Since its origins, the digital utopia has maintained the image of a dematerialisation at the intersection of cybernetic system theory and counter-cultural influences, heavily masking the material and technical infrastructures on which the digital world depends. The public internet of the early days opened an unlimited horizon. In the mid-1990s, the apprehension of digital city life remained full of uncertainties. The information and communication technologies revolution, the digital new economy and the technification of society made the apprehension of urban futures and technological landscapes obsolete. The ‘hybrid’ or fractal epithet dominated territorial perceptions, just like the image of a ‘drape’ that has returned many times to describe this informational continuum that disturbed traditional spatial references. The line was blurred between ‘centre’ and ‘outskirts’, ‘infrastructure’ and ‘superstructure’, ‘soft’ and ‘hard.’ In his work La ville territoire des cyborgs, the architect, engineer and historian of technologies Antoine Picon asked the question: ‘Are we headed toward a sort of virtual environment on the planetary scale, a sort of giant electronic game that only fleetingly lets the awareness of an exterior material reality subsist?’ In the 1990s, it was truly impossible to answer this question as the large monopolies of the web were still inexistent, and the territorial footprint of the digital in its material form was unpredictable. Data decentralisation was a prospect that was as credible as that of the ultra-centralisation to come. How far would the explosion of the hybrid, of the virtual go?

Would the heaviness of infrastructures of the last century, the ‘large-scale’ and ‘centralising’ characteristics of Large Technical Systems, make way for a more hybrid and lighter, decentralised and distributed, economical in terms of resources, and peer-to-peer infrastructural future? This has in no way been the case. The illusion of an unconstrained information evaporated; centralisation won the day. In the digital’s technical and material system, data centre fortresses are the central infrastructures of our reticular territoriality. The analysis of their energy impact through their electricity and spatial consumption demonstrates a material reality that is still largely underestimated.

Several authors have analysed the evolution of telecom networks from the socio-spatial angle, which has made it possible to better understand their link with metropolitan developments. The complexity of the development of metropolitan telecoms implies a group of contextual and historic processes, interests and practices at several spatial scales, made more complex by the emergence of the internet. It is from this viewpoint of the urban externalities of the digital economy that spatial effects are the most often discussed, leaving the physicality of the infrastructures and their energy consumption in the background. The invisibility of the infrastructures is often mentioned as a strategy connected to the issue of data confidentiality and respect for private life, endangered with the development and use of predictive systems for the purpose of monitoring and controlling populations and for economic interests that focus on advertising profiling. Digital sobriety, when it is tackled, most often concerns the use of services and electronic devices that require large amounts of raw materials, including the extraction of rare minerals and ores needed for their functioning. The subject of the electricity consumption of data centres and the multiplication of the infrastructures required for their operation has not been studied despite the fact that these places are witnesses to the energy-territorial destablisations induced by the rebound effects of the ‘all-digital’. The energy density of these ‘boxes’ has no equivalent in the history of urban planning. Data centres are disruptive to local energy systems, and their accumulation in urban areas, similar to their propagation in rural environments, is a major concern for spatial and energy planning.

In All Territories: Uncontrolled Spatial and Energy Growth

In the twenty-first century, digital infrastructures are one of the most important sources of electricity consumption. The growth of the digital economy has created a great deal of enthusiasm, driven by an increase in internet traffic, the explosion of data exchanges and the exponential development of the Cloud. By 2030, according to Gartner, the Internet of Things will reach up to 125 connected devices (not including tablets and PCs). According to Intel, the autonomous vehicle is itself a connected object that will produce four terabytes of data per day to be processed and stored. The growth of the digital sector seems to have no limits. And despite the smart city’s promise of sustainability, there is no evidence that digital technology will save energy. As a new stage in the network’s urbanism, the digital city has often been analysed in terms of use and practices, services and events, leaving the spatial and energy impact of its material infrastructures in the background.

A cornerstone of the digital technical system, the data centre, which comprises warehouses hidden away in rural areas is only a part of the phenomenon. As Rem Koolhaas observed in an 2016 article in Flaunt Magazine, over ‘the last 20 years an immense proliferation of boxes totalling more than 14 million square feet has emerged in this landscape – a combination of high-tech warehouses, factories and data centers of at least 300 feet wide, and from 1,200 to 2,600 feet long – organised in a number of competing grids’. Beyond rural areas, today there continues to be an increasing number of data centres in urban areas. Edge computing requires small data centres to meet the needs of 5G. To this end, Google has announced the purchase of a site in the centre of London. This is already the case in the centre of New York City and elsewhere, where there are large data centres, or in central Paris and the nearby suburbs. This raises the question of the scale and architectural form of these infrastructures, their programmatic hybridity and their urban integration. Data centres offer a wide range of typologies: standardised and flexible ‘boxes’, emblematic architectures on high-tech campuses and converted buildings such as old offices, industrial sites or obsolete commercial areas. Visible, discreet or even hidden data centres can be found throughout city centres and in surrounding territories, rural areas, remote and desert regions – and even under the sea.

