Photo: panumas nikhomkhai / Pexels
AI data centers are saving water on-site and spending it at the power plant
New cooling systems cut local water use, but the extra electricity they require pushes consumption upstream to thermal power generation
As Bloomberg reports, the newer cooling technologies spreading through AI facilities save water at the site itself. They also raise total water use once you count the power plants keeping those facilities running.
Where the water actually goes
Data centers have traditionally relied on evaporative cooling. Water absorbs heat from the equipment, then evaporates into the air, taking that heat with it.
The newer approaches include closed-loop liquid cooling and immersion cooling. Closed-loop systems circulate the same fluid repeatedly instead of evaporating it away. Immersion cooling goes further, submerging hardware directly in a cooling liquid.
Both can reduce or completely eliminate on-site water use for evaporation. The problem is electricity. Liquid and closed-loop systems may increase electricity demand by 6-35% during peak conditions compared with evaporative cooling. More power means more generation, and many thermal power plants use substantial water for their own cooling.
The numbers behind the shift
Lawrence Berkeley National Laboratory estimated that US data centers directly consumed about 17.5 billion gallons of water in 2023. Indirect consumption tied to producing the electricity those data centers use was approximately 12 times higher than the direct cooling figure.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
AI is making the math harder. Older data center facilities typically ran server racks at 6-10 kW. AI server racks can now exceed 100 kW.
Looking ahead, total water footprints could grow significantly by 2030 under AI expansion scenarios. In those projections, electricity-related water use makes up the majority, at around 50-72% of total consumption.
How the hyperscalers stack up
The industry tracks this with a metric called water usage effectiveness, or WUE. It measures how many liters of water a facility uses per kilowatt-hour of energy.
AWS reported a WUE of 0.12 liters per kWh. Microsoft anticipates reaching 0.27 liters per kWh by fiscal 2025.
Notably, WUE focuses on water used at the facility. It does not capture the water consumed at the power plant supplying the electricity.