Huawei Unveils World’s First NPO-Based AI SuperPoD With 4,096 Processors
Huawei has introduced the Atlas 960E SuperPoD, a new large-scale AI computing system designed for training and inference workloads involving … Read More The post Huawei Unveils World’s First NPO-Based AI SuperPoD With 4,096 Processors appeared first on ProPakistani .
Huawei has unveiled the Atlas 960E SuperPoD, a groundbreaking large-scale AI computing system. Introducing on September 17 at HUAWEI CONNECT 2026 in Shanghai, the Atlas 960E is the industry's first SuperPoD designed around Near-Packaged Optics, or NPO. The system is tailored for training and inference workloads involving increasingly large AI models, potentially approaching 10 trillion parameters.
The Atlas 960E integrates Huawei's UnifiedBus technology, also known as Lingqu, with the company's Hi-ONE optical system. Engineered for smooth operation, the architecture employs unified memory addressing, allowing computing resources across the system to function like a single computing environment instead of separate machines. A single Atlas 960E SuperPoD can accommodate up to 4,096 Ascend AI processors, enabling delivery of up to 8 EFLOPS of FP8 performance and 16 EFLOPS of FP4 performance.
Huawei's Hi-ONE NPO optical engine is a key component of the system, supporting up to 7.2 terabits per second of transmission capacity and featuring an integrated light source. The Atlas 960E employs around 5,500 Hi-ONE units, replacing approximately 48,000 conventional 800G optical modules. This innovative design brings significant power savings, cutting power consumption by over 550kW. Furthermore, the system claims to double fault-free operating time and achieve a 99.8% system availability.
Huawei has broader plans to extend the SuperPoD architecture beyond AI accelerators, with an upgraded TaiShan 950 SuperPoD supporting up to 4,096 nodes and providing a 256TB unified memory pool. Additionally, the OceanStor M900 offers petabyte-scale key-value caching for AI workloads. The company also intends to connect multiple Atlas 960E SuperPoDs into larger AI clusters using UnifiedBus and RoCE networking, with potential to connect up to 512,000 AI processors.
The Atlas 960E SuperPoD is slated for launch in Q3 2027 and will be complemented by new generations of Ascend AI processors. The Ascend 960 DT is planned for release in Q1 2027, followed by the Ascend 960 PR in Q3 2027. Huawei then aims to introduce the Ascend 970 in 2028, and finally, the Ascend 980 in 2029.
Written by urgent.news from ProPakistani's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
