Huawei unveils Ascend 960 SuperPoD computing cluster, targeting Nvidia

Huawei today launched the Ascend 960 SuperPoD computing cluster at the 2026 Connect conference. It also introduced Near-Packaged Optics (NPOs) and an upgraded UnifiedBus interconnect technology as AI tech alternatives to Nvidia products.

Ascend 960 SuperPoD is an ultra-large-scale AI computing cluster in the Huawei world that will work as a single massive logical machine using new-gen tech tools.

UnifiedBus is a key component often used to develop advanced AI systems in the absence of cutting-edge semiconductors. Meanwhile, NPO is a placement structure enabling higher speeds and reducing energy consumption.

Both UnifiedBus Interconnect technology and Hi-ONE optical system tools allow a SuperPoD to connect with 4000 processors simultaneously. Thus, it expands high-speed optical connections more efficiently into the computing system.

These technological tools show Huawei’s constant efforts for an independent computing power infrastructure, breaking the US tech dominance on top-grade chips.

David Wang Tao, the Rotating and Acting Chairman at Huawei, said:

“Supernodes are an inevitable choice for building ultra-large-scale data centers. Huawei is set to begin mass production of the world’s first NPO product with an integrated light source.”

Huawei further posted a paper to arXiv, saying that it is working on a 256K-node-class single AI computing framework to extend its AI architecture far beyond an individual SuperPoD. It described the system as one machine instead of a network of over 200,000 separate machines, built over UnifiedBus 2.0.

Coming to the SuperPoD, Atlas 950 has 8192 Ascend NPUs, while the Atlas 960 SuperPoD runs on 15488 Ascend chips, covering 220 cabinets and 2200 square meters.

Atlas 960 SuperPoD will likely accelerate training and inference performance by three to four times over the previous-generation SuperPoD tech solution.

Huawei has also shared Ascend 960 chip release plans, saying that the OEM is running ahead of schedule and will continue to follow an upgrade cycle for AI chipsets, guided by the Tau Scaling Law that can take performance to the next level.

(Image Credits: Huawei Blog)