DeepSeek’s recent struggle with using Huawei’s AI chips for training its R2 model highlights a growing challenge in the Chinese AI industry. Initially aiming to utilize domestic technology, DeepSeek found Huawei’s AI chips insufficient, prompting a shift to NVIDIA’s more robust offerings.
Despite efforts to boost local chip development due to U.S. export controls, domestic alternatives like Huawei’s Ascend 910C are falling short. These chips face issues with stability and slower interconnect technologies, lacking crucial features like NVIDIA’s CUDA tech stack, which is essential for high-performance AI training. This has resulted in delays for DeepSeek, pushing the R2 model’s launch beyond its intended May release.
China’s reliance on NVIDIA’s superior AI ecosystem seems unavoidable at this point, even as Beijing urges firms to adopt homegrown solutions. This switch also comes amid investigations into potential security risks associated with foreign chips, including allegations of embedded tracking mechanisms.
As DeepSeek prepares to debut its R2 AI model, the situation underscores the significant gap that still exists between domestic and international tech capabilities. For now, companies are driven back to NVIDIA, highlighting the critical role of advanced chip technology in the rapidly evolving AI landscape. This dynamic will be pivotal to watch as China’s AI firms navigate the complex terrain of technology development and international competition.