There are three worldwide trends of data centres. The first is the growth of data centres in all territories. On one hand, American and Chinese Big Tech firms are continuing their development and construction of hyperscale data centres, preferably in rural and semi-urban territories where land and energy are inexpensive. On the other hand, medium-size data centres are concentrated in metropolitan spaces, near internet backbones and exchange points. There is an urban concentration of flows through colocation data centres that are also entrances to Clouds and a geographic decentralisation of stored data with hyperscale data centres. Areas with a low-density population are becoming heavy users of electricity. One example is Prineville, a city of ten thousand inhabitants in Oregon, where electricity consumption has risen from 10–500 MW since the arrival of large Facebook and Apple data centres in 2012.The second trend is the growing interdependence of the data centres. IT configurations accompany complementary spatial sitings on territories between rural/semi-urban and urban territories. It is a whole complementary system that is being established, adapting to the multitude of uses and needs of our new digital life. The third major trend is centralisation. The centralised large technical system has prevailed over the distributed and micro-system that was at the origin of the internet’s history. This is partially due to the monopolistic nature of Big Tech.

Rising Electricity Consumption

Data centres consume a growing amount of energy even if hardware and software are increasingly efficient. Academic research on data centre energy consumption and energy-saving computing centres started at Virginia Tech in 2000. For data centres, new buildings have reasonably efficient power usage effectiveness (PUE), but their electricity consumption nonetheless is constantly growing. A 2015 study by Anders Andrae and Tomas Edler at the Huawei R&D centre in Stockholm estimated that the digital sector consumed seven percent of worldwide electricity in 2013. The data centres themselves represented two percent of the worldwide total. Their projections reach a maximum of 13 percent of worldwide electricity consumed by data centres in 2030, and 51 percent for the IT sector in its totality. This figure is approximately nine percent in France today. The Shift Project think tank recently revised this worst-case scenario downward but estimates that the digital sector could represent 25 percent of worldwide electricity in 2025 (five percent for data centres), without making a projection for 2030. While these figures show a trend, the reliability of these dizzying prognoses remains abstract and remote. To measure the energy impact of data centres, what could be more tangible than talking to the electricity distribution and transmission operators and the municipal services of the cities affected by this peak?

Data centre operators need to communicate with those in the development and energy sectors, particularly electricity transmission and distribution system operators. This is not a simple task, because over the last ten years, installations have been carried out on a case-by-case basis, without a global and strategic vision. This poses many problems for energy and urban operators. Requests from mid-size and large data centres are on average for 20, 40 or 60 MW.

If the arrival of digital infrastructures on territories is equivalent to the arrival of the major infrastructures of the twentieth century, they are more complex to grasp because they are less visible in their technical and architectural materiality. The network prevails and there is no digital architecture, unlike that built for electricity or the transport system. This absence has several explanations. The first data centres were not thought of as an autonomous programme, but as an outgrowth of the network, a proliferating object. They are the extensions of server cabinets and computer rooms originally integrated into office buildings and research complexes. The rapid development of the internet in the 1990s also contributed to a poorly anticipated and opportunistic infrastructural multiplication with locations in old telephone exchanges or in obsolete service buildings, before being developed by the data centre industry as a standardised real estate product.

Pressure and Destabilisation on Urban Electricity Systems

The great amount of electrical power needed by data centres is a relatively opaque subject: distribution system operators (DSO) and transmission system operators (TSO) cannot talk about it, given the obligation of discretion and business confidentiality. Data centres attempt to capture a large amount of power to put the brakes on the competition. They generally request and reserve more electricity than they consume and ask for more power than they consume, which blocks other consumers. The principle of an electrical connection is the same in most European countries. This is a tradition inherited from public service: ‘first come, first served’, with no customer discrimination. It can be a data centre, hospital or housing development. A customer asks for a certain number of megawatts, and the DSO is obliged to supply the electricity requested. But data centres, most often, operate under a maximalist consumption hypothesis, basing their request on how much power will be required when the data centre is full. This sometimes takes several years, and, occasionally, never happens. In dense localities, this overbooking creates conflicts of use with other urban developments. One example is the city of Marseille where a major international data centre’s operator reserved a huge amount of electricity (90 MW), leaving no electricity for the municipality’s electric buses. The mayor of Marseille had to negotiate to recover the 7 MW required. In 2023, in large metropolises, an additional power demand of at least 1,000 MG on average will be necessary but this figure could be greater. The figure translates to one gigawatt, equivalent to the average electrical power of a reactor in a modern nuclear power plant. In Europe, the cities where the presence of data centres is the highest and still growing are Frankfurt, London, Amsterdam, Paris and Dublin. Some try to limit growth – Amsterdam, for example, became the first city in Europe to suspend installation with a moratorium in 2019.

