AI sustainability is often framed as a carbon problem. It is also a water problem. Modern data centres rely on water-intensive cooling — and as AI workloads grow, water use becomes a strategic and reputational issue, especially in water-stressed regions.

What research has highlighted

Academic analysis has drawn attention to water consumption associated with model training and inference, including cooling requirements. One widely cited study is hosted on arXiv: University of California, Riverside (arXiv, 2023). For broader context on data centre growth and electricity drivers, see the International Energy Agency’s reporting on electricity demand and data centres: IEA: Electricity 2024.

Why it matters for organisations

Water becomes relevant in three ways:

• Operational resilience: water constraints can limit data centre operations.
• Risk and reputation: water use can trigger community pushback and policy pressure.
• Location strategy: where compute runs affects sustainability outcomes.

Practical actions

• Ask for transparency: request sustainability reporting from vendors that includes water.
• Optimise workloads: right-size models; reduce unnecessary inference.
• Consider geography: prefer regions with cleaner grids and lower water stress where possible.

As AI adoption scales, sustainability diligence will expand. Leaders who track water now will be ahead of the next wave of scrutiny.