Data Center Types

From chips in cars to gigawatt AI factories · One flagship per type

Data centers range from a chip in a car to gigawatt AI campuses. Each page here covers one type with a flagship site: Tesla's fleet, SpaceX Starmind in orbit, Akamai at the edge, Compass in Red Oak, Eli Lilly on-premises, Digital Realty in Northern Virginia, El Capitan, and SpaceX's Colossus 2.

Data center types by power per site on a log scale, from a phone or car chip at a few watts to AI factories at hundreds of megawatts to gigawatts, including orbital satellites at up to about 150 kW, edge sites, enterprise, colocation, HPC and hyperscale campuses POWER PER SITE, BY TYPE (LOG SCALE) 1 W 1 kW 1 MW 1 GW Vehicles and Devices Orbital Edge Modular Enterprise Colocation HPC Clusters Hyperscaler AI Factories
Types
9
Largest site
Colossus 2, 946 MW
Hyperscale share
48%
Smallest
Watts, in devices
Hub
Types

Types

Nine Types and How They Are Built

Each Types page defines one kind of data center, names a flagship site and lists the operators that matter.

Data center types and flagships
PageFlagshipSize
Vehicles and DevicesTesla fleet, Cybercab and OptimusWatts to a few kW per device
Orbital and Space Data CentersSpaceX Starmind~120-150 kW per satellite, from 2027
Edge Data CentersAkamai Inference Cloud; Duos Edge AI, TexaskW to ~10 MW
Modular Data CentersCompass and Schneider, Red Oak, TXModules to 3.5 MW; blocks to 50 MW
Enterprise Data CentersEli Lilly's LillyPod, IndianapolisServer rooms to campuses
Colocation Data CentersDigital Realty, Northern VirginiaCabinets to whole campuses
HPC ClustersEl Capitan, Lawrence Livermore~10-40 MW
Hyperscaler Data CentersFairwater, Rainier, Tembo100s of MW to GW
AI FactoriesSpaceX Colossus 2~400 MW to 5 GW
Deployment ModelsSelf-build to neocloudCost per MW, time to power

How to tell them apart

Size, Owner and Workload Separate the Types

  • Size: from watts in a car to gigawatts at an AI factory; the chart above spans nine orders of magnitude.
  • Owner: the company itself (enterprise), a landlord renting space and power (colocation), a cloud platform (hyperscaler) or a single AI developer (AI factory).
  • Workload: 64-bit simulation for many users (HPC), training and inference for a few huge jobs (AI factory), or low-latency inference near users (edge and devices).
  • How it is built: self-built, leased, prefabricated or financed through a joint venture; see Deployment Models and Business Models.

Shift

Capacity Is Moving to Hyperscalers and AI Factories

Hyperscalers held 48% of global capacity in 2025 and Synergy Research forecasts 67% by 2031, while company-owned capacity falls from 32% to 19%. The largest single sites are now AI factories: five US campuses run more than 400 MW each. At the other end, inference is spreading to edge sites, cars, robots and, in early tests, satellites.

Record

Dataset Record

Dataset
Index of data center types with flagship sites; data on each linked page
Sources
Listed on each Types page
Last verified
October 2026