Huawei Debuts CloudMatrix 384 AI System to Rival Nvidia/ Newslooks/ WASHINGTON/ J. Mansour/ Morning Edition/ Huawei has unveiled its CloudMatrix 384 AI system at the World Artificial Intelligence Conference in Shanghai, marking a direct challenge to Nvidia’s most advanced platform, the GB200 NVL72. The system, built with 384 Ascend 910C chips, is seen by some analysts as capable of outperforming Nvidia on certain metrics. Huawei aims to dominate China’s AI sector despite facing U.S. export restrictions.
Huawei AI Breakthrough + Nvidia Rivalry Quick Looks
- Huawei unveils CloudMatrix 384, a high-performance AI computing system
- Public debut at WAIC 2025 in Shanghai, drawing major attention
- Designed to compete with Nvidia’s GB200 NVL72, its most powerful system
- Uses 384 Ascend 910C chips, compared to Nvidia’s 72 B200 chips
- Outperforms Nvidia on select metrics, according to SemiAnalysis
- Employs “supernode” architecture for ultra-fast chip interconnectivity
- Operational on Huawei Cloud, confirmed by Huawei Cloud CEO in June
- Huawei positioned as China’s top domestic AI chip supplier
- Nvidia CEO Jensen Huang acknowledged Huawei’s rapid AI advancement
- Comes amid continued U.S. export controls on Huawei
Deep Look
Huawei Unleashes CloudMatrix 384, Aimed at Nvidia’s AI Dominance
SHANGHAI, July 26 – In a bold move to reshape the competitive AI computing landscape, Huawei Technologies has publicly launched its CloudMatrix 384 AI computing system at the World Artificial Intelligence Conference (WAIC) in Shanghai. The powerful new platform is being hailed by experts as a direct rival to Nvidia’s GB200 NVL72, the most advanced AI system currently available from the U.S.-based tech giant.
Huawei’s system generated significant buzz at the event, with throngs of attendees crowding around the company’s exhibition booth to get a glimpse of what could be China’s most formidable answer to U.S. chip technology.
AI Arms Race: Huawei vs. Nvidia
Announced earlier in April, the CloudMatrix 384 utilizes 384 of Huawei’s new Ascend 910C chips. By comparison, Nvidia’s GB200 NVL72 harnesses just 72 B200 chips. While Nvidia’s chips are generally more powerful on an individual basis, Huawei compensates through large-scale parallel computing and innovative system design, according to leading semiconductor analysis firm SemiAnalysis.
“Huawei now has AI system capabilities that could beat Nvidia,” said Dylan Patel, founder of SemiAnalysis.
This claim, though not confirmed by Huawei directly, underscores a significant shift in the AI computing market, particularly in China, where Huawei has emerged as the go-to domestic supplier amidst continued U.S. trade restrictions.
A System Built for Scale and Speed
The CloudMatrix 384’s supernode architecture allows the 384 Ascend 910C chips to interconnect at exceptionally high speeds, providing seamless performance across massive datasets and AI models. This architecture enables system-wide efficiency and computational throughput that could match or even exceed Nvidia’s offerings in certain applications, SemiAnalysis reports.
Huawei staff at the WAIC declined to elaborate on the specs, but Huawei Cloud CEO Zhang Pingan confirmed in June that the system is already operational on the company’s cloud platform.
U.S. Sanctions Can’t Stop Huawei’s AI Push
Despite being blacklisted and facing strict export controls from the United States, Huawei has managed to develop and deploy advanced AI hardware, a feat that has caught the attention of competitors. Nvidia CEO Jensen Huang, in a Bloomberg interview this past May, acknowledged Huawei’s rapid progress, calling the CloudMatrix system an impressive development.
Huawei’s presence at WAIC and its technological advancements illustrate the company’s resilience in the face of geopolitical pressure, and its strategic focus on AI infrastructure, especially as China continues to push for self-sufficiency in semiconductors and critical technologies.
You must Register or Login to post a comment.