Huawei Atlas 960E SuperPoD challenges Nvidia in AI infrastructure race

Editorial illustration: Four densely packed black server cabinets connect through overhead copper pipes to a cream cooling unit. A smaller two-cabinet installation stands behind them on a raised concrete platform.

In brief

  • Atlas 960E SuperPoD unveiled September 17 at HUAWEI CONNECT 2026 in Shanghai with 4,096 Ascend NPUs.
  • System delivers 8 EFLOPS for FP8 and 16 EFLOPS for FP4 computations targeting 10 trillion parameter models.
  • Hi-ONE optical design cuts power consumption by 550+ kilowatts per pod versus predecessor versions.
  • System achieves 99.8% availability with mean time between failures doubled versus prior models.
  • Larger Atlas 960 variants expected Q4 2027; Ascend 960 program ahead of schedule.

Performance and Scale

The Atlas 960E delivers 8 EFLOPS for FP8 computations and 16 EFLOPS for FP4, with up to 1 petabyte of High Bandwidth Memory. Compared to the previous Atlas 950 SuperPoD, Huawei claims 2.3 to 4 times better performance on training and inference tasks for workloads at scale. This performance jump positions the Atlas 960E as a credible alternative for data center operators evaluating infrastructure investments.

The scale ambitions are substantial. Huawei is targeting a single AI computing framework scalable to 256,000 nodes, with SuperCluster configurations eventually exceeding one million NPUs. Larger versions of the Atlas 960 platform are expected to arrive in the fourth quarter of 2027.

Energy Efficiency and Reliability

A critical differentiator lies in power consumption. The Atlas 960E replaces roughly 48,000 conventional 800G optical modules with approximately 5,500 Hi-ONE Near-Packaged Optics units, a redesign that cuts power consumption by more than 550 kilowatts per pod. In an era where energy costs and grid constraints are reshaping AI infrastructure decisions, this efficiency gain carries material weight.

Reliability metrics also improved. System availability reaches 99.8%, and mean time between failures doubles compared to predecessor models. Higher uptime reduces operational friction and total cost of ownership, factors that weigh heavily in enterprise procurement.

Development Timeline

The Ascend 960 development program is reportedly running ahead of its original schedule, signaling momentum in Huawei's push to narrow the gap with entrenched competitors. The announcement at HUAWEI CONNECT 2026 marks a public statement of capability—one that signals Huawei's confidence in its roadmap and its ability to deliver at scale.