China Telecom deploys the first domestic TPU cluster - Zhonghao Xinying brings silicon to service

China Telecom deploys the first TPU cluster in Hangzhou based on Zhonghao Xinying's Chana chips. The Chinese sovereign compute gains a third silicon vendor with a scale customer.

In plain terms - China Telecom, ZTE and Zhonghao Xinying stand up China's first domestic TPU cluster in Hangzhou - 1,024 Chana chips, 400 PFLOPS today, 10K PFLOPS planned on the second-gen Xuyu TPU. Sovereign compute goes from spec sheet to installed base.

Le fait

The Hangzhou branch of China Telecom, with ZTE and Zhonghao Xinying, deploys the first Chinese TPU cluster based on Chana accelerators developed by Zhonghao Xinying (Pandaily, July 28, 2026). Cluster of 1,024 chips for 400 PFLOPS, with an expansion planned to 10,000 PFLOPS on the 2nd generation of TPU ("Xuyu"). This is another step in the construction of an autonomous Chinese compute stack (fil china-sovereign-compute) - after Huawei Atlas 950 SuperPoD (publi #1165) and Biren/ZTE optical supernodes (publi #1361), the ecosystem gains a third silicon vendor with a scale customer.

Notre lecture

Three angles. Vertical: China Telecom is a state infrastructure operator - the deployment signals that Zhonghao Xinying has passed the production test, not just the POC. Horizontal: unlike Huawei (vertically integrated, from silicon to datacenter), Zhonghao Xinying is an independent foundry; ZTE provides the integration. This is the condition for the formation of a real domestic AI chip market - getting out of the vendor-unique model. Trajectory: the announced transition from 400 to 10,000 PFLOPS (×25) on the 2nd generation is ambitious but not far-fetched - it's the order of magnitude of a scale-out packaging/interconnect rather than a silicon miracle.

À surveiller

• The next industrial customer (Alibaba, Baidu, Tencent) ready to switch from a single Huawei supplier to a multi-vendor mix.
• The measured performance of the Hangzhou cluster on real workloads (training, inference).
• The execution of the Xuyu roadmap (2nd generation) to reach the announced 10,000 PFLOPS.