Skip to content
Water,
Actually

Compare one scope at a time

Put AI water use in context.

Start with one disclosed estimate for text prompts. Then explore other kinds of water use in separate lanes—because cooling water, tap water, supply-chain water, and national totals do not measure the same thing.

Build your water receiptRead the comparison rules

A simple scenario

How many similar text prompts?

This multiplies a single published per-prompt estimate. It is useful for understanding volume, but it is not a universal measurement of every AI model or request.

prompts
Advanced details: what this calculation includes

Included

The single source-backed operational water estimate shown here, multiplied by your prompt count.

Not included

Model training, hardware manufacturing, household water, product footprints, or national data-center totals.

Five different questions

Keep the water lanes separate.

Browse source-backed examples inside their own measurement context. Values are shown as reported—not converted into a leaderboard and never added into a mixed total.

Operational and cooling waterWater used or consumed while a computing service runs. The result depends on the facility, cooling system, workload, location, and time period.
Absolute Volume· Consumption

Gemini Apps median text prompt

0.26 mL

per median text prompt

Direct Operational · Google global production fleet · per-event · company-estimated

How we know

Company-estimated May 2025 median using 2024 fleet-average water-use effectiveness. Not independently verified or representative of every model, prompt, modality, location, or future performance.

Google Cloud: Measuring the environmental impact of AI inferenceCompany disclosure · 2025
Calculation notes

No unit conversion; Google reports 0.26 mL per median Gemini Apps text prompt.

Absolute Volume· Mixed Or Unclear

ChatGPT average query

0.32 mL

per average query

Mixed Or Unclear · Not disclosed · per-event · company-estimated

How we know

Public CEO estimate with no supporting measurement period, geography, accounting boundary, or technical methodology. Stored as 0.32 mL rather than a falsely precise conversion.

Sam Altman: The Gentle SingularityCompany disclosure · 2025
Calculation notes

0.000085 U.S. gal × 3,785.411784 mL/U.S. gal = 0.32176 mL; normalized to 0.32 mL to match source precision.

Electricity-related waterWater consumed where electricity is generated. This is an upstream grid boundary, not water used directly at a data center.
Absolute Volume· Consumption

U.S. data centers — electricity

2.2 billion L

all U.S. data centers, average day in 2023

Upstream Electricity · United States electricity grid mix at data center locations · per-day · independently-modeled · All data centers; not AI-specific

How we know

This is upstream water consumed to generate electricity for all U.S. data centers, not direct facility water and not AI alone. It must not be summed or ranked with direct-operation records.

Lawrence Berkeley National Laboratory: 2024 United States Data Center Energy Usage ReportAcademic research · 2024
Calculation notes

800 billion L/year × 1,000,000,000 L/billion L × 1,000 mL/L ÷ 365 days/year

Direct, everyday waterWater delivered at the point of use for activities such as showering, washing clothes, or running an appliance.
Absolute Volume· Delivered

dishwasher cycle

12.1 L

per cycle

Household Delivered · United States · per-event · derived · Non-AI context record

How we know

Energy derived from max 240 kWh/year and 215 cycle/year assumption; water uses 3.2 gal/cycle criterion.

ENERGY STAR: Dishwashers Key Product CriteriaGovernment source · 2023
Calculation notes

Normalized from seed research (basis: per cycle, energyWh=1116.3, waterMl=12113.3).

Absolute Volume· Delivered

shower hot water

37.9 L

per shower, hot water only

Household Delivered · United States · per-event · derived · Non-AI context record

How we know

DOE table lists shower at 10 gallons/use. Energy is calculated assuming 65°F temperature rise and no heater losses.

DOE Energy Saver: Reduce Hot Water Use for Energy SavingsGovernment source · 2026
Calculation notes

Normalized from seed research (basis: per shower, hot water only, energyWh=1588.8, waterMl=37854.1).

Absolute Volume· Delivered

clothes washer load

53 L

per load

Household Delivered · United States · per-event · derived · Non-AI context record

How we know

Full-sized ENERGY STAR washers use 14 gallons/load vs 20 gallons for standard machine.

ENERGY STAR: Clothes WashersGovernment source · 2025
Calculation notes

Normalized from seed research (basis: per load, waterMl=52995.8).

