AI Model Water Footprint Calculator

Combine workload energy, facility WUE, and grid water intensity to quantify the total liters of water attributable to an AI training or inference run, with optional adjustments for reclaimed water and renewable supply.

Total electrical energy drawn by the workload over the reporting window.
Liters of make-up water consumed on-site per kWh of IT energy.
Upstream water withdrawn to supply one kWh of electricity from the grid mix.
Defaults to 0%. Portion of on-site use offset by reclaimed or recycled water.
Defaults to 0%. Fraction of energy matched with renewables assumed to have negligible upstream water intensity.

Validate reported water intensities against the latest facility operations data and power purchase agreements before publishing sustainability disclosures.

Examples

  • 550,000 kWh workload, 0.20 L/kWh WUE, 1.10 L/kWh grid intensity, 15% reclaimed, 40% renewable ⇒ Total AI model water footprint: 456,500.00 liters
  • 120,000 kWh workload, 0.40 L/kWh WUE, 0.50 L/kWh grid intensity, optional fields blank ⇒ Total AI model water footprint: 108,000.00 liters

FAQ

What counts as reclaimed water in this context?

Include tertiary-treated effluent, harvested rainwater, or closed-loop cooling water that displaces freshwater withdrawals. Only count the share verified for the workload's reporting window.

How should I estimate grid water intensity?

Use regional water intensity factors published by utilities or lifecycle databases. Many U.S. ISOs report liters withdrawn per kWh; average data from the relevant balancing authority for the period you are analysing.

Does renewable matching always drive upstream water intensity to zero?

Geothermal and concentrated solar power consume water, but most corporate renewable portfolios rely on wind and PV resources with negligible withdrawals. Adjust the renewable share input if your contracts include water-intensive resources.

Can I apply the result to per-inference metrics?

Yes. Divide the total liters by the number of inferences processed during the window to express water per query, or by tokens generated for model benchmarking.

Additional Information

  • On-site water consumption equals IT energy multiplied by water usage effectiveness (WUE), net of any reclaimed water share.
  • Upstream water captures withdrawals associated with electricity generation; renewables with negligible water intensity reduce this contribution.
  • Total footprint sums net on-site and upstream liters so you can attribute water use to specific AI workloads.

Related calculators