What is an AI data center?

The definition

There is no single legal definition, but federal policy has offered two. Executive Order 14141 defined an AI data center as one used primarily to develop or operate artificial intelligence, as the Congressional Research Service notes[4]; that order was revoked in July 2025[5]. Its replacement, Executive Order 14318, speaks instead of data center projects needing more than 100 megawatts of new load dedicated to AI training, inference, simulation or synthetic data generation[6].

Two of those words describe most of the work. Training is the long, compute-heavy job of building a model from data. Inference is running the finished model to answer requests, such as a chatbot replying to a question. Both need the same kind of hardware, and the International Energy Agency describes both as happening in large, power-hungry facilities[3].

What’s inside: GPUs and accelerated servers

A conventional data center is filled mostly with general-purpose servers built around CPUs. An AI data center is built around accelerators, chiefly graphics processing units. CRS says GPUs are generally considered better than CPUs at AI computing, and that today’s systems pass data between hundreds of GPUs spread across many servers at gigabit-per-second speeds[4].

Those accelerated servers are what changed the energy picture. Lawrence Berkeley National Laboratory found that around 2017, GPU servers for AI became a big enough part of the country’s server stock that total U.S. data center electricity use started climbing again[7], and that the growth of accelerated servers more than doubled total data center energy demand between 2017 and 2023[7].

How an AI data center differs from a regular one

The table explains most of what neighbors notice. A campus planned in hundreds of megawatts often comes with new transmission lines or substations, sometimes its own power plant, and a cooling system sized to match. The power tracker lists the utility deals and on-site generation behind tracked projects, and how data centers are cooled covers the cooling options in detail.

Why so many are being built

The IEA calls AI the most important driver of the growth it projects in global data center electricity use through 2030[3]. The money behind it is visible in company guidance: Alphabet raised its 2026 capital spending forecast to $195 billion to $205 billion[11] and said the vast majority of its second-quarter spending went to technical infrastructure for AI[11]; Meta expects $130 billion to $145 billion[12] and Amazon about $220 billion[13].

Federal policy has pushed in the same direction. Executive Order 14318 made the rapid buildout of AI data centers and the infrastructure that powers them a priority of the administration, including by easing federal regulatory burdens[5]. The full account of what is driving the boom covers cloud demand and state tax incentives as well.

The largest AI data centers we track

The Census counts a project as an AI data center only when its own record shows it: a cited source that describes the project as AI, or a developer or tenant whose business is building AI models, named in a cited source. That makes the count a floor, since many cloud campuses run AI work without saying so. Today it stands at 42 active projects in 20 states. These are the ten largest by published capacity; the live AI data center count lists every one with the evidence behind it.

Capacity is each project’s published planned figure, usually the full build-out, not what is running today. Open a project for its sources.

Common questions

What is an AI data center?

A data center used mainly to train or run artificial intelligence models. It is built around GPU servers linked in large numbers, packs far more power into each rack than a conventional data center, and usually needs liquid cooling and a very large power supply.

How is an AI data center different from a regular data center?

Mainly in power density and scale. An industry executive quoted by IEEE Spectrum put the average rack at about 8 kilowatts and AI racks at up to 100. The IEA says a conventional data center may be 10 to 25 megawatts, while a hyperscale, AI-focused data center can be 100 megawatts or more.

How much electricity does an AI data center use?

The International Energy Agency says a typical AI-focused data center uses as much electricity as 100,000 households, and the largest under construction in 2025 will use 20 times as much.

Why do AI data centers need liquid cooling?

Because AI racks give off far more heat than ordinary ones. Liquid can carry about 1,000 times as much heat as air, according to the National Laboratory of the Rockies, which makes racks of 60 kilowatts or more practical.

How many AI data centers are in the U.S.?

There is no official count. The Census keeps a live count of tracked AI data center projects, each listed with the evidence that it is built for AI.

Sources

1. Next-Gen AI Needs Liquid Cooling
   IEEE Spectrum · 2025-10-13
   Source excerpt

   “The average power density in a rack was around 8 kW … For AI, that's growing to 100 kW per rack. That's an order of magnitude.”

2. Understanding the energy-AI nexus – Energy and AI
   International Energy Agency · 2025-04-10
   2 source excerpts

   “A hyperscale, AI-focused data centre can have a capacity of 100 MW or more, consuming as much electricity annually as 100 000 households.”

   “A conventional data centre may be around 10-25 megawatts (MW) in size.”

3. Executive summary – Energy and AI
   International Energy Agency · 2025-04-10
   3 source excerpts

   “A typical AI-focused data centre consumes as much electricity as 100 000 households, but the largest ones under construction today will consume 20 times as much.”

   “Training and deploying AI models takes place in large and power-hungry data centres.”

