Super Micro Computer, Inc. (hereinafter referred to as Supermicro), a global IT solutions company, has begun providing high-density, high-efficiency infrastructure for hyperscale data centers and AI factories.


Supermicro announced on the 15th that it is introducing a new liquid-cooled server solution equipped with NVIDIA's latest HGX B300 platform.

The newly released products include two types: a 4U front I/O water-cooled system and a 2-OU (OCP) water-cooled system, each supporting standard 19-inch EIA racks and 21-inch OCP ORV3 specifications.

With this solution, Supermicro delivers both the high-density GPU performance and energy efficiency required to build hyperscale data centers and AI factories.

“Demand for AI infrastructure is exploding globally,” said Charles Liang, CEO of Supermicro. “Our NVIDIA HGX B300-based liquid-cooled solution is the industry’s most compact design, supporting up to 144 GPUs in a single rack. “Our proven DLC technology can significantly reduce cooling costs and energy consumption,” he emphasized.

He continued, “Supermicro’s Data Center Building Block Solutions (DCBBS) enable customers to rapidly deploy large-scale AI infrastructure by reducing time to market, delivering the highest performance per watt, and supporting end-to-end integration from design to deployment.”

The new 2-OU (OCP) system is designed to the 21-inch OCP ORV3 standard, allowing for up to 144 GPUs per rack.

This helps hyperscale and cloud companies achieve both space efficiency and serviceability.

Additionally, maintenance and expandability have been enhanced with blind-mate manifold connections, modular GPU/CPU tray architecture, and advanced liquid cooling technology.

In particular, it can efficiently process high-load AI workloads in a smaller area with eight NVIDIA Blackwell Ultra GPUs (up to 1,100W per GPU).

A single ORV3 rack supports up to 18 nodes and 144 GPUs, and can scale to 1,152 GPUs with NVIDIA Quantum-X800 InfiniBand switches and Supermicro in-row CDUs.

Another new product, the 4U front I/O water-cooled system, is designed to fit the existing 19-inch EIA rack form factor.

Based on Supermicro's DLC technology, it removes up to 98% of system heat through liquid cooling, maintaining stable performance even in large-scale AI factory environments.

This solution is optimized for high-density learning and inference clusters and offers various benefits, including enhanced energy efficiency, reduced noise, and improved serviceability.

Supermicro's HGX B300 solution features 2.1TB of HBM3e GPU memory per system.It supports large-scale model learning by equipping Mori.

Additionally, with NVIDIA ConnectX-8 SuperNICs, network throughput can be increased up to 800 Gb/s, enabling rapid processing of various AI workloads such as agentive AI applications, base model training, and multimodal inference.

Supermicro's DLC-2 technology reduces data center energy consumption by up to 40% and reduces water usage through 45°C hot water cooling. Eliminating the need for traditional chilled water and compressors significantly improves operational efficiency.

This launch further strengthens Supermicro's NVIDIA Blackwell architecture-based product line.

It has been certified by NVIDIA by applying NVIDIA AI Enterprise and Run:ai software along with the existing GB300 NVL72, HGX B200, and RTX PRO 6000 Blackwell Server Edition.

Customers can easily build flexibly scalable AI infrastructure, from single nodes to full-stack AI factories.