Rethinking data centre cooling: cutting water use with liquid cooling and alternatives

Overview

The Conversation explains how data centres powering AI workloads rely on water-intensive cooling systems, and it surveys alternatives such as direct-to-chip liquid cooling and evaporative cooling, highlighting the trade-offs between water use and energy consumption. The piece places cooling in the broader context of water scarcity and energy systems, and it points to Ireland as a case study for water management in data infrastructure.

• Data centres currently use water to carry away waste heat through chilled water loops and cooling towers.
• Direct-to-chip liquid cooling can reduce water use and enable heat reuse, but retrofitting can be costly and complex.
• Evaporative cooling leverages latent heat to cut electricity needs but increases water demand.
• Electricity generation itself consumes water, so reducing energy use can be a net water-saver in some scenarios, with rainwater harvesting offered as a local water-reduction strategy.

Introduction

There’s a growing demand for data centres to support artificial intelligence workloads, which has accelerated the construction of new facilities around the world. The Conversation explains that cooling these facilities is energy-intensive and heavily water-dependent, creating a tension between reliability, energy costs, and water scarcity.

The cooling challenge in data centres

Traditionally data centres use air cooling, where servers emit heat into room air that is circulated by fans. The air then passes through heat exchangers before being discharged. The process is energy-intensive because air is a relatively poor coolant, so large volumes of air must be moved and the cooling plant and fans consume substantial electricity. Water, by contrast, has superior thermal properties, making it an attractive medium for transferring and removing heat when used in chilled water loops and heat exchangers. Evaporative cooling can then reject heat to the environment via cooling towers, where water evaporates and the remaining water is cooled for recirculation.

Water's cooling advantages and limitations

Water’s high specific heat capacity and latent heat of vaporization underpin its effectiveness as a cooling medium. These properties enable water to transport large amounts of energy with relatively small temperature changes or by exploiting evaporation to achieve significant cooling. However, using water comes with risks such as potential leaks and the environmental and logistical implications of water withdrawal and consumption, especially in arid regions or during droughts.

Direct-to-chip cooling versus conventional air cooling

Direct-to-chip cooling involves sealing pipes that bring liquid coolant directly to CPUs and GPUs to extract heat at the source. This method can substantially reduce the total cooling infrastructure and potentially enable waste-heat reuse. Yet retrofitting an existing, air-driven data centre to a liquid-cooled configuration presents costs, downtime, and compatibility considerations, meaning a full transition is not universally feasible. In cases where direct cooling is deployed, the remaining air cooling elements may be retained for certain equipment classes, maintaining a hybrid approach.

Evaporative cooling and water use

Evaporative cooling remains widely used because it can reduce electrical energy consumption for cooling. Still, it consumes water to evaporate some of the water used in the process. The Conversation notes that while evaporative cooling can lower electricity demand, the overall water footprint must be considered alongside the energy footprint, since electricity generation also requires significant water resources in many energy mixes.

Global and regional perspectives

A recent UN report warns that water used by AI data centres could reach levels comparable to the everyday water needs of billions of people by 2030. The article also cites Ireland as an example where data centres account for a large share of electricity consumption, about a quarter of metered usage, underscoring the interlinked challenges of energy and water in data infrastructure. There is also discussion of rainwater harvesting as a possible strategy to lessen pressure on water networks for data centres.

What is the path forward?

To address the water–energy nexus, the piece argues for a staged approach: adopt liquid cooling where feasible to reduce water intake and enable heat reuse, continue with evaporative cooling where it provides net energy savings, and pursue water-management strategies such as rainwater capture when local conditions permit. Ultimately, reducing electricity demand through efficient cooling and smarter design can lessen the water footprint even when evaporative cooling is used, but context matters, including climate and water availability.

Conclusion

The Conversation emphasizes that data centres are at the intersection of technology, energy systems, and water resources. A nuanced mix of cooling strategies—tailored to local water availability, energy costs, and upgrade cycles—can help reduce water consumption while maintaining the high performance required by AI workloads.