Guest post by: Sebastian Leape, CEO of Natcap
Data centres have become one of the more visible environmental consequences of the AI boom. Their appetite for electricity is rising sharply. Some use millions of litres of water a day. New projects are large enough to strain power grids that were never designed for such concentrated demand.
The concern is real, but the debate is often framed too crudely.
A data centre that consumes a lot of electricity or water is not necessarily equally damaging everywhere. A megawatt-hour drawn from a low-carbon, unconstrained grid has very different consequences from one drawn from a fossil-fuel-heavy system already struggling to meet demand. A litre consumed in a water-abundant catchment in Finland is not environmentally equivalent to one consumed from a stressed aquifer in Arizona during a drought.
That makes the geography of the data-centre boom at least as important as its scale. The question is not simply how much energy and water data centres use, but whether they are being built in places that can support that demand.
Big globally, bigger locally
Data centres consumed around 1.5% of global electricity in 2024. The International Energy Agency expects their demand to more than double to roughly 945TWh by 2030, just under 3% of global consumption. AI is the largest contributor to that growth.
Those figures are significant, but not apocalyptic. Data centres are not about to overtake heavy industry, transport or buildings as the dominant consumers of energy. Even in 2030, the IEA expects data centres to account for only around a tenth of the growth in global electricity demand.
The difficulty is that data centres are unusually concentrated.
A conventional facility might require 10-25MW of power. A large AI-oriented facility can exceed 100MW, and projects under construction are already far larger. Unlike millions of electric vehicles or air conditioners, that demand arrives at a single point on the grid.
Ireland shows what this can mean. Data centres’ share of national electricity demand rose from 5% in 2015 to 22% in 2024. On the basis of currently contracted demand, Ireland’s energy regulator expects that share to reach 31% by 2034. New facilities must now provide generation or storage capacity alongside their proposed connections.
The source of the electricity matters too. The IEA expects renewables to meet around half of the additional global demand from data centres over the coming decade, but natural gas will also expand materially to serve them.
A megawatt-hour consumed on a clean grid with spare capacity is therefore a very different proposition from one consumed on a carbon-intensive or constrained grid. The source of that power also shapes the emissions, air pollution and pressure on ecosystems associated with generating it.
Water is more local still
Water produces more dramatic headlines, but it is an even more local issue.
Data centres use water directly for cooling and indirectly through electricity generation. A large water-cooled facility can consume several million litres a day.
Yet the volume alone says little about environmental harm.
If a facility draws reclaimed water in a wet region, its impact may be relatively modest. A more efficient facility drawing potable water from a stressed aquifer can be far more problematic.
That distinction matters because a large share of new capacity is being built in places where water is already scarce. The World Resources Institute estimates that two-thirds of US data centres built or under development since 2022 are located in water-stressed areas, including parts of Arizona, the Colorado River Basin and Texas.
Data centres do not have a uniform water problem. They have a water problem where large new users are added to already constrained systems. Water withdrawals can affect aquatic habitats, streams, wetlands and aquifers even before supplies to people become constrained, and the temperature and quality of water returned to the environment may also matter.
Energy and water dominate the operational debate, though they are not the whole footprint. A data centre may occupy a relatively small area of land, but the infrastructure required to serve it can extend much further. New roads, substations and transmission lines can convert or fragment habitat well beyond the campus itself, while construction requires large quantities of concrete and steel.
The same is true upstream. Servers, cooling equipment and electrical infrastructure depend on semiconductor plants, mines and mineral-processing facilities, while the electricity consumed by the facility ultimately depends on power stations and their fuel and material supply chains. The environmental footprint therefore extends far beyond the data-centre campus.
A location that works at one scale may fail at the next
If geography matters this much, have data centres actually been built in the right places?
Historically, that was not really what site selection was trying to achieve.
Data centres clustered where fibre was plentiful, customers were close, land was available and power connections could be obtained quickly. Tax incentives and planning regimes helped too. Even in 2026, property adviser JLL describes “speed to power” as the primary criterion in site selection, ahead of community support, latency and proximity to customers.
This made commercial sense. It produced enormous clusters around Northern Virginia, Dublin, London, Frankfurt, Amsterdam, Dallas and Phoenix.
It did not necessarily produce environmentally optimal ones.
That does not mean these places were obvious mistakes. The more important point is that a location which makes sense at one scale may stop making sense at the next.
Dublin is the clearest example. Its connectivity and concentration of technology companies made it an obvious hub. But once data centres consume more than a fifth of national electricity, another large facility has consequences that the first one did not.
Northern Virginia shows the same effect at greater scale. Exceptional fibre connectivity and proximity to customers made it the world’s leading data-centre cluster, but continued concentration means each additional campus adds to an already substantial burden on the power system.
Phoenix presents a different trade-off. Its land, connectivity and business environment are attractive. Its long-term water constraints are not.
The problem is cumulative pressure. Developers often assess projects one at a time. Grids and water systems experience them together.
New locations are not necessarily better
Congested grids and scarce land are pushing developers into new markets. That creates an opportunity to place capacity where resources are cleaner and less constrained, but new does not automatically mean better. Texas offers abundant land and renewable-energy potential alongside water stress, extreme heat and grid constraints; Spain combines strong renewable resources with water scarcity. The Nordics benefit from cooler climates and relatively low-carbon electricity systems, but grid capacity remains finite.
Latency, resilience, sovereignty and customer proximity limit how freely workloads can move. More flexible workloads, including some AI training and batch processing, can go where electricity is cleaner and water less constrained. The industry is dispersing, but not yet systematically towards better environmental locations.
The next constraint is geography
Efficiency improvements still matter. Better chips, higher server utilisation and improved cooling allow more computing to be done with less energy and water. But efficiency per unit of compute does not guarantee lower total demand if computing grows faster, nor can efficiency fully compensate for poor location. A highly efficient facility can still place significant pressure on a constrained grid or a stressed catchment.
The industry has traditionally asked whether a site has enough land, power and connectivity, and whether it can be developed cheaply and quickly. Increasingly, it will also need to ask whether the surrounding system can cope. Can the local grid absorb another 100MW, or another gigawatt? What generation will actually meet that demand? Is water plentiful throughout the year? How many other facilities are competing for the same resources?
These questions matter because the scale is changing. Data centres used to be one load among many. In some places they are becoming a defining feature of the power system.
Globally, their footprint is meaningful and growing quickly, but remains modest relative to agriculture, heavy industry, transport and fossil-fuel production. Locally, the picture can be very different. A new facility in a cool region with clean power and plentiful water is a fundamentally different proposition from one added to an overstretched grid in a dry basin.
That does not make the first facility environmentally benign. Its servers still require minerals and semiconductors; its electricity still requires generation and grid infrastructure; and connecting it may require new transmission lines, substations and roads. Siting is only part of the picture: some of the footprint sits in the infrastructure and supply chains that support the facility.
The industry has spent years making its machines more efficient. The next challenge is more basic: putting them, and the infrastructure that supports them, in the right places.
