Epoch AI - Frontier Data Centers - November 2025

Free

Rigorous public data on the world's largest AI data centers.

FreeFree tier

Type

Open Source

Founded

2021

Company

Epoch AI

Visit Website GitHub (38 stars)

About Epoch AI - Frontier Data Centers - November 2025

Epoch AI's Frontier Data Centers database provides rigorous, publicly accessible data on the world's largest AI data centers. The dataset includes independent estimates of power, compute, and cost built from high-resolution satellite imagery, permits, and expert knowledge of data center construction. Every data point is vetted by the Epoch AI team. The database currently covers 74 sites, representing 12.1 million H100-equivalents of compute and 11.4 GW of IT power across 15 owners. Users can explore the data through an interactive map, customizable graphs (with filters by owner, user, and country), and a timeline from Q1 2023 to Q1 2030. The data is downloadable and licensed under CC-BY.

Key Features

Independent estimates of power, compute, and cost for AI data centers

Built from high-resolution satellite imagery, permits, and construction knowledge

Every data point vetted by team of experts

Covers 74 sites with 12.1M H100-equivalents and 11.4 GW IT power

Interactive map with geographic visualization

Customizable graph and table with filters by owner, user, country

Timeline from Q1 2023 to Q1 2030 with snapshot view

Data downloadable under CC-BY license

Pros & Cons

Pros

Rigorous, vetted, independent estimates based on multiple data sources
Transparent methodology using satellite imagery and public permits
Free and openly accessible data with regular updates
Comprehensive coverage of the largest AI data centers globally

Cons

Limited to the 74 largest data centers; does not cover smaller facilities
Data points are estimates, not exact measured values
Requires some technical understanding to interpret compute and power metrics
Timeline projections (to 2030) involve inherent uncertainty

Best For

Understanding AI infrastructure scaling and resource allocationTracking compute and power usage of major AI data centers over timeResearch on drivers and bottlenecks of AI progressEconomic modeling and impact analysis of AI developmentPolicy and investment decision-making related to AI hardware

FAQ

What is an AI data center according to Epoch AI?

Epoch AI defines an AI data center as a collection of one or more buildings located near each other (typically less than 10 km apart), with a shared hardware owner or facility operator, running hardware specialized for AI such as GPUs or custom chips like Google's TPUs. These data centers may be used to experiment on, train, and deploy AI models.

What are the largest AI data centers by IT power?

The largest known AI data center by IT power is Colossus 2, owned by SpaceXAI in Memphis, TN, drawing 946 MW. It is followed by Anthropic-Amazon New Carlisle (Amazon, 910 MW), Microsoft Fairwater Atlanta (Microsoft, 636 MW), Meta Prometheus (Meta, 631 MW), and OpenAI Stargate Abilene (Oracle, 421 MW).

How much power do the covered AI data centers consume?

The 74 AI data centers cover a combined 11.4 GW of IT power capacity. Total facility power is 20–50% higher due to cooling and other infrastructure overhead, pushing the real-world footprint to roughly 14.8 GW, which is more than New York City's peak demand of 11 GW. Average power consumption is typically 60–80% of capacity.

Where are these AI data centers located?

The United States has the most large-scale AI data centers, concentrated in Texas (13 sites), Virginia (6 sites), Ohio (5 sites), Nebraska (4 sites), and Iowa (4 sites). International facilities include Huawei Horinger in Hohhot, China and DayOne Nusajaya in Johor Bahru, Malaysia.