Huawei Takes Aim at Nvidia with Atlas 960E SuperPoD

Huawei has announced the Atlas 960E SuperPoD, a computing cluster designed to rival Nvidia in the AI infrastructure market. The system is capable of delivering up to 8 EFLOPS for FP8 computations and 16 EFLOPS for FP4, with up to 1 petabyte of High Bandwidth Memory across its full configuration.

The Atlas 960E features a unified memory-addressing scheme that allows every chip to access the same pool of memory, rather than passing data back and forth across slow interconnects. This design choice enables improved performance compared to previous models, with Huawei claiming between 2.3 and 4 times better results on training and inference tasks for large models.

One of the key benefits of the Atlas 960E is its reduced power consumption. The system uses approximately 5,500 Hi-ONE Near-Packaged Optics units instead of traditional optical modules, cutting power consumption by over 550 kilowatts per pod. This improvement in efficiency can provide significant cost savings for data center operators.

Huawei's ambitions extend beyond the Atlas 960E, with plans to develop a single AI computing framework scalable to 256,000 nodes and potentially exceeding one million NPUs in larger configurations. The company is expected to release larger versions of the Atlas 960 platform in the fourth quarter of 2027.