White paperFree to read2026

Water Usage at AI Scale: Insights from a 100 MW Comparative Analysis

Schneider Electric · Fabian Janischowsky, Amir Ibraheem et al.

Schneider Electric WP 220 treats data center water consumption as a design choice rather than a fixed consequence of size or climate. It simulates four 100 MW designs in Dallas and Paris to show how WUE collapses as TCS temperature rises.

On this page

Overview

The study compares four designs: a traditional air-cooled site, an AI factory with liquid cooling at 32 C TCS, the same factory at 45 C TCS, and a 45 C version with right-sized equipment. Each is simulated in Dallas and Paris. The scenario basis is 188 kW liquid-cooled server racks, 86.6 MW of liquid plus 13.4 MW of air-cooled load, and networking racks of 20/40 kW cooled by air.

The effect of each step is given separately. Moving from air to liquid (32 C TCS) lowers on-site water about 8% in Dallas, and PUE about 7% (cooling overhead about 63% lower). Raising TCS from 32 C to 45 C cuts WUE 66 to 82% depending on location while improving PUE only about 0.6%. Example values: Dallas 0.42 to 0.22 L/kWh, Paris 0.12 to 0.06 L/kWh. Right-sizing the equipment for 45 C reduces WUE about 50% against the 32 C baseline.

The paper also supplies typical WUE ranges: about 1.7 to 4 L/kWh with cooling towers and about 0.2 to 1.5 L/kWh with adiabatic dry coolers. It shows adiabatic assist is poor value at high TCS temperature: in Paris it improved PUE by only about 0.13% while raising water use from 0 to 0.07 L/kWh. The conclusion is that water use is not a fixed price; cooling architecture and liquid temperature can shift it by 66 to 82%. The limit is that only two climates and four schemes are modelled, so other climates need their own calculation.

Key points

  1. Typical WUE is about 1.7 to 4 L/kWh with cooling towers and about 0.2 to 1.5 L/kWh with adiabatic dry coolers.
  2. Air to liquid (32 C TCS): on-site water about 8% lower in Dallas; PUE about 7% lower (cooling overhead about 63% lower).
  3. TCS 32 to 45 C: WUE down 66 to 82%, PUE down about 0.6% (cooling overhead about 13%).
  4. Right-sizing for 45 C: WUE down about 50% and PUE about 0.5% versus the 32 C baseline.
  5. Dallas WUE falls from 0.42 to 0.22 L/kWh; Paris from 0.12 to 0.06 L/kWh.
  6. In Paris at high TCS, adiabatic assist cut PUE only about 0.13% but raised water from 0 to 0.07 L/kWh.
  7. Scenario basis: 188 kW liquid-cooled racks; 86.6 MW liquid plus 13.4 MW air-cooled load.

Using it in design

  • Compute and display WUE (L/kWh IT) next to PUE for every cooling design.
  • Use default WUE ranges: tower 1.7 to 4, adiabatic dry cooler 0.2 to 1.5, dry cooler without adiabatic 0.
  • When TCS is 40 C or higher, recommend switching off adiabatic assist and show the water-versus-energy delta.
  • For AI factories default the heat split to about 87% liquid / 13% air; add a water-stressed-site flag that steers toward high-temperature liquid with dry coolers.

Monthly newsletter · free

Keep your reading list current

Once a month: new library sources, changes to Turkish regulations and upcoming data centre events.

SoSo CAD

Check cooling capacity alongside your design

Review zone heat loads, available capacity and the capacity balance after the largest unit is lost in SoSo CAD.

Request a demo
Zone heat loads and capacity balance in the SoSo CAD cooling workspace

Who should read it

Designers planning AI campuses in water-constrained regions, and decision makers who must commit to a WUE target.

Related standards

  • ISO/IEC 30134-9 (WUE)
  • The Green Grid WUE
  • ASHRAE TC9.9 Liquid Cooling Guidelines

Publication details

Authors
Fabian Janischowsky; Amir Ibraheem; Maria A. Torres Arango
Publisher
Schneider Electric
Year
2026
Document number
WP 220
Document language
English
Source type
Manufacturer
Access
Free to read
Licence
Copyright retained; citation only — the publisher’s terms apply to the original document

Citation

Fabian Janischowsky; Amir Ibraheem; Maria A. Torres Arango (2026). Water Usage at AI Scale: Insights from a 100 MW Comparative Analysis. Schneider Electric. WP 220. https://download.schneider-electric.com/files?p_Doc_Ref=SPD_WP220_EN

All content and rights belong to the respective authors and publishers. SoSo CAD provides only the citation, its own summary and a link to the source, and is not responsible for the sources' content.