TechFEATURE

Why AI Data Centers Guzzle Water — And How WUE Measures It

林紀旭 James LinEditor-in-Chief
Published · Updated
AI data centers draw water mainly through evaporative cooling towers, and Microsoft reports an average Water Usage Effectiveness (WUE) of 0.30 liters per kilowatt-hour across its facilities. The IEA attributes about 60% of that footprint to power generation and 40% to on-site cooling, together equal to 6,500 households' water use. One Oregon city found nearly 30% of its municipal water tied to Google's data centers, which tripled consumption in five years. Microsoft's chip-level cooling design now eliminates cooling water entirely, while Google has pledged to replenish more freshwater than it consumes by 2030.

Why Do AI Data Centers Use Water in the First Place?

Data centers primarily consume water through evaporative cooling towers, which spray water onto internal surfaces to accelerate heat transfer, and that water must be constantly replenished as it evaporatesCITE:E1. This physical process is the root cause of what is commonly called "direct" water use. But the footprint doesn't stop at the facility gate: the International Energy Agency (IEA) estimates that data center water use splits roughly 60% indirect (consumed at power plants generating the electricity) and 40% direct (consumed on-site for cooling), with the combined total equal to the water consumption of 6,500 householdsCITE:E7.

How Is Water Usage Effectiveness (WUE) Actually Measured?

Water Usage Effectiveness (WUE) measures a data center's water consumption relative to the electricity it uses, expressed in liters per kilowatt-hourCITE:E2. It is the water-focused counterpart to energy efficiency metrics, letting operators track sustainability performance across facilities on a common unit. Microsoft has disclosed that its data centers operated with an average WUE of 0.30 liters per kilowatt-hour over its last fiscal yearCITE:E3, giving the industry a concrete benchmark for what "efficient" cooling looks like in practice.

How Does Data Center Water Use Strain Local Communities?

Data center cooling water is often drawn from local municipal utilities, and one city in Oregon found that nearly 30% of its total water consumption was attributable to Google's data centersCITE:E5. That same reporting found the water consumption from those facilities had tripled over a five-year periodCITE:E5. The case illustrates how a company-wide average like WUE can coexist with concentrated strain on a single community's water supply.

Can Data Centers Eliminate Cooling Water Entirely?

Microsoft's next-generation data center design, launched in August 2024, uses chip-level cooling that consumes zero water and produces zero evaporation during the cooling processCITE:E6. By replacing evaporative cooling towers with precise chip-level temperature control, Microsoft states this design avoids the need for more than 125 million liters of water per year, per data centerCITE:E6. Water is still used for administrative purposes such as restrooms and kitchens, but the cooling process itself no longer requires itCITE:E6.

What Have Tech Giants Pledged on Water Replenishment?

Google has committed to replenishing more freshwater than its data centers and offices consume, on average, by 2030CITE:E4. Rather than targeting zero consumption on-site, Google's framing is net-positive: replenishment volume exceeding withdrawal volume across its global portfolio.

Key Figures at a Glance

MetricValueSource
Microsoft average WUE (last fiscal year)0.30 L/kWhCITE:E3
IEA data center water use split60% indirect (power plants) / 40% direct (on-site)CITE:E7
Combined water use, IEA estimateEqual to 6,500 householdsCITE:E7
Oregon city's water tied to Google data centers~30% of city consumptionCITE:E5
Growth in that city's data center water useTripled over 5 yearsCITE:E5
Microsoft chip-level cooling water savings>125,000,000 liters/year per data centerCITE:E6
Google freshwater replenishment targetReplenish more than consumed, on average, by 2030CITE:E4

What This Means

The numbers point to a gap between how water use is measured and where the strain actually lands. Microsoft's 0.30 L/kWh average and its zero-water chip-level cooling design address the 40% direct-use share the IEA identifiesCITE:E7CITE:E6, but neither figure touches the 60% indirect share consumed at power plantsCITE:E7. Meanwhile, the Oregon case shows that even where a company-wide average looks efficient, a single municipality absorbed nearly 30% of its water supply and saw that draw triple in five yearsCITE:E5. Google's 2030 replenishment pledge is a portfolio-level, net-positive targetCITE:E4, which is a different accounting frame than a facility-level metric like WUE or a zero-water cooling redesign — none of the three approaches (efficiency metrics, replenishment pledges, cooling redesigns) is shown in the evidence to directly offset the other two.

