Alibaba unveils a 10,000-card intelligent computing cluster powered by its Zhenwu AI chips from the T-Head unit, marking a major domestically built AI infrastructure deployment in China.

The Shaoguan, Guangdong data centre represents the first Zhenwu-powered system of this scale in the Greater Bay Area, with potential to run models containing hundreds of billions of parameters.

Plans call for expanding the system to 100,000 chips to cut costs and boost resource utilization as demand for large-scale AI computing grows.

Alibaba emphasizes that the cluster delivers roughly 30% higher training and inference efficiency and nearly tenfold improvement in single-card throughput, with pay-as-you-go access for SMEs via China Telecom.

The project is already being used in healthcare and advanced manufacturing, and Alibaba plans to broaden access to smaller businesses through China Telecom’s platform and per-card or hourly pricing.

Beyond hardware, Alibaba envisions deeper software collaboration within China’s AI ecosystem, with government services and data sovereignty driving adoption.

Industry adoption spans healthcare, manufacturing, and government services, where data sovereignty and security considerations accelerate domestic AI infrastructure deployment.

The Shaoguan initiative aligns with a broader push to support cloud computing and applications across sectors like healthcare and advanced materials.

The effort fits Beijing’s five-year plan to build intelligent computing infrastructure, high-performance resources, and ultra-large AI clusters nationwide.

China’s total computing power reached about 962,000 petaflops by mid-2025, underscoring a shift toward expandable systems and domestic chip-led design.

The expansion toward 100,000 chips reflects a strategy to scale capacity, reduce costs, and optimize utilization amid rising demand for large-scale AI computing.

The project is marketed as fully domestic, using Chinese-made chips to support a unified system with ultra-low latency around four microseconds.