we know it makes way for distributed networks primarily based on intermittence.

Steered by the University of California, Berkeley, the study Datagravity provides information on how Forbes’ 2,000 largest worldwide companies exchange data. One of the results is acceleration and densification in a ‘snowball effect’ around 50 or so cities in the world. The data goes where the data is. There is a major data concentration and centripetal aggregation dynamic in proportions that create perplexity. This phenomenon does not mean that there are no data centres in the region; there are, more and more, and they are getting larger. However, these are not the same data; it is mostly about distribution and collection.

The continuation of the digital growth dynamic is explained by the evolutions in the technologies deployed (optronics, wavelength, etc), but also by the creation in 2022 of new undersea optical fibre cables: the Peace, 2Africa, Ellalink, Maréa, Dunan and Amitié cables. The Chinese Peace cable linking Pakistan to the Prado beaches in Marseille in the south of France represents, with these 320 terabytes per second, itself, the double of the 15 undersea cables connected in the large Marseille data centres. The 2Africa cable rolled out by Facebook will also revolutionize African connectivity and the interconnection with the Middle East and Europe. Over 45,000 km long, it has become the world's longest, linking three continents, almost the equivalent of the Earth's circumference. This 160-terabyte 2Africa cable (upgradable to 320 terabytes) will open the 4G and 5G channels.

States’ sovereignty is increasingly based on their capacity to control digital interference, protect themselves from the pirating of public services and businesses, and fight against the theft of strategic data. China and Russia have created national and closed Internets with this in mind. The creation and support of heavyweight digital actors like Alibaba, Tencent and Xiaomi in China, and Amazon and Meta in the United States, also serve these objectives. In Europe, most countries, including France, have repeatedly raised the question of their digital sovereignty.

The subject of secure data storage is at the heart of this concern. A new strategy was consequently adopted in France in 2021 through the label ‘Cloud de confiance’ and the ‘Cloud au centre’ strategy. French actors today use Cloud services while making use of a hosting offer of the data centre type. The two options are activated given the variety of needs, notably of the large local communities or public structures, and both of them can be public or private. The rules and offers for the government administrations are different from those of local communities. For both types, the use of Cloud solutions or data centres operated by actors outside Europe is, however, increasingly raising questions surrounding the location and sovereignty of the data. The protection of public data is a strategic issue and foreign intrusions in national affairs, like Russia in the 2016 US elections, has reactivated this issue in France. Consequently, the French state has, for example, prohibited (with exceptions) the use of Microsoft 365 by the ministries in a circular published in September 2021.

In April 2021, Amazon Web Services France was denied authorization to build a data centre of 33,100 m2 in Bretigny-sur-Orge in the northern part of the Essonne department. It was a first in France. The order signed by the prefect of the Île-de-France specified that the request to build this complex for use as a data storage centre was incompatible with the orientations of the regional development policy, in particular with the master plan of the Île-de-France region and the regional plan for climate, air and energy. According to this same source,1 the Amazon project did not include any analysis of vacancy or the densification potential of the existing urban fabric so that the necessity of artificializing this 9.7-hectare space could be evaluated. It was moreover noted that no solution for recycling waste heat was proposed by the digital industrialist. Over the last ten years, the monopolizing of real estate and energy by large private digital companies has become a major concern for the public authorities that are confronted with conflicts of use. In Île-de-France, there are 168 data centres for 1152 sites and an electricity consumption that exceeds 1 GW (1 GW = 1,000 MW), the equivalent of the electricity production of the two pressurized-water reactors of the nuclear power plant in Nogent-sur-Seine. Buoyed by the digitalization of the economy, the explosion of data exchanges, the Cloud and the connected Internet of Things (IoT), digital infrastructures will be one of the most important areas of electricity consumption in the 21st century.

Despite the contribution of science fiction at the beginning of the century that had anticipated a number of its abuses and promises, notably on the use of personal data and the limits of prediction, the conceptual success of the smart city marked the 2000s. Both a technical imaginary dimension and a group of connected objects, this global infrastructural product, created by industrialists and the Cisco and IBM engineers, then relayed by the information technology giants, dominated economic and urban forecasting. The dream of order was no longer achieved by a plan created by the urbanist-designer, but by the predictability of engineers from the IT sector whose programs made it possible to imagine homeostatic and self-referential systems. Their determination to incarnate this urban performance ideal through a new neighbourhood, a new head office or a new city, seems, however, to have raised their claim to compete with the local communities a notch, with the platformization of urban management towards which the data converge by means of embedded sensors in our mobile phones and connected objects.

