DeepSeek, a new player in the AI wars from China, has captured international attention with its remarkable cost-efficiency claims that challenge American tech giants. The company’s open-source chatbot has secured the number one position across app stores in 51 nations, but now another interesting development has come to light. The Large Language Model (LLM) is using Huawei chips to operate.
Technical specifications show that while DeepSeek’s R1 large-language model was initially developed using Nvidia’s H100 processors, it now employs Huawei’s Ascend 910C chip for day-to-day operations.
According to technology analyst @Dorialexander, this shift is particularly significant because the Ascend chips, focused solely on inference rather than training, require substantially less GPU processing power.
The existing Ascend 910C, while serviceable, has shown performance constraints that make it less than ideal for intensive AI training operations. To overcome current performance limitations, Huawei is developing its next-generation 920C chip, designed to rival Nvidia’s top-tier Blackwell B200 processor in AI applications.
This goes to show how Chinese tech giants are not only surviving but thriving despite U.S. trade restrictions, as evidenced by DeepSeek’s recent achievements. The company’s ability to develop a competitive AI model at a fraction of the cost of its American counterparts has not just put it on the top charts in app stores but also wiped away trillions of dollars from Wall Street.
While U.S. export controls have limited access to advanced Nvidia chips like the H100 and the upcoming Blackwell series, Chinese companies have adapted by using alternative solutions, including lower-performance processors and domestic chip development.
