Huawei admits its AI chips lag behind U.S. rivals, emphasises progress through clustering

Huawei founder Ren Zhengfei has publicly tempered expectations around the company’s semiconductor ambitions, stating that its flagship AI chip, Ascend, remains “one generation behind” U.S. competitors such as Nvidia. Amid heightened U.S.–China trade negotiations in London focused on export controls, Ren argued that Washington has overstated Huawei’s chip prowess. He acknowledged technical limitations—such as performance inefficiencies and the difficulty of clustering multiple chips—but highlighted Huawei’s strategy to deploy cluster computing and chip stacking to bridge the gap. The company’s new CloudMatrix 384 AI server, which networks 384 Ascend processors via optical interconnects, is a key element of this approach and is currently being tested by partner firms.

Huawei is investing heavily in research and development, committing around Rmb180 billion (≈ $25 billion) annually, with roughly one-third directed at basic theoretical research. Still, Ren admitted that Huawei faces a significant challenge in building an ecosystem comparable to Nvidia’s developer-friendly Cuda platform. He pointed to China’s robust electricity and telecom infrastructure as an advantage in supporting large-scale AI deployment domestically. While Huawei asserts improvements in chip yields and cluster performance, analysts caution that its single-chip performance and developer support remain behind U.S. counterparts. The company is leveraging its domestic position—boosted in part by U.S. export bans on Nvidia—to advance its AI chip ambitions, though experts say closing the technological divide will require sustained investment and ecosystem development.