Huawei has introduced a new computing architecture designed to make one million processors operate as a single computer, a move that could reshape the global AI infrastructure race and reduce reliance on Western chip technology. The announcement came during HUAWEI CONNECT on September 17, 2026, where Eric Xu, Huawei's Rotating Chairman, and Dr. Liao Heng, Chief Scientist of HiSilicon, detailed the Peerium Computing Architecture and the UnifiedBus (UB) interconnect to international media.
The Peerium architecture achieves strong scaling to the million-processor level through nested parallelism, unified memory addressing, and peer-to-peer interconnect. It extends the parallelism paradigm of Turing with Nested Bulk Synchronous Parallel (Nested BSP) and overruns the von Neumann single-machine architecture and the master-slave design that has prevailed for decades. Because available industry technologies could not support this scale, Huawei invented UnifiedBus, a high-speed bus using an open protocol that connects CPUs, NPUs, memory, SSDs, interface cards, and switches without limit. This enables peer-to-peer interconnect across compute, storage, and networking.
The first product built on Peerium is the Atlas 950 SuperPoD, based on the Ascend 950 chip. A SuperCluster with 256,000 computing cards is currently being deployed and tested. Dr. Liao explained that 256,000 is not arbitrary: it supports training foundation models already in the 5-trillion-parameter range, and over the next two years, six or seven frontier AI labs in China aim to train models of 10 to 40 trillion parameters. He added that 200,000 cards is a conservative number given China's national power grid plans for data centers.
Eric Xu said the Ascend 950PR is already available for AI inference, while SuperPoDs using the Ascend 950DT for training are under testing, with large-scale supply expected by the end of 2026 or early 2027. He noted that testing results for training are promising and that supply capacity, not persuasion, will determine adoption. Xu also stated that Ascend has surpassed NVIDIA in China based on Huawei's data, though he acknowledged that collecting market share data is difficult.
On international expansion, Xu said Huawei lacks capacity to fully serve markets outside China, though several countries have strong demand and are receiving limited supply. He called China's push for chip self-sufficiency an inevitable path forward, emphasizing that assured supply matters even if chips are less advanced. The full Q&A session is available at https://www.apmultimedianewsroom.com/multimedia-newsroom/huawei-pioneers-a-new-computing-architecture-for-the-ai-era-making-one-million-processors-work-as-one-computer-2.
The development matters because it challenges the dominant master-slave computing paradigm and could accelerate AI model training at unprecedented scales. For the industry, it signals a potential shift in AI hardware supply chains and competition, particularly as China pursues semiconductor independence. For readers, it means the AI era may see more powerful, domestically produced computing infrastructure that could influence global technology standards and access.
