AI compute is concentrating in data-centre towns like Ulanqab, moving intellectual output out of office towers and leaving city office property oversupplied.

A vlogger went to Ulanqab in Inner Mongolia and found the grassland covered in data-centre construction sites, with tens of thousands of builders at work. The city already has more than 100 data centres, and almost every major Chinese IT company has a facility there.
Why Ulanqab
It is close to Beijing: the round-trip fibre delay is only 4.2 milliseconds. Wind and solar power are so abundant that electricity is cheap. The average temperature is 4.3°C, so for ten months a year outside air alone can cool the servers. Land was initially cheap too, a few tens of thousands of yuan per mu.
What stuck with me was the vlogger's closing line: office towers will keep losing value, because intellectual output is moving into these data centres.
Office towers are not competing with other office towers
These halls mainly run large models. If compute can be called remotely from anywhere, why pay prime-city rent for office space? The data centre is the new office tower, the main place where intellectual work gets produced.
Many cities spent recent years building huge central business districts. That concept may not survive; a compute town like Ulanqab is the CBD of the AI era. Office rents are heading down.
One item in the same issue fits neatly: a company that went all-in on AI coding for a year saw its codebase triple, with no one able to explain how most of it works. Of its two most active programmers, one quit and one was laid off. Management has recently begun hinting that people may hand-write a little more code.
Why it matters
The fate of office towers is set not by landlords but by where people produce value. Once intellectual work moves into server halls, CBDs become landmarks of the previous economy, and both rents and city plans have to be recalculated.



Curated from high-quality sources, with concise summaries and key takeaways.