Alibaba launches 10,000-card AI cluster as China ramps up tech push

Alibaba and China Telecom are moving ahead with a new data centre project in southern China, powered entirely by the e-commerce giant’s in-house AI chips, as Beijing steps up efforts to build domestic computing infrastructure.

The facility, unveiled on Tuesday, will be equipped with 10,000 of Alibaba’s Zhenwu semiconductors, designed for both AI training and inference. The system is capable of supporting models with hundreds of billions of parameters, placing it among the most advanced computing setups currently in operation.

Deployed at a data centre in Shaoguan, Guangdong province, the cluster is described as a “fully domestic” project and represents the first Zhenwu-powered system of this scale in the Greater Bay Area. Alibaba said the chips can operate as a unified system, enabling the cluster to function like a single supercomputer with ultra-low latency of around four microseconds.

The rollout highlights how China’s leading technology firms are accelerating development of proprietary AI chips and infrastructure as the country pushes for self-reliance in critical technologies.

In recent years, Washington has tightened restrictions on China’s access to advanced semiconductor technologies, including AI chips produced by Nvidia. The curbs have prompted Chinese firms to speed up the development of local alternatives across both hardware and infrastructure.

Alibaba Group Holding has been advancing its chip ambitions through its T-Head semiconductor unit, while continuing to expand its position in cloud computing. The company now operates across the full AI stack, from chip design to data centre construction and model development, with services delivered through its cloud division.

Cloud computing has remained one of Alibaba’s fastest-growing segments in recent quarters, supported by rising demand for AI workloads. Across China, investment in large-scale data centres using domestic technologies has picked up pace.

A similar project went online last month in Shenzhen, where a 10,000-card cluster built on Huawei’s Ascend 910C chips began operations, signalling a coordinated effort among Chinese firms to scale local computing power.

Unlike their U.S. counterparts, companies such as Meta and Microsoft, which are expected to collectively spend hundreds of billions of dollars on AI infrastructure this year, Chinese players have taken a more targeted approach. Investment has focused on sectors expected to generate measurable returns, including industrial applications and enterprise services.

Scaling AI infrastructure for real-world deployment

The Zhenwu-powered cluster adds to growing evidence that China is shifting from experimentation to large-scale deployment of AI systems. Demand for computing power continues to rise as industries integrate AI into production and services.

According to Charlie Zheng, chief economist at Samoyed Cloud Technology Group Holdings, the rollout of domestic clusters signals a transition from “hardware replacement” to “software collaboration” across China’s AI sector.

He noted that adoption has been fastest in government services and urban governance, where requirements around data sovereignty and system security remain particularly strict.

“The sector’s rigid demand for data sovereignty and security has driven the fastest deployment,” Zheng said.

Alibaba stated that the cluster delivers around 30% higher efficiency in training and inference tasks, while single-card throughput has increased nearly tenfold. The system has already been deployed in areas such as healthcare and advanced manufacturing.

Access to the cluster is being extended to small and medium-sized enterprises through China Telecom’s platform, with usage priced on a per-card or hourly basis.

Looking ahead, Alibaba plans to expand the system to 100,000 chips, a move expected to reduce costs further and improve overall resource utilisation as demand for large-scale AI computing continues to build.