DeepSeek to expand computing power with new data center in China

Chinese company DeepSeek plans to build a large artificial intelligence data center in Inner Mongolia. The project is intended to provide the developer with additional computing power for training and running new AI models.

The project was reported by Bloomberg, citing informed sources. The cost of the facility, the number of servers, its computing power, and the expected launch timeline have not yet been disclosed.

DeepSeek's plans confirm that the company intends to expand its own infrastructure. In April, it posted vacancies for a data center engineer and a manager responsible for launching the project, construction of the facility, and its commissioning in Inner Mongolia. This was the company's first public indication of the location of its computing site.

The future engineer is to maintain and optimize servers and network equipment. Among the advantages for candidates, the company listed experience in operating large clusters of artificial intelligence accelerators in technology corporations or research centers.

Inner Mongolia is considered an attractive region for hosting data centers due to access to large volumes of electricity and a relatively cool climate. Low temperatures help reduce cooling costs for equipment, which consumes a significant portion of the energy of computing complexes.

Building its own infrastructure could reduce DeepSeek's dependence on third-party cloud capacity providers. This is especially important when developing large language models, which require thousands of specialized processors for training and daily operation.

The project is being implemented amid U.S. restrictions on supplying China with the most powerful artificial intelligence accelerators. In February, a U.S. administration official told Reuters that DeepSeek may have used Nvidia Blackwell processors, allegedly located in a data center in Inner Mongolia. The company has not publicly confirmed this information, and the Reuters source did not disclose what Washington's assessment was based on.

DeepSeek gained international recognition after releasing models that showed high performance with relatively low claimed training costs. When creating DeepSeek-V3, the company used a cluster of 2,048 Nvidia H800 processors and applied architectural solutions that allow more efficient use of computing resources.

The new data center points to DeepSeek's transition to a more capital-intensive development model. Its own computing capacity may allow the company to release new models more frequently, increase their scale, and compete with the largest American and Chinese AI developers.