Huawei Launches AI CloudMatrix 384 with 2.6x Lower FLOP per Watt to Compete with Nvidia's GB200 NVL72

7 posts • GPT (4o mini)

Published Apr 16, 2025, 12:53 PM

Huawei has introduced its AI CloudMatrix 384 rack-scale system, which is designed to compete with Nvidia's GB200 NVL72. While the CloudMatrix 384 is noted to be less power-efficient, with a performance metric of 2.6 times lower FLOP per Watt, this is not seen as a limiting factor in China due to the country's abundant power resources. The system utilizes Huawei's Ascend 910C AI accelerator and is positioned as a powerful domestic solution in China's tech landscape. As production ramps up for the Ascend 910C, the competitive landscape is shifting, with Nvidia's H100 and H20 systems currently dominating existing clusters. Analysts suggest that while Huawei's new offering is impressive, Nvidia's products will likely maintain a strong presence in the short to medium term. The introduction of the CloudMatrix 384 is part of a broader trend in the ongoing tech war between the U.S. and China, with Huawei's advancements prompting discussions on the implications for global semiconductor supply chains and technology competition.

More breaking stories on DeepNewz — updated live.

#Huawei #CloudMatrix 384 #Nvidia #CloudMatrix #China #Ascend 910C #H100 #H20

Related Prediction Markets

Mexico tariffs on China in effect before July?

Canada tariffs on China in effect before July?

•

View market suggestions→

Sources

semianalysis.com(4)digitimes.com techmeme.com scmp.com

Ray Wang

@rwang07

SemiAnalysis: Huawei is making waves with its new AI accelerator and rack scale architecture. Meet China’s newest and most powerful Chinese domestic solution, the CloudMatrix 384 built using the Ascend 910C. This solution competes directly with the GB200 NVL72, and in some https://t.co/OVtoOnlIj4

Paul Triolo

@pstAsiatech

Huawei debuts 'nuclear-grade' AI supernode, but power-hungry design puts cooling to test https://t.co/fkEsy9LnOm

Ray Wang

@rwang07

With Huawei’s 910C—competitive in both compute and memory—now ramping production, the key question is how quickly the market space left by Nvidia will be filled. In the short to medium term, Nvidia’s H100 and H20—already widely deployed in existing clusters—will likely remain in https://t.co/0nTuSaEIVZ