Core Insights - SuperX, an AI infrastructure solutions provider listed on NASDAQ, has launched the SuperX GB300NVL72 system, a rack-level AI supercomputing platform designed to overcome the physical and computational limits faced by trillion-parameter large models [1][2] - The system utilizes NVIDIA's GB300Grace Blackwell Ultra chips and marks a transition from traditional air cooling and AC power distribution to a liquid cooling rack era, emphasizing the importance of power and cooling systems as core components of next-generation AI workloads [1] Group 1 - The SuperX GB300 system is designed for large-scale horizontal expansion, connecting 72 Blackwell Ultra GPUs into a single massive GPU system, achieving 1.8 exaFLOPS of FP4 AI computing power within a single rack [2] - The advanced liquid cooling design is essential for sustaining the operation of this high-performance system, as it efficiently removes the significant heat generated by the chips, ensuring continuous 24/7 operation [2] - The SuperX GB300NVL72 system targets several key areas in future AI computing, including large-scale sovereign AI, exascale scientific computing, and industrial digital twins in sectors like automotive, high-end manufacturing, and energy [2]

Core Insights - SuperX AI Technology Limited has launched the SuperX GB300NVL72 system, a revolutionary AI supercomputing platform that marks a shift from traditional air cooling to advanced liquid cooling technology, addressing the exponential growth in AI computational demands [1][3][6] - The system is powered by NVIDIA's GB300Grace Blackwell Ultra super chip, designed to overcome physical and computational limits faced by large-scale AI models, providing up to 1.8 exaFLOPS of AI performance [1][4][6] Group 1: System Features - The SuperX GB300NVL72 system integrates a high-performance liquid cooling system with an 800VDC power supply, which is crucial for ensuring stability and efficiency at high power levels [3][6] - The architecture combines 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs, creating a unified memory pool that eliminates I/O bottlenecks, thus enhancing performance for large-scale models [4][6] - The system is designed for large-scale horizontal expansion, achieving groundbreaking performance metrics and redefining industry standards for AI training and inference [6][7] Group 2: Market Applications - The SuperX GB300NVL72 system targets several key areas, including large-scale and sovereign AI infrastructure, providing a foundation for national-level AI capabilities and public cloud services [7] - It is also positioned for high-performance computing (HPC) applications in scientific research, such as high-energy physics and climate modeling, where it can significantly accelerate data processing and complex calculations [7] - In industrial sectors, the system supports the creation of high-fidelity digital twins, enabling accurate simulations of factories, power grids, and cities by leveraging its powerful data processing and parallel computing capabilities [7][8]