TLDR
- Alibaba and China Telecom launched a 10,000-chip AI data center in Shaoguan, Guangdong province, powered by Alibaba’s own Zhenwu semiconductors.
- The cluster is the first Zhenwu-powered deployment of this scale in China’s Greater Bay Area, designed to train AI models with hundreds of billions of parameters.
- The data center promises 30% higher training and inference efficiency, with single-card throughput nearly 10 times higher than previous generations.
- Alibaba plans to scale the cluster to 100,000 chips, with computing access available to small businesses via China Telecom’s platform.
- The launch follows a similar 10,000-chip Huawei Ascend 910C cluster that went live in Shenzhen last month.
Alibaba (BABA) and China Telecom have officially launched a 10,000-chip AI computing cluster in Shaoguan, Guangdong province. The cluster is powered entirely by Alibaba’s own Zhenwu AI semiconductors, developed by its T-Head chip design unit.
Alibaba Group Holding Limited, BABA
The announcement marks the first time Zhenwu chips have been deployed at this scale in the Greater Bay Area. Alibaba Cloud described the move as a step toward moving China’s AI computing “from high-end performance breakthroughs to large-scale industrial implementation.”
ALIBABA LAUNCHES DATA CENTER WITH 10,000 OF ITS OWN CHIPS AS CHINA RAMPS UP AI PUSH
— First Squawk (@FirstSquawk) April 8, 2026
The cluster uses a next-generation high-performance networking architecture that lets all 10,000 chips operate as a single supercomputer. Alibaba says this delivers 30% higher training and inference efficiency, with single-card throughput up nearly 10 times versus older systems.
The system can train models with hundreds of billions of parameters â putting it in the same class as some of the largest AI models being developed globally.
Latency is rated at 4 microseconds, which Alibaba attributes to the network architecture tying the chips together. That figure is key for enterprise AI workloads where response time matters.
China’s Push for Domestic AI Infrastructure
The launch is part of a broader national effort. Beijing included intelligent computing infrastructure in its 15th five-year plan last month, and a State Council AI action plan from August called for an optimized buildout of computing resources across China.
By the end of June last year, China’s total computing power stood at 962,000 petaflops â 21% of world capacity, up 73% year-on-year, according to the China Academy of Information and Communications Technology.
The Shaoguan cluster has already been deployed across healthcare and advanced manufacturing use cases. Small and medium-sized businesses can access computing time through China Telecom’s platform, paying by the card or by the hour.
Alibaba also confirmed plans to scale the cluster from 10,000 to 100,000 chips. That expansion is aimed at lowering costs and improving overall resource efficiency.
Context: Huawei and the Domestic Chip Race
The launch follows a similar milestone last month, when China’s first 10,000-card intelligent computing cluster â built with Huawei’s Ascend 910C chips â went live in Shenzhen.
That cluster carries 11,000 petaflops of computing capacity and has been combined with a separate 3,000-petaflop cluster activated in 2024. Shanghai is also building out a 10,000-card cluster through a subsidiary of state-owned INESA, compatible with multiple domestic chip types.
While Chinese chips still trail Nvidia in raw individual performance, Beijing’s strategy relies on large-scale cluster architecture and efficient networking to close the gap.
U.S. export restrictions on Nvidia chips have accelerated China’s domestic chip development timeline. Alibaba’s T-Head unit has been a central part of that push, alongside Huawei.
BABA was trading up 7.79% on the day of the announcement, with after-hours gains adding a further 0.82% on its Hong Kong-listed stock (728-HK).







