- Ana Pinheiro Privette, University of Illinois, Urbana-Champaign
As artificial intelligence (AI) becomes increasingly central to environmental services, its own environmental footprint – particularly water use – remains poorly understood. This paper examines the water footprint of AI through the lens of data centers, considering three key dimensions: direct water consumption from cooling systems, indirect water use embedded in the electricity that powers these facilities, and the indirect water embodied in hardware production. Training and deploying large-scale AI models demand immense computational power, which drives high energy use and substantial heat generation. To dissipate this heat, many data center facilities rely on water-intensive cooling systems that draw heavily on freshwater sources. While data centers’ global water use is modest compared to major sectors such as agriculture, manufacturing, and energy, concerns are mounting about their rapid expansion and concentrated local impacts. Hyperscale facilities, in particular, can intensify localized water stress, especially in arid or water-scarce regions, creating competition with residential, industrial, and agricultural users and placing additional pressure on already fragile water systems. These risks highlight the urgency of aligning AI growth and data center development with local and regional water resource planning, environmental policy, and infrastructure resilience. This paper identifies technological and system-level barriers that hinder sustainable AI deployment and proposes strategies to align AI development and deployment with sustainable, water-conscious growth. Key priorities include fostering technological innovation and optimization (e.g., advancing cooling technologies, improving efficiency through model compression, chip design, algorithmic improvement, and operational and network optimization), strengthening water resource management practices (e.g., applying circular water economy principles such as wastewater reuse, water recovery, and closed-loop cooling, while better integrating data center planning into local and regional water management), and enhancing transparency through standardized water-use reporting at both the infrastructure and AI model levels. Ultimately, the paper calls for a multidisciplinary, systems-level approach to AI governance that aligns technological advancement with sustainable water stewardship, ensuring innovation supports – rather than undermines – longterm water security and resilience.