But ten years after the first international Smarter Cities Challenge, in spring 2020, the failure of one of the most celebrated smart city projects of Sidewalk Labs (a subsidiary of Alphabet, Google's parent company), launched in 2017 in Toronto, marked a turning point.

For distributed and decentralized solutions, the absence of backup infrastructures (backup generators, batteries) and large air-conditioning systems, but also electrical equipment like undulators, reduces the expense of the initial material (production), integrated energy used and energy consumed in the event of a breakdown. On the other hand, a breakdown on a distributed storage system can generate the mobilization of more machines on the peer-to-peer network and cause additional energy consumption.

A question that is central to these alternatives is the energy impact of distributed storage and data centres of proximity. Several hypotheses are possible, but there is very little research on the subject at this stage to compare decentralized and distributed systems to today's dominant, centralized and globalized system. Nonetheless, several indications seem to be in favour of alternatives.

For distributed storage as for local data centres, visualizing the physical spaces occupied by the servers favours a better-adjusted functioning: the storage capacities are not oversized, the servers are not on standby like zombie servers … These storage modes avoid the drawbacks of dematerialization – that impression of having access to infinite storage spaces, without any impact on the environment. Storage is also carried out more locally (not on continents far from the users) and the data travel less on communication networks; we can therefore suppose that less energy is used on the network.

Two approaches coexist, each favouring a different strategy: on the one hand, the first approach starts with distributed protocols and the Internet technical system to transform the landscape of the actors and improve access to the digital (preceding part) and, on the other, the second approach first of all changes he actors’ typology: associative and cooperative Internet Service Providers (ISPs), decentralized on the territory. These actors will then help technically modify the digital system by the adaptation to more local scales, and a less centralized organization. Thus, whereas the distributed systems proposed by alternative actors aim to radically transform the Internet architecture, the projects presented here are more involved in a decentralized architecture.

Communicating freely: networks in the hands of civil society

The associative ISPs present many advantages: coverage of territories poorly served by the classic operators, notably rural zones; development of local skills and indirect jobs; accessible rates; involvement of volunteers and users, a social link; and commitment to data protection. Their growth, however, is limited by the number of volunteers available, which makes repairs potentially more complicated in the case of service interruption.

In France, in addition to the four main ISPs (Bouygues, Free, Orange and SFR), there are the associative and cooperative ISPs. French Data Network (FDN) was the first. Created in 1992, it was the first structure to provide Internet access for the general public when the web was first established, before France Telecom in 1996.

new spatial entity, data centres have consolidated the territories on a continuous digital landscape. Whereas the habitability of dense urban zones depends increasingly on their digital networks, data centres appear as a new piece of the ‘energy landscape’ linking the national or even transnational scale to the local scale. The recent debates and publications on the energy transition and the adaptation of cities to climate change have led to rethinking the role of energy production systems, notably in environments with a high population density. The relationship of dependence of urban centres vis-à-vis the worldwide ‘energy hinterland’ has undergone modifications over the last 15 years. The search for and integration of more decentralized energy production systems have hybridized the traditional functioning of the electricity grids. Densely populated zones are becoming increasingly involved in producing renewable energy sources, through integrated micro-electricity production plants, but also micro-grids, which is notably the case in New York and London. The relationship of interdependence between architecture, infrastructures and the territory in a world with limited resources has forced reconsiderations and redirections that cannot be limited to questioning how digital infrastructures can be better integrated and pooled. The debate on the ideal digital society is not the object of this work, which has limited itself to giving an account of data centres’ spatial and energy impact as revelatory of an untenable situation over the long term.

The projects that follow, aiming at making the digital technical system evolve towards more diversified forms favouring a control of and decrease in electricity consumptions, resilience to extreme climate events, small-scale network infrastructures and more collaborative functioning. A part of these initiatives has been rolled out through distributed and peer-to-peer modalities, transforming the Internet's technical system at its core, while others aim at taking control over the infrastructures of the existing system to self-manage on more local scales, or, notably for the French state, to take control for geopolitical reasons.

