Huawei Unveils Powerful Ascend 960 SuperPoD Computing Cluster to Challenge Nvidia in AI Market

Huawei has officially launched its cutting-edge Ascend 960 SuperPoD computing cluster during the 2026 Connect conference, marking a major milestone in high-performance artificial intelligence infrastructure. Designed to compete directly with leading global hardware like Nvidia products, this massive computing cluster introduces advanced hardware solutions and next-generation interconnect technologies aimed at reshaping the future of independent AI infrastructure and data centers worldwide.

• ✨ Huawei unveiled the powerful Ascend 960 SuperPoD computing cluster at the 2026 Connect conference.
• ✨ The system features advanced Near-Packaged Optics (NPOs) and upgraded UnifiedBus interconnect technology.
• ✨ Atlas 960 SuperPoD integrates 15,488 Ascend chips across 220 cabinets to significantly accelerate AI workloads.
• ✨ Huawei is driving forward an independent computing infrastructure strategy to reduce reliance on foreign technology.

Next-Generation Technologies Powering the Ascend 960 SuperPoD

During the major industry event, Huawei showcased the ultra-large-scale computing cluster alongside Near-Packaged Optics (NPOs) and an upgraded UnifiedBus interconnect technology. These innovations serve as high-performance alternatives to standard industry solutions, offering developers advanced tools to build massive AI systems even in challenging hardware environments. You can check out more details about Huawei Ascend developments to understand the broader roadmap.

The UnifiedBus Interconnect technology and Hi-ONE optical system tools allow a single SuperPoD to seamlessly connect with 4,000 processors simultaneously. This architecture expands high-speed optical connections efficiently throughout the entire computing system, highlighting Huawei's persistent commitment to creating an independent computing power infrastructure that bypasses traditional tech dominance.

Leadership Vision and Massive Scale Improvements

David Wang Tao, the Rotating and Acting Chairman at Huawei, emphasized the strategic importance of these installations, stating that supernodes represent an inevitable choice for constructing ultra-large-scale data centers. Furthermore, Huawei announced that it is ready to initiate mass production of the world's first NPO product equipped with an integrated light source.

Technical papers published by the company indicate ongoing research into a 256K-node-class single AI computing framework. While previous models like the Atlas 950 featured 8,192 Ascend NPUs, the new Atlas 960 SuperPoD scales up dramatically with 15,488 Ascend chips spanning 220 cabinets over 2,200 square meters. This substantial hardware upgrade is projected to boost training and performance efficiency significantly compared to previous generations.

What is the Huawei Ascend 960 SuperPoD?

The Ascend 960 SuperPoD is an ultra-large-scale AI computing cluster designed to operate as a single massive logical machine using advanced optical and bus interconnect technologies.

How many Ascend chips does the Atlas 960 SuperPoD use?

The Atlas 960 SuperPoD runs on 15,488 Ascend chips, housed across 220 cabinets covering an area of 2,200 square meters.

What is the purpose of UnifiedBus technology?

UnifiedBus is a key interconnect technology that enables thousands of processors to link together seamlessly, allowing massive data centers to function as one unified computing system.

How does the new SuperPoD compare to older models?

The new Atlas 960 SuperPoD delivers roughly three to four times the training and inference performance improvement over previous-generation SuperPoD technology solutions.

🔎 Ultimately, Huawei's aggressive push into large-scale AI clusters with the Ascend 960 SuperPoD signals a transformative shift in the global technology landscape. By combining innovative optics, scalable interconnect frameworks, and massive chip counts, Huawei continues to solidify its position as a dominant force capable of challenging established industry standards and driving independent technological advancement.