Huawei Unveils Next-Gen Ascend Chips to Fuel the World’s Most Powerful Computing Clusters
Chinese tech giant Huawei made a significant announcement regarding its Ascend chip lineup during the Huawei Connect 2025 event held in Shanghai. Eric Xu, the deputy chair of the Huawei board, emphasized that 2025 has been an important year for the company, marking the launch of the DeepSeek-R1 chip in January as a pivotal moment for its future. However, he acknowledged the challenges faced by China in semiconductor manufacturing, particularly when compared to global standards.
In response to ongoing trade restrictions, Huawei is advancing its technology infrastructure and has decided to open-source various software components, including the openPangu foundation AI models and the Mind series SDKs.
New Ascend Chips
Huawei revealed plans to produce three new series of the Ascend chip family: the 950, 960, and 970. The Ascend 950PR and 950TO chips will be manufactured from the same die and will support low-precision data formats like FP8, with the 950 offering performance of one PFLOP (one quadrillion floating-point operations) and the MXFP8 reaching two PFLOPs. Notably, the 950 chips are set to provide 2 TB/s interconnect bandwidth, representing a 2.5 times increase over the previous Ascend 910C version, with the 950PR slated for release in Q1 2026.
The Ascend 960 is scheduled for a Q4 2027 release and is designed to offer double the computing power, memory capacity, and bandwidth of the 950. The most advanced of the new series, the Ascend 970, is expected to launch in Q4 2028, aiming for an astounding interconnect bandwidth of 4 TB/s and up to 8 PFLOPs.
SuperPods of NPUs
Huawei’s strategy includes deploying SuperPoDs, which are clusters of compute resources, beginning with the Atlas 950 SuperPoD powered by the new Ascend 950DT chips, available in Q4 2026. This SuperPoD is anticipated to dramatically outperform competitor NVIDIA’s NVL144 system, boasting over 56 times more NPUs than GPUs while delivering nearly seven times the processing capacity.
General Computing Chips
For broader computing applications, Huawei plans to introduce two variants of the Kunpeng 950 processors in Q1 2026, featuring exceptional core counts and threads to meet diverse computing needs. Additionally, the first general-purpose computing SuperPoD, based on the Kunpeng 950, is also set to launch in early 2026.
Open-source Connectivity Protocol
The new NPU and general computing SuperPoDs will utilize the UnifiedBus 2.0 protocol, improving interconnectivity among SuperPods and enabling the formation of SuperClusters. The initial result of this effort will be the Atlas 950 SuperCluster, designed to outperform existing solutions, including xAI’s Colossus, the leading computing cluster at present. Huawei intends to further enhance their offerings with the Atlas 960 SuperCluster, featuring over a million NPUs capable of delivering 4 ZFLOPS in FP4 by 2027.
Eric Xu concluded that these advancements demonstrate Huawei’s commitment to meeting the increasing demand for robust computing solutions now and in the future.
Discover the pinnacle of WordPress auto blogging technology with AutomationTools.AI. Harnessing the power of cutting-edge AI algorithms, AutomationTools.AI emerges as the foremost solution for effortlessly curating content from RSS feeds directly to your WordPress platform. Say goodbye to manual content curation and hello to seamless automation, as this innovative tool streamlines the process, saving you time and effort. Stay ahead of the curve in content management and elevate your WordPress website with AutomationTools.AI—the ultimate choice for efficient, dynamic, and hassle-free auto blogging. Learn More
