SuperX AI Technology Limited (NASDAQ: SUPX) today announced the SuperX GB300 NVL72 System, a liquid-cooled, rack-scale AI supercomputing platform designed to tackle the growing demands of trillion-parameter AI models. Powered by NVIDIA’s GB300 Grace Blackwell Ultra Superchips, the system combines extreme compute density, energy efficiency, and advanced infrastructure for modern data centers.
A single GB300 NVL72 rack achieves up to 1.8 exaFLOPS of FP4 AI compute, making it a milestone in high-density AI performance. Traditional air-cooled data centers and AC power distribution struggle to support such concentrated compute workloads, making 800VDC power delivery and integrated liquid cooling not just efficiency improvements, but operational necessities.
Key Hardware Advantages
The GB300 system integrates 72 NVIDIA Blackwell Ultra GPUs with 36 NVIDIA Grace CPUs in a 2:1 GPU-to-CPU ratio. High-bandwidth connections (900GB/s chip-to-chip links) unify 2,304GB HBM3E GPU memory with 17TB LPDDR5X CPU memory, enabling large-scale model training without I/O bottlenecks. The system also supports 4 NVIDIA NVLink connectors and 800Gb/s InfiniBand XDR networking, ensuring ultra-low-latency scaling across clusters.
Liquid Cooling & Rack-Scale Design
Advanced liquid cooling enables 24/7 continuous operation, maintaining peak performance while optimizing energy efficiency. The system scales horizontally, linking up to 72 GPUs in a single NVL72 rack, and is designed to integrate seamlessly with SuperX’s Prefabricated Modular AI Factory, which includes cooling and 800VDC infrastructure to support high-density deployments.
Applications & Market Positioning
The GB300 NVL72 system is ideal for organizations tackling next-generation AI workloads and high-performance scientific computing:
- Hyperscale & Sovereign AI: National AI infrastructure, cloud providers, and enterprise AI factories for massive multi-modal and large language models.
- Exascale Scientific Computing: Government and research institutions handling physics, materials science, or climate simulations.
- Industrial Digital Twins: Automotive, manufacturing, and energy sectors building high-fidelity digital twin simulations.
Technical Highlights (Per Rack):
- CPUs: 36 NVIDIA Grace (144 Arm Neoverse V2 cores)
- GPUs: 72 NVIDIA Blackwell Ultra
- Memory: 2,304GB HBM3E + 17TB LPDDR5X
- Peak AI Performance: ≈1.8 ExaFLOPS FP4
- Networking: NVLink 1.8TB/s, ConnectX 8 OSFP 800Gb/s, BlueField 3 DPU 400Gb/s
- Dimensions: 48U NVIDIA MGX Rack (2296mm H × 600mm W × 1200mm D)
SuperX GB300 NVL72 represents a critical leap for enterprises, research institutions, and governments needing exascale compute with energy-efficient density. The system is designed to serve as the foundation for next-generation AI infrastructure, combining high-density hardware, liquid cooling, and optimized DC power delivery into a single, integrated solution.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
