Most of us meet artificial intelligence through a screen. We upload a photo or prompt a chatbot, and the response appears as if from nowhere, an illusion reinforced by the metaphor of “the cloud”. In reality, however, each of these actions runs through physical buildings: data centers full of servers, linked to power grids, cooling systems and fiber networks. These in turn depend on power plants, wind and solar farms, water, land, minerals and human labor. What feels weightless online is anchored in a vast physical geography of infrastructures.

The geography of the cloud is growing at breakneck speed. According to the International Energy Agency (IEA), electricity use by data centers is set to roughly double, from 485 terawatt hours (TWh) in 2025 to around 950 TWh in 2030, while power used specifically by AI-focused facilities is set to triple. The IEA also warns that growth is running into physical limits: new transformers can take two to three years to arrive, and grid connections five to ten years in many places (IEA, 2026). Yet many data centers remain difficult to read from the outside, appearing as secured technical buildings, or “black boxes” whose wider connections are largely invisible.

Data centers are increasingly becoming a strategic front in the current AI race. In the United States, the White House’s America’s AI Action Plan and an accompanying executive order make physical infrastructure a core pillar of national strategy, fast-tracking federal permitting and opening federal land for data centers exceeding 100 megawatts (White House, 2025). Similarly, the European Commission’s AI Continent Action Plan aims to at least triple EU data center capacity within five to seven years (European Commission, 2025). Meanwhile, China has elevated computing networks into one of six core national infrastructure pillars under its 2026–2030 Five-Year Plan framework, on par with water networks and power grids, wiring its data center clusters directly into regional energy systems (State Council, 2026).

This growth is creating a planning problem as much as a technical one. Computing demand can shift within months, while grids, heating networks, land-use plans and public institutions evolve much more slowly. As demand speeds up, planners, architects, governments and communities increasingly face the question of where computing capacity should be built, which resources it should draw on, and what it should contribute to the places that host it.

My work on cloud infrastructure began during my doctoral fieldwork in Zhangjiakou, where I watched data centers rise across a landscape already being reshaped by wind and solar farms, regional development schemes, and infrastructure for the 2022 Beijing Winter Olympics. In 2024, an Enabling Grant from the ETH Zurich Department of Architecture allowed me to turn these observations into a research project, The Production of Cloud. After developing the initial framework and assembling a consortium across Lucerne University of Applied Sciences and Arts, the University of Virginia, the National University of Singapore, and four industry and non-profit partners, the project secured funding from the Swiss Network for International Studies for 2025–2027. Co-coordinated with Prof. Milica Topalović at the Chair of Architecture of Territory at ETH Zurich, our work bridges architecture and urban studies with building technology, energy engineering, and public policy.

We focus on three distinct cloud regions: Zurich, a dense and tightly regulated European hub; Northern Virginia, the sprawling, market-driven epicenter of the US cloud; and Zhangjiakou, a state-planned Chinese computing corridor, closely tied to Beijing’s digital demand and to renewable energy development. Looking at these three territories side by side shows how the same global push for computing power plays out across vastly different systems of land ownership, electricity governance, and local planning. As AI-driven data center expansion raises mounting socio-environmental concerns worldwide, keeping these regions in conversation helps us identify shared pressures, understand local differences, and test what practical lessons can travel between them.

Teaching and visual fieldwork keep this research anchored on the ground. Since autumn 2025, more than 600 students have participated in my cloud-focused lectures, seminars, and design studios at ETH Zurich and TU Munich, tracing how everyday digital habits translate into demands on land, energy, and local governance. Alongside traditional mapping, we work with artists, photographers, and filmmakers in China and Switzerland to document the everyday landscapes and communities that rarely show up in infrastructure datasets. Together, these methods are helping a new generation of architects and urban researchers learn to read AI infrastructure critically and to explore possible responses through design.

China makes the territorial scale of cloud infrastructure especially visible. The “East Data, West Computing” program, launched nationally in 2022, established eight national computing hubs and ten data center clusters. It is designed to link computing demand in major eastern urban regions with places that offer more land and energy resources (NDRC, 2022). Northern China has emerged as a strategic region for this expansion.

