Particle.news

Download on the App Store

Huawei Debuts CloudMatrix 384 Supernode System at World AI Conference to Challenge Nvidia

Engineered with high-speed interconnects across 384 Ascend 910C chips, the system goes live on Huawei Cloud as Chinese AI firms launch model-chip alliances

Image
A person visits the World Artificial Intelligence Conference in Shanghai, China July 26, 2025.  REUTERS/Go Nakamura/ File Photo
Image

Overview

  • Huawei unveiled the CloudMatrix 384 at the World Artificial Intelligence Conference in Shanghai, marking its first public showcase of the clustered AI platform.
  • The system packs 384 self-developed Ascend 910C chips into a proprietary supernode architecture designed to accelerate inter-chip communication.
  • U.S. research firm SemiAnalysis reports that CloudMatrix 384 outperforms Nvidia’s GB200 NVL72 on select benchmarks by leveraging large-scale chip clustering and system-level innovations.
  • Huawei Cloud CEO Zhang Pingan confirmed in June that the platform is already running on Huawei’s cloud infrastructure, enabling immediate enterprise access.
  • China’s AI firms formed the Model-Chip Ecosystem Innovation Alliance and the Shanghai General Chamber of Commerce AI Committee in response to U.S. export restrictions, aiming to build a fully integrated domestic chip-to-model ecosystem.