Login to Continue Learning
The Chinese AI industry has long relied on NVIDIA and its ecosystem. When recent US export controls were implemented, Beijing accelerated the development of domestic AI chip alternatives, prompting companies like Huawei to present solutions such as the Ascend 910C. However, these homegrown alternatives have not been successful. According to a report by The Financial Times, DeepSeek’s R2 model launch was delayed because it could not be trained using Chinese AI chips; instead, the firm is now switching back to NVIDIA.
The initial expectation for the R2 model was a May release. Unfortunately, Huawei’s AI chips encountered stability issues and slower interconnect technologies, particularly the lack of a viable CUDA tech stack replacement. As a result, DeepSeek has reverted to using NVIDIA’s technology to ensure a more robust and efficient training process.
Huawei’s AI chips are currently causing difficulties for Chinese firms, necessitating the adoption of NVIDIA’s H20 AI chips. This shift creates new demand for NVIDIA in regional markets but also complicates matters due to ongoing investigations into potential security backdoors in AI chips entering China. The Chinese government has reportedly advised domestic AI firms not to use Team Green’s chips, citing location tracking mechanisms.
DeepSeek’s R2 model is set to debut soon, and it will be interesting to see its performance. However, the report indicates that China’s domestic alternatives are still far behind NVIDIA, meaning local AI firms must continue relying on NVIDIA’s technology for now.
