Example facilities
Real AI data centers with what's publicly known about their insides. Every fact carries its source; supplier placements come from reporting, not disclosures, unless the source says otherwise. Use these as worked examples of the catalog in one building.
xAI's GPU supercluster for training Grok, built inside a former Electrolux appliance factory in south Memphis and famous for its build speed: the first 100,000 GPUs came online in 122 days. A second campus (Colossus 2) is under construction nearby.
The flagship campus of the OpenAI–Oracle–SoftBank Stargate venture, built and operated by Crusoe on roughly a thousand acres outside Abilene. Oracle owns the hardware and OpenAI consumes the compute; the first buildings were energized about a year after ground-break.
Microsoft's AI campus on the former Foxconn site in Mount Pleasant — three buildings and 1.2 million square feet on 315 acres, running Nvidia GB200 racks on closed-loop liquid cooling. Microsoft links it to a sister Fairwater site in Atlanta over a dedicated AI WAN.
Meta's largest AI campus, planned for up to 5 GW of compute on 2,250 acres in northeast Louisiana, with Entergy building three combined-cycle gas plants and a new substation to feed it. Investment guidance has grown from about $10 billion at announcement to more than $50 billion.
Amazon's purpose-built training campus for Anthropic in northern Indiana — 30 planned buildings designed around AWS's own Trainium2 chips rather than GPUs, brought from ground-break to a half-million-chip cluster in about a year.
Nebius's European flagship, expanded from 25 MW to 75 MW, cooled with outside air and feeding recovered server heat into the town's district heating network — a working example of the heat-reuse chain in the catalog.
One of Google's oldest and largest US campuses, running since the late 2000s and still expanding — a reference point for what a classic hyperscale campus looks like before the AI-era designs, and for how one grows across two decades.