AI Water Transparency Report

We think a public, cited record of what AI companies disclose about water should exist. It didn't, so we started one.

Last updated: 2026-03-22

Google (Gemini)

Water Usage Yes

Reports total water consumption annually (8.1B gallons in 2024, up 88% since 2019). Reports fleet-wide WUE (~1.1 L/kWh).

View source →

Energy Usage Yes

Reports total energy consumption and fleet-wide PUE (~1.10). Publishes carbon-free energy percentage.

View source →

Per-Facility Data Partial

Some facility-level data available (e.g., Council Bluffs 1B gallons in 2024), but not systematically published for all sites.

View source →

AI-Specific Data Yes

Published per-query water estimate for Gemini (0.26 ml) via Google Cloud Blog (2025).

View source →

Water Pledge Yes

Pledged to replenish 120% of water consumed by 2030. Replenished ~4.5B gallons (~64%) in 2024.

View source →

Updated 2026-03-22

Microsoft (OpenAI)

Sustainability report

Water Usage Yes

Reports total water consumption annually (6.4M cubic meters in 2022, 34% YoY increase). Explicitly attributed increase to AI infrastructure.

View source →

Energy Usage Yes

Reports total energy consumption. Does not consistently report average PUE across fleet.

View source →

Per-Facility Data No

Does not publish per-facility water or energy breakdowns.

Checked 2026-03-22 — no public disclosure found

AI-Specific Data No

No published water or energy data specific to AI workloads or OpenAI models.

Checked 2026-03-22 — no public disclosure found

Water Pledge Yes

Pledged to be water positive by 2030 — replenish more water than consumed globally.

View source →

Updated 2026-03-22

Amazon (AWS)

Sustainability report

Water Usage Partial

Reports water use but with less granularity than peers. Official total figures are less clear than Google or Microsoft.

View source →

Energy Usage Partial

Reports renewable energy progress (100% renewable by 2025 goal). Only market-based emissions reported, not location-based. Does not report PUE.

View source →

Per-Facility Data No

Does not publish per-facility water or energy data.

Checked 2026-03-22 — no public disclosure found

AI-Specific Data No

No published water or energy data specific to AI/ML workloads.

Checked 2026-03-22 — no public disclosure found

Water Pledge Yes

Pledged to be water positive by 2030. Some AWS regions use zero-water cooling (e.g., Ireland, Sweden).

View source →

Updated 2026-03-22

Meta (LLaMA)

Sustainability report

Water Usage Yes

Reports total water consumption (813M gallons in 2023, 95% from data centers). Reports WUE (~0.26 L/kWh), among the lowest in industry.

View source →

Energy Usage Yes

Reports total energy consumption and renewable energy progress.

View source →

Per-Facility Data Yes

Some facility-level detail published (e.g., Prineville, OR uses zero water for cooling).

View source →

AI-Specific Data No

No published water or energy data specific to AI workloads (LLaMA training, Reels inference, etc.).

Checked 2026-03-22 — no public disclosure found

Water Pledge Yes

Committed to net-zero emissions across value chain by 2030; water stewardship investments ongoing.

View source →

Updated 2026-03-22

Anthropic (Claude)

No report published

Water Usage No

No sustainability report or water usage data published as of March 2026.

Checked 2026-03-22 — no public disclosure found

Energy Usage No

No energy consumption data published. Relies on AWS and Google Cloud infrastructure.

View source →

Per-Facility Data No

Not applicable — Anthropic does not operate its own data centers.

Checked 2026-03-22 — no public disclosure found

AI-Specific Data No

No per-model or per-query energy or water data published.

Checked 2026-03-22 — no public disclosure found

Water Pledge No

No water-related sustainability commitments published.

Checked 2026-03-22 — no public disclosure found

Updated 2026-03-22

Perplexity

No report published

Water Usage No

No sustainability report or water usage data published as of March 2026.

Checked 2026-03-22 — no public disclosure found

Energy Usage No

No energy consumption data published. Relies on third-party cloud infrastructure (AWS, Azure, CoreWeave).

View source →

Per-Facility Data No

Not applicable — Perplexity does not operate its own data centers.

Checked 2026-03-22 — no public disclosure found

AI-Specific Data No

No per-query energy or water data published.

Checked 2026-03-22 — no public disclosure found

Water Pledge No

No water-related sustainability commitments published.

Checked 2026-03-22 — no public disclosure found

Updated 2026-03-22

OpenAI (ChatGPT)

No report published

Water Usage Partial

CEO disclosed ~0.3 ml direct water per average ChatGPT query (mid-2025), but no total water consumption figures or methodology published. Excludes indirect water from electricity generation.

View source →

Energy Usage Partial

CEO disclosed ~0.34 Wh per average ChatGPT query (mid-2025). No total energy consumption figures published. Runs on Microsoft Azure infrastructure.

