Huawei Predicts Major Shift to Ascend AI Chips for Model Training in 2027
Eric Xu Zhijun, the rotating chairman of Huawei Technologies, anticipates a significant domestic transition towards its artificial intelligence computing infrastructure for model training in the upcoming year. During his address at the Huawei Connect 2026 conference held in Shanghai on Thursday, Xu expressed confidence that much of the AI model training will utilize the SuperPoD or SuperCluster powered by the Ascend 950DT starting next year. He noted that Huawei's proprietary Ascend chips have surpassed Nvidia in market share within China, although he did not disclose exact numbers. Despite facing supply challenges, the Ascend processors are increasingly competing with Nvidia in the Chinese market, highlighting Huawei's expanding influence in the country's AI landscape and potential effects on the global AI chip industry.
Key facts
- Huawei expects a major domestic shift to its AI computing infrastructure for model training in 2027.
- Rotating chairman Eric Xu Zhijun made the prediction at the Huawei Connect 2026 conference in Shanghai on Thursday.
- Xu said AI model training will be based on SuperPoD or SuperCluster using Ascend 950DT starting next year.
- Huawei's Ascend chips have overtaken Nvidia in domestic market share, according to Xu.
- Xu did not provide specific figures for the market share.
- Ascend processors are gaining ground on Nvidia in China despite supply constraints.
- The shift is expected to occur next year, which is 2027.
- The conference took place in Shanghai.
Entities
Institutions
- Huawei Technologies
- Nvidia
Locations
- Shanghai
- China