A more recent development is suandian xietong (算电协同), the coordination of computing power and electricity. Pilot work launched in 2024 in Beijing-Tianjin-Hebei, the Yangtze River Delta, Inner Mongolia and Ningxia explores how computing demand can respond more closely to energy supply, including renewable electricity and systems that combine generation, grids, demand and storage (NDA, 2025; NDRC, 2025). In Ulanqab, China’s first data center project integrating a direct green electricity supply with generation, grid, demand and storage is already connected to the grid and generating power (NDA, 2026). Here, computation starts to become part of energy planning itself. The timing of computing workloads, renewable generation, storage and transmission can increasingly be considered together.

This broader perspective also guides my fieldwork in Zhangjiakou and Inner Mongolia, where I study how data centers become part of existing energy, settlement and regional development landscapes. By combining mapping, interviews, photography and documentary work, the research examines how new computing infrastructure interacts with local land use, livelihoods and public planning, and what kinds of territorial change accompany its expansion.

Switzerland presents a different spatial condition. Here, data centers have to fit into a dense, urbanized landscape where land, power networks, heating systems and municipal plans already serve many competing uses.

A 2026 study commissioned by the Swiss Federal Office of Energy (SFOE) estimates that Swiss data centers used about 2.08 TWh of electricity in 2024. By 2030, this could rise to between 2.5 and 3.2 TWh, roughly 4.3 to 5.6 percent of the country’s 2024 electricity use (SFOE, 2026). So far, AI plays only a minor part in this growth. Most of it comes from companies moving their IT out of in-house data centers and into commercial colocation facilities and the cloud. The study also found that large commercial data centers run at around 30 percent of their installed capacity, and that contracted grid capacity can take several years to be put to real use. This raises a practical question: how much infrastructure should be committed before future demand is clear?

Our ongoing ETH research-design studios treat the Swiss cloud as both a spatial and a governance question (NSL, 2026). In autumn 2025, students traced electricity and heat flows, fiber networks, ownership, investment and planning processes through site visits, interviews, film and mapping. In spring 2026, they examined the Zurich data center cluster through energy demand, waste heat, land use, digital sovereignty and data ecologies. Design then became a way to test how architectural and territorial strategies might respond to these conditions, through scenarios for energy ceilings, new siting criteria, municipal data labs and local energy systems. The semester culminated in a collective Territorial Constitution: a combined spatial plan and regulatory charter proposing how the region might govern its digital infrastructure in the public interest.

Our preliminary fieldwork and comparative research across China, Europe, and the United States suggest that data center development is shaped as much by spatial planning, energy systems, and governance as by technology itself. For cantons, municipalities, utilities, operators, and research institutions in Switzerland, three priorities offer constructive ways forward:

• Grow in phases. Demand forecasts are uncertain, while grid connections, buildings, and land commitments last for decades. Tying new capacity to evidence of actual use lowers the risk of locking in infrastructure that sits half empty long after today’s hardware becomes obsolete.
• Build local reciprocity into early planning. Waste heat, grid investment and local services can benefit host communities when they are planned with municipalities, utilities and nearby users from the start. Land allocation, grid costs, heat reuse, ownership and community benefit belong early in planning and permitting.
• Establish transparent public criteria. Switzerland can define openly where cloud infrastructure belongs, what obligations it carries, and when extra capacity serves a wider public interest.

Everyday users of AI have a part to play too. Each prompt, image or video we generate draws on the servers, electricity and water described above. Awareness can start small: asking whether a task really needs a large AI model, favoring providers that are open about their energy use, and treating computing power as a shared resource with real costs.

Continued exchange on AI infrastructure between Switzerland, China and other countries can help keep these questions open. This matters all the more as rapidly expanding computation creates shared environmental and territorial challenges across very different political systems, on a planet we all share.

The Production of Cloud is a research project funded by the Swiss Network for International Studies (SNIS) at ETH Zurich, coordinated by Prof. Milica Topalović and Dr. Yiqiu Liu. Project partners: Prof. Ali Fard (University of Virginia), Prof. Adrian Altenburger (Lucerne University of Applied Sciences and Arts), Prof. Naomi C. Hanakata (National University of Singapore).

ETH studio teaching team, autumn 2025: Martin Kohlberger, Yiqiu Liu, Milica Topalović, Jakob Walter, Jan Westerheide. Teaching team, spring 2026: Dorothee Hahn, Martin Kohlberger, Yiqiu Liu, Milica Topalović, Jakob Walter, Jan Westerheide. Student assistant: Leon Schade