Embedded and lifecycle waterWater counted across a product's supply chain. These footprints may combine rainfall, irrigation, and water-quality estimates.
Absolute Volume· Embedded Mixed

Cup of coffee

130 L

per 125 mL cup

Product Lifecycle · Global product average · lifecycle · independently-modeled · Non-AI context record

How we know

Embedded agricultural water footprint, not direct tap water consumed while brewing.

Water Footprint Network Product Gallery: CoffeeIndependent nonprofit · 2011
Calculation notes

Normalized from seed research (basis: per 125 mL cup, waterMl=130000).

Absolute Volume· Embedded Mixed

Cup of tea

30 L

per 250 mL cup

Product Lifecycle · Global product average · lifecycle · independently-modeled · Non-AI context record

How we know

Embedded agricultural water footprint, not direct tap water consumed while brewing.

Water Footprint Network Product Gallery: TeaIndependent nonprofit · 2011
Calculation notes

Normalized from seed research (basis: per 250 mL cup, waterMl=30000).

Absolute Volume· Embedded Mixed

Beef serving

1,746.3 L

per 1/4 lb serving

Product Lifecycle · Global product average · lifecycle · independently-modeled · Non-AI context record

How we know

Derived from 15,400 L/kg global average and 0.113398 kg serving. Mostly green water; local impact varies.

Water Footprint Network Product Gallery: BeefIndependent nonprofit · 2012
Calculation notes

Normalized from seed research (basis: per 1/4 lb serving, waterMl=1746329.2).

Absolute Volume· Embedded Mixed

Cotton shirt

2,500 L

per 250 g cotton shirt

Product Lifecycle · Global product average · lifecycle · independently-modeled · Non-AI context record

How we know

Global average; local irrigation and blue-water share vary widely.

Water Footprint Network Product Gallery: CottonIndependent nonprofit · 2011
Calculation notes

Normalized from seed research (basis: per 250 g cotton shirt, waterMl=2500000).

National and system contextSystem totals for a sector, grid, or country. These figures describe infrastructure at scale, not one AI request.
Absolute Volume· Consumption

U.S. data centers — direct

180.8 million L

all U.S. data centers, average day in 2023

Direct Operational · United States · per-day · independently-modeled · All data centers; not AI-specific

How we know

This is direct cooling-related consumption for all U.S. data centers, not AI alone. A national annual estimate can hide large differences by site, season, cooling design, and local water stress.

Lawrence Berkeley National Laboratory: 2024 United States Data Center Energy Usage ReportAcademic research · 2024
Calculation notes

66 billion L/year × 1,000,000,000 L/billion L × 1,000 mL/L ÷ 365 days/year

Absolute Volume· Consumption

Crop irrigation

286.5 billion L

CONUS average day, water years 2010–2020

Direct Operational · Conterminous United States (CONUS) · per-day · government-modeled · Non-AI context record

How we know

Modeled mean for the conterminous United States across water years 2010–2020. This is consumptive use, not total withdrawal, and it predates the 2023 data-center estimate.

USGS: Water use across the conterminous United States, water years 2010–20Government source · 2025
Calculation notes

75,698 Mgal/day × 1,000,000 gal/Mgal × 3,785.411784 mL/gal

Absolute Volume· Consumption

Public supply

16 billion L

CONUS average day, water years 2010–2020

Direct Operational · Conterminous United States (CONUS) · per-day · government-modeled · Non-AI context record

How we know

Modeled mean for the conterminous United States across water years 2010–2020. Public supply serves many end uses; this is consumptive use, not deliveries or total withdrawal.

USGS: Water use across the conterminous United States, water years 2010–20Government source · 2025
Calculation notes

4,219 Mgal/day × 1,000,000 gal/Mgal × 3,785.411784 mL/gal

Absolute Volume· Consumption

Thermoelectric power

11 billion L

CONUS average day, water years 2010–2020

Direct Operational · Conterminous United States (CONUS) · per-day · government-modeled · Non-AI context record

How we know

Modeled mean freshwater consumptive use for the conterminous United States across water years 2010–2020. Thermoelectric withdrawals are much larger because most withdrawn water returns.

USGS: Water use across the conterminous United States, water years 2010–20Government source · 2025
Calculation notes

2,904 Mgal/day × 1,000,000 gal/Mgal × 3,785.411784 mL/gal

The useful takeaway

A small per-query estimate and a meaningful local data-center impact can both be true. Scale, water source, season, and watershed matter.

Build a receipt Browse evidence