Supermicro, Inc. (NASDAQ: SMCI) has announced full production availability of its end-to-end AI data center Building Block Solutions powered by the NVIDIA Blackwell platform, a significant milestone in providing highly efficient and scalable infrastructure for advanced AI, machine learning, HPC, cloud, storage, and 5G/Edge applications. The new portfolio features a comprehensive suite of systems that offer both air-cooled and liquid-cooled configurations, each designed with superior thermal management capabilities to support a range of CPU options. These systems include the innovative NVIDIA HGX B200 8-GPU solutions, which are available in both 4U and 10U form factors and incorporate newly developed cold plates alongside a powerful 250kW coolant distribution unit. This advanced cooling architecture, which utilizes both liquid-to-liquid (L2L) and liquid-to-air (L2A) techniques, more than doubles the cooling capacity of previous models while preserving a compact 4U footprint. In a 42U rack, eight systems can be installed to accommodate 64 NVIDIA Blackwell GPUs, with configurations scaling up to 12 systems in a 52U rack that collectively house 96 GPUs. Enhanced by vertical coolant distribution manifolds that free up valuable rack space, these solutions deliver unmatched efficiency and rapid time-to-deployment, all while integrating a full suite of data center management software, comprehensive rack-level integration, and cluster-level validation services.
Supermicro’s latest offerings extend to a redesigned 10U air-cooled system that features a modular GPU tray optimized for eight 1000W TDP Blackwell GPUs, delivering up to 15 times the inference performance and triple the training capabilities compared to earlier designs. The portfolio also encompasses a liquid-cooled 4U system that builds upon the success of previous NVIDIA HGX H100/H200 platforms, further enhancing cooling efficiency and serviceability with advanced tubing designs and innovative cold plate technology. Additionally, the new rack-scale designs support multiple configurations with 42U, 48U, and 52U options, facilitating dense architectures that can scale from 64 GPUs in a single rack to 768 GPUs across nine racks. These systems are purpose-built to integrate with high-speed networking solutions such as NVIDIA Quantum-2 InfiniBand and NVIDIA Spectrum-X Ethernet, enabling non-blocking, scalable compute fabrics that support demanding AI workloads. This hardware is complemented by native support for NVIDIA AI Enterprise software, which streamlines the development and deployment of production-grade AI pipelines, while NVIDIA NIM microservices ensure secure and efficient access to the latest AI models whether deployed in data centers, cloud environments, or workstations.
These advancements underscore Supermicro’s commitment to delivering sustainable, cutting-edge data center solutions that meet the escalating demands of modern AI applications. The comprehensive approach not only addresses the physical challenges of thermal management and high-density GPU configurations but also integrates end-to-end services ranging from detailed system design and rigorous testing to professional support and global delivery. By seamlessly incorporating advanced liquid-cooling technologies, robust network connectivity, and sophisticated management software, Supermicro is well-positioned to enable enterprises to harness the transformative power of AI with unprecedented efficiency and scalability. This integrated solution set, bolstered by worldwide manufacturing capabilities in San Jose, Europe, and Asia, demonstrates a clear vision for a future where AI-driven innovation can thrive in an environment optimized for both performance and sustainability.