Amazon’s proposed „Project Blue” data center in Tucson sparked resident backlash over projected water consumption of 1.2 million gallons daily in a region experiencing its worst megadrought in 1,200 years. A new Nature Sustainability study quantifies the concern: AI-driven data center expansion could strain water resources in precisely the states attracting them.
Geographic Mismatch: Tax Breaks vs. Resource Reality
Data centers powering ChatGPT and enterprise AI consumed 460 terawatt-hours globally in 2024, equivalent to Spain’s total electricity use. Evaporative cooling consumes 1.8 liters per kilowatt-hour, translating to 660 million gallons daily across U.S. facilities.
The study analyzed 50 U.S. states using a „suitability index” weighing electricity costs, renewable energy, disaster risk, and water stress. States ranking highest for tax incentives—Arizona, Nevada, Utah—rank lowest in water availability. Arizona offers 75% property tax abatements for investments exceeding $250 million, attracting Meta and Google despite the Colorado River operating at 35% capacity.
Industry Response and Regulatory Lag
Google claims its latest facilities reduce water consumption by 20% through air cooling in moderate climates. However, Greenpeace’s 2024 report found only 18% of hyperscale operators disclose water consumption data.
Each GPT-4 query requires 10x the server resources of a Google search per Epoch AI’s analysis. The authors conclude: „Without coordinated policy intervention, AI’s environmental footprint could undermine climate commitments in water-stressed regions.”
Sources:
– Fast Company: Data Center Environmental Study
– Nature Sustainability: Data Center Suitability Research
– IEA: Global Electricity Report 2024
– Greenpeace: Clicking Clean 2024 Report
– Epoch AI: AI Compute Trends Analysis