NVIDIA's newly detailed Vera Rubin NVL72 system delivers approximately 10 times more tokens per megawatt than its current Grace Blackwell NVL72. The system also offers about one-tenth the cost per million tokens of the GB200 NVL72, reframing electricity as the primary scarce resource.
NVIDIA's Vera Rubin system shifts the compute bottleneck from silicon to power, making energy efficiency a critical factor for datacenter buildouts.