DeepSeek, a Chinese technology company known for its cutting-edge artificial intelligence models, plans to construct a large-scale data center in Inner Mongolia. According to Bloomberg, an AI data center campus with a capacity of 1 GW will be built in Ulanqab, a prefecture-level city located 350 km northeast of Beijing.
Expansion Strategy and Location
The choice of Ulanqab is not accidental: the region has an average annual temperature of around +5 °C, creating ideal conditions for natural server cooling and reduced energy consumption. This makes the location attractive not only to DeepSeek but also to other IT giants, such as China Mobile. According to sources, projects from more than a dozen companies are already planned in Ulanqab.
DeepSeek expects to bring at least part of the new campus's capacity online by the end of next year or early 2028. The company plans not only to build its own facilities but also to lease additional capacity from partners.
Technological Equipment and Competition
While the exact equipment specifications have not yet been disclosed, NVIDIA chips dominate the industry. However, under sanctions, Huawei is the undisputed leader in China. There are also rumors that DeepSeek is developing its own accelerators, although they will likely be intended for model training rather than inference.
The DeepSeek project is part of a massive race with Western companies. For comparison: industry leaders such as OpenAI and Anthropic are investing tens of billions of dollars in infrastructure with a capacity of 3–5 GW. Nevertheless, the 1-GW campus will become the largest project of its kind in China. Competitor Z.AI (formerly Zhipu) is also building a 1-GW facility but plans to equip it exclusively with domestic accelerators.
Economic Pressure and Geopolitics
DeepSeek has become one of the key players in China's strategy to capture the global AI market. The company shocked the industry by presenting a model with advanced characteristics created with relatively modest computing resources. Moreover, Chinese services offer prices 10 times or more lower than American counterparts, threatening the business models of Western giants.
The US administration expresses concern that DeepSeek is bypassing export restrictions by using NVIDIA Blackwell chips, which are not officially supplied to China. American companies also accuse the Chinese competitor of "distillation"—using their models to train their own analogues.
Investments and the Future
The project in Inner Mongolia may signal that Chinese companies are beginning to increase financial support for AI infrastructure, which has historically been dominated by Silicon Valley. According to estimates by NVIDIA CEO Jensen Huang, creating a 1-GW data center with advanced accelerators will require approximately $50 billion in investment.
DeepSeek is preparing for an initial public offering (IPO) and may file an application before the end of 2026. Previously, the company raised a record $7 billion in its latest funding round, valuing its capitalization at approximately $50 billion. Alibaba Group Holding, MiniMax Group, and Moonshot, which debuted with the Kimi K3 model, have also joined the race for leadership.
