Supermicro unveiled ten new AI rack models designed to simplify large-scale GPU cluster setups. These racks, weighing up to 5,500 pounds, come pre-assembled and pre-tested, aiming to cut the usual delays between hardware delivery and operation. The lineup falls under the company's Data Center Building Block Solutions (DCBBS) portfolio, highlighting a strategy to streamline data center construction.
The racks offer multiple sizes, including 44OU, 48U, 48OU, and 52U, where “O” variants feature open designs for enhanced cooling. Each rack supports NVIDIA’s Vera Rubin/GB300 platform and complies with OCP ORv3 standards, boasting a static load capacity that surpasses typical industry levels at 2,500 kg.
Built for Intense Workloads and Efficiency
Liquid cooling is embedded directly into the racks, capable of handling up to 120 kW, which is key for maintaining performance in high-density AI compute environments. Before shipping, each unit undergoes shock and vibration testing to ensure reliability right out of the box.
According to Supermicro, their production capacity reaches 3,000 advanced racks monthly, with two-thirds equipped for liquid cooling. Manufacturing spans key locations in Silicon Valley, the Netherlands, and Taiwan. CEO Charles Liang emphasizes the importance of DCBBS as a modular foundation, extending beyond individual servers to full data center builds.
