I kind of missed this news from May 5th: OpenAI's frontier models are being trained on Stargate Abilene and Microsoft's Fairwater in Atlanta. (https://openai.com/index/mrc-supercomputer-networking/) and from Nov 12th 2025: Microsoft can do a multi-cluster training with their Fairwater datacenters in Atlanta, Wisconsin and future sites, "which all work together to support OpenAI, the Microsoft AI Superintelligence Team, Copilot capabilities and other leading AI workloads" (https://news.microsoft.com/source/features/ai/from-wisconsin-to-atlanta-microsoft-connects-datacenters-to-build-its-first-ai-superfactory/) It seems likely that GPT-6 is or was already trained on between ~220k and ~520k GB200s. That's enough for a 15-30T model.