DeepSeek delays R2 model launch due to Huawei chip issues
What's the story
DeepSeek, a Chinese artificial intelligence (AI) start-up, has postponed the launch of its highly-anticipated R2 model. The delay comes after training attempts with Huawei's Ascend chips were unsuccessful. According to the Financial Times, persistent technical issues during the training process prompted DeepSeek to switch to using NVIDIA chips for training, while continuing to use Ascend chips for inference.
National strategy
China's push for Huawei's Ascend chips
China has been pushing its local developers to use Huawei's Ascend chips, especially after US chip exports to China became a major flashpoint this year. However, despite the push from Huawei and Beijing, NVIDIA's H20 remains the most popular AI chip in China. The US had even allowed NVIDIA to resume sales of its H20 chips in China last month.
Security worries
Security concerns over American-made chips
Despite the resumption of sales, Beijing has raised security concerns over the use of American-made chips in critical Chinese AI development. This comes as DeepSeek's R1 model, which shook up the AI sector earlier this year, was developed largely using H20 chips. Other major Chinese AI developers such as ByteDance and Alibaba Group have also used these chips. Earlier this week, China had asked local developers to justify their orders of NVIDIA chips.