Huawei launches the Atlas 960E SuperPoD at HUAWEI CONNECT 2026, supporting AI models with up to 10 trillion parameters

Huawei unveiled the Atlas 960E SuperPoD to challenge the global AI infrastructure landscape at HUAWEI CONNECT 2026 in Shanghai, declaring it the industry's first SuperPoD built on NPO architecture, designed to support Foundation Models with up to 10 trillion parameters.

HUAWEI CONNECT 2026 officially opened in Shanghai, China, with David Wang, Huawei's Deputy Chairman of the Board and Rotating Chairman, delivering a keynote titled "Advancing the Agentic World, Building a Solid Silicon Foundation," reflecting the company's vision of laying a silicon foundation for the rapidly growing AI era.

SuperPoD strategy

Huawei's AI strategy centers on computing power, aiming to create value from hardware and build AI infrastructure through SuperPoD and SuperCluster systems, alongside promoting open systems and open source. As Foundation Models (large AI models that serve as the base for various applications) approach the 10-trillion-parameter scale, using a SuperPoD becomes an essential approach. Huawei's design philosophy connects every component within a cluster using a single protocol through peer-to-peer connectivity, supporting memory access across servers and combining short-reach copper with optical communication systems for long-distance data transmission.

Hi-ONE and the Atlas 960E

Huawei has applied its optical technology expertise to the SuperPoD with the launch of Hi-ONE, the world's first mass-produced NPO-based Optical Engine, supporting data transfer speeds of up to 7.2 Tbit/s, and the only NPO in the industry with a built-in light source.

The Atlas 960E SuperPoD, powered by UnifiedBus and Hi-ONE, is the industry's first SuperPoD built on NPO architecture, featuring an Orthogonal architecture and fully liquid-cooled design, with unified memory addressing via UnifiedBus. This allows a single SuperPoD to scale up to 4,096 NPUs (AI processing units), delivering up to 8 EFLOPS of compute at FP8 precision and 16 EFLOPS at FP4 precision.

Integrating 5,500 units of Hi-ONE into the system reduces the need for 48,000 units of 800G optical modules that would normally be required to connect all the NPUs, resulting in significant power savings.

Source: Techhaus Thailand