Huawei, the Shenzhen-based telecom and computing company, has launched the industry’s first near-packaged optics (NPO)-based AI computing system, alongside the upgraded TaiShan 950 SuperPoD and the OceanStor M900 context memory storage system.

Huawei said a single Atlas 960E SuperPoD can scale to 4,096 neural processing units (NPUs). It also said its agentic SuperCluster, interconnected with UnifiedBus, can scale up to 1 million NPUs.

The company added that its Compute Architecture for Neural Networks (CANN) software stack has moved to sustained, community-driven open-source development, and that Ascend now spans more than 90 third-party open-source projects.

The launch comes as the US export controls target China’s access to advanced semiconductors, and the US authorities have issued guidance on the use of Huawei chips for AI.

It also follows reports that China has banned foreign AI chips in state-funded data centers, a policy that could direct more domestic AI infrastructure demand to local suppliers.

🔗 Source: Huawei

Recent Huawei developments