Overview
- Microsoft says Atlanta is live as the first Fairwater site, with the Mount Pleasant, Wisconsin campus being brought online to operate as one unified supercluster.
- The company connects the sites across roughly 700 miles via a dedicated AI Wide Area Network and new fiber links to run single, large AI workloads in real time and cut training from months to weeks.
- Fairwater centers deploy Nvidia GB200 NVL72 rack‑scale systems scaling to hundreds of thousands of Blackwell GPUs to provide massive training capacity.
- A two‑story layout with direct‑to‑chip, closed‑loop liquid cooling increases GPU density, trims cable runs and latency, and minimizes water use.
- Microsoft positions the system for partners including OpenAI, Mistral AI and xAI, while ramping capital outlays and planning to double its global data‑center footprint within about two years.