AI Resource Use Dashboard

Public-source reference estimates for AI-related energy, carbon, and water use, shown alongside comparable national, company, and per-query context.

This dashboard combines measured benchmarks, derived estimates, and company-reported disclosures. Use the methodology page for scope notes and comparability limits.

Dataset ages: 1.1y old · 4mo old · 4mo old

Estimated AI Energy Use
56TWh/year

415 TWh total data centers (2024)

Estimated AI Carbon Emissions
34Mt CO2eq/year

Range: 24-44 Mt (est. 2025)

Estimated AI Water Use
539billion liters/year

Range: 312.5-764.6 B liters (est. 2025)

Per-User Reference Scale

Google Cloud(2025)· 9mo old
30

Per-query reference values

Search is included only as a rough comparison baseline from the source dataset; the per-query values shown below are specific to Google's 2025 AI inference study.

0.24 Wh
Energy per query
0.03g
CO2 per query
0.26 ml
Water per query (5 drops)

Annualized estimate (30 queries/day)

2.6
kWh/year
= 0.0 EV charges
0.33
kg CO2/year
= 0.8 miles driven
2.8
liters/year
= 11.8 showers

Scaling note

With an estimated 1 billion+ AI queries per day worldwide, small per-query values can aggregate into large system-wide totals. Switch to the national view for broader comparisons.

Global Data Center Map

Company Reports(2025)· 0d old

Major cloud data center locations where AI workloads run. Larger dots indicate higher capacity clusters.

Northern Virginia (Ashburn) (aws)Dallas/Fort Worth (aws)The Dalles, Oregon (google)Quincy, Washington (microsoft)Phoenix, Arizona (aws)Council Bluffs, Iowa (google)Dublin, Ireland (aws)Amsterdam, Netherlands (aws)Frankfurt, Germany (aws)London, UK (multi)Singapore (aws)Tokyo, Japan (aws)Sydney, Australia (aws)Mumbai, India (aws)São Paulo, Brazil (aws)Stockholm, Sweden (aws)
Largest hub
Large
Medium
Northern Virginia (Ashburn) is the world's largest data center cluster — over 300 data centers within a 30-mile radius
5
Europe
4
Asia Pacific
3
US West
2
US Central
1
US East
1
South America

Model Energy Efficiency

Benchmarks + normalized estimates(2025)· 5mo old

Energy consumption per 1,000 queries on NVIDIA H100 where public benchmark data exists. Some entries are normalized estimates assembled from public disclosures and should be treated as directional rather than canonical provider measurements.

Visible rows in this view: 5 measured benchmarks and 9 normalized estimates.

Measured benchmark
Normalized estimate

Interpretation Note

Model choice can shift energy use materially across tasks, with visible benchmark gaps of 10-200x. A Mistral 7B normalized estimate is 15 Wh/1K, while the DALL-E 3 normalized estimate is 3,200 Wh/1K. Those differences are useful for comparison, but exact values should be read as directional unless marked as measured.

Company-Reported Operations Comparison

Sustainability Reports(2024)· 11mo old

Values below come from company sustainability disclosures. Reporting boundaries and accounting methods differ, so treat this section as directional comparison rather than a perfectly standardized ranking.

Google
PUE: 1.09
Renewable: 66%
Microsoft
PUE: 1.18
Renewable: 100%
Amazon (AWS)
PUE: 1.15
Renewable: 100%
Meta
PUE: 1.08
Renewable: 100%