Huawei Technologies Co. has introduced its latest solution to enhance computing power by bundling large numbers of AI chips together, aiming to compete with Nvidia Corp.’s industry-leading technology. The Shenzhen-based firm announced on Thursday that its new SuperPod system can support linking up to 15,488 graphics cards equipped with Huawei’s Ascend-branded AI chips.
The company also reported building a super cluster with computing power equivalent to roughly 1 million graphics cards. The move comes amid China’s push to reduce reliance on Nvidia hardware, with the government recently instructing major tech firms not to use Nvidia’s RTX Pro 6000D—a high-end semiconductor often repurposed for AI workloads.
Huawei’s SuperPod appears to be an upgrade of its efforts to rival Nvidia’s NVLink technology, which enables high-speed communication between chips in a server. The system is particularly important given that Huawei’s most advanced Ascend chips currently offer lower individual performance compared with Nvidia’s cutting-edge AI silicon. By linking thousands of chips together in clusters, Huawei seeks to compensate for that gap.
Earlier this year, Huawei founder Ren Zhengfei told the state-owned People’s Daily that while Huawei lags behind the U.S. in single-chip output, “we can still get the results we want by compensating with cluster-based computing.”
In addition to the SuperPod announcement, Huawei revealed a roadmap for new AI chips over the next three years. The lineup includes the Ascend 950PR set for early 2026, the Ascend 950DT in late 2026, Ascend 960 in late 2027, and Ascend 970 in late 2028, according to Chinese media reports.
The developments signal Huawei’s intensified push to establish domestic alternatives to Nvidia, reflecting China’s broader strategic goal of reducing dependency on foreign AI technology while advancing its own semiconductor capabilities.
