Largest operating AI data centers, mid-2026

What this dataset measures

The five facilities with the highest observed IT power capacity in Epoch AI's Frontier Data Centers data as of its August 31, 2026 update, the facilities a claim about 'the biggest AI data centers' refers to.

Geography

United States

Unit of observation

One AI data center campus at one estimate date

Update cadence

Irregular; revise when Epoch AI updates its estimates

Curator

FactHub Research Team

Curated data tables

Largest operating AI data centers, mid-2026

One row per facility: Epoch AI's IT power estimate and FactHub's derived annual energy and household equivalents.

Unit of observation: One data center campus

largest_ai_dc.csv text/csv · 560 bytes

All 5 rows 8 columns Download the file

All 5 rows

facility	owner	location	it_power_mw MW	estimate_date	annual_gwh_full_it_load GWh per year	households_full_it_load homes	households_half_it_load homes
Colossus 2	xAI (listed by Epoch AI as SpaceXAI)	Memphis, TN	946	2026-06-15	8287.0	767951	383976
Anthropic-Amazon New Carlisle	Amazon	New Carlisle, IN	910	2026-06-03	7971.6	738727	369364
Microsoft Fairwater Atlanta	Microsoft	Fayetteville, GA	636	2026-06-14	5571.4	516297	258148
Meta Prometheus	Meta	New Albany, OH	562	2026-08-01	4923.1	456225	228112
OpenAI Stargate Abilene	Oracle	Abilene, TX	421	2026-05-23	3688.0	341763	170882

8 field definitions

Field	Type	Meaning	Units	Value type
facility	string	Facility name as Epoch AI lists it.	—	observed
owner	string	Owner as Epoch AI lists it.	—	observed
location	string	City and state.	—	observed
it_power_mw	number	Estimated IT power capacity available to chips, servers and other IT equipment, excluding cooling.	MW	estimated
estimate_date	date	Date of the capacity estimate.	—	observed
annual_gwh_full_it_load	number	Electricity the IT load would use running at full capacity for a year.	GWh per year	derived
households_full_it_load	integer	Average U.S. homes using the same electricity in a year, at full IT load.	homes	derived
households_half_it_load	integer	Same, if the IT load averaged half its capacity.	homes	derived

Sources

1. Epoch AI, 'The record for power capacity in a single data center has doubled every 10 months' (Sep. 4, 2026), with its Frontier Data Centers data IT power capacity estimates and estimate dates for each facility

   Retrieved October 26, 2026. CC BY 4.0. Allowed with attribution to Epoch AI

   Reliability: Academic. Epoch AI is an independent nonprofit research group that estimates data center power from satellite imagery, permits and public documents. It does not own or operate data centers.

2. U.S. Energy Information Administration, 'How much electricity does an American home use?' Average annual electricity sold per U.S. residential customer (10,791 kWh, 2022)

   Retrieved October 26, 2026. Public domain (U.S. government work). Unrestricted

   Reliability: Government. EIA is the statistical agency of the U.S. Department of Energy; the figure comes from utility sales reports.

Limitations and reproducibility

Fully reproducible: Every derived column is arithmetic on the observed IT power and the EIA household figure: annual GWh = MW x 8,760 hours / 1,000; households = MW x 8,760,000 kWh / 10,791 kWh, rounded to the nearest home; half-load households = full-load households / 2, rounded.

• IT power capacity is Epoch AI's estimate from satellite imagery, permits and disclosures, not metered consumption.
• IT power excludes cooling and power conversion, so whole-facility draw at a given IT load is higher.
• Sites rarely run at full capacity all year; the half-load column illustrates a lower utilisation and is not a measured value.
• The household benchmark is 2022 electricity sold per U.S. residential customer; homes with rooftop solar consume more than they buy.