Huawei Begins Shipping CloudMatrix 384 AI Clusters to Chinese Clients; Priced Triple Compared to NVIDIA’s GB200 NVL72 System

Huawei has made significant strides with its CloudMatrix 384 AI cluster, gaining traction among leading Chinese technology firms. These companies are expressing enthusiasm for the capabilities of this new server, particularly in comparison to NVIDIA’s solutions.

Huawei’s CloudMatrix 384 AI Cluster: A Game-Changer in China’s Tech Landscape

The tech giant Huawei is not holding back on its ambition to enhance its AI computing portfolio, especially given NVIDIA’s shifting influence within the Chinese market. Reports suggest that Huawei’s CloudMatrix 384 AI cluster has been developed solely using in-house resources, positioning it to compete robustly against NVIDIA’s offerings. Recently, the Financial Times revealed that ten major clients have begun utilizing the CloudMatrix 384 AI server, marking a pivotal advancement in Huawei’s aspirations within the AI sector.

Though the identities of these clients remain under wraps, they are recognized as key players in Huawei’s ecosystem. The CloudMatrix 384 has been highlighted as a viable competitor to NVIDIA’s flagship AI server, the GB200 NVL72, indicating a shift in China’s self-reliance when it comes to advanced computational resources.

In terms of specifications, the CloudMatrix 384 (CM384) AI cluster is engineered with 384 Ascend 910C chips arranged in an “all-to-all topology”configuration. Remarkably, Huawei’s design integrates five times more Ascend chips than NVIDIA’s GB200. This configuration enables the CloudMatrix to achieve an impressive 300 PetaFLOPS of BF16 computing performance—nearly double that of the GB200 NVL72. Nevertheless, a significant consideration is the power consumption; the CloudMatrix 384 is anticipated to use 3.9 times more power than the GB200 NVL72, resulting in considerably lower performance per watt across various AI workloads.

On the pricing front, the CloudMatrix 384 AI cluster commands a price tag of around $8 million, substantially exceeding the cost of NVIDIA’s GB200 NVL72 configuration—nearly three times as much. This pricing strategy suggests that Huawei is not primarily focused on offering a lower-cost alternative but rather on delivering a high-performance product developed entirely from its own resources, thereby directly competing with Western technological dominance.

Source & Images