These DLC-2 systems not only achieve faster time-to-deployment but also provide substantial operational cost benefits, including reduced noise levels as low as 50dB.

The introduction of these systems supports future NVIDIA HGX B300 platforms, offering customers a compelling upgrade path and potential recurring revenue opportunities.

Charles Liang, CEO of Supermicro, emphasized the company's commitment to delivering tailored solutions quickly to meet customer needs in the rapidly evolving AI landscape.

Supermicro has expanded its NVIDIA Blackwell system portfolio by introducing new 4U Direct Liquid-Cooled (DLC-2) systems and enhanced 8U air-cooled models, specifically designed to support large-scale AI training and cloud-scale inference workloads.

The new 4U DLC-2 system is optimized for efficiency, offering up to 40% data center power savings, a 40% reduction in water consumption, and up to 98% heat capture, which significantly enhances operational efficiency.

Supermicro's 8U air-cooled system is designed for environments without liquid cooling, offering flexibility in system memory configuration while maintaining high cooling performance.

The front I/O design of these systems simplifies deployment and management of AI infrastructure by relocating networking components and storage bays to the cold aisle, enhancing thermal efficiency.

Both systems are equipped with high-performance NVIDIA ConnectX-7 NICs and BlueField-3 DPUs, designed to support significant scalability and robust networking capabilities in AI data centers.

Both systems support dual-socket Intel Xeon 6700 Series processors and feature 32 DIMM slots with a maximum capacity of 8TB, along with 8 NVIDIA ConnectX-7 NICs and 2 NVIDIA Bluefield-3 DPUs, ensuring robust performance.

The 4U liquid-cooled system features front-accessible components and supports up to 180GB of HBM3e memory per GPU, while the 8U air-cooled system maintains similar specifications without requiring liquid cooling infrastructure.

The NVIDIA Blackwell platform provides significant performance improvements, delivering up to 15 times faster inference and three times faster training compared to previous generations of GPUs.

With these additions, Supermicro's portfolio now includes eight different Blackwell-based system configurations, showcasing their adaptability to diverse customer needs in the evolving AI infrastructure market.