Report: Huawei Working on Million-GPU Training Cluster That Skips HBM

teortaxesTex · x · 2026-11-19

According to hints from Zartbot on WeChat (relayed via translation, unverified), Huawei is developing a million-GPU-scale training cluster system that does not rely on HBM, with some progress reportedly made.

If true, this would be a major departure from the prevailing assumption that HBM bandwidth gates large-scale training — especially significant amid HBM supply constraints. Treat as rumor; no official confirmation.

Related event: Rumor: Huawei Secretly Building Million-GPU Training Cluster Without HBM(2 posts)→

Original post →

More from Infra

Infra channel →