Core Insights - SuperX has launched a modular AI factory to address key challenges in traditional AI data center construction, including long build times, high costs, significant energy consumption, and limited scalability [1][2] - The modular AI factory aims to reduce the construction time of AI data centers from the traditional 18-24 months to under 6 months, enabling faster deployment of computing power [1] Delivery Speed - The solution allows for all core modules, including computing, cooling, and power units, to be prefabricated, integrated, and tested in the factory, requiring only quick assembly on-site [1][2] - This improvement in delivery speed enables AI companies to convert computing power into business value more rapidly [1] Computing Density - The core computing unit, SuperX NeuroBlock, supports up to 24 NVIDIA GB300NVL72 units with a power capacity of 3.5MW, achieving seven times the density of traditional solutions [2] - The integration with high-density liquid cooling and HVDC power systems significantly reduces the physical space required for the same computing density [2] Energy Efficiency - The solution employs high-voltage direct current (HVDC) technology, enhancing end-to-end power supply efficiency to over 98.5% [2] - The overall Power Usage Effectiveness (PUE) is reduced to 1.15, which is a 23% decrease in total energy consumption compared to traditional air-cooled systems with a PUE of approximately 1.5 [2] Flexibility - The modular deployment allows for a shift from complex on-site construction to standardized module production in the factory [2] - Clients can start with a single module and seamlessly scale from "1 to N" based on business needs, showcasing the flexibility of the solution [2] Innovation - The modular AI factory represents a fundamental shift from "complex customization" to "standardized products," enabling plug-and-play capabilities [2] - This innovative solution is expected to drive the transformation of traditional data centers into AI factories [2]
SuperX发布模块化AI工厂 部署周期有望缩短至6个月内
Zheng Quan Shi Bao Wang·2025-10-03 04:08