A Data Center That Could Power a Small City
DeepSeek intends to add a full gigawatt of computing capacity for its next AI data center.
Ulanqab's naturally cold climate helps with a simple but expensive problem: keeping AI servers cool. The average annual temperature there is just 4 degrees Celsius, which dramatically cuts the cost of running the air conditioning that AI chips need.
DeepSeek is not alone in picking that spot. Local government approvals show more than a dozen companies have projects planned in Ulanqab. China Mobile already owns infrastructure in the region, and other Chinese AI players like Z.AI (formerly Zhipu) is also planning a 1-gigawatt data center there - though it will use only Chinese-made chips.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
The Dollars Behind the Gigawatt
Building a data center this big does not come cheap. Nvidia CEO Jensen Huang estimated that equipping a 1-gigawatt facility with top-tier AI accelerators would cost roughly $50 billion.
American AI firms such as OpenAI and Anthropic have been at the forefront, investing huge sums - tens of billions - into computing power for artificial intelligence. Chinese AI services generally cost one-tenth or less than their US equivalents. Even so, some US data center projects are even bigger. Facilities sized at 3 gigawatts and 5 gigawatts are under development in the United States.
The Race for AI Dominance
DeepSeek's massive data center comes amid an intensifying global race for AI supremacy. While US projects like those from OpenAI and Anthropic are larger in scale, Chinese firms are able to offer services at a fraction of the cost due to lower infrastructure expenses and government support. The choice of Ulanqab, with its cold climate, highlights how location can reduce operational costs. Additionally, DeepSeek raised a record $7 billion in a fundraising round that valued it at roughly $50 billion, even as it faces accusations of using restricted technology and copying model outputs.
What It Means for Your Portfolio
DeepSeek is already laying the groundwork for a stock market debut, potentially submitting its IPO filing within this year. According to the Trump administration, DeepSeek bypassed export controls by employing Nvidia's Blackwell chips - equipment that is essentially prohibited from being sold to China - at a facility in Inner Mongolia. OpenAI and Anthropic have both accused DeepSeek of distillation, meaning it uses the outputs generated by their AI models to train its own technology with similar capabilities.
This expansion underscores the escalating competition between US and Chinese AI firms. While American companies like OpenAI and Anthropic have invested tens of billions, Chinese firms benefit from lower costs and government backing. DeepSeek's planned data center, along with similar projects by Z.AI, signals China's determination to scale up AI infrastructure rapidly, despite US export restrictions. The cold climate of Ulanqab further reduces operational expenses, giving Chinese players a cost advantage that may intensify the global race.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
