Huawei has introduced the CloudMatrix 384 Supernode, a computing system positioning itself as a significant player in the AI chip market. Early reports indicate its performance rivals that of Nvidia’s offerings, potentially altering the competitive landscape amid continuous AI advancements and ongoing US sanctions against China.
Described as a powerful and innovative product, the CloudMatrix 384 Supernode is reported to deliver 300 petaflops of computing power, surpassing Nvidia’s NVL72 system’s 180 petaflops. This system is designed to overcome the computational bottlenecks associated with complex AI models, aiming to directly challenge Nvidia’s current market dominance.
The CloudMatrix infrastructure, initially unveiled in September 2024, was developed to address China’s growing domestic needs. The 384 Supernode configuration achieves a throughput of 1,920 tokens per second, reportedly matching the performance of Nvidia’s H100 chip using components manufactured within China.
This AI hardware achievement is particularly significant considering the restrictions imposed on Huawei as a US Entity List member. These sanctions have limited its access to US semiconductor technologies and design tools, pushing Huawei to innovate alternative solutions and rely on its domestic supply chains.
Critical to the CloudMatrix 384’s performance is Huawei’s high-speed interconnect technology, a direct alternative to Nvidia’s NVLink. According to the SCMP, Huawei is collaborating with SiliconFlow to implement the CloudMatrix 384 Supernode to support DeepSeek-R1.
Supernodes, like the CloudMatrix 384, offer expanded resources, including CPUs, NPUs, bandwidth, storage, and memory, to enhance cluster computing performance and accelerate the training of AI models.
Huawei’s initiatives are part of a broader effort by Chinese tech companies to establish a robust domestic AI computing infrastructure. For example, Alibaba Group has announced a substantial investment of ¥380 billion in computing resources and AI infrastructure over the next three years.
The emergence of alternatives to Nvidia’s hardware could address current limitations in AI development by increasing available computing capacity and providing developers with more diverse options. As of now, Huawei has not officially responded to requests for comments on these developments.
Against the backdrop of intensifying US-China tensions in the tech sector, Huawei’s CloudMatrix 384 Supernode represents a step toward greater technological self-reliance. If its performance claims are confirmed, it marks a milestone for Huawei in achieving computing independence despite ongoing sanctions, and highlights a larger trend of investment in domestic AI infrastructure.
Photo by Dzenina Lukac on Pexels
Photos provided by Pexels