AINEWS
China's 2021 plan to move AI compute to its cooler, emptier west has the country on course to nearly double data center capacity in five years, while U.S. and European towns fight new builds.
When a country wants to nearly double its data center capacity in five years, the harder question is rarely chips. It is siting. And on that question, China and the West are now running different races.
China's 2021 "Eastern Data, Western Computing" plan answered it by directing the country's AI compute to its cooler, emptier west. Five years in, the buildout is visible on a hillside in Guizhou province, where Huawei's largest data center sits inside the Gui'an New Area, a planned district that climbs above Guiyang. The campus is designed, deliberately, to look like a row of European hill towns, with clock towers, a multi-arched bridge over a man-made lake, and what AFP describes as a "barren mountain" transformed into a server park. AFP reported this month that China is on course to roughly double its national data center capacity by 2031, driven by east-coast AI demand routed westward to renewable and thermal baseload.
The logic is four-part, and each part does real work. The climate is cooler at altitude, which cuts cooling load. Land is cheap, available, and politically uncontested. Western provinces offer renewable and thermal baseload that the grid can dispatch, framed as national energy-AI coordination by the Jamestown Foundation. And the political permission to build is granted at the center, not negotiated block by block with county councils, which is where most U.S. and European data centers now stall.
Local shopkeepers and residents told AFP they had seen jobs, transport links, and two new universities arrive with the corridor. The contrast AFP draws is with the NIMBY record in the U.S. and Europe, where water draw, substation noise, and grid load have stalled projects in Loudoun County, the Dutch polder, and the Irish midlands. The Gui'an story does not need to be sold to its neighbors. It has been sited, in the language of permitting, on their behalf.
Dissent around central government policy in China is tightly controlled, and the residents-on-record positivity is shaped by that context. Local tradeoffs (water draw, land use, power allocation, the displacement of smallholders) do not surface in the same way they do in a Virginia planning hearing. The contrast is real, but so is the cost of suppressing the friction that would otherwise make it visible.
The "Eastern Data, Western Computing" plan is the global AI buildout with one variable removed: local political permission. The West treats that variable as a constraint. China treats it as an input.
The Jamestown analysis reads the plan as energy-AI coordination rather than pure industrial policy: the grid routes eastern demand to western supply, and the western provinces get a compute corridor in return. That framing separates two claims AFP's dispatch fuses, that China is building more and that the West is building less. The first is a function of permits and capital. The second is also a function of permits, with a different answer attached.
Two caveats. The "nearly double" figure in the AFP dispatch traces to wire reporting; it would land harder against a primary capacity dataset, and the visible material is sufficient to angle on but not to assert quantitatively. And the Western slowdown is not solely a story of obstruction: the local concerns over water and grid load that have stalled projects are real engineering questions, not just politics. Treating them as politics alone would flatter the Chinese model and miss why the Western answer is slower.
The plan's five-year test, by 2031, is whether a top-down siting regime can absorb a doubling of national compute without producing the political friction the West has surfaced in microcosm. The first five years suggest it can. The next five will show whether the local costs have a quieter way of surfacing, or whether the absence of public friction is itself the result.