Sourced evidence

Fairwater Atlanta power/networking architecture: ~140kW/rack, ~1,360kW/row (per-unit, not total site), two-tier 800Gbps Ethernet/SONiC backend, AI WAN fiber buildout

Microsoft's official Nov 12, 2025 blog post on the Fairwater Atlanta AI superfactory (the second site in its connected AI Superfactory network with Wisconsin) discloses power and networking architecture: rack-level power density of approximately 140kW and row-level of approximately 1,360kW -- Microsoft presents these explicitly as per-rack/per-row design figures, not a disclosed total campus power number, and no total-site MW figure should be inferred from them. Microsoft describes a "4 nines availability at 3 nines cost" grid-power strategy. Networking is described as a two-tier Ethernet/SONiC-based scale-out backend operating at 800Gbps. This differs in emphasis from Microsoft's own Sept 2025 Wisconsin post, which describes both InfiniBand and Ethernet fabrics operating at 800Gbps at that site -- a documented discrepancy between Microsoft's two official posts on networking-fabric composition across its two Fairwater sites, preserved here rather than reconciled by inference. Microsoft also discloses an AI WAN backbone effort, citing 120,000+ miles of new fiber added in the prior year to support inter-site and long-haul AI traffic.

Type
Company filing or release
Published
November 12, 2025
Confidence
85/100
Filed under
NVIDIA

Supports these links

Cite this

InsightNodes, "Fairwater Atlanta power/networking architecture: ~140kW/rack, ~1,360kW/row (per-unit, not total site), two-tier 800Gbps Ethernet/SONiC backend, AI WAN fiber buildout" (NVIDIA). Source: Microsoft Official Blog, November 12, 2025, https://blogs.microsoft.com/blog/2025/11/12/infinite-scale-the-architecture-behind-the-azure-ai-superfactory/. https://insightnodes.com/claims/e/nvidia-microsoft-fairwater-atlanta-networking-power-2025 (accessed October 29, 2026).

This page has a permanent address — link to it from notes, articles or models. Not investment advice.