Super Micro Computer on Tuesday introduced new 4U and 2-OU liquid-cooled NVIDIA HGX B300 systems aimed at hyperscale data centers and emerging AI factory deployments. The platforms extend Supermicro’s NVIDIA Blackwell portfolio and ship as part of its Data Center Building Block Solutions (DCBBS), targeting higher GPU density, improved power efficiency, and faster rack-scale deployment.
The new 2-OU HGX B300 system aligns with the 21-inch OCP Open Rack V3 (ORV3) specification and supports up to 144 NVIDIA Blackwell Ultra GPUs per rack. Each node integrates eight GPUs operating at up to 1,100W TDP and uses blind-mate liquid manifolds and modular trays to support serviceability at scale. A single ORV3 rack accommodates up to 18 nodes, while eight compute racks combined with NVIDIA Quantum-X800 InfiniBand networking and Supermicro in-row coolant distribution units scale to 1,152 GPUs in a single SuperCluster unit.
For traditional data center environments, Supermicro also introduced a 4U liquid-cooled HGX B300 system designed for standard 19-inch EIA racks. The platform supports up to 64 GPUs per rack and captures up to 98% of system heat using Supermicro’s DLC-2 direct liquid-cooling technology. Both form factors deliver up to 2.1TB of HBM3e memory per system and double cluster-level fabric throughput to 800Gbps using NVIDIA ConnectX-8 SuperNICs with either Quantum-X800 InfiniBand or Spectrum-4 Ethernet.
• 2-OU (OCP) HGX B300 system supports up to 144 GPUs per 21-inch ORV3 rack
• 4U HGX B300 system targets 19-inch EIA racks with up to 64 GPUs per rack
• Direct liquid-cooling captures up to 98% of system heat and supports 45°C warm-water operation
• Up to 2.1TB of HBM3e memory per system enables larger model training and inference
• Integrated 800Gbps networking accelerates agentic AI, foundation model training, and multimodal inference
• Platforms ship as validated L11/L12 rack-scale solutions under Supermicro DCBBS
“With AI infrastructure demand accelerating globally, our new liquid-cooled NVIDIA HGX B300 systems deliver the performance density and energy efficiency that hyperscalers and AI factories need today,” said Charles Liang, president and CEO of Supermicro.
🌐 Analysis
Supermicro’s HGX B300 expansion reflects an industry-wide shift toward rack-scale, liquid-cooled AI infrastructure as GPU power envelopes approach and exceed 1kW. The focus on ORV3 compatibility and 800Gbps networking aligns closely with hyperscaler roadmaps that emphasize serviceability, standardization, and fabric-level performance rather than node-level benchmarks alone. Competitors including Dell, HPE, and Lenovo are also expanding liquid-cooled Blackwell platforms, but Supermicro’s emphasis on dense OCP racks and pre-validated SuperCluster configurations positions it strongly for near-term AI factory deployments.
