DeepSeek is building a gigawatt of compute in Inner Mongolia, Bloomberg reports

The company everyone filed under “cheap models” is building a gigawatt of compute in Inner Mongolia.

Bloomberg reported on July 30, 2026 that Hangzhou-based DeepSeek is adding one gigawatt of compute in Ulanqab, Inner Mongolia, about 350 kilometers northwest of Beijing. The plan is to build its own campus while leasing additional capacity from others, with part of it expected online by the end of 2027 or the start of 2028.

The open question: whose chips

The processors have not been settled, according to Bloomberg’s sources. The options on the table are Nvidia hardware, Huawei silicon, DeepSeek’s own future chips, or some combination of the three.

That undecided chip question is its own story. A gigawatt campus commits DeepSeek to a hardware direction for years, and the choice between US and domestic silicon is as much a geopolitical bet as a technical one.

How big is a gigawatt here

For scale: the project would be larger than any AI facility a Chinese company currently operates, and still smaller than the 3GW and 5GW campuses under development in the US. Chinese rival Z.ai is reportedly pursuing a 1GW project of its own.

DeepSeek has also been advertising data center engineering roles in the region, which is a decent tell that the plan is real rather than aspirational.

Efficiency and scale, not either-or

The story people told about DeepSeek was efficiency: better results from less compute, the lab that proved you did not need a hyperscaler’s budget to compete. This is the same company deciding that efficiency and scale are not an either-or.

If the most compute-efficient frontier lab in China thinks it still needs a gigawatt, that says something about where everyone believes the frontier is heading. Does China close the compute gap this decade? Projects like this one are how the answer gets built.