Advertisement
Chinese AI start-up DeepSeek has postponed the launch of its new R2 artificial intelligence model after running into problems with Huawei’s Ascend chips. The company had planned to release the model in May, but persistent technical issues forced it to switch to Nvidia chips for training while keeping the Ascend chips for running the finished model.
The delay shows how hard it is for Chinese AI firms to move away from U.S. technology. Beijing has been urging companies to use Huawei’s chips as tensions with Washington grow and U.S. chip exports face restrictions. But Nvidia’s H20 chip remains the most popular AI chip in China because of its performance.
Training an AI model, which means feeding it huge amounts of data to improve its abilities, needs far more computing power than inference, which is when the model is used to generate answers. The Ascend chips struggled with the training part, making Nvidia’s chips the better choice for now.
DeepSeek’s earlier R1 model, which shook up the AI industry this year, was mostly trained on Nvidia H20 chips. Big Chinese tech names like ByteDance, Tencent, and Alibaba also rely on them. Even though the U.S. allowed Nvidia to restart sales of H20 chips to China in July, Beijing has raised security concerns over using American-made chips for critical AI projects.
The company is now behind schedule compared to rivals who have already rolled out their latest models. Chinese media reports suggest the R2 model could finally be released in the next few weeks.