When Amsterdam Says Stop: Digital and Urban Renewal Planning

Amsterdam is a unique example in Europe of strong and structured planning to guide the location of data centres. The majority (75 percent) of Dutch data centres are located in the metropolitan region around Amsterdam. In total, this region currently houses 100 data centres facilities (Amsterdam, Almere, Aalsmeer, Haarlem, Hoofddorp/Schiphol, Purmerend). Others are in the pipeline. For example, CyrusOne recently announced the opening of the company’s first data centre campus at Schiphol, with an expected capacity of 72 MW.

The problem of the location of data centres was apparent as early as 2011. At the time, the electricity network was already saturated. These data centres were not located in industrial or residential zones, but in zones reserved for businesses. Faced with this uncontrolled growth, in July 2019, two cities – Haarlemmermeer and Amsterdam – launched an initiative to stop the construction of new data centres. The aim was to pause further building to better integrate them. The entire problem was the availability of energy and land. A new regulatory policy is currently being drawn up with the objective of defining minimum conditions: mixed use, architectural quality, high-rise construction to limit the impact on the land and landscape integration, the absence of fences and the accessibility of the building’s surroundings. At the same time, reflections are being undertaken on urban heating networks in connection with heat recovery from the data centres. However, the heating network has not been developed in the country, which means that the cost of the infrastructure will be high.

An unprecedented amount of expertise on the subject in the Netherlands was produced. Commissioned by the public authorities, these studies are the result of an association between energy operators, the data centre industry and urban planning departments. The report produced by the city of Amsterdam with the electricity distribution operator Liander identifies the impacts on the electricity network through 2050. It includes a complete study on the location and electricity availability of source substations. Data centres currently have the greatest impact on the power grid, with a current load of approximately 750 MW. In 2050, this will be two-and-a-half to five times higher depending on the scenario. It will increase to 2,000 MW (in the low scenario) and to 4,000 MW (in the high scenario). In the median scenario, the load is about 2,900 MW in 2050. The document states that 37 percent of the total electricity demand will come from data centres.

At present there are 26 substations in Amsterdam, and the capacity of the current grid is insufficient to supply electricity to future urban developments. To solve this, the network will have to be extended by adding, before 2030, six to eight substations on new sites across the city. Each site means land to find and equip. In parallel to the work being done at the municipal, at a national level officials are looking at regional electricity pockets with 1 GW. The two scales – national and local – are linked. If a national injunction favours data centre developments at the 1 GW pocket level, there may be fewer data centres in Amsterdam. If this is the case, the number of substations may be reduced. At the same time the Spatial Datacentre Strategy Roadmap 2030 REOS, launched by the Ministry of the Interior, CRE and the regional level has given rise to work focusing on the spatialisation of the energy and land impact of data centres. Three scenarios are presented: super concentration, regional and deconcentration, and dispersion. The report recommends a development in the Amsterdam metropolitan area as well as an expansion in the region outside the cities of Amsterdam and Haarlemmermeer. For example, this could be in Almere and where offshore wind power could be developed (on the west coast) and, additionally, south of Amsterdam (in the Rotterdam–The Hague metropolitan area).

Agriport, in the north of the country, is an example of a hyperscale hub. It is 40km from Amsterdam, in an area where space and energy are available. It is a zone of 1,000 hectares that was created in 2006 with huge agricultural greenhouses, logistics warehouses and, later, data centres. There is also good connectivity on the site, with five fibre optic cables to the Amsterdam internet exchange point. In 2013, Microsoft announced an investment project in a 37ha hyperscale data centre. Two of four sectors of the Microsoft campus have been completed. This year, Google announced the acquisition of new large plot of land to extend its first data centre. One is already operational. CyrusOne is also interested in a 30ha site in the same area.

In 2011, a private electricity micro-grid was installed, one of the few in the country, which connects greenhouses, logistics areas, offices and data centres. Thirty-five gas cogeneration plants are currently connected to this micro-grid, with a total production capacity of 135 MW. This micro-grid is also connected to the TenneT electricity network with a 240 MW capacity. The TSO has started the construction of a second 480 MW source substation for further developments. Heat recovery is not yet operational but remains under study. Separately, in a more urban location, the Sciences Park campus in the heart of the city of Amsterdam houses three large data centres owned by Digital Realty and Equinix.Inspite of architectural efforts for a better integration into the urban landscape, a strong reflection to bring data centres back into the energy landscape puzzle is underway in the Netherlands. Based on this case, but also on other international examples, some prospects for energy pooling still need to be developed.

