Supermicro has begun shipping its Data Center Building Block Solutions based on the NVIDIA Vera Rubin platform. The flagship NVL4 configuration integrates 1,152 Rubin GPUs and 576 Vera CPUs across eight liquid-cooled racks, operating within a 3.2 MW power envelope. The system utilizes the company's DLC-2 direct liquid cooling technology, which manages a thermal load of up to 362 kW per rack.
The Vera Rubin architecture succeeds the Blackwell generation and features HBM4 memory alongside Arm-based Vera CPUs. Beyond the NVL4, Supermicro has introduced configurations for NVL72 and HGX NVL8 architectures. Full production volume for the entire product line is anticipated for the second half of 2026. The company has expanded its manufacturing capacity in the US and Taiwan to support production.
The NVL4 architecture is designed to scale from a 5 MW deployment to gigawatt-class clusters. Supermicro provides these blueprints as standardized, pre-validated systems to accelerate cluster deployment. Following the announcement, Supermicro shares rose by approximately 15 to 17 percent. The company competes with firms such as Dell and HPE in the AI infrastructure market.
