Bloomberg today cited people familiar with the matter as saying that DeepSeek is planning to build a large AI data center in Inner Mongolia. This ambitious infrastructure project shows that the Chinese artificial intelligence startup is expanding its AI computing power deployment.
According to people familiar with the matter, who spoke on condition of anonymity, DeepSeek plans to add 1GW (gigawatt) of computing power in Ulanqab, Inner Mongolia. In addition to its plan to build its own data center, the company also intends to lease additional computing power from other companies.
Two of the people familiar with the matter said that some of the computing capacity at the DeepSeek data center will come online by the end of next year or early 2028. It is currently unclear what chips the project will use. For reference, NVIDIA is currently the industry standard for AI data centers, while Huawei is a representative company in China's data center sector.
DeepSeek released a new batch of job postings in April. In addition to Hangzhou and Beijing, positions based in Ulanqab, Inner Mongolia, appeared for the first time. The two related positions are Senior Data Center Delivery Manager and Senior Data Center Operations and Maintenance Engineer, with salaries of 15-30K · 14 months' pay.
Ulanqab has an average annual temperature of approximately 4°C. Its naturally cold environment can reduce the cooling requirements of high-power AI servers, making it an ideal location for major companies to build data centers.
NVIDIA CEO Jensen Huang once estimated that a 1GW-scale data center equipped with the most advanced AI accelerators could cost as much as 50 billion US dollars to build (note: approximately 338.682 billion yuan at the current exchange rate). However, this cost can vary greatly depending on the construction site and the type of AI chips. As is widely known, the construction costs of AI data centers in China are generally lower than those in the United States.
DeepSeek did not immediately respond to Bloomberg's request for comment due to working hours.