Azure readies datacenters for NVIDIA Rubin platform at CES 2026

Microsoft / Azure announced at CES 2026 that Azure datacenters are already engineered and validated to deploy NVIDIA’s next-generation Vera Rubin NVL72 platform across existing and future Fairwater AI superfactories.

• Main announcement & deployment details: Azure will integrate NVIDIA Vera Rubin NVL72 racks across Fairwater sites (named current examples: Wisconsin and Atlanta, plus future locations). Rubin hardware claims 50 PF NVFP4 per chip and 3.6 EF NVFP4 per rack (a five times jump versus GB200 NVL72), and Azure states it has redesigned rack architecture, power, cooling, and topology (including NVLink and ConnectX-9 support) for immediate deployability.
• Background & platform validation: Azure cites years of co-design with NVIDIA (Ampere, Hopper, GB200/GB300 NVL72 deployments), operation of the world’s largest commercial InfiniBand deployments, and platform components such as NVIDIA NVLink (~260 TB/s), NVIDIA ConnectX-9 (1,600 Gb/s), Azure Boost, CycleCloud/AKS orchestration, liquid-cooling Heat Exchanger Units, HSM silicon offloads, and regional AI superfactory design as the basis for validated integration.