Alibaba launched a 10,000-chip AI cluster using its own chips

Alibaba and China Telecom have launched a data center in Shaoguan, Guangdong province, built around 10,000 of Alibaba's Zhenwu AI chips — the first deployment of that scale in the Greater Bay Area, Alibaba's cloud unit said Tuesday.

Capable of handling models reaching into the hundreds of billions of parameters, the Zhenwu processors are built to handle both AI training and inferencing workloads. Alibaba develops the chips through its T-head unit. The two companies said the facility is expected to expand to 100,000 chips and can serve industries from healthcare to advanced materials.

Alibaba Cloud characterized the cluster as "fully domestic" and used the occasion to declare that the country's computing buildout had shifted from chasing benchmark milestones to achieving what it called "large-scale industrial implementation."

The launch comes as Chinese technology companies accelerate efforts to build domestic AI infrastructure. Last month, a computing cluster built with Huawei's Ascend 910C AI chips went online, according to CNBC.

U.S. export restrictions on advanced semiconductors, including Nvidia $NVDA AI chips, have pushed Chinese companies to develop homegrown alternatives. The company's cloud unit — a consistently fast-growing segment in recent quarters — spans the full stack from chip design and data center construction to proprietary AI models that are offered to customers as commercial services, according to CNBC.

The chip restrictions have reshaped both sides of the market. China has directed its top technology firms, including Alibaba, to stop purchasing Nvidia chips, while separately Beijing issued guidance barring foreign-made AI chips from data center projects funded even partly by the state. The domestic alternative push reflected in the Shaoguan facility is a direct consequence of that pressure.