First Nvidia Vera Rubin Racks Arrive in Microsoft Data Centers
Satya Nadella posted photos of the first serial Nvidia Vera Rubin racks arriving at Microsoft data centers, and Nvidia confirmed the platform has entered full-scale production. Vera Rubin is the rack-scale AI platform that succeeds Blackwell: each NVL72 system packs 72 Rubin GPUs and 36 Vera CPUs on the Arm architecture, fully liquid-cooled with modular, cable-less trays that install in about five minutes instead of two hours.
Nvidia says that versus GB200 NVL72, the platform trains models with four times fewer GPUs and cuts the cost per million tokens for agentic inference roughly tenfold.
Related: Nvidia Puts $7 Billion Behind Open-Weight Models to Rival DeepSeek and Kimi, Nvidia RTX Spark Superchip: Technical Details