Huawei Ascends to SuperPods and SuperClusters

At the Huawei Connect 2025 event in Shanghai last week, Chinese technology Huawei explained its plans for the next generations of its Ascend chip line, and released details on its deployment in pods, and world-leading compute clusters.

Speaking at the Huawei Connect conference, deputy chair of Huawei board, Eric Xu, said 2025 had been a memorable year, and claimed the debut of DeepSeek-R1 in January was a turning point for Huawei. He acknowledged what industry commentators have previously confirmed; China will be likely to lag behind in semiconductor manufacturing processing, “for a relatively long time.”

In the face of tariffs, politically-motivated rhetoric, and trade embargoes, Huawei will remain ahead by advancing infrastructure design and associated technologies, producing powerful clusters connected by an open-sourced protocol. It’s also decided to open-source several large pieces of previously proprietary software, including the openPangu foundation AI models and its Mind SDKs.

The new Ascend chip range

The company plans to produces three new series of the Ascend chip, which it’s calling the 950, 960, and 970.

The Ascend 950PR and 950TO will provide additional support for low-precision data formats, including FP8: the 950 model will deliver a PFLOP of performance, and at MXFP8, deliver two PFLOPs. (A PFLOP is one thousand trillion floating point calculations per second.)

The company says the 950 will offer better vector processing, and be capable of granular memory access, processing 128 byte chunks, down from 512 bytes.

The Ascend 950 chips will offer 2 TB/s interconnect bandwidth, two-and-a-half more than the current Ascend 910C, which went into production for Huawei customers in May this year. The 950PR will be available in the first quarter of 2026, with the Ascend 950DT following Q4.

A year later in Q4 2027, the Ascend 960 will offer twice the computing power, memory access bandwidth, memory capacity, and number of interconnect ports as its stablemate, the 950. It will support Huawei’s proprietary HiF4 data format, which brings greater precision than other FP4 technologies, the company says.

The end of 2028 will see the emergence of Huawei’s most capable chip, the Ascend 970. Xu told the conference, “We’re still working on some of its specs, but our general goal is to push all of its specs much higher.” He said the Ascend 970 series will be able to communicate via interconnect bandwidth of 4TB/s, burn through 8 PFLOPs of FP4, and will come with an even larger memory capacity than previous iterations of the Ascend range.

(Eric Xu, Huawei

SuperPods of NPUs

Rather than concentrating on producing chips that are more powerful than its main competitor, Nvidia, Huawei’s strategy is to offer hyperscalers clusters of ready-made compute in the form of SuperPoDs. These will begin to appear Q4 2026 in the form of the Atlas 950 SuperPoD loaded with Ascend 950DT chips.

Competitor NVIDIA’s NVL144 system (the closest SuperPod competitor) is set to launch mid- to late-2026. Huawei claims that its first SuperPoD will deliver nearly seven times its processing power, and have 56.8 times more NPUs than the number of GPUs in the NVL144. Nvidia’s next-generation NVL576, set to be released in 2027, will still be out-performed by the Atlas 950 SuperPoD, Huawei says.

General computing (non-AI focused) chip range

For general computing, Huawei plans to release two models of its Kunpeng 950 processors at the beginning of 2026. They will comprise of 96 cores & 192 threads, or 192 cores & 384 threads in the faster model. At the same time, Huawei plans to release what Xu called, “the wold’s first general-purpose computing SuperPoD,” the Kunpeng 950-based TaiShan 950.

Open-source connectivity protocol, UnifiedBus 2.0

Both the NPU and ‘general computing’ SuperPoDs will use Huawei’s UnifiedBus 2.0, which is an update on the existing UnifiedBus 1.0. Version 1.0 of UnifiedBus is the interconnection technology used by the Atlas 900 A3 SuperPoD, which appeared in March this year, and already has more than 300 installations in DCs and institutions around the world.

UnifiedBus 2.0 will be licensed as open-source, and the technical specifications are to be released immediately to the developer community. UnifiedBus 2.0 will cut its teeth in the new generations of SuperPods, and be the connection protocol for clusters of SuperPods, inevitably named SuperClusters.

The first SuperCluster will be the Atlas 950 SuperCluster, which will offer 2.5 times more NPUs and 1.3 times more computing power than xAI’s Colossus cluster, currently the world’s most powerful.

Towards the end of 2027, Huawei will launch the Atlas 960 SuperCluster, which will contain over a million NPUs and deliver 4 ZFLOPS in FP4 (with a ZFLOP representing 10^{^}21 floating point operations per second) via a single virtualised compute instance. “SuperPoDs and SuperClusters powered by UnifiedBus are our answer to surging demand for computing, both today and tomorrow,” Xu said.

(Image source: “Coronal Mass Ejection on the Sun” by Blatant Views is licensed under CC BY 2.0.)

Want to learn more about Data Centres from industry leaders? Check out Data Centre Congress taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.

CloudTech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.