At HUAWEI CONNECT on September 17, 2026, Huawei unveiled its Peerium Computing Architecture, a new computing paradigm designed for the AI era that can make one million processors work as one computer. The announcement, made during a Q&A session with Eric Xu, Huawei's Rotating Chairman, and Dr. Liao Heng, Chief Scientist of HiSilicon, signals a major shift in how large-scale AI infrastructure is built and deployed. The architecture achieves strong scaling to the million-processor level through nested parallelism, unified memory addressing, and peer-to-peer interconnect, extending the parallelism paradigm of Turing with Nested Bulk Synchronous Parallel (Nested BSP). It also overruns the master-slave design paradigm that has prevailed for decades, according to the company.
The implications are significant. As AI models grow to trillions of parameters, the ability to train them efficiently requires massive computing clusters. Huawei's Peerium architecture, paired with its UnifiedBus (UB) interconnect, allows one million processors to function as a single larger computer, fundamentally changing the economics and scalability of AI training. The Atlas 950 SuperPoD, a product built upon this architecture, is already being deployed and tested with 256,000 computing cards. Dr. Liao noted that such scale is necessary to train foundation models in the 10 to 40 trillion parameter range, which several Chinese frontier AI labs are targeting over the next two years.
Huawei's progress also carries geopolitical weight. Eric Xu stated that Ascend has surpassed NVIDIA in market share in China, based on the data collected. He emphasized that China's push for chip self-sufficiency is an inevitable path forward, driven by a sense of urgency to avoid dependence on foreign supply. While Huawei does not have plans to expand the Atlas 950 SuperPoD internationally in a fully-fledged way due to limited capacity, it is conducting tests and providing limited supply to several countries with strong demand. This suggests that Huawei's advancements could influence global AI infrastructure competition, particularly as the US and China diverge on AI development and risk management.
Xu also addressed the recent calls by leading US AI model providers to slow down AI development, noting that the market feeling about AI risks is weaker in China. He believes that Chinese model providers need to speed up to reach a level where they can also feel those risks, but stressed the need to balance AI development with risk management, ensuring AI for good, not for evil. The full Q&A session is available here.
With the Ascend 950DT SuperPoDs entering large-scale supply at the end of this year or early next year, Huawei's Peerium architecture could accelerate AI model training in China and beyond, while reducing reliance on foreign chip technology. The development underscores a broader trend of technological decoupling and the pursuit of sovereign AI capabilities. As reported by Media OutReach Newswire, this announcement positions Huawei at the forefront of a new computing era, where the ability to scale to a million processors as one computer may become a key differentiator in the global AI race.
