Huawei’s rotating chair Eric Xu has urged Chinese AI companies to accelerate development to catch up with US labs, contrasting with Silicon Valley calls for a slowdown over existential risks. Speaking at a press conference, Xu said Chinese researchers must increase development speed to also see AI dangers, then strike a balance between progress and risk. His comments followed Anthropic and OpenAI’s call for a coordinated slowdown.

Xu acknowledged Huawei remains “not as advanced” as US rivals but said export controls have pushed customers toward domestic tech, emphasizing self-sufficiency in chips and supply chains. Huawei unveiled updates to its AI hardware, including the Ascend 960 DT chip for training (launching Q1 2027) and an inference chip (Q3 2027), each delivering twice the performance of its current flagship. The company aims to connect up to 4,096 Ascend processors in data centre systems, up from 1,024 previously, to compensate for weaker individual chips due to TSMC restrictions.

Huawei claims Ascend has overtaken Nvidia in China market share, though reliable estimates are difficult. Xu noted six or seven Chinese AI labs plan to train models with 10tn to 40tn parameters in the next two years, a significant increase from current sizes. Huawei’s chief scientist Heng Liao said CUDA’s importance has diminished as new programming tools reduce Nvidia’s software advantage.