View source →

Per-Facility Data No

Does not operate its own data centers (runs on Microsoft Azure). No facility-level data published. Stargate project campuses are under construction.

View source →

AI-Specific Data Partial

Published per-query energy and water estimates for ChatGPT via CEO statement, but figures are unverified, exclude training costs, and definition of 'average query' is unclear.

View source →

Water Pledge Partial

Stargate project commits to closed-loop cooling and water positivity at new campuses via partner Vantage, but no company-wide water positive pledge published by OpenAI itself.

View source →

Updated 2026-03-22

xAI (Grok)

No report published

Water Usage No

No water usage data published. Memphis Colossus facility draws ~1.3M gallons/day from the aquifer per reporting by environmental groups, but xAI has not disclosed figures itself.

View source →

Energy Usage No

No energy consumption data published. Operates its own data centers with gas turbines and battery storage, but has not disclosed total energy usage.

View source →

Per-Facility Data No

Operates the Colossus facility in Memphis, TN, but publishes no environmental data for it. Environmental impact information comes from third-party investigations and permit filings.

View source →

AI-Specific Data No

No per-model or per-query energy or water data published for Grok models.

Checked 2026-03-22 — no public disclosure found

Water Pledge Partial

Building an $80M water recycling plant to use treated wastewater instead of aquifer water at Memphis facility, expected operational late 2026. No formal company-wide water positive pledge.

View source →

Updated 2026-03-22

Mistral (Le Chat)

Sustainability report

Water Usage Yes

Published total water consumption for Mistral Large 2: 281,000 m3 over 18 months (training + inference). Conducted ISO 14040/44-compliant lifecycle analysis with Carbone 4 and ADEME.

View source →

Energy Usage Yes

Published energy data and greenhouse gas emissions (20.4 kt CO2e for Mistral Large 2 training + 18 months of use) via comprehensive lifecycle analysis.

View source →

Per-Facility Data No

Operates a 40 MW facility in Essonne, France and is building data centers in Sweden, but does not publish per-facility environmental breakdowns.

View source →

AI-Specific Data Yes

Published per-query environmental impact: 45 ml water, 1.14g CO2e, and 0.16 mg antimony-equivalent resource depletion per 400-token Le Chat response.

View source →

Water Pledge No

No water positive or replenishment pledge. Committed to ongoing environmental reporting and contributing to international AI sustainability standards.

Checked 2026-03-22 — no public disclosure found

Updated 2026-03-22

Cohere

No report published

Water Usage No

No sustainability report or water usage data published as of March 2026.

Checked 2026-03-22 — no public disclosure found

Energy Usage No

No energy consumption data published. Published a policy primer on AI energy efficiency but disclosed no data about its own operations. Relies on third-party cloud infrastructure.

View source →

Per-Facility Data No

Not applicable — Cohere does not operate its own data centers.

Checked 2026-03-22 — no public disclosure found

AI-Specific Data No

No per-model or per-query energy or water data published.

Checked 2026-03-22 — no public disclosure found

Water Pledge No

No water-related sustainability commitments published.

