Glossary
gigawatt-class cluster
An AI training facility whose power draw is measured in gigawatts rather than megawatts, the scale at which siting decisions become grid-and-permitting problems rather than real-estate ones.
A training cluster whose continuous power draw is one billion watts or more. Historic data centers ran at 10 to 50 MW; modern AI training clusters announced in 2024 to 2026 are sized at 1 to 5 GW, with multi-site campuses targeting 5 to 10 GW combined. Stargate’s first Abilene site is planned in this class; xAI’s Colossus has already crossed 100K H100s on a single power feed.
Mechanically the shift moves the bottleneck. Below ~500 MW the limit is rack-level cooling and building permits. Above 1 GW the limit is grid interconnect: utility build-out lead times of 4 to 7 years, transmission corridor approvals, sometimes new generation capacity specifically procured to serve one customer. That is why hyperscalers are signing 20-year nuclear PPAs and building behind-the-meterinfrastructureA power arrangement where generation sits on the same side of the utility meter as the load, letting a data center draw directly from the plant and bypass the grid. Open full entry generation.
In the open-source AI sovereignty argument this is the most centralizing layer of all. A few entities can afford GW-class buildouts; the rest of the field works at MW scale or rents capacity from those who can.