In addition to its Ascend AI chip lineup, Huawei has also unveiled its next-generation Kungpeng 950 & 960 chips for general-purpose computing.
Huawei Aims To Deliver 192 Cores With Its Kungpeng 950 Chips By Q4 2026, Over 256 Cores With Kungpeng 960 Series By Q1 2028
During the Huawei Connect 2025 event, held in China, Huawei unveiled its next-generation AI & HPC product portfolio. These include two main families, the Ascend AI accelerators and Kungpeng chips for general-purpose computing.
It looks like Huawei is going all-out with its next-gen Kungpeng series. The Kungpeng 950 series, which is slated for launch by Q4 2026, is said to offer 96 cores / 192 threads high-performance, and 192 cores / 384 threads high-density CPU variants. This marks a 50% increase for the high-performance variant and a massive 2.4x increase for the high-density variant. These chips will feature the latest dual-threaded Unix-Core architecture and will offer compatibility plus support on the next-gen Huawei SuperPods.
The following is also unveiled, labeled as the Huawei Kungpeng 960 series. These are slated for launch by Q1 2028, and will include 96 cores / 192 threads high-performance, and over 256 cores / 512 threads for the high-density chip variants. According to Huawei, the new architecture for the 960 series is expected to deliver a 50% performance improvement per core and is designed for AI hosts and databases. The higher-density chips will be optimized for virtualization, containers, and big data warehouses.
Once again, these chips will go hand in hand in powering Huawei's next-gen fleet of SuperPods and SuperPod Clusters. The Atlas 950 SuperPod is expected to feature 8192 accelerators, while the Atlas 960 SuperPod is expected to feature 15,488 accelerators. These Pods will be combined into massive Super-Clusters that would exceed 500,000 / 1,000,000 accelerators in total.
In terms of technical aspects, the Atlas 950 SuperPoD will offer:
- 8192 Ascend 950 Acclerators
- Up To 8 Exaflops of FP8 Compute
- Up To 1152 TB of Memory
- Up To 16.3 PB/s Bandwidth
- Total Training Throughput: 4.91m TPS
- Total Inference Throughput: 19.6m TPS
The Atlas 960 SuperPoD will offer:
- 15,488 Ascend 960/950DT Accelerators
- Up To 30 Exaflops of FP8 Compute
- Up To 60 Exaflops of FP4 Compute
- Up To 434 PB/s Bandwidth
Meanwhile, the TaiShan 950 SuperNode will feature a total of 16 nodes with 32 Kungpeng 950 series CPUs, and a maximum of 48 TB of memory.
Chinese authorities have been pushing hard to switch data centers and AI clusters to domestic hardware. This means that there's a bigger need for domestic companies and manufacturers such as Huawei to accelerate their AI and chip roadmap if they have to keep up with the huge demand. As such, the coming years will make for an interesting landscape in China where Huawei tries to give the local market a reason to select its chips over foreign solutions.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