Checked 2026-03-22 — no public disclosure found

Updated 2026-03-22

Provider	Water Usage	Energy Usage	Per-Facility Data	AI-Specific Data	Water Pledge
Google (Gemini)	Yes Reports total water consumption annually (8.1B gallons in 2024, up 88% since 2019). Reports fleet-wide WUE (~1.1 L/kWh). View source →	Yes Reports total energy consumption and fleet-wide PUE (~1.10). Publishes carbon-free energy percentage. View source →	Partial Some facility-level data available (e.g., Council Bluffs 1B gallons in 2024), but not systematically published for all sites. View source →	Yes Published per-query water estimate for Gemini (0.26 ml) via Google Cloud Blog (2025). View source →	Yes Pledged to replenish 120% of water consumed by 2030. Replenished ~4.5B gallons (~64%) in 2024. View source →
Microsoft (OpenAI)	Yes Reports total water consumption annually (6.4M cubic meters in 2022, 34% YoY increase). Explicitly attributed increase to AI infrastructure. View source →	Yes Reports total energy consumption. Does not consistently report average PUE across fleet. View source →	No Does not publish per-facility water or energy breakdowns. Checked 2026-03-22	No No published water or energy data specific to AI workloads or OpenAI models. Checked 2026-03-22	Yes Pledged to be water positive by 2030 — replenish more water than consumed globally. View source →
Amazon (AWS)	Partial Reports water use but with less granularity than peers. Official total figures are less clear than Google or Microsoft. View source →	Partial Reports renewable energy progress (100% renewable by 2025 goal). Only market-based emissions reported, not location-based. Does not report PUE. View source →	No Does not publish per-facility water or energy data. Checked 2026-03-22	No No published water or energy data specific to AI/ML workloads. Checked 2026-03-22	Yes Pledged to be water positive by 2030. Some AWS regions use zero-water cooling (e.g., Ireland, Sweden). View source →
Meta (LLaMA)	Yes Reports total water consumption (813M gallons in 2023, 95% from data centers). Reports WUE (~0.26 L/kWh), among the lowest in industry. View source →	Yes Reports total energy consumption and renewable energy progress. View source →	Yes Some facility-level detail published (e.g., Prineville, OR uses zero water for cooling). View source →	No No published water or energy data specific to AI workloads (LLaMA training, Reels inference, etc.). Checked 2026-03-22	Yes Committed to net-zero emissions across value chain by 2030; water stewardship investments ongoing. View source →
Anthropic (Claude)	No No sustainability report or water usage data published as of March 2026. Checked 2026-03-22	No No energy consumption data published. Relies on AWS and Google Cloud infrastructure. View source →	No Not applicable — Anthropic does not operate its own data centers. Checked 2026-03-22	No No per-model or per-query energy or water data published. Checked 2026-03-22	No No water-related sustainability commitments published. Checked 2026-03-22
Perplexity	No No sustainability report or water usage data published as of March 2026. Checked 2026-03-22	No No energy consumption data published. Relies on third-party cloud infrastructure (AWS, Azure, CoreWeave). View source →	No Not applicable — Perplexity does not operate its own data centers. Checked 2026-03-22	No No per-query energy or water data published. Checked 2026-03-22	No No water-related sustainability commitments published. Checked 2026-03-22
OpenAI (ChatGPT)	Partial CEO disclosed ~0.3 ml direct water per average ChatGPT query (mid-2025), but no total water consumption figures or methodology published. Excludes indirect water from electricity generation. View source →	Partial CEO disclosed ~0.34 Wh per average ChatGPT query (mid-2025). No total energy consumption figures published. Runs on Microsoft Azure infrastructure. View source →	No Does not operate its own data centers (runs on Microsoft Azure). No facility-level data published. Stargate project campuses are under construction. View source →	Partial Published per-query energy and water estimates for ChatGPT via CEO statement, but figures are unverified, exclude training costs, and definition of 'average query' is unclear. View source →	Partial Stargate project commits to closed-loop cooling and water positivity at new campuses via partner Vantage, but no company-wide water positive pledge published by OpenAI itself. View source →
xAI (Grok)	No No water usage data published. Memphis Colossus facility draws ~1.3M gallons/day from the aquifer per reporting by environmental groups, but xAI has not disclosed figures itself. View source →	No No energy consumption data published. Operates its own data centers with gas turbines and battery storage, but has not disclosed total energy usage. View source →	No Operates the Colossus facility in Memphis, TN, but publishes no environmental data for it. Environmental impact information comes from third-party investigations and permit filings. View source →	No No per-model or per-query energy or water data published for Grok models. Checked 2026-03-22	Partial Building an $80M water recycling plant to use treated wastewater instead of aquifer water at Memphis facility, expected operational late 2026. No formal company-wide water positive pledge. View source →
Mistral (Le Chat)	Yes Published total water consumption for Mistral Large 2: 281,000 m3 over 18 months (training + inference). Conducted ISO 14040/44-compliant lifecycle analysis with Carbone 4 and ADEME. View source →	Yes Published energy data and greenhouse gas emissions (20.4 kt CO2e for Mistral Large 2 training + 18 months of use) via comprehensive lifecycle analysis. View source →	No Operates a 40 MW facility in Essonne, France and is building data centers in Sweden, but does not publish per-facility environmental breakdowns. View source →	Yes Published per-query environmental impact: 45 ml water, 1.14g CO2e, and 0.16 mg antimony-equivalent resource depletion per 400-token Le Chat response. View source →	No No water positive or replenishment pledge. Committed to ongoing environmental reporting and contributing to international AI sustainability standards. Checked 2026-03-22
Cohere	No No sustainability report or water usage data published as of March 2026. Checked 2026-03-22	No No energy consumption data published. Published a policy primer on AI energy efficiency but disclosed no data about its own operations. Relies on third-party cloud infrastructure. View source →	No Not applicable — Cohere does not operate its own data centers. Checked 2026-03-22	No No per-model or per-query energy or water data published. Checked 2026-03-22	No No water-related sustainability commitments published. Checked 2026-03-22

This report tracks what major AI providers publicly disclose about their water consumption, energy usage, and environmental commitments. Every claim is sourced from published sustainability reports and official statements.

We track five dimensions of transparency: total water usage, energy usage, per-facility reporting, AI-specific environmental data, and water stewardship pledges. Read the full methodology.

If you spot an error, open an issue. If a provider has published data we missed, submit a PR.