Artificial Intelligence and the Hidden Water Footprint of Data Centres: Cooling, Consumption and Environmental Sustainability

Abstract

Abstract The rapid expansion of artificial intelligence (AI) has created an unprecedented demand for computational infrastructure. Behind applications such as generative AI, image generation, machine translation, and automated decision-making are large data centres containing thousands of high-performance processors. While discussions of AI's environmental impact have traditionally focused on electricity consumption and carbon emissions, its water footprint is receiving increasing attention. AI workloads generate substantial heat because of the high power density of modern computing equipment, making effective cooling essential for reliable operation. Many data centres use cooling systems that consume water through evaporation, while additional water is associated with electricity generation and semiconductor manufacturing. This paper examines how AI data centres operate, why their cooling systems require water, and how the expansion of AI may affect freshwater resources. It adopts a secondary-research methodology based on peer-reviewed research, institutional reports, and publicly available environmental assessments. Particular attention is given to the distinction between water withdrawal and water consumption, because treating these measures as interchangeable can produce misleading conclusions. Existing research indicates that global AI-related water withdrawal could reach approximately 4.2–6.6 billion cubic metres annually by 2027 under the scenarios examined by Li et al. (2025). At the same time, technological developments such as direct-to-chip liquid cooling, closed-loop systems, dry cooling, improved data-centre siting, and water recycling provide opportunities to reduce freshwater dependence. The paper argues that the sustainability of AI cannot be assessed through energy and carbon metrics alone. Water availability must become an important consideration in the planning, operation, and governance of future AI infrastructure.