China’s Sugon unveils new AI infrastructure to rival platforms from Nvidia and Huawei

An eflop, for exa floating-point operations per second, is a unit used to measure the performance of supercomputers and high-performance AI hardware that could calculate at least one quintillion floating-point operations per second.

“The scaleX 10,000-card supercluster is a large-scale computing infrastructure designed for complex scenarios such as trillion-parameter models and AI for science,” said Sugon senior vice-president Li Bin, in a statement. This infrastructure leverages the company’s “expertise in the research and development of large-scale computer systems”.

Some of the scaleX supercluster’s capabilities have already surpassed those of Nvidia’s Rubin Ultra NVL576 rack, according to Li. The new Nvidia platform is expected to be introduced in 2027.

Growing competition in the AI infrastructure arena underscores China’s high demand for greater computing power to process data and perform various digital tasks.