Overview
- Microsoft says the first live Fairwater node connects Atlanta with Mount Pleasant, Wisconsin as a single training complex over high‑speed, dedicated fiber.
- Fairwater facilities are purpose‑built for unified AI workloads with two‑story layouts, closed‑loop liquid cooling, and capacity to integrate hundreds of thousands of Nvidia GPUs per site.
- The platform will train and serve next‑generation models for OpenAI, Mistral AI, and xAI alongside Microsoft’s own systems.
- Technical reporting notes the Atlanta site will deploy Nvidia GB200 NVL72 rack systems, and coverage indicates the cross‑datacenter link began operating in October.
- Microsoft reported more than $34 billion in capex last quarter and plans to double its global datacenter footprint within two years as rivals scale projects such as AWS’s Project Rainier.