
Nvidia’s Vera Rubin: a modular, power-efficient AI data-center rack
Nvidia unveiled Vera Rubin, its next AI rack-scale system set to ship in H2 2026, boasting about 10x the performance per watt of the Grace Blackwell platform while consuming roughly twice the total power. The 72 Rubin GPUs and 36 Vera CPUs reside in a fully liquid-cooled, modular rack designed for quick swaps and easier upgrades. Competition from AMD’s upcoming Helios and in-house cloud chips from Amazon and Alphabet adds pressure, but Meta plans to deploy Rubin by 2027. Pricing is reportedly around $3.5–$4 million per rack, signaling strong pricing power as AI data-center demand grows. Analysts show a Strong Buy with a notable upside potential.













