In a bold move that could reshape the global AI hardware race, Chinese tech giant Huawei has unveiled a powerful new computing system that reportedly surpasses Nvidia’s top AI chip infrastructure in performance. The system, dubbed the CloudMatrix 384 Supernode, is being hailed as a major technological milestone—especially considering the export restrictions Huawei continues to face.
Table of Contents
A Chinese Contender to Nvidia’s Reign
According to Chinese media reports cited by the South China Morning Post, the CloudMatrix 384 Supernode delivers 300 petaflops of computing power—outperforming the 180 petaflops achieved by Nvidia’s NVL72, which debuted in March 2024. Local outlets have described Huawei’s hardware as a “nuclear-level product,” a nod to its potentially game-changing capabilities.
The CloudMatrix system aims to directly challenge Nvidia’s longstanding dominance in AI acceleration hardware. Huawei first introduced the CloudMatrix framework in September 2024, targeting China’s rapidly growing demand for domestic AI infrastructure. The 384 Supernode marks the most advanced version of this system yet, offering breakthrough performance for training and deploying massive AI models.
Reports suggest the Supernode achieves a throughput of 1,920 tokens per second and matches the performance of Nvidia’s flagship H100 chips—all while using components developed within China. This development could represent a crucial step toward achieving technological independence in AI computing.
Sanctioned but Undeterred
What makes Huawei’s latest leap even more remarkable is that it was achieved under intense technological sanctions. Since being added to the U.S. Entity List, Huawei has been cut off from cutting-edge American chipmaking technology and software tools, forcing the company to innovate domestically.
Central to this breakthrough is Huawei’s version of Nvidia’s NVLink, a high-speed interconnect architecture that enables multiple GPUs to function as a unified processing system. Nvidia’s NVL72 includes a 72-GPU array linked through NVLink to deliver ultra-fast inference for models with over a trillion parameters. Huawei’s new Supernode is said to offer similar scalability and performance—without relying on U.S. technologies.
The CloudMatrix 384 is being deployed in collaboration with SiliconFlow, a Chinese AI infrastructure startup, to support DeepSeek-R1, a reasoning-focused AI model developed by DeepSeek in Hangzhou. These Supernodes are equipped with higher-level resources than conventional systems—such as upgraded CPUs, NPUs (neural processing units), network capacity, and storage—which significantly accelerate the training of complex foundational models.
Part of a National AI Push
Huawei’s achievement is not an isolated event. It forms part of a broader national strategy in China to boost AI infrastructure and reduce reliance on foreign technologies. In February, Alibaba Group announced a record-breaking investment of 380 billion yuan (approx. $52.4 billion) over three years to expand its own AI and computing infrastructure—a move that underscores the country’s commitment to building a resilient, self-sufficient tech ecosystem.
The rise of Huawei’s AI hardware could also serve a global purpose by increasing competition in the AI chip market. For years, Nvidia’s dominance has meant limited options and high costs for developers and enterprises in need of powerful AI computing tools. The emergence of high-performance alternatives from China might not only diversify supply chains but also ease some of the computational bottlenecks currently slowing AI innovation worldwide.
A Step Toward Self-Sufficiency
Despite skepticism and a lack of official comment from Huawei, the CloudMatrix 384 Supernode signals a clear intent: China aims to lead in AI infrastructure using its own tech. If benchmark claims hold true, Huawei has proven that domestic innovation can thrive under restrictions—and even challenge global leaders like Nvidia.
As geopolitical tensions over technology persist, Huawei’s hardware launch could mark a turning point in the AI arms race. It also reflects a broader industry trend toward building localized, high-performance AI solutions that can compete on a global scale.
(With AI inputs)