The Hydro-Digital Paradox: Water Scarcity in the Age of Artificial Intelligence

Abstract

The development of artificial intelligence is rapidly accelerating, and the demand for computational infrastructure is similarly growing. While these rapidly expanding data centers, known as hyperscale data centers, are of crucial importance for both instruction and operation of advanced AI models, the environmental consequences of this water consumption remain largely invisible to end users. While data centers remain a small fraction of total water consumption, increasing AI capabilities have led to massive increases in water consumption, in many cases, amounting to water consumption similar to that of thousands of households. Hyperscale facilities are always located in urban areas and often drought areas, which creates a degree of competition for scarce resources while raising questions of environmental concerns and community well-being. From a structural standpoint, the cooling of AI infrastructure has significant water requirements and is routinely accomplished via evaporative cooling, which represents one of the most water-intensive activities in the digital economy. As AI becomes more and more integrated into the global system--economically and socially--the issues raised by the need for physical infrastructure that sustains this activity are an urgent challenge for policymakers, technology, and environmental advocates.