At HUAWEI CONNECT on September 17, 2026, Huawei unveiled a new computing architecture called Peerium, designed to make one million processors operate as a single computer. The announcement, detailed in a roundtable with Rotating Chairman Eric Xu and HiSilicon Chief Scientist Dr. Liao Heng, marks a significant shift in how AI infrastructure can be scaled. The architecture underpins the newly announced Atlas 950 SuperPoD, which is built on the Ascend 950 chip. Unlike conventional supernodes that connect thousands of processors, Peerium achieves strong scaling to the million-processor level through nested parallelism, unified memory addressing, and peer-to-peer interconnect. This extends the parallelism paradigm of Turing with Nested Bulk Synchronous Parallel (Nested BSP) and overruns the von Neumann single-machine architecture, challenging the master-slave design paradigm that has prevailed for decades.
The key enabler is UnifiedBus, an open-protocol high-speed bus that connects CPUs, NPUs, memory, SSDs, interface cards, and switches without a scaling limit. By achieving peer-to-peer interconnect across compute, storage, and networking, UnifiedBus fundamentally overthrows the traditional master-slave model. A SuperCluster with 256,000 computing cards is currently being deployed and tested. Dr. Liao explained that this scale is not arbitrary: it supports training foundation models with up to 40 trillion parameters, which several frontier AI labs in China are targeting over the next two years. Additionally, China's national power grid planning makes 200,000 cards a conservative estimate for a single data center campus.
Eric Xu noted that the Ascend 950PR is already on the market for AI inference, while the Ascend 950DT for training is under testing, with large-scale supply expected by late 2026 or early 2027. He stated that testing results for training are promising and that many AI model training workloads will likely rely on SuperPoDs using the 950DT. However, supply capacity remains a bottleneck. Xu emphasized that Huawei does not plan a full international expansion due to insufficient capacity to meet domestic demand, though limited testing and supply are ongoing in several countries with strong demand. He also claimed that Ascend has surpassed NVIDIA in market share in China, based on available data.
On the global AI race, Xu differentiated China's open-source model ecosystem from the more opaque, compute-rich US landscape. He suggested that Chinese model providers need to accelerate to reach a level where they feel similar risks from AI development, while stressing the need to balance development and risk management, ensuring AI for good. Regarding chip self-sufficiency, Xu called it an inevitable path for China, emphasizing that even if chips are less advanced, assured supply is crucial for a country of 1.4 billion people and an industrial powerhouse.
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 announcement was distributed via Media OutReach Newswire.
