Huawei Chips to Power DeepSeek AI Soon

A new contender has entered the AI race. DeepSeek, a Chinese AI model, is making waves by claiming to surpass US competitors in cost efficiency.

Its open-source chatbot has already dominated the App Store in 51 countries, and now it has been revealed that Huawei’s Ascend 910C AI chip is powering its operations.

The DeepSeek R1 LLM (large-language model) was initially trained using Nvidia’s H100 GPUs but now relies on Huawei’s Ascend 910C for inference—the process of generating responses from a trained AI model.

This shift highlights China’s push for self-reliant AI infrastructure, reducing dependence on US chipmakers.

According to industry sources, Ascend chips are not used for training, so their GPU power demands are lower.

However, the 910C’s performance is still not on par with Nvidia’s H100, limiting its ability to handle AI training at scale.

Huawei is set to address this with the upcoming 920C chip, which aims to rival Nvidia’s Blackwell B200, the top-tier AI chipset.

DeepSeek’s rapid rise in the AI space signals China’s growing AI ambitions, especially as US restrictions limit access to cutting-edge Nvidia hardware.

Whether Huawei’s AI chips can keep up with global competitors remains to be seen, but DeepSeek’s App Store dominance and open-source approach suggest it’s already gaining traction.