Microsoft Announces Fairwater, Its AI Superfactory Located in Atlanta

Microsoft has announced Fairwater, which the company calls as AI superfactory, located in Atlanta, Georgia. The company says that it has been built to power the next era of hyperscale AI computing. The facility combines NVIDIA Blackwell GPUs, custom networking, and advanced power management.

Fairwater achieves 4×9 availability at 3×9 cost, thanks to highly available grid power and next-gen efficiency systems. Instead of relying on costly backup generators, Microsoft developed software and GPU-based energy controls that stabilize power use without waste.

These systems dynamically manage load, introduce supplementary workloads, and integrate on-site storage to maintain grid balance as AI workloads surge.

It’s worth noting that Fairwater data center runs a unified GPU cluster powered by NVIDIA Blackwell accelerators. Each rack packs up to 72 GPUs connected via NVLink, providing 1.8 TB of bandwidth and 14 TB of pooled memory per GPU. Interestingly, that’s among the highest compute densities ever achieved.

Microsoft has also redefined AI networking with a two-tier, 800 Gbps Ethernet-based system and its SONiC switch OS. Partnering with NVIDIA and OpenAI, the company introduced a Multi-Path Reliable Connected (MRC) protocol to improve data flow, telemetry, and congestion control for ultra-reliable AI performance.

Fairwater is part of Microsoft’s broader AI WAN optical network. For the uninitiated, it connects Azure AI data centers nationwide. With 120,000 new fiber miles added last year, this network allows seamless interconnection between supercomputers across the U.S., giving developers access to a unified AI infrastructure optimized for scale, flexibility, and cost-efficiency.

If you’d like to deep dive into everything, feel free to check this blogpost.