Model the power, water and carbon of an AI campus
Size an AI or hyperscale campus from 5 to 250 MW, pick a cooling architecture and grid region, and see annual facility energy, on-site and grid-related water, and location-based Scope 2 emissions, side by side for all four cooling options.
PUE: The Green Grid / ISO/IEC 30134-2WUE: The Green Grid / ISO/IEC 30134-9Scope 2: GHG Protocol location-basedGrid factors: EPA eGRID 2023, SEAI
Your facility
Critical IT load (servers, GPUs, storage, network).
Average annual draw as a share of IT capacity. AI training clusters often run high.
Cooling architecture
Total facility energy divided by IT energy. Global survey average: 1.54 (Uptime Institute, 2025).
On-site cooling and humidification water per kWh of IT energy. Often-cited US average: 1.8 L/kWh.
eGRID 2023 SRVC: 596 lb CO2e/MWh
Advanced assumptions
Water consumed at power plants per kWh delivered. 4.5 L/kWh is the US average implied by LBNL (2024). Replace it with a local value if you have one.
Illustrative weighting (1 = low stress). Check WRI Aqueduct for your exact site.
Annual results
Facility energy
0
Scope 2 emissions (location-based)
0
On-site water (WUE)
0
On-site + grid water
0
On-siteGrid (Scope 2 water)
Switching cooling technology: water and carbon compared
Each card uses that technology’s typical PUE and WUE with your capacity, utilisation and grid region. Differences are against your current settings. Results are planning estimates, not measured values.
Go deeper: the data center water and energy series
Pillar guideGreen data center cooling: immersion vs direct-to-chipPUE, WUE and grid trade-offs of each cooling architecture, with sourced benchmarks.Read the guide →
Web storyAI’s secret water crisisWhy data centers drink so much water, and where it hurts most, in 11 slides.View the story →
Next toolScope 1, 2 and 3 GHG Emissions EngineAdd backup generators, refrigerants and supply chain to your Scope 2 result.Open the engine →
Methodology and sources
- IT energy = IT capacity × 8,760 h × utilisation.
- Facility energy = IT energy × PUE (The Green Grid; ISO/IEC 30134-2).
- On-site water = IT energy × WUE (The Green Grid WUE; ISO/IEC 30134-9). 1 US gallon = 3.785 L.
- Grid water (“Scope 2 water”) = facility energy × grid water intensity at power plants.
- Scope 2 emissions = facility energy × grid emission factor (GHG Protocol Scope 2 Guidance, location-based). Market-based results with PPAs or certificates will differ.
| Cooling preset | PUE | WUE (L/kWh) | Basis |
|---|---|---|---|
| Evaporative cooling towers | 1.30 | 1.80 | 1.8 L/kWh is the US average WUE most often cited in industry guidance |
| Air-cooled chillers (closed loop) | 1.50 | 0.10 | No evaporation; PUE close to the 1.54 global survey average (Uptime Institute 2025) |
| Direct-to-chip liquid | 1.15 | 0.20 | Typical design point with dry coolers; Microsoft reports a 0.30 L/kWh fleet WUE (FY24) and zero-water chip-level cooling in new designs |
| Single-phase immersion | 1.08 | 0.10 | Typical design point; heat rejected through dry coolers (near-zero evaporation; 0.10 is the slider floor) |
| Region | Emission factor | Source | Default stress weighting |
|---|---|---|---|
| Northern Virginia (Dominion zone, PJM) | 0.270 t CO2e/MWh | EPA eGRID 2023, subregion SRVC (596.3 lb/MWh) | 1.5 |
| Phoenix, Arizona | 0.320 t CO2e/MWh | EPA eGRID 2023, subregion AZNM (706.2 lb/MWh) | 4.0 |
| Texas (ERCOT) | 0.334 t CO2e/MWh | EPA eGRID 2023, subregion ERCT (736.6 lb/MWh) | 2.5 |
| Ireland (Dublin) | 0.198 t CO2/MWh | SEAI electricity conversion factor (197.8 g/kWh) | 1.0 |
| US national average | 0.350 t CO2e/MWh | EPA eGRID 2023, US (770.9 lb/MWh) | 2.0 |
- Grid water intensity default (4.5 L/kWh): derived from Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (about 800 billion litres of indirect water for 176 TWh in 2023). The same default is used for Ireland because no Irish figure is published; edit it under Advanced assumptions.
- Context: LBNL estimates US data centers directly consumed about 66 billion litres (17.4 billion gallons) of water in 2023.
- Water-stress weightings are illustrative and do not change the litres or gallons shown. Use WRI Aqueduct for site-specific baseline water stress.
AM
Methodology reviewed by Dr. Ahmad Mahmood, PhD (Environmental Systems)
Sustainability Awakening · last checked 24 October 2026. All calculations run in your browser; nothing you enter is stored.
Sustainability Awakening · last checked 24 October 2026. All calculations run in your browser; nothing you enter is stored.
Frequently asked questions
What is a good PUE for an AI data center?
PUE is total facility energy divided by IT energy. The 2025 Uptime Institute survey found a global average of 1.54. New liquid-cooled AI halls are typically designed for about 1.1 to 1.2, and Google reports a fleet-wide PUE close to 1.09.
What is WUE and how is it calculated?
Water Usage Effectiveness (WUE), defined by The Green Grid and ISO/IEC 30134-9, is the litres of water used on site for cooling and humidification divided by the kWh used by IT equipment. An often-cited US average is 1.8 L/kWh; Microsoft reported a fleet average of 0.30 L/kWh for FY24.
Which data center cooling method uses the least water?
Closed-loop designs that reject heat through dry coolers, such as direct-to-chip liquid cooling, immersion and air-cooled chillers, use far less on-site water than evaporative cooling towers. Air-cooled chillers usually need more electricity, which raises Scope 2 emissions and the water used at power plants.
Why include water used at power plants?
Thermal power plants consume water for cooling, so every kWh a data center draws carries an indirect water footprint. Lawrence Berkeley National Laboratory estimates this indirect footprint for US data centers at about 800 billion litres in 2023, far more than the 66 billion litres used on site.