At the World Artificial Intelligence Conference (WAIC), Hangzhou proudly unveiled Eastern China's inaugural locally-produced TPU thousand-card intelligent computing cluster. This milestone project, a collaborative effort among Hangzhou Telecom, ZTE, and Zhuhai Serecore Technology Co., Ltd., marks the first significant deployment of domestically-manufactured TPU clusters within the China Telecom network.

Built upon the 'Taize®' intelligent computing platform, the cluster's 2.0 version AI server boasts two high-performance CPUs and eight AI chips based on Zhuhai Serecore Technology's 'Xuyu®' TPU architecture, collectively delivering a staggering 7.168P of computing power. Remarkably, a single super node can directly link up to 2048 Xuyu chips, enabling the training of large-scale models with trillions of parameters.

On the software front, the cluster has undergone extensive optimization to seamlessly integrate with large models such as the comprehensive Qwen series, DeepSeek, and GLM. This ensures robust, autonomous computing power support for a wide range of applications, including large model training, AI inference, and scientific computing.