DeepSeek은 내몽고 데이터센터에서 Nvidia 하드웨어를 대체하기 위해 Huawei AI 가속기를 대규모로 도입할 계획이다.

Chinese artificial intelligence startup DeepSeek plans to deploy at least 160,000 high-end AI accelerators from Huawei Technologies Co. at a large data centre under construction in Inner Mongolia, potentially creating one of the largest known clusters of Huawei AI chips. According to a Bloomberg report citing people familiar with the matter, DeepSeek intends to use Huawei’s next-generation Ascend 950DT processors to operate its AI models. However, the installation schedule will depend largely on Huawei’s manufacturing capacity.

Despite Huawei designing and positioning the Ascend 950DT chips for demanding training workloads, DeepSeek is not currently planning to use them for model training. The startup had previously explored training its AI models using Huawei processors but has continued to rely on Nvidia accelerators for the critical training process, Bloomberg reported. Instead, the new hardware will be deployed primarily for running or inferring AI models.

DeepSeek is seeking additional supplies of Huawei’s most advanced chips, but production constraints could complicate efforts to meet the startup’s requirements. Huawei is expected to produce only a few hundred thousand Ascend 950DT chips this year because of shortages of key components, including advanced memory. The company must also fulfill demand from other domestic customers while attempting to ship limited quantities to overseas markets. As a result, fulfilling DeepSeek’s order could take more than a year, according to people familiar with the situation cited by Bloomberg.

If the project proceeds as planned, the Inner Mongolia facility will become a cornerstone of DeepSeek’s expanding AI infrastructure. The startup is reportedly in discussions to raise billions of dollars to finance this expansion. A deployment of more than 160,000 Huawei Ascend processors at a single site would substantially exceed other publicly known clusters built on the Chinese company’s AI chips. By the time the Inner Mongolia facility becomes operational, China could potentially host several AI computing clusters of a similar scale, according to the report.

The move could mark a significant step in China’s efforts to reduce its dependence on Nvidia Corp. for advanced processors used to develop and run artificial intelligence models. While a data centre containing more than 100,000 AI chips would place DeepSeek’s infrastructure ambitions broadly in line with facilities increasingly built by major Western technology companies, Chinese facilities would continue to operate with processors that remain less powerful than Nvidia’s leading accelerators. Nvidia’s AI chips remain the global industry standard for training and running sophisticated AI models. Their availability in China has been severely restricted by US export controls, with sales of certain advanced processors requiring Washington’s approval. Such a project represents an important development in Beijing’s push to build an AI ecosystem that does not depend on Nvidia technology.