On July 20, 2026, at the World Artificial Intelligence Conference (WAIC), China unveiled its first homegrown TPU thousand-card intelligent computing cluster in Eastern China, specifically in Hangzhou. This cutting-edge cluster, a collaborative effort among Hangzhou Telecom, ZTE, and Zhuhai Chiplink Technology, marks the first large-scale deployment of domestically produced TPU clusters within the China Telecom ecosystem.

Built upon the foundation of the 'Taize®' intelligent computing platform, the cluster is engineered to deliver robust computing power for a variety of high-demand scenarios, including large-scale model training, AI-driven inference, and scientific computing tasks. The 'Taize®' 2.0 server, a cornerstone of this cluster, integrates two high-performance CPUs with eight 'Xuyu®' AI chips, which feature a TPU architecture developed by Zhuhai Chiplink Technology. This configuration culminates in an impressive total computing capacity of 7.168 petaflops (P).

Remarkably, a single super node within this cluster can directly link up to 2,048 Xuyu chips, enabling the training of large models with parameters reaching into the trillions. At the software level, extensive adaptations have been made to ensure seamless compatibility with prominent large models, including the entire Qwen series, DeepSeek, and GLM, thereby enhancing the cluster's versatility and applicability in diverse AI applications.