AI training and inference
Training and serving large AI models. 119 data centres on EveryDatacenter so far.
AI data centres train and serve machine-learning models. Training a large language model uses tens of thousands of GPUs or other accelerators for weeks or months in one place; serving it ("inference") runs in many data centres closer to users. Both need far more power per rack than ordinary servers, and AI is now the main driver of new data centre building.
Training and inference
Training needs the largest single clusters, joined by very fast networks, so it goes to gigawatt campuses such as Stargate at Abilene, Colossus in Memphis and Meta's Hyperion in Louisiana. Inference needs less power in one place but runs all the time and grows with users, so it spreads into cloud regions and colocation halls. The IEA expects AI to be the main reason data centre power use doubles by 2030.
National AI capacity
Many governments now fund AI computing at home. India's IndiaAI Mission, approved in 2024, rents tens of thousands of GPUs from Indian firms such as Yotta, E2E Networks and others for start-ups and researchers. Saudi Arabia set up Humain in 2025, the UAE is building a 5 GW AI campus in Abu Dhabi, and Europe funds AI factories at its supercomputer centres.
Famous AI training and inference data centres
Further reading
Photo credits on this page
The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.
- CoreWeave: Wikideas1, CC0, source
- LNCC: Trabalho próprio., Public domain, source