Huawei expands Ascend chip line to challenge Nvidia
China’s tech company Huawei revealed its plans to expand its chip and server capacities, aiming to reduce reliance on foreign manufacturers and build its own AI ecosystem.
At the annual Huawei Connect conference in Shanghai, rotating chairman Eric Xu outlined a roadmap to release new versions of the company’s AI chips on an annual cycle, with each generation expected to double computing power.
Huawei says it has already developed its own high-speed memory with high bandwidth, a field long dominated by South Korean players such as SK Hynix and Samsung. The company pointed to the Ascend 910C, which began shipping to major Chinese clients in May 2025 and confirmed updates for its Kunpeng server processors aimed at the domestic market.
Next year, the manufacturer plans to release the Ascend 950 in two variants, followed by the Ascend 960 in 2027 and the Ascend 970 in 2028.
Huawei is developing large-scale server clusters, dubbed “super-nodes.” These designs will connect thousands of chips to accelerate the processing of various types of data (video, audio data, texts). The Atlas 950 super-node, scheduled for late 2026, will link 8,192 Ascend chips. The Atlas 960, designed for 15,488 chips, is planned for the end of 2027.
The announcements underscore China’s push to reduce dependence on U.S. technology and deepen its rivalry with Washington in AI and semiconductors. Chinese authorities have recently tightened rules on AI chip purchases, banning major firms such as ByteDance and Alibaba from buying Nvidia-made processors.
