Key Highlights
- Alibaba partnered with China Telecom to activate a 10,000-chip AI computing facility in Shaoguan, Guangdong, utilizing Alibaba’s proprietary Zhenwu processors.
- This represents the inaugural large-scale deployment of Zhenwu technology in the Greater Bay Area, capable of training massive AI models with parameter counts in the hundreds of billions.
- The facility delivers 30% enhanced training and inference performance, with individual card throughput approximately 10 times greater than earlier models.
- Expansion plans call for scaling the infrastructure to 100,000 chips, with computing resources accessible to smaller enterprises through China Telecom’s service platform.
- This deployment follows a comparable 10,000-chip Huawei Ascend 910C facility that became operational in Shenzhen the previous month.
Alibaba (BABA) has teamed up with China Telecom to bring online a 10,000-chip AI processing cluster in Shaoguan, located in Guangdong province. The entire infrastructure runs on Alibaba’s internally developed Zhenwu AI processors, created by its T-Head semiconductor division.
Alibaba Group Holding Limited, BABA
This unveiling represents a watershed moment as the first large-scale Zhenwu chip deployment throughout the Greater Bay Area region. Alibaba Cloud characterized the initiative as advancing China’s AI computing capabilities “from cutting-edge performance achievements to widespread commercial deployment.”
The infrastructure incorporates an advanced high-speed networking framework that enables all 10,000 processors to function cohesively as one unified supercomputing system. Alibaba reports this configuration produces 30% improved training and inference performance, with individual card output increasing nearly tenfold compared to legacy platforms.
The installation possesses the capability to train algorithmic models containing hundreds of billions of parameters — positioning it alongside the most sophisticated AI systems currently under development worldwide.
Network latency measures just 4 microseconds, which Alibaba credits to the sophisticated network infrastructure connecting the processors. This metric proves critical for commercial AI applications where rapid response times are essential.
Beijing’s Drive Toward Indigenous AI Computing Capacity
This rollout forms part of a comprehensive national strategy. Chinese authorities incorporated intelligent computing infrastructure into their 15th five-year development blueprint last month, while an August State Council AI directive emphasized optimized distribution of computational resources nationwide.
As of the conclusion of June last year, China’s aggregate computing capacity reached 962,000 petaflops — representing 21% of global capacity and marking a 73% annual increase, per data from the China Academy of Information and Communications Technology.
The Shaoguan installation has already been implemented across healthcare and sophisticated manufacturing applications. Small and medium-sized enterprises can obtain computing resources via China Telecom’s infrastructure, purchasing access on a per-card or hourly basis.
Alibaba has additionally revealed intentions to expand the cluster from its current 10,000 chips to 100,000 units. This expansion strategy aims to reduce operational costs while enhancing overall resource utilization.
Background: Huawei and China’s Semiconductor Competition
This launch arrives one month after China’s inaugural 10,000-card intelligent computing cluster — constructed with Huawei’s Ascend 910C processors — commenced operations in Shenzhen.
That installation provides 11,000 petaflops of processing capacity and has been integrated with an additional 3,000-petaflop cluster activated during 2024. Shanghai is simultaneously constructing a 10,000-card facility through an INESA state-owned subsidiary, designed for compatibility with various domestic processor architectures.
While Chinese semiconductors continue to lag behind Nvidia in individual chip performance metrics, Beijing’s approach emphasizes extensive cluster configurations and optimized networking infrastructure to narrow the competitive gap.
American export controls on Nvidia processors have expedited China’s domestic semiconductor development schedule. Alibaba’s T-Head division has emerged as a cornerstone of this initiative, operating alongside Huawei.
BABA shares advanced 7.79% during regular trading on the announcement day, with extended-hours activity adding another 0.82% to its Hong Kong-listed shares (728-HK).