For More Energy Pooling and Political Arbitration

How can territories obtain better benefits from data centres? There are two main concerns: first, is the reduction or sharing of infrastructural redundancy or heat recovery sufficient and, second, in a context of profound transformation of energy systems with the development of renewable energies and the privatisation of the electricity sector, is it not up to the public authorities to regulate? National policies are still expected, as was the case for the organisation and rationalisation of the major infrastructures of the twentieth century.

Reducing infrastructural redundancy could benefit the territories. Data centres, whatever their type, have doubled infrastructures in case of a breakdown: a second or a third electricity source connected to the network, batteries and backup generators (very rarely used). This redundancy requirement could become a shared resource, as Portland General Electric in Oregon is doing with a smart grid connecting 85 different customers, including five data centres. The electricity company can mobilise the backup generators in case of a breakdown on its network (it represents 120 MW in total). In exchange, Portland General Electric manages maintenance and buys fuel for all the generators. Nonetheless, the problems of air pollution linked to the use of fuel oil generators must not be overlooked. Another example is Microsoft in Wyoming. The company shares its data centre’s backup gas generators with the local power company Black Hills to produce electricity during grid peaks.

At a time when the public sector lacks the financial resources to invest in infrastructure, private Big Tech companies appear as new energy actors. There is competition between the digital actors and the historic electricity operators, especially in the United States. GAFAM companies, like Microsoft, are accelerators of renewable energy use, in particular solar and wind. Facebook and Apple consequently co-invested with Pacific Corp, an energy operator in Oregon and northern California, to create solar farms in the vicinity of their data centres. Apple also created the Apple Energy subsidiary, accredited by the Federal Energy Regulatory Commission (FERC), and all major digital companies have developed specialised competencies in this sphere.

Data centres are increasingly developing their energy autonomy through onsite or nearby production infrastructures. Facebook and Microsoft, for example, are heavily involved in the development of micro-grids following the example of Microsoft’s data centre park project in Colorado, which will produce 200 MW onsite. In Europe, private micro-grids are banned or heavily regulated. This is different than in the United States where the distribution sector has been deregulated, and where there are some interesting examples.

Heat recovery, which is often cited as a perspective, remains complex to implement. The waste heat discharged by data centres can be a source of energy recovered for other uses, but there are two types of difficulties. The first is economic. Economic profitability models, project interests and timelines diverge between data centre developers that commit to returns on investments over very short periods (around two years), whereas the contracts for heating networks must make commitments for 25 years. Second, there is a technical curb that limits initiatives. In fact, it is preferable for an operator to envisage recovering heat when the data centre is first built because intervention and work on an existing installation could disturb its operations and would often be too costly. The connectable distance and temperature needed must be studied in detail (heating networks in the Île-de-France have temperatures generally between 60°–110°C, while data centres have temperatures of 40°–50°C). In Stockholm, the great success of data centre heat recovery is the result of a strong national policy and the commitment of municipal energy operators. The Data Centre Parks programme, a municipal project developed in 2016, combines an energy, digital and real-estate strategy. On lands belonging to the city, served by a heating and cooling network from the Exergi municipal company, Stockholm proposes a long-term lease to data centre operators to set up, offering them free cooling in return for their waste heat.

Many initiatives exist and are being implemented at municipal or national levels by teams committed to building a digital public service. However, these initiatives remain fragmented and are confronted with national regulations that are often restrictive. As the digital industry continues its infrastructural growth, we can note the need to think together about an energy and spatial planning strategy; to renew urban planning tools to better regulate installations; to promote an articulation of the national regional and metropolitan scales for data centre clusters; the implementation of energy sobriety and energy-saving measures; more constraints from network operators on data centres; the use of waste heat where possible; the sharing of backup generators and other types of pooling; but also more decentralised visions of the digital world that are still insufficiently debated.

Today, architecture and urban planning tends to minimise the energy and spatial impact of the digital world on cities, the territory and climate. The digital sector does not increase, it transforms. It is not an urban exoskeleton that can be put on and taken off, but a pervasive system that is gradually modifying every territory. Supporting territories in accessing a free and open internet is as essential as a broader reflection on this new infrastructure that has become a central piece of the ‘energy landscape’. The data centres that are being built today will determine our digital practices for decades to come. Architects and urban planners need to take advantage of these objects to better measure the environmental and energy impact of technical choices with regard to the expected social value, and to collectively move toward more reasoned, sober, and decreasing digital practices.