Chinese tech companies have started using Huawei's CloudMatrix 384 AI cluster. Ten clients have already added the system to their data centers. These customers find that the server performs well compared to NVIDIA products. The names of these buyers remain secret, but they already use other Huawei products. This marks a major step forward for Huawei's AI plans.
The CM384 cluster contains 384 Ascend 910C chips in an all-to-all setup. Huawei packed five times more chips than NVIDIA uses in its GB200 server. The Chinese system delivers 300 PetaFLOPS of BF16 computing power, almost double what NVIDIA offers. It needs almost four times more electricity than the GB200 NVL72, making it less efficient. Each cluster costs about $8 million, three times more than NVIDIA's option.
Huawei created this system to free Chinese companies from depending on foreign technology. The company built every part itself instead of relying on Western parts. The high price shows Huawei cares more about independence than saving money. Chinese firms can run powerful AI without American chips. Local tech leaders seem happy with how the system runs their AI tasks.
The CM384 cluster contains 384 Ascend 910C chips in an all-to-all setup. Huawei packed five times more chips than NVIDIA uses in its GB200 server. The Chinese system delivers 300 PetaFLOPS of BF16 computing power, almost double what NVIDIA offers. It needs almost four times more electricity than the GB200 NVL72, making it less efficient. Each cluster costs about $8 million, three times more than NVIDIA's option.
Huawei created this system to free Chinese companies from depending on foreign technology. The company built every part itself instead of relying on Western parts. The high price shows Huawei cares more about independence than saving money. Chinese firms can run powerful AI without American chips. Local tech leaders seem happy with how the system runs their AI tasks.
