Citat:
Ursprungligen postat av
Katki
Har inte behövts något enormt datacenter för ai hittills och skulle inte behövas framöver heller.
Klart det behövs. Det har körts i befintliga datacenter till stor del. Tror knappast att världens största företag lägger miljarders dollar på nybyggen om det inte behövs:
Meta's "Hyperion" (Louisiana) — a $27 billion development covering 2,250 acres, initially delivering 2 GW and scaling to 5 GW over several years, making it the world's largest planned AI datacenter.
Meta's "Prometheus" (Ohio) — expected to reach 1 GW when it comes online in 2026, making it one of the world's first gigawatt-scale AI datacenters.
Amazon Project Rainier (Indiana) — built on 1,200 acres of farmland, with Amazon spending $11 billion on the project. It will use 2.2 GW of electricity (equivalent to 1 million households) and is specifically intended for training Anthropic's Claude models.
Microsoft "Fairwater" (Wisconsin & Atlanta) — the Wisconsin campus houses more than 150,000 Nvidia GB200 GPUs with a current draw exceeding 350 MW, with plans to push beyond 2 GW.
xAI "Colossus 2" (Memphis) — Elon Musk's successor to Colossus 1, containing over 110,000 Nvidia GB200 GPUs, pulling 350–400 MW and planning to scale past 1 GW by mid-2026.
Stargate / Oracle (Abilene, Texas) — the flagship Stargate campus launched its first two buildings in September 2025 with 1.2 GW capacity and plans for more than 450,000 Nvidia GPUs. The full Stargate national program targets up to 5 GW across ~20 US locations.
Google (Columbus, Ohio) — Google's Columbus sites top 1 GW of compute capacity, making it one of the top AI clusters in the world.