   “Data centre electricity consumption is set to more than double to around 945 TWh by 2030. ... AI is the most important driver of this growth, alongside growing demand for other digital services. ... In the United States, data centres account for nearly half of electricity demand growth between now and 2030.”

4. Data Centers and Cloud Computing: Information Technology Infrastructure for Artificial Intelligence (IF12899)
   Congressional Research Service (via EveryCRSReport.com) · 2025-02-05
   2 source excerpts

   “In E.O. 14141, the term AI data center means a data center used primarily to develop or operate AI”

   “a GPU is generally considered better than a CPU to handle AI computational tasks ... Cutting-edge chip technologies also support high-speed (at the gigabit-per-second level) GPU-to-GPU data communication among hundreds of GPUs across multiple servers”

5. Accelerating Federal Permitting of Data Center Infrastructure (Executive Order 14318)
   Federal Register / Executive Office of the President · 2025-07-23
   2 source excerpts

   “Executive Order 14141 of January 14, 2025 (Advancing United States Leadership in Artificial Intelligence Infrastructure), is hereby revoked.”

   “These plans include artificial intelligence (AI) data centers and infrastructure that powers them, including high-voltage transmission lines and other equipment. It will be a priority of my Administration to facilitate the rapid and efficient buildout of this infrastructure by easing Federal regulatory burdens.”

6. Accelerating Federal Permitting of Data Center Infrastructure
   The White House · 2025-07-23
   Source excerpt

   “a facility that requires greater than 100 megawatts (MW) of new load dedicated to AI inference, training, simulation, or synthetic data generation. ... a Data Center Project or Covered Component Project for which the Project Sponsor has committed at least $500 million in capital expenditures as determined by the Secretary of Commerce”

7. 2024 United States Data Center Energy Usage Report
   Lawrence Berkeley National Laboratory · 2024-12-19
   3 source excerpts

   “In 2017, the overall server installed base started growing and Graphic Processing Unit (GPU)-accelerated servers for artificial intelligence (AI) became a significant enough portion of the data center server stock that total data center electricity use began to increase again”

   “Most notably, the rapid growth in accelerated servers has caused current total data center energy demand to more than double between 2017 and 2023, and continued growth in the use of accelerated servers for AI services could cause further substantial increases by the end of this decade.”

   “Water-cooled chillers are widely used in data centers, owing to their high efficiency and capacity to manage substantial cooling requirements. These systems use water-cooled condensers to extract heat from the system, subsequently releasing it into the environment via cooling towers.”

8. Uptime Institute Global Data Center Survey 2025
   Uptime Institute · 2025-07
   Source excerpt

   “Average modal density excluding these outliers comes to 7.5 kW in our 2025 survey, up from 6.8 kW in 2024. … More than 80% of the operators responding to our survey say their facility has no racks above 30 kW — about the same share as last year.”

9. Data Centers and Their Energy Consumption: Frequently Asked Questions (R48646)
   Congressional Research Service · 2025-08-26
   Source excerpt

   “High-performance computing (HPC) equipment has necessitated cooling methods that are thermodynamically closer to the chips and intercept the energy before it has substantially raised the room air temperature. These direct liquid cooling technologies can address the higher power of HPC.”

10. High-Performance Computing Data Center Warm-Water Liquid Cooling
    National Laboratory of the Rockies (formerly National Renewable Energy Laboratory) · 2025-12-04
    Source excerpt

    “Liquid has approximately 1,000 times the cooling capacity of air. … And rack power densities of 60 kilowatts (kW) per rack or more can be achieved using warm-water liquid cooling, which requires less floor space.”

11. Earnings call transcript: Alphabet beats Q2 2026 estimates, shares fall on capex surge
    Investing.com · 2026-07-22
    2 source excerpts

    “We are updating our full-year 2026 CapEx guidance range to $195 billion-$205 billion, up from our previous estimate of $180 billion-$190 billion.”

    “CapEx was $44.9 billion in the second quarter, with the vast majority of this spent in technical infrastructure to support our investments in AI. Approximately 60% of our investment in technical infrastructure this quarter was in servers, and 40% was in data centers and networking equipment.”

12. Meta Reports Second Quarter 2026 Results (Exhibit 99.1)
    Meta Platforms, Inc. (SEC filing) · 2026-07-29
    Source excerpt

    “We anticipate 2026 capital expenditures, including principal payments on finance leases, to be in the range of $130-145 billion, narrowed from our prior outlook of $125-145 billion.”

13. Andy Jassy said Amazon will spend $220 billion this year—and still won't have enough capacity to meet demand
    Fortune · 2026-07-30
    Source excerpt

    “During the call, Jassy told investors that Amazon now expects to devote $220 billion to capital expenditures in 2026, up from its prior estimate of $200 billion, owing to higher memory costs.”

Each source was opened and checked against the excerpt shown. Company statements are identified as such. Figures are as published; where sources differ, each is shown.