DeepSeek is planning a major expansion of its artificial intelligence computing infrastructure with the development of a gigawatt-scale data centre in Inner Mongolia, according to people familiar with the matter, marking one of the most ambitious infrastructure projects undertaken by a Chinese AI company.
The Hangzhou-based startup is seeking to add one gigawatt of computing capacity in Ulanqab, approximately 350 kilometres northwest of Beijing. According to the people, who requested anonymity because the information is private, DeepSeek intends to construct its own new facility while also leasing additional computing capacity from other companies. More than a dozen firms already have projects planned in the city, based on a third-party compilation of local government approvals.
The company is aiming to bring at least part of the facility’s capacity online by the end of next year or early 2028, two of the people said. It remains unclear which semiconductor technology will power the project. Nvidia Corp.’s chips remain the global industry standard for AI data centres, while Huawei Technologies is China’s leading domestic competitor in the sector. DeepSeek did not immediately respond to an emailed request for comment outside regular office hours.
The planned investment reflects a broader effort by Chinese AI companies to strengthen computing infrastructure as competition with leading US developers intensifies. While companies such as OpenAI and Anthropic have invested tens of billions of dollars in expanding AI computing capacity, DeepSeek’s proposed one-gigawatt facility would exceed the scale of any AI data centre currently operating under a Chinese company, although it would remain smaller than the three-gigawatt and five-gigawatt projects being pursued by American firms.
China’s Z.AI, formerly known as Zhipu, is also developing a one-gigawatt AI project that it plans to equip entirely with Chinese-made chips, highlighting a wider domestic push to expand AI infrastructure.
DeepSeek has previously indicated its ambitions to develop significantly greater computing capabilities. A current job advertisement for a data centre position, based in either Hangzhou, Beijing or Inner Mongolia, states that the successful applicant would be directly involved in building a gigawatt-scale AI infrastructure campus.
Ulanqab offers several advantages for large-scale computing facilities. With an average annual temperature of around four degrees Celsius, the city’s naturally cold climate reduces the energy required to cool power-intensive AI servers, making it an attractive location for operators such as China Mobile.
Led by Liang Wenfeng, DeepSeek has become a central player in China’s efforts to compete globally in artificial intelligence. The company attracted widespread attention early last year after unveiling an AI model that it said achieved cutting-edge performance using only a fraction of the computing resources required by competing models.
The startup has also emerged as a leading provider of low-cost, open-source AI services that have increased competitive pressure on American companies including OpenAI and Anthropic. Chinese AI services are often priced at one tenth, or less, of comparable US offerings. That competition intensified further this month when China’s Moonshot introduced its Kimi K3 model, which the company says delivers benchmark performance comparable to leading US systems. Alongside Moonshot and Z.AI, companies including Alibaba Group Holding and MiniMax Group form part of an increasingly competitive Chinese AI sector.
Bloomberg News has previously reported that DeepSeek has begun preparations for an initial public offering and could file as early as this year, following a record US$7 billion fundraising round that valued the company at approximately US$50 billion.
The company also faces scrutiny from the United States. According to people familiar with the matter, the Trump administration believes DeepSeek has circumvented export restrictions by using Nvidia’s Blackwell processors, which are effectively prohibited from being sold to China, in a data centre in Inner Mongolia. OpenAI and Anthropic have also accused DeepSeek of using distillation, a technique involving the outputs of their AI models to train systems with similar capabilities.
DeepSeek’s proposed expansion may indicate that Chinese AI companies are securing the financial resources needed to build advanced computing infrastructure, an area historically dominated by Silicon Valley firms through access to US capital markets. Nvidia Chief Executive Jensen Huang has estimated that constructing a one-gigawatt AI data centre equipped with the most advanced AI accelerators could require around US$50 billion in investment, although overall costs vary significantly depending on location and the chips deployed. AI data centres in China are generally considerably less expensive to build than equivalent facilities in the United States.
