Data Centers Have a Water Problem Nobody Wants to Talk About

The AI infrastructure boom is generating enormous amounts of heat — literally. Every GPU cluster running large language models produces thermal output that must be managed, and the dominant method remains water-based cooling. A standard 100-megawatt data center consumes approximately two million liters of water every day, according to the International Energy Agency. Globally the sector consumes more than 560 billion liters annually, with projections pointing toward 1.2 trillion liters by 2030. That number would be concerning in any geography. It becomes genuinely alarming when you map where the data centers are actually being built. More than 160 new AI data centers have been constructed across the United States in the past three years, and close to two-thirds of them are located in regions classified as having high or extremely high water stress.

The five states that account for 72% of new data center construction in water-stressed regions are Virginia, Arizona, Texas, California, and Illinois. Texas alone offers a particularly vivid case study. The Houston Advanced Research Center estimates that data centers in the state will consume 49 billion gallons of water in 2025, a figure projected to reach 399 billion gallons annually by 2030 — equivalent to drawing down Lake Mead, the country's largest reservoir, by more than 16 feet in a single year. This is occurring in a state where reservoirs and groundwater are already declining, Corpus Christi is approaching emergency usage restrictions, and communities are actively competing for what water remains. OpenAI has selected Abilene, Texas, as the anchor site for its 1.2-gigawatt Stargate data center campus, adding a facility of extraordinary scale to a region already under significant hydrological stress.

The community-level consequences are beginning to surface with increasing frequency. In Fayette County, Georgia, residents noticed declining water pressure throughout 2025, which eventually led investigators to discover unauthorized industrial-scale water connections feeding a 615-acre data center campus developed by QTS, a Blackstone-owned firm. In Tucson, Arizona, the city council unanimously rejected an Amazon-linked data center project after determining it would immediately become the single largest consumer of the municipal water utility. In Cedar Rapids, Iowa, state officials discovered 40 unpermitted wells at a QTS data center site. These are not isolated incidents. They reflect a structural pattern in which water has historically been treated as a secondary consideration in data center siting decisions, subordinate to power availability, real estate cost, and regulatory environment.

The industry is beginning to respond, though the scale of the response remains modest relative to the challenge. Microsoft, Google, and Meta have all committed to reducing water consumption and have invested in alternative cooling technologies including cold plates, single-phase immersion, and two-phase immersion systems that can reduce water usage by 31 to 52% over their operational lifetimes. MSCI analysis indicates that approximately 30% of data centers currently under construction are in regions where water scarcity is projected to intensify by 2050, suggesting that the infrastructure being built today will face escalating resource constraints within its operational lifespan. Water is becoming a material risk factor for data center operators, and investors evaluating hyperscaler and cloud infrastructure stocks would be wise to begin treating it as such.

Key Takeaway: The concentration of AI data centers in water-stressed regions represents a growing operational and regulatory risk that is underpriced in current infrastructure valuations and will intensify as AI compute demand scales toward 2030.

Sources