Alibaba Deploys 10,000 Zhenwu Chips in Data Center

Alibaba Cloud and China Telecom have inaugurated a new data center in southern China equipped with 10,000 of Alibaba’s self-developed Zhenwu AI semiconductors. The facility targets both model training and inference and is sized to host very large models — the company says it can support models with “hundreds of billions” of parameters. Alibaba will combine its chip designs, cloud services and in-house models to populate this cluster for commercial cloud offerings and internal development.

The announcement sits squarely in the context of Beijing’s drive for semiconductor self-sufficiency after U.S. policies tightened exports of advanced AI chips. Chinese hyperscalers and chip vendors are accelerating vertical stacks: designing chips (Zhenwu), building data center capacity, and training large models to offer end-to-end domestic alternatives. The CNBC report notes similar recent deployments, such as a Huawei cluster using Ascend 910C processors, signaling parallel efforts across major Chinese tech players.

The site hosts 10,000 Zhenwu accelerators and explicitly supports models on the scale of hundreds of billions of parameters, indicating both significant aggregate FLOPS and memory/networking capacity. Alibaba’s corporate profile — chip design, data center engineering, and model development sold through its cloud division — positions it to monetize the stack for Chinese enterprises and government customers. The partner on this project is China Telecom, a national carrier with deep data-center and network reach.

This is not merely a marketing deployment: it represents growing onshore compute supply that can shift where large-scale model training and inference happen for Chinese customers. For ML engineers and ops teams, domestic accelerators at this scale change cost, latency, data governance, and dependency calculations when deciding where to train or deploy models. For global practitioners, it highlights an accelerating bifurcation in AI compute ecosystems driven by policy and supply-chain controls.

Performance benchmarks and software ecosystem maturity (compilers, frameworks, optimized libraries) for Zhenwu; how Alibaba integrates this capacity into cloud offerings and pricing; and whether China Telecom expands such partnerships. Also watch comparative energy efficiency, interconnect design, and support for training workflows at multi-hundred-billion-parameter scale.