Research Publications

Artificial intelligence (AI) is becoming part of everyday life, shaping how people work, communicate, search for information, and make decisions. Public attention has rightly focused on AI’s energy demands, the expansion of data centers, and the carbon implications of powering large-scale computation. Yet one critical resource remains far less visible: water. This lack of visibility is a matter of concern not only for public awareness, but also for data transparency. Recent analysis of major technology company sustainability reports found that disclosure of water-related data center metrics remains limited and inconsistent, with AI-specific water metrics and indirect water consumption rarely disclosed by companies (de Vries-Gao 2026).

Behind AI’s rapid growth is a massive and expanding physical infrastructure that must be cooled, powered, maintained, and continuously upgraded. Data centers generate substantial heat, and keeping them operational often requires significant water use, either directly through cooling systems or indirectly through the electricity that powers computation. At the same time, AI is increasingly being applied in water management itself, supporting demand forecasting, system optimization, leak detection, climate analysis, and real-time operational decision-making. This dual role makes AI especially important for water resources planning: it can help improve water management, but it can also intensify pressure on the very systems it is intended to support.

AI’s water footprint extends beyond direct thermal management. It includes operational water use for data center cooling, indirect water use associated with electricity generation, and embodied water use in hardware manufacturing and supply chains (Herrera et al. 2025; Barnett-Itzhaki 2026). This broader framing matters because AI is not merely a digital technology. It is a resource-dependent infrastructure system connected to energy production, industrial supply chains, land use, and local watersheds.

Clear terminology also matters. Water withdrawals refer to the amount of water taken from a source, while water consumption refers to the portion that is not returned to the same watershed in a usable form or within a relevant time frame. This distinction is important because evaporative cooling can convert withdrawals into consumptive use by transferring water to the atmosphere. Data centers are often evaluated using water usage effectiveness (WUE), but reported WUE values can vary substantially with climate, cooling technology, water source, operating practices, and reporting boundaries (Lei et al. 2025). As a result, a facility may appear efficient under a narrow metric while still contributing to local water stress, especially when water is used during dry periods or in constrained basins.

This editorial highlights critical gaps and misconceptions in how AI’s hidden water footprint is understood, measured, and incorporated into planning and policy, and calls for greater transparency, consistency, and action in water resources management.