Dropmatch is a US nonprofit. Recognition under section 501(c)(3) is in progress.Check a certification
Home / The Standard

The Dropmatch Standard

How the water footprint of AI usage is measured, verified, matched and proven. Version 1, draft.

Version 1.0 draftStatus In draftingPublic comment 30 daysLicense Open

1. What is counted

Two kinds of water consumed, counted at the location where the model was served.

Scope 1 Cooling water at the data center + Scope 2 Water behind the electricity = Footprint About 10x scope 1 alone Full lifecycle figures (hardware, training) are shown as the upper bound.

Why both: a vendor's cooling-only figure understates the footprint by about ten times. One median text prompt, six published answers:

  • 0.12 mLGoogle, accelerators only
  • 0.26 mLGoogle, cooling only
  • 0.32 mLOpenAI, converted from the published figure
  • 1.3 mLGoogle's energy plus US grid water
  • 16.9 mLLi et al., scope 1 plus scope 2, US average
  • 45 mLMistral, full lifecycle

Bar lengths are illustrative. The ten to one match covers the whole ladder.

2. Tokens to liters

Usage is counted in tokens, as providers bill it, and converted with published factors per model tier and region.

QuantityVersion 1 factorBasis
Output tokensabout 15 L per millionAbout 2.9 Wh per 1,000 tokens at 0.30 L/kWh on site plus 4.35 L/kWh for the grid (Jegham et al., Azure)
Input tokensabout 0.5 L per millionAbout 0.05 Wh per 1,000 tokens, same factors
Small modelsabout a thirdGPT-4.1 nano, about 2.4 mL per 1,000 tokens
Reasoning modelsan upliftAbout 30 times the energy on average (AI Energy Score v2)

Method: EcoLogits (server energy times on-site WUE plus PUE times off-site water intensity). Reference: ISO 14046. The derivation and sensitivity range are published with each version.

3. Verifying usage

Usage comes from data the provider generated, and every pull is reconciled to the company's accounts-payable list of AI vendors, so every account is counted.

Attested

Under $5M revenue

Usage exports plus a signed attestation. Liters uplifted 25%.

API-verified

$5M to $100M

Monthly pulls from billing data, gateway keys or the signed collector you run yourself.

Audited

Over $100M

Limited assurance under ISO 14064-3, by your own auditor.

Companies keep their administrative API keys. Read-only exports, gateway keys and the signed collector carry the data.

4. The match

The floorEvery certification covers the full footprint.
The standard tierAt least ten liters funded for every liter, with a minimum annual commitment.

5. Delivered water

Confirmed from outsideChlorine tests in household water, GPS and photos, flow sensors.
Counted conservatively20 liters per person per day, actual users, uptake discounts, under VWBA 2.0 and the WASH Benefits framework.
A flowLiters and person-years are counted within the year delivered.
Retired onceSerial-numbered in the public registry, credit split among funders.

6. The badge

Says what was fundedVolume, place, year, verified. Displayed only against a live registry entry.
Reported where it was fundedWater funded in one basin is reported as water funded there, as a contribution.

7. Governance

Public and free. Open to every organization that meets it, on a cost-based fee schedule. Developed with experts and stakeholders, a thirty-day comment period per version, certification decisions independent of fundraising, a published complaints and revocation procedure.

Version history

VersionDateStatus
1.0 draftOctober 2026In drafting. Public comment opens with the pilot cohort.