Microsoft has introduced a new class of AI infrastructure with the launch of its Fairwater 2 datacenter in Atlanta.

Now operational, the facility is directly connected to an earlier Fairwater site in Wisconsin, forming the foundation of what the company describes as its first AI superfactory. Rather than functioning as isolated facilities, these sites are designed to work together as a single computing environment.

A dedicated fiber-optic network allows data to move between locations with minimal delay, supporting large-scale AI training jobs that span multiple states.

The two datacenters follow a shared design built to support high-throughput AI computing.

Each facility uses custom rack architecture and advanced GPUs, paired with a liquid cooling system that significantly reduces water consumption. This setup is built to handle large-scale AI workloads, including training for OpenAI models and powering Microsoft’s own AI services like Copilot.

Built on several years of infrastructure development, the design aims to maintain reliable performance under increasing demand.

Why It Matters: AI development is pushing infrastructure to handle larger models and heavier workloads. Datacenters now require faster training speeds, increased throughput, and more efficient energy use. Microsoft’s Fairwater network meets these needs by connecting facilities through a system built specifically for AI computing across multiple locations.

• Connected Datacenters That Function as One: Fairwater sites in Atlanta and Wisconsin are joined by Microsoft’s AI Wide Area Network, a dedicated fiber backbone designed to keep data moving quickly across long distances. This connection allows the datacenters to share GPU workloads in real time, maintaining steady performance and avoiding the delays that can occur when systems operate in isolation.

• Purpose-Built Hardware in a High-Density Layout: Each site uses NVIDIA GB200 NVL72 systems, with GPUs grouped into rack-scale clusters to support fast, direct communication. The datacenters are built with a two-story layout that increases chip density while keeping internal cables short, which helps reduce latency and improve overall efficiency during AI training.

• A Network Tailored to AI Traffic: The fiber infrastructure supporting Fairwater spans more than 120,000 miles and has been reconfigured to support the demands of AI computing. By optimizing the network protocol and layout, Microsoft has created a system that can move large volumes of training data between sites without the congestion seen in traditional networks.

• Cooling Systems Designed for Efficiency and Conservation: To manage the heat produced by dense GPU clusters, Microsoft developed a closed-loop liquid cooling system that uses very little water. The system relies on an initial fill, roughly equal to the yearly water use of about 20 homes, and only requires replacement if water quality changes. This approach supports high-performance computing while reducing ongoing resource use.

• Reliable Power Without Traditional Backup Systems: The Atlanta site runs on a reliable utility grid connection, removing the need for generators, UPS systems, or dual-corded power distribution. This setup reduces both cost and complexity while maintaining a high level of availability. It provides a straightforward and efficient way to support continuous AI operations.

Go Deeper -> From Wisconsin to Atlanta: Microsoft connects datacenters to build its first AI superfactory – Microsoft

Microsoft reveals new datacenter for Atlanta, will be world’s first ‘AI super factory’ – WSBTV