Alibaba opens AI centre with domestic chips

Alibaba and China Telecom have opened an artificial intelligence data center in Shaoguan, Guangdong, built around 10,000 Alibaba-designed Zhenwu chips. (cnbc.com) The facility was announced on April 8, 2026, and Alibaba said the chips are built for both training and inferencing, the two main jobs behind modern artificial intelligence systems. China Telecom will own and operate the site, while Alibaba supplies the processors and cloud stack. (cnbc.com) Training is the expensive part where a model learns from huge amounts of data; inference is the cheaper part where the finished model answers questions or generates text and images. Alibaba said this cluster can support models with hundreds of billions of parameters, the internal settings that shape how a model behaves. (cnbc.com) Alibaba’s cloud unit said the 10,000-chip system uses a networking setup with 4-microsecond latency, meaning the chips are linked tightly enough to act more like one large machine than 10,000 separate ones. That is the basic requirement for training very large models without losing speed as work moves between chips. (scmp.com) The project lands as Chinese companies are building more of their own artificial intelligence hardware after Washington tightened chip export controls in October 2022 and expanded them again on October 17, 2023. The U.S. rules target advanced computing chips and semiconductor manufacturing equipment tied to large-scale artificial intelligence and supercomputing. (cset.georgetown.edu) Alibaba has been preparing for this shift for years through T-Head, its chip-design arm, which also introduced the Yitian 710 server processor in October 2021 for Alibaba Cloud data centers. That earlier chip was aimed at general cloud computing, while Zhenwu is being deployed here for artificial intelligence workloads. (alibabagroup.com, cnbc.com) The company is also trying to tie chips, cloud infrastructure and models into one business. Alibaba said on March 19, 2026, that its Cloud Intelligence Group revenue rose 36% year over year to RMB43.284 billion, or about $6.19 billion, and that T-Head’s proprietary graphics processing units had entered scaled production. (alibabagroup.com) That same March update said Alibaba’s Qwen model family had passed 1 billion cumulative downloads on Hugging Face as of January 21, 2026, and the consumer-facing Qwen app had more than 300 million monthly active users as of February. More users and larger models both raise demand for computing power, which is why companies keep adding clusters like this one. (alibabagroup.com) Alibaba is not alone in the buildout. South China Morning Post reported that a 10,000-card cluster using Huawei Ascend 910C chips went online in Shenzhen in late March, showing that China’s domestic artificial intelligence infrastructure push now spans multiple companies, not just one cloud provider. (scmp.com) Alibaba and China Telecom said the Shaoguan site is expected to expand from 10,000 chips to 100,000. If that happens, the opening announced this week will look less like a one-off launch and more like a template for how Chinese artificial intelligence capacity is being built. (cnbc.com)