As artificial intelligence (AI) spreads, data centres are sharply increasing their water use, and changing cooling infrastructure has emerged as a key task for the industry.

International Energy Agency (IEA) data cited by IT outlet TechRadar on July 7 (local time) put the data centre industry’s annual water consumption at about 560 billion litres. A recent report projects the figure could rise to as much as 1.2 trillion litres by 2030.

Cooling methods are at the centre of the problem. Data centres use large volumes of water to reduce heat generated by servers and semiconductors, and many facilities still rely on open evaporative cooling systems. In this setup, heated water is sprayed onto cooling pads and heat is reduced through evaporation, requiring continuous replenishment of water lost to evaporation.

Closed-loop liquid cooling has been presented as an alternative. It repeatedly circulates the same water and removes heat through a heat exchanger without evaporating water. Because the same water is continuously recirculated, it can ease usage to some extent under normal operating conditions.

Improving cooling efficiency is also directly linked to water savings. Some data centres currently spend up to one-third of total electricity use on heat management due to inefficient cooling. In the process, fans cool the entire data centre space and individual equipment is also cooled separately, increasing the burden of both power and water. Alternatives being discussed include direct node cooling and direct chip cooling.

Direct chip cooling attaches cold plates to heat-generating components such as GPUs and runs coolant through them to absorb heat directly. Liquid-cooling technologies, including direct node cooling, can remove up to 98 percent of heat generated by servers. Data centre operators can reduce energy demand and lower pressure on local water resources by handling heat at the processor level instead of cooling hot air throughout the room.

The expansion of AI infrastructure is accelerating the shift. GPUs used for generative AI can use 5 to 10 times more energy than CPUs, and cooling burdens are rising as high-density designs pack more components into tight spaces using technologies such as 3D silicon stacking. As a result, high-temperature-water-based closed-loop cooling has been presented as a water-saving alternative that can replace evaporative cooling while maintaining performance even in high rack-density environments. DreamWorks Animation introduced this method for high-performance computing, raising performance by 20 percent and reducing cooling requirements.

Waste-heat reuse has also been cited. Most of the electricity used by data centres turns into heat, but it has often been treated as a byproduct to remove and discard. That heat can become a resource supplied to residential areas, commercial buildings and district heating.

Some projects in Ireland and Scandinavia already reuse data centre waste heat for heating homes and workplaces. A structure that supplies discarded heat back to local communities can help reduce reliance on fossil fuels and contribute to cutting energy costs.

Site selection strategies are also needed to avoid locations that put heavy pressure on local freshwater supplies. Forty-five percent of IT experts said current data centre design does not support sustainability goals.

Reducing data centre water use is linked to shifting overall operating structures rather than a single technology. The article says efforts need to move together to cut open evaporative cooling, expand closed-loop circulation systems and chip-level cooling, use waste heat as an external resource and redesign site selection. With AI infrastructure continuing to grow, changing cooling methods and water-use structures is expected to remain a yardstick for judging the sustainability of the data centre industry.