There is a familiar habit in the tech world of treating efficiency as a kind of reset button. If machines get faster, smaller, and smarter, then the resources behind them are assumed to shrink too. That idea is now being tested more uncomfortably. A recent United Nations report has pulled together a range of projections about artificial intelligence and the infrastructure behind it, suggesting that the physical cost of digital systems is moving in the opposite direction to the one often assumed. The picture it paints is not about sudden collapse or alarmist scarcity, but a steady expansion of demand for electricity, cooling water, land, and raw materials. Some of those demands are already visible in the growth of data centres, others sit further ahead in forecasts that stretch to the end of the decade.
Why AI's growing network of data centres is driving a surge in electricity demand
Most people still experience AI as something weightless, a chat window, or a tool that appears instantly on screen. Behind that simplicity sits a growing network of warehouses filled with servers that never really pause. These data centres are becoming larger and more numerous, spreading across regions where power supply and land access make expansion easier.
The UN report draws attention to how quickly this infrastructure has grown. At present, global data centres collectively consume electricity on a scale comparable with mid-sized nations. Saudi Arabia is often used as a reference point in these comparisons, not because of any direct link, but because it gives a sense of proportion that is easier to picture than abstract gigawatts.
The expectation is not that this plateaus. Instead, demand curves continue upwards as more systems are trained, deployed, and repeatedly queried by billions of users.
The hidden water footprint behind AI’s cooling systems
Electricity is only part of the story. Servers generate heat in large volumes, and that heat has to be managed continuously. This is where water enters the system in a way most users rarely think about.
Cooling systems in data centres rely on significant water supplies, either directly or through electricity-intensive refrigeration cycles. In many regions, this water competes with municipal and agricultural needs, although the competition is not always visible at ground level. It happens through contracts, supply chains, and regional planning decisions that are not widely discussed outside technical circles.
The report suggests that if current growth patterns continue, the water required for cooling AI-related infrastructure could reach volumes that begin to sit awkwardly alongside basic human consumption needs at a global scale. It is less a prediction of literal shortage in every location and more a comparison meant to show scale.
How lower costs are expanding AI’s global footprint
There is a persistent belief that newer AI models will ease pressure on infrastructure. Smaller models, better chips, improved software design, all of that is expected to reduce the cost per task. On paper, that logic holds.
The complication appears when usage is taken into account. When something becomes cheaper and easier to access, it tends to be used more often and in more places than originally planned. The UN report leans on an old economic idea to describe this pattern, known as the Jevons paradox. It came from observations about coal use in industrial Britain, where improved efficiency did not reduce consumption, but expanded it.
AI appears to be following a similar path. Lower costs per query or per image generation do not necessarily reduce overall demand. They often unlock new applications, some of them trivial, others embedded into large systems that run continuously.
Growing share of AI in global electricity consumption by the end of the decade
By the end of the decade, projections in the report place AI-related electricity use at a significantly larger share of global consumption than today. Around a few percent of the world’s total power demand is mentioned in some scenarios. That might sound small at first glance, but in global energy terms, it represents a large industrial sector being added almost on top of existing ones.
Associated emissions depend heavily on how electricity is generated in different regions, but the report notes that without cleaner grids, the growth of data infrastructure risks locking in higher carbon output. Water use follows a similar pattern of regional dependence, concentrated in areas where data centres cluster rather than evenly spread across countries.
There is also a land footprint that tends to be overlooked. Large facilities require not just buildings, but buffer zones, substations, and transport access. When multiplied across hundreds of sites, the spatial demand becomes more noticeable than the individual projects suggest.
How AI data centres are concentrated in a few key regions
A quieter point in the report concerns geography. AI infrastructure is not evenly distributed. A relatively small number of countries host the majority of high-capacity cloud and AI systems, with the United States and China accounting for a large share of global capacity.
That concentration shapes more than economics. It also determines where energy demand rises most sharply and where environmental pressure is felt most directly. Meanwhile, many countries remain primarily consumers of AI services rather than hosts of the infrastructure itself.
Follow Us On Social Media