DeepSeek plans 1-gigawatt Inner Mongolia AI data center with early 2028 capacity

DeepSeek is planning a 1-gigawatt AI data center in Ulanqab, Inner Mongolia, with partial operations targeted by the end of next year or early 2028, marking a major expansion of the Chinese startup’s computing infrastructure. The company intends to build its own facility while also leasing equipment from other firms, and has reportedly posted related jobs. The type of AI semiconductors to be used has not been finalized, with Nvidia GPUs the industry standard and Huawei chips seen as a major domestic alternative in China. The project underscores how competition between U.S. and Chinese AI companies is shifting from model performance toward control of physical infrastructure, power and chip supply. Ulanqab’s low temperatures and established data center base make it attractive for large-scale AI server cooling and electricity efficiency. Bloomberg said the site would be the largest among currently operational Chinese company facilities at that scale, though still smaller than the 3-5 gigawatt projects being pursued by some U.S. firms. DeepSeek recently completed a $7 billion funding round at a valuation of about $50 billion, and Bloomberg separately reported it could file for an IPO as early as this year. The company, led by CEO Liang Wenfeng, gained prominence after unveiling an AI model that delivered cutting-edge performance with fewer computing resources than rivals, helping position it as a low-cost, open-source challenger to OpenAI and Anthropic. The project also comes amid scrutiny from the Trump administration over whether DeepSeek may have used Nvidia Blackwell processors, whose sale to China is effectively banned, while OpenAI and Anthropic have alleged the company used distillation techniques based on their models’ outputs.