Beyond the question of storage, turning to the network infrastructure is necessary to grasp all the advantages, motivations and approaches of the Internet's associative and cooperative actors. It is entirely in the interest of citizen Internet Service Providers (ISPs) that aim to provide a connection not only to the Internet in remote zones or those abandoned by the operators (citizen Wi-Fi networks, mesh networks), but also to a neutral, unfiltered, unsorted Internet through, for example, a Virtual Private Network (an encrypted tunnel), a system widely used, notably for political reasons in China, Iran and Egypt, for example. The entire structure of the infrastructural system is thus rethought. On this basis, the question of alternative storage (distributed on user terminals or in nearby data centres) comprises a second stage that these ISPs wish to attack, as shown by the Spanish example of Guifi.net.

Despite the improvement in the energy efficiency of the components, the global electricity consumption of data centres has substantially increased. In academic and industrial research, the energy efficiency of the digital in its use phase is a fully-fledged research topic. In 1965, the cofounder of Intel, Gordon Moore, established in an article that the power of information systems doubles every 18 months. More recently, at Stanford University, Jonathan Koomey and his colleagues demonstrated that the quantity of energy that a machine needs to carry out a given number of calculations decreases by a factor of two every year and a half. However, energy dissipation (notably thermal) and also the multiplication of machines keep growing, as anticipated by the physicist Rolf Landauer in the 1970s. Research on the energy impact linked to the computing of frugal data centres started in the early 2000s with the work of Wu Feng and his team at Virginia Tech. In the framework of the Super Computing in Small Spaces project, the researchers developed Green Destiny: Low-Power Supercomputer (2001), which succeeded in putting supercomputers into production in reducing costs, the heat emitted, the energy consumed and the space required. In France, at the INRIA, the EcoInfo group studied the impacts and rebound effects of these different phases. As for the buildings themselves, new constructions are generally very energy-efficient as the data centre operators have the financial resources to invest in very upmarket equipment (undulators, batteries, air-conditioners, controllers, etc).

Data centres are the most tangible spaces of the digital technical system on the territory, but remarkable architectures remain rare, far from an architectural tradition that is found in train stations, telephone exchanges and hydroelectric or nuclear power plants of specific formal and aesthetic characteristics. At first glance, most data centres are boxes, hangars, sometimes camouflaged or decorated. The acceleration of flows has replaced storage architecture (brick warehouses and other stylized infrastructures) with more ordinary structures, boxes of all sorts, like containers or huge hangars for storage of material or data.

The first Google data centre in The Dalles (Oregon) or that of Amazon Web Services, not far from Umatilla and Port Morrow (Figures 3.1 to 3.4) can be compared to this warehouse or factory typology.

The absence of architecture could be explained by three reasons. The first is that data centres are the almost organic extension of the server cabinets and computer rooms originally included in office buildings and university and research complexes. Storage was not thought of as an autonomous program, but as an outgrowth of the network. The change in scale happened gradually: a much larger cabinet, then a dedicated computer room. The initially decentralized character of the Internet favoured, moreover, many actors, and uncoordinated developments, without a global view and without any architectural consideration.

Amsterdam was the first European city to take hold of the subject of digital infrastructure spatialization and to propose updated planning. Since 2018, many evaluation reports and documents commissioned by the public authorities have involved all the actors: data centre industrials, electricity transmission and distribution operators, as well as the metropolis’ urbanism departments. We can cite the report by the city of Amsterdam, Vestigingsbeleid data centers, de Amsterdamse regio als Green Data Port, from 2013, and also especially Datacenter Spatial Strategy, 2030 Roadmap for the Growth of Datacenters in the Netherlands (REOS) in 2019. This global strategy is notably based on cooperation at the national (under the direction of the Ministry of the Interior) and regional levels. This report gave rise to a work bringing to the forefront the spatialization of the energy and land impact of data centres, a document unique in Europe.

The REOS1 presented a new national spatialization of data centres in three illustrated scenarios: (1) metropolitan concentration; (2) regional cluster; and (3) territorial decentralization. The report recommends a two-stage development, first within the metropolitan region (with a mixture of scenarios 1 and 2). It then suggests that the digital infrastructures extend to where maritime wind farms are being developed (on the west coast) and south of Amsterdam (in the Rotterdam/The Hague metropolitan region). It also recommends an expansion of the existing hyperscale sites of Hollands Kroom and Eemshaven.