Study: Data center water needs emerging as new AI ‘bottleneck’

A new academic study warns that the growing water demands of data centers needed to train and run artificial intelligence models could be a major challenge for the AI industry. Researchers from UC Riverside, Caltech, and RIT note that data centers often require large amounts of water for cooling systems, sometimes consuming millions of gallons per day. This water demand can exceed the available supply in the areas where data centers are being built, leading to the need for new infrastructure, operational delays, or switching to less water-intensive but more power-hungry cooling methods. The issue could particularly impact major AI companies like OpenAI, Anthropic, Meta, and Alphabet that are heavily investing in data center capacity to fuel the growth of their AI models and services. OpenAI recently scrapped plans to expand its data center in Abilene, Texas due to water concerns. Previous research has focused on the aggregate water usage of data centers nationwide, but this new study highlights how the local water capacity of individual regions may be a key constraint on future AI development. The researchers argue policymakers need to pay closer attention to the water implications of the surging data center construction needed to support AI.