Huawei debuts CloudMatrix 384 AI system, touts performance edge over Nvidia

At the World Artificial Intelligence Conference (WAIC) in Shanghai, Huawei Technologies introduced its groundbreaking AI computing platform, the CloudMatrix 384, marking a significant milestone in the competition for AI infrastructure supremacy in China. The CloudMatrix 384 system, which integrates 384 of Huawei’s advanced 910C chips, has garnered considerable attention from industry experts and conference attendees. Analysts are already positioning it as a formidable rival to Nvidia’s cutting-edge GB200 NVL72 system, which utilizes 72 B200 chips. While Huawei representatives at the conference refrained from divulging detailed specifications, the company touts its innovative “supernode” architecture, which facilitates rapid interconnectivity between chips, thus enhancing overall system performance even if individual chip capabilities are comparatively lower. This launch comes at a time when China is intensifying efforts to decrease its dependency on foreign AI technology. With stringent U.S. export regulations impacting its operations, Huawei has emerged as a prominent domestic player in the AI chip and computing markets. Nvidia's CEO, Jensen Huang, acknowledged this competition earlier in the year, noting Huawei's swift advancements and specifically mentioning the CloudMatrix system. Further emphasizing its commitment to AI, Huawei Cloud CEO Zhang Pingan revealed in June that the CloudMatrix 384 is already operational within the company's cloud infrastructure. Analysts suggest that the system's advantages stem more from its architectural design and scalability than from sheer chip performance. Despite the excitement surrounding the CloudMatrix 384, Huawei maintained a level of secrecy during the event, with no live demonstrations or performance benchmarks presented, leaving many inquiries from journalists unanswered.