Huawei’s Atlas 960E and OceanStor M900: System-Level Engineering to Bypass U.S. Chip Bans

Huawei launched the Atlas 960E SuperPoD AI cluster, which packs 4,096 Ascend NPUs with proprietary optical interconnects, alongside the OceanStor M900 context memory storage built to offload large Key-Value caches, according to Pandaily. Huawei positions the two systems as a full-stack domestic alternative for high-performance LLM training and ultra-low-latency inference, engineering around Western hardware constraints through near-package optics and specialized KV cache storage. The releases indicate U.S. export controls are accelerating China's development of customized AI data center ecosystems.

Huawei’s Atlas 960E and OceanStor M900: System-Level Engineering to Bypass U.S. Chip Bans Huawei has launched the Atlas 960E SuperPoD AI cluster, featuring 4,096 Ascend NPUs and proprietary optical interconnects, alongside the OceanStor M900 context memory storage designed to offload massive Key-Value caches. Dick Weisinger · September 24, 2026 · 2 min read · Source: Pandaily · Issue 106 East Asian Technology Intelligence Japan & China tech news — translated, contextualized, and delivered for Western readers. Free. Unsubscribe anytime. This story ran in Issue 106, alongside three other stories. AI & Machine Learning Huawei has launched the Atlas 960E SuperPoD AI cluster, featuring 4,096 Ascend NPUs and proprietary optical interconnects, alongside the OceanStor M900 context memory storage designed to offload massive Key-Value caches. Together, these systems provide a full-stack, domestic alternative for both high-performance LLM training and ultra-low-latency inference. Collectively, they demonstrate Huawei’s ability to engineer around Western hardware constraints by optimizing optical networking and storage architecture. For Western observers, these releases show that Huawei is not merely replicating American hardware, but actively innovating in system-level architecture—such as near-package optics and specialized KV cache storage—to overcome chip-level manufacturing limitations. This proves that export controls are accelerating China’s development of highly customized, resilient AI data center ecosystems.