Huawei Links Thousands of AI Chips Into Computing Clusters, Challenging NVIDIA's Dominance

South Korean newspaper Chosun Ilbo reported on July 28 that Huawei has succeeded in linking thousands of its self-developed Ascend AI chips into large-scale computing clusters, directly challenging NVIDIA's dominance in the AI training hardware market.

According to the report, Huawei has achieved efficient parallel computing across thousands of chips through its Ascend series interconnect technology. This cluster-scale deployment capability represents a significant milestone in China's pursuit of self-sufficient AI infrastructure.

Since U.S. export controls blocked Huawei from accessing NVIDIA's most advanced AI chips — including the H100 and B200 series — the company has accelerated development of its Ascend chip line and pushed toward large-scale deployment. Interconnecting thousands of chips at scale is a formidable engineering challenge, requiring high-bandwidth, low-latency communication, balanced load distribution, and system-level stability.

Huawei has previously demonstrated Ascend cluster capabilities at various industry events. The Chosun Ilbo report provides independent confirmation that Huawei has moved beyond single-chip performance benchmarks and reached the deployment stage for production-scale clusters.

For the global AI chip market, this development intensifies the competitive landscape. While NVIDIA still holds advantages in raw performance and its CUDA ecosystem on premium training hardware, Huawei's cluster solution offers Chinese AI companies an alternative path that bypasses export restrictions.

Key questions remain: independent third-party benchmarks of Ascend cluster training throughput and power efficiency are not yet available. The industry will be watching for (1) how well mainstream AI frameworks integrate with Ascend clusters, (2) the linear scaling efficiency at thousands-of-chips scale, and (3) adoption pace by Chinese cloud providers and AI enterprises.