Nvidia has a serious fight on its hands. Huawei recently took the stage at the World Artificial Intelligence Conference to showcase its strongest computing cluster yet, a system designed to challenge the peak of American silicon dominance. According to reports from Chinese technology outlets IT Home and KuaiKeji, the physical hardware is called the Atlas 950 SuperPoD. It represents a massive leap in processing capabilities, boasting performance numbers that dwarfed comparable hardware from US competitors.
The claimed performance gap is wider than most analysts expected. Research published by Bank of China International indicates that the Atlas 950 SuperPoD offers 6.7 times the total computing power of Nvidia and its NVL144 system. That competitor platform uses 144 graphics cards. Huawei is moving quickly to prove its hardware can run massive AI models with over 100 billion parameters. This public display is unusual for the company. Historically, Huawei kept its Ascend chip details quiet, choosing to sell directly to government agencies and private enterprise clients without sharing public benchmarks.
The secret behind this raw speed lies in how the chips talk to each other. Traditional setups waste valuable time waiting for data to travel between separate processors. Huawei solved this bottleneck by using a proprietary interconnect protocol called Lingqu alongside an updated SuperPoD architecture. This design binds the AI chips into a single, tightly packed computing unit. Data latency drops. Effective processing power spikes because the chips do not have to wait on slow transfer speeds.
Beside the flagship system, the company also brought out a physical version of its Atlas 850E air cooled SuperPoD. This hardware is aimed at businesses that cannot afford to rebuild their data centers. Upgrading to liquid cooling is incredibly expensive. Huawei engineered a self developed thermal management setup that fits right into standard server cabinets. Businesses can deploy this system in their existing air cooled server rooms without investing in costly liquid infrastructure.
