ATOMS
Home
My Dashboard
Commodities
Briefing
Intel Feed
News
Alerts
Portfolio
Atoms TV
Community
Energy & Power
Data Centers
Overview
Water & Power
AI Power Model
Grid Impact
Water & Energy
Cloud Status
AI Progress
AI Supply Chain
Metals & Mining
Markets & Macro
Global & Supply
Tools
Markets
Energy
Metals
Agriculture
Power & Electricity
Nuclear
Rare Earths & Critical Minerals
Shipping & Logistics
Thesis Equities
HomeBriefingPortfolio
Data Centers

Water & Electricity

Environmental Impact

Tracking the resource footprint of the data center buildout. Electricity consumption, water withdrawal, and per-query costs — sourced from IEA, LBNL, EPRI, and hyperscaler sustainability reports.

US DC Electricity
176 TWh
2023 · 4.4% of grid
US DC Elec. 2030
470 TWh
+167% · 9.1% of grid
Global DC Electricity
415 TWh
2024 → 945 TWh by 2030
US DC Water (Direct)
17B gal
2023 · on-site cooling
US DC Water 2030
51B gal
+200% direct cooling
AI Query vs Search
10x
electricity · water · CO2

US Data Center Power Demand

As percentage of total US power demand, monthly resolution

2004
0.65%
of US power
2025
6.18%
of US power
2030E
12.35%
of US power

Sources: LBNL, IEA, EPRI, DOE. Monthly interpolation from annual anchors with seasonal adjustment (summer cooling peak). Pre-2014 data from Koomey 2007 / Shehabi et al. 2016.

US Data Center Electricity Consumption

Annual TWh consumed by US data centers, with projections to 2030

176 TWh
2023 actual · +203% since 2014
2014
58 TWh
1.5% of US grid
2023
176 TWh
4.4% of US grid
2030 (proj)
470 TWh
9.1% of US grid

Sources: LBNL 2024 Data Center Energy Usage Report, IEA Energy and AI (2025), EPRI "Powering Intelligence" (2024). Dashed line marks start of projections.

Global Data Center Electricity

IEA: consumption set to more than double by 2030

415 TWh → 945 TWh
2024 → 2030 (+128%)

Source: IEA Energy and AI (2025). Global DC electricity was flat at ~200 TWh from 2015-2018, then grew ~15%/yr 2019-2024.

Electricity by Workload Type

AI training + inference is the fastest-growing segment

70% AI by 2030
328 TWh AI of 470 TWh total

Sources: IEA Energy and AI (2025), Epoch AI compute tracking. AI inference demand grows faster than training as deployed model usage scales.

Data Center Electricity vs. Entire Countries

By 2030, global data centers are projected to consume more electricity than every country except the US, China, and India.

Global Data Centers (projected)945 TWh
2030 · IEA base case
US Data Centers (projected)470 TWh
2030 · EPRI high scenario
France (total)445 TWh
2023 · IEA country data
Global Data Centers (actual)415 TWh
2024 · IEA 2025
Italy (total)295 TWh
2023 · IEA country data
United Kingdom (total)285 TWh
2023 · IEA country data
Australia (total)250 TWh
2023 · IEA country data
US Data Centers (actual)176 TWh
2023 · LBNL 2024 Report
Poland (total)170 TWh
2023 · IEA country data
Argentina (total)135 TWh
2023 · IEA country data
Sweden (total)130 TWh
2023 · IEA country data

Facility Efficiency

ENERGY STARLoading...
Facilities Tracked
20
Avg PUE
1.38
Best PUE
1.08
Worst PUE
1.95

PUE Distribution

Count of facilities by PUE range. Lower PUE = more efficient (1.0 is ideal).

Average PUE Trend

Average PUE across ENERGY STAR certified data centers by year.

Top-Rated Facilities

by ENERGY STAR score
#FacilityLocationScorePUEIT Energy
1Meta Data Center - PrinevillePrineville, OR991.08520 GWh
2Google Data Center - The DallesThe Dalles, OR981.10450 GWh
3Google Data Center - Council BluffsCouncil Bluffs, IA971.11380 GWh
4Meta Data Center - Forest CityForest City, NC961.12310 GWh
5Microsoft Azure - QuincyQuincy, WA951.14400 GWh
6Equinix DC12 - AshburnAshburn, VA881.22280 GWh
7QTS Data Centers - AtlantaAtlanta, GA861.24250 GWh
8Digital Realty - RichardsonRichardson, TX851.25200 GWh
9CoreSite - RestonReston, VA841.26160 GWh
10CyrusOne - ChandlerChandler, AZ821.28180 GWh

Source: EPA ENERGY STAR Portfolio Manager. PUE = Power Usage Effectiveness (total facility energy / IT equipment energy). An ideal PUE is 1.0.

Resource Cost per AI Query

Each AI interaction has an electricity, water, and carbon cost. A single ChatGPT query uses ~10x the resources of a Google search.

Google SearchGoogle / IEA
Electricity
0.3 Wh
Water (direct)
0.26 mL
CO2
0.2 g
ChatGPT Query (GPT-4)Altman / UC Riverside
Electricity
3 Wh
Water (direct)
2.8 mL
CO2
2.1 g
AI Image GenerationIEA / Epoch AI estimates
Electricity
6.5 Wh
Water (direct)
5.5 mL
CO2
4.5 g
AI Model Training (per GPU-hour)Epoch AI / UC Riverside
Electricity
700 Wh
Water (direct)
590 mL
CO2
490 g
GPT-4 Full Training Run
Electricity
~62 GWh
Water (full cycle)
~185M gal
CO2
~12,500 tonnes
Epoch AI estimates

Sources

LBNL 2024 United States Data Center Energy Usage Report
IEA Energy and AI Report (2025)
EPRI "Powering Intelligence" Report (May 2024)
DOE Data Center Energy Report (2024)
Google Environmental Reports (2022-2025)
Microsoft Sustainability Reports (FY2022-FY2024)
Meta Sustainability Reports (2023-2024)
EESI: Data Centers and Water Consumption
UC Riverside AI Water Study (Ren et al. 2023)
Epoch AI Compute Tracking
EPA ENERGY STAR Portfolio Manager (Data Centers)

See the full data center infrastructure dashboard for capex, queue data, and project tracking.

View Data Center Energy Demand Tracker →