📊 Evidence

FAQ

Why Do AI Data Centers Use Water in the First Place?

Data centers primarily consume water through evaporative cooling towers, which spray water onto internal surfaces to accelerate heat transfer, and that water mu…

How Is Water Usage Effectiveness (WUE) Actually Measured?

Water Usage Effectiveness (WUE) measures a data center's water consumption relative to the electricity it uses, expressed in liters per kilowatt-hourCITE:E2.

How Does Data Center Water Use Strain Local Communities?

Data center cooling water is often drawn from local municipal utilities, and one city in Oregon found that nearly 30% of its total water consumption was attribu…

Can Data Centers Eliminate Cooling Water Entirely?

Microsoft's next-generation data center design, launched in August 2024, uses chip-level cooling that consumes zero water and produces zero evaporation during t…

📎 Sources

  1. congress.gov
  2. datacenters.microsoft.com
  3. microsoft.com
  4. datacenters.google

Related data

Author's Take林紀旭 James Lin

The most telling number in this dataset isn't Microsoft's 0.30 L/kWh average — it's the IEA's 60/40 split showing most of a data center's water footprint sits invisibly at the power plant, not the cooling tower. That means facility-level fixes like Microsoft's zero-water chip-level cooling only address the smaller (40%) direct-use slice, while the larger indirect share tracks the electricity mix, not the cooling architecture. The Oregon case matters for the same reason: a company can report an efficient global WUE average while one municipality absorbs a tripling of local draw in five years and ends up supplying nearly 30% of its water to a single operator's facilities. The indicator worth watching is whether chip-level, zero-water cooling designs like Microsoft's actually get deployed at scale across new builds — that's the one lever in this evidence that shrinks the direct-use number rather than just reallocating or pledging to offset it.

林紀旭 James LinEditor-in-Chief

Related

BRIEF

GreenTrans Unveils GT5X, GT3X Quadruped Robots, Targets 100% Taiwan-Made Content by 2027

GreenTrans (綠捷), the robotics subsidiary of China Motor (中華車), unveiled quadruped robots GT5X and GT3X at SEMICON Taiwan 2026, targeting 100% Taiwan-made content by 2027. The robots combine an in-house-designed control unit and battery management system, NVIDIA's Jetson Orin and Isaac Lab platforms, and a new LFP battery developed with Formosa Smart Energy (台塑新智能), while GreenTrans's inspection robots are already deployed in semiconductor fabs.

EffectStory 編輯部 ·
BRIEF

Nvidia Confirms $12.93 Billion Acquisition of Hugging Face

Nvidia confirmed on September 3, 2026 that it agreed to buy Hugging Face for $12.93 billion, exactly $12,930,300,000, gaining the open-source AI hosting platform used by over 18 million developers. CEO Jensen Huang pledged the platform will stay open, with no Nvidia compute required to build on or deploy through it.

EffectStory 編輯部 ·
BRIEF

NVIDIA to Subscribe US$3.5 Billion of MediaTek's Record US$3.9 Billion Convertible Bond

NVIDIA will subscribe US$3.5 billion of MediaTek's US$3.9 billion offshore convertible bond offering, the largest such issuance in Taiwan's capital market history, deepening cooperation in AI infrastructure, edge AI computing, and automotive platforms while marking NVIDIA's first major investment in a Taiwanese company.

EffectStory 編輯部 ·