Amazon has released its first-ever water usage report for its global data centers, touting impressive numbers: 2.5 billion gallons (~9.5 billion liters) of water consumed in 2025, with a water usage effectiveness (WUE) of 0.12 liters per kilowatt-hour, a 2% improvement year-over-year despite expanding operations. The report includes comparison charts showing Google, Microsoft, and Meta all consuming more water per kWh, leading Amazon to claim it is the industry's most water-efficient operator.
The Fine Print: Missing Indirect Water Footprint
The 0.12 liters/kWh figure looks stellar against the industry average of 0.84 liters/kWh. However, the report's fine print reveals that it excludes indirect water consumption from power plants supplying electricity to data centers, as well as water used during new facility construction. Thermal power plant cooling water typically accounts for the largest share of a data center's total water lifecycle footprint — far exceeding on-site operational use. Amazon's 2.5 billion gallons only represent direct costs; the indirect ledger could be much thicker.
Furthermore, 0.12 is an annual average that masks peak-demand periods and facilities in water-stressed regions. A pretty annual average does not show the real burden during the hottest weeks or at the driest sites. Another issue: the comparison data unfavorable to Google focuses specifically on Google's Gemini AI data centers, while Amazon reports across all its data center operations. The denominators are different — pitting a specific AI workload against an entire business portfolio makes the chart methodologically questionable.
Cooling Technology: Air-Cooling Works, but Total Consumption Still Rises
Amazon explains its low WUE: about 90% of the time, facilities use air cooling (no water); evaporative water cooling kicks in only during the hottest hours, and servers are designed with higher heat tolerance. This aligns with industry trends — Microsoft is rolling out closed-loop, zero-evaporation cooling systems that save over 125 million liters per facility per year. But can efficiency gains keep pace with compute demand? A structural tension emerges: every company shows improving efficiency curves, yet absolute total electricity and water use continues to climb. One study warns that by 2030, AI data center water consumption could equal the annual drinking water needs of 1.3 billion people.
Water vs. Electricity: Local Resource Competition Intensifies
2.5 billion gallons may seem manageable globally, but water is not dispatched across grids like electricity. It is drawn from specific local watersheds and community supplies. A data center in a water-scarce county, even with a stellar WUE, can still compete with residents and agriculture for the same aquifer or river rights. That's why local governments like Seattle hit pause — global averages do not settle local water bills.