NVIDIA Unveils Vera Rubin Architecture With 10-Fold Performance Gain, Tackling AI Energy Bottleneck

Vera Rubin is NVIDIA's rack-scale successor to Blackwell, designed for agentic AI factories. The platform integrates Vera CPUs, Rubin GPUs, HBM4, NVLink 6 and an all-liquid-cooled design. It is intended to lower energy use and cost per token while uniting global suppliers including TSMC, Foxconn and Wistron around a common specification.

Jensen Huang announced at GTC Taipei on June 1, 2026, that Vera Rubin had entered full-scale production, with partner products expected to launch in the second half of the year. The NVL72 combines 72 Rubin GPUs with 36 Vera CPUs, delivering up to 10 times Blackwell's inference throughput per watt and cutting cost per token by 90%. NVIDIA also reiterated its April 14, 2025 plan to produce up to $500 billion of AI infrastructure in the United States with partners over four years.