Water Usage & Resource Demands

Data centre cooling requires significant amounts of water, and AI workloads - which generate more heat than traditional computing - are intensifying this demand. Estimates suggest that training a large language model can consume millions of litres of water for cooling. Google reported that its global water consumption increased by 20% in 2023, a rise attributed partly to AI demand. This is particularly concerning in water-stressed regions, where data centres compete with agricultural, industrial, and domestic water needs. Some data centres use evaporative cooling, which is energy-efficient but water-intensive. Others use closed-loop systems that recirculate water but require more energy. The choice involves trade-offs between water use, energy use, and cost. Beyond water, the hardware underpinning AI has its own resource demands. GPUs and AI chips require rare earth elements, cobalt, lithium, and other materials with their own environmental and ethical extraction concerns. The rapid hardware upgrade cycle in AI - where chips become obsolete within a few years as performance demands increase - creates electronic waste challenges. For a complete picture of AI's environmental impact, you need to look beyond energy and carbon to consider water, materials, and waste across the entire lifecycle. These factors are less well-measured and less frequently discussed, but they're increasingly relevant as AI infrastructure scales.