Supermicro says it is shipping NVIDIA Vera Rubin NVL72 liquid-cooled racks

On 23 September 2026 Super Micro Computer (NASDAQ: SMCI) said it is now shipping NVIDIA Vera Rubin NVL72 racks integrated with Supermicro’s Data Center Building Block Solutions (DCBBS) and its DLC-2 direct liquid-cooling stack. The company press release, carried on its site and PR Newswire, frames the move as production delivery of liquid-cooled rack-scale systems rather than a future roadmap announcement.

What each rack contains

According to Supermicro, a Vera Rubin NVL72 rack combines 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs as one liquid-cooled machine: eighteen 1U compute trays (four Rubin GPUs and two Vera CPUs each) connected through nine sixth-generation NVIDIA NVLink switch trays. The company cites 216 TB/s of scale-up bandwidth, 20.7 TB of HBM4 memory and up to 54 TB of LPDDR5X per rack. NVIDIA’s own platform pages describe Vera Rubin NVL72 as a rack-scale system for training and agentic inference, with liquid cooling as a design assumption for the higher thermal load versus prior generations.

Cooling and deployment package

Supermicro says it validates every rack with a full fluid loop — cold plates through manifolds to cooling towers — and offers in-row cooling distribution units rated at 1.8 MW per CDU with N+1 redundancy, plus optional rear-door heat exchangers. Its DCBBS “Blueprint” defines a bill of materials from about 5 MW to gigawatt scale; one Scalable Unit example lists 1,152 Rubin GPUs and 331 TB of HBM4 across 16 compute racks with matched power, cooling, storage and networking. CEO Charles Liang said customers can order a Scalable Unit and receive production-ready systems with end-to-end integration.

Context and limits

Partner shipping of Vera Rubin hardware has been a rolling 2026 story; Supermicro’s 23 September release is a vendor confirmation of NVL72 rack shipments with a full liquid-cooling stack. Performance claims such as tokens-per-megawatt improvements versus Blackwell come from NVIDIA and cloud partners in separate materials and are not independently verified in this article. Availability, lead times and pricing remain customer-specific.