Just a few hours after Chinese AI R1 by DeepSeek hit the public, the AI space is changing. As the US reacted to R1’s rise, Nvidia lost $600 billion, the biggest one-day drop for any company in U.S. history. The presence of R1 is intensifying the rivalry in the AI space. This new Chinese AI model claims to offer major cost-efficiency advantages over US options. To make it even better, it is open-source and has gained widespread popularity in a few days. This system is now in the top spot in the App Store across 51 countries. However, what powers DeekSeek AI?
Why abandon the Nvidia chip for Huawei?
Earlier reports about the R1 AI reveal that the system was trained using Nvidia’s H100 GPU. However, new information now reveals that training the system is all Nvidia gets from it. According to @Dorialexander, the bigger headline should be the fact that R1 runs inference on Huawei Ascend 910C. Inference is the process of generating responses from a trained model. Thus, the Huawei chip will do most of the job. So, why did DeepSeek not stick to Nvidia chips?
There are a couple of reasons for this but the most obvious one is cost. Nvidia chips are well more expensive than Huawei’s equivalent. The inference requires less computational power thus the Ascend 910C provides a more affordable option for this stage of the process.
Also, the US export ban on advanced Nvidia chips has limited access to the latest Nvidia hardware in China. By using Huawei’s chips, DeepSeek reduces its reliance on US technology, aligning with China’s broader push for greater self-sufficiency in AI infrastructure.
Huawei’s ambitions and future plans
Despite its role in inference, the Ascend 910C has performance limitations that make it less suitable for training large-scale AI models. To address this, Huawei is working on the Ascend 920C. This chip is expected to compete with Nvidia’s upcoming Blackwell B200, a next-generation chipset designed for AI applications.
Huawei’s AI chip development is part of China’s broader effort to reduce dependence on US technology in the semiconductor and AI sectors. The country has been investing heavily in domestic AI hardware to counter US sanctions. It is trying to maintain a competitive position in the global AI race.
DeepSeek’s use of Huawei chips shows a move toward China’s own AI tools. While Nvidia still leads in AI training, the rise of Huawei chips for running tasks could take some of Nvidia’s cake. The new Ascend 920C will try to match Nvidia’s best chips, which may make the AI chip race tight. If Huawei succeeds in closing the performance gap, it could significantly alter the dynamics of AI hardware. This will be more noticed in markets where Chinese companies are looking for alternatives to US technology.