SHANGHAI: Huawei has introduced its Atlas 960E SuperPoD at Huawei Connect 2026, presenting it as a large-scale computing system for training and inference workloads.
According to Huawei’s product announcement, the system is built around its Ascend processors and can be configured with up to 4,096 NPUs. The company said the platform is designed to support work on very large AI models.
Performance figures are company claims
Huawei cited up to 8 EFLOPS at FP8 precision and 16 EFLOPS at FP4 precision for the system. Those are Huawei’s stated specifications, rather than independently verified benchmark results.
The company also described the Atlas 960E as an NPU-based SuperPoD. Its announcement focuses on scale, interconnection and cooling for data-centre deployments.
Why the launch matters
The announcement arrives as firms and governments seek more options for AI infrastructure. Hardware launches are important for the AI supply chain, but their commercial impact depends on availability, software support, pricing and how customers assess performance in real workloads.
Source: Huawei official announcement.
Featured image: AI-generated editorial illustration.