Product Not yet confirmed elsewhere 1 publisher2 min readPublished Aug 21, 2026, 6:45 PM PDT

Ulanqab has pledged 12.5GW of AI data centers, and its water goes off at night

Cold winters, cheap coal-and-wind power and spare land are deciding where China's AI compute lands. In Ulanqab, water is the item nobody put on the qualifying list.

The Product Desk

How we use AI Send a correction

What happened

Chinese companies have pledged data centers totalling an estimated 12.5 gigawatts in Ulanqab, Inner Mongolia, according to a Goldman Sachs research note.
More than 70 percent of those capacity commitments were announced in just the past year.
DeepSeek is reported to be building a large facility in the city, alongside ByteDance, Alibaba and Xiaohongshu.
Last month the city's water company shut several waterworks for seven hours each night to blunt peak demand.
Two dedicated fiber cables built in 2017 and 2019 cut average latency below five milliseconds, enough to serve inference and not just training.

Why it matters

constraint Cooling water, not accelerators or megawatts, is the ceiling here: a municipality that cannot already cover resident demand has nothing spare to allocate to a queue of new load.
capability Sub-five-millisecond fiber means the cheap-power hinterland can host live inference for eastern users, so low-cost siting is no longer reserved for latency-indifferent training jobs.
cost Residents carry the first bill, in overnight supply interruptions, for load whose owners' cooling draw is concentrated in the warm months.
precedent If winter temperature, power price and land are what qualify a site, the next round of capacity announcements will be read against grid and water permits rather than chip counts.

Goldman Sachs is counting pledges, and pledges are cheap ^[8]. The pace inside its note is the harder number: roughly 8.75 gigawatts of that capacity was announced in a single twelve-month window ^[24], in a city of about 1.5 million people ^[6]. That is around 8.3 kilowatts of promised capacity per resident ^[25]. The total sits 25 percent above the 10 gigawatts OpenAI's Stargate project expects to reach at completion ^[23], and Stargate carries a $500 billion price tag ^[1].

None of that was decided by silicon. Ulanqab sits high on the Inner Mongolian Plateau with long cold winters, so cooling draws less power ^[11], and electricity in Inner Mongolia is cheaper than almost anywhere else in China, the product of fast wind and solar growth alongside abundant coal ^[12]. Huawei built the city's first data center in 2016 and Apple followed three years later ^[17]; in 2021 the region was named a main hub of the state's "Eastern Data, Western Compute" program ^[18]. For years the distance from the populous coast left those buildings doing backup storage ^[19]. Andrew Stokols of Singapore Management University told Wired that the rise of AI in 2022 brought the realization that remote sites could be well used for model training ^[20], which runs for months and tolerates latency ^[21].

So the qualifying list is cold air, cheap electrons, fiber and land. Water is not on it. Ulanqab gets about 14 inches of rain a year, roughly as dry as Denver ^[13], and the local government already cannot supply enough water for residents, before most of the announced projects are running ^[14]. The defense of the buildout is seasonal: weather data from the local government shows the centers need additional cooling water in only two months of the year ^[16]. That is the municipality's own figure, and it is the figure holding up the pledge column.

The ownership change is what turns a water question into a corporate one. Wired frames the Ulanqab spending as the first time Chinese AI companies have put real money into infrastructure they own rather than renting compute from cloud providers ^[2], after years of spending far less on physical plant than their American peers ^[4]. Renting is exposure to a price list. Owning the buildings is exposure to a municipal water table and a grid interconnection queue, in a jurisdiction where, Wired reports, growth has been driven more by commercial demand than by government direction, unlike other Chinese hubs ^[5]. Nearly 100 facilities have opened or broken ground there since 2016 ^[7], which means the operators arriving now are inheriting a water budget already spoken for.

What to watch

Whether the pledged gigawatts convert into energized capacity, and how far the commitment count drifts from what is actually running.
Any Ulanqab water quota, or a requirement for dry or closed-loop cooling, which would rewrite the cost case for siting there.
Whether other Eastern Data, Western Compute hubs start posting comparable pledge volumes, or Ulanqab stays the outlier.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence54
Adoption71
Hype gap+24
Incentives63
Confidence52

Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
OpenAI's $500 billion Stargate Project is set to reach 10 gigawatts of total capacity when complete.
ReportedSupported2 sources— create a free account to open them View cited source
[2]
For the first time, Chinese AI companies are making big investments in their own infrastructure rather than renting compute from cloud companies.
ReportedSupportedSource: Wired2 sources— create a free account to open them View cited source
[3]
DeepSeek is reportedly building a massive AI data center in Ulanqab, as are ByteDance, Alibaba and Xiaohongshu.
ReportedSupported2 sources— create a free account to open them View cited source
[4]
For years Chinese AI companies have spent far less on building physical infrastructure than their American peers, despite developing popular models with impressive capabilities.
ReportedSupported2 sources— create a free account to open them View cited source
[5]
Compared with other data center hubs in China, Ulanqab's growth appears to have been driven more by commercial demand than by government direction.
ReportedSupportedSource: Wired2 sources— create a free account to open them View cited source
[6]
Ulanqab is a city in Inner Mongolia home to about 1.5 million people.
ReportedSupportedView cited source
[7]
Nearly 100 data centers have been opened or begun construction in Ulanqab since 2016.
ReportedSupportedView cited source
[8]
Chinese companies have pledged projects in Ulanqab with a combined estimated capacity of 12.5 gigawatts, according to a Goldman Sachs research note published last week.
ReportedSupportedSource: Goldman Sachs research note, via WiredView cited source
[9]
Goldman Sachs describes Ulanqab as one of the fastest growing compute clusters in Asia.
ReportedSupportedSource: Goldman Sachs research note, via WiredView cited source
[10]
Over 70 percent of the total capacity commitments in Ulanqab have been announced in just the last year.
ReportedSupportedSource: Goldman Sachs research note, via WiredView cited source
[11]
Ulanqab sits at high elevation on the Inner Mongolian Plateau and has long, cold winters, so data centers there do not need to use as much energy to stay cool.
ReportedSupportedView cited source
[12]
Electricity is cheaper in Inner Mongolia than almost anywhere else in China, driven by strong growth in wind and solar energy and an abundant supply of coal.
ReportedSupportedView cited source
[13]
Ulanqab is about as dry as Denver, getting only roughly 14 inches of rain each year.
ReportedSupportedView cited source
[14]
The local government in Ulanqab is already struggling to provide enough water to meet resident demand, before many of the planned data center projects are up and running.
ReportedSupportedView cited source
[15]
Last month the local water company in Ulanqab was forced to turn off several waterworks for seven hours each night to mitigate peak demand.
ReportedSupportedView cited source
[16]
Weather data from the local government of Ulanqab shows the data centers require additional water for cooling during only two months of the year.
ReportedSupportedSource: local government of Ulanqab, via WiredView cited source
[17]
Huawei built its first data center in Ulanqab in 2016, and Apple followed three years later.
ReportedSupportedView cited source
[18]
In 2021 the area was designated one of the main hubs of the Chinese government project dubbed "Eastern Data, Western Compute", which aims to build data centers in China's western hinterlands.
ReportedSupportedView cited source
[19]
Because they sit far from China's populous eastern coast, these data centers initially faced high latency when transferring data to most users and were largely relegated to backup storage.
ReportedSupportedView cited source
[20]
Andrew Stokols, a professor at Singapore Management University who studies China's compute infrastructure, said that with the rise of AI in 2022 there was the realization that remote data centers could be well-utilized for model training.
ReportedSupportedSource: Andrew Stokols, Singapore Management University, quoted by WiredView cited source
[21]
A training run for an AI model can take months and does not require much real-time tinkering, so latency is less of an issue.
ReportedSupportedView cited source
[22]
Ulanqab is connected by two dedicated fiber optic cables built in 2017 and 2019 that reduced average latency to less than five milliseconds, fast enough to support real-time data exchanges such as AI inference.
ReportedSupportedView cited source
[23]
Ulanqab's 12.5 gigawatts of pledged capacity is 2.5 gigawatts, or 25 percent, more than Stargate's 10 gigawatt endpoint.
DerivedSupported2 sources— create a free account to open them
[24]
More than 70 percent of 12.5 gigawatts implies roughly 8.75 gigawatts of Ulanqab capacity was pledged within the past year.
DerivedSupported1 source— create a free account to open them
[25]
12.5 gigawatts spread across about 1.5 million residents is roughly 8.3 kilowatts of pledged capacity per resident.
DerivedSupported1 source— create a free account to open them

Sources

1 independent publisher whose own reporting we read for this story.

wired.com
1 article · August 21, 2026
The Unlikely Place at the Center of China’s AI Boom

Ask Clarity

Have a question this piece left open? The answer comes from its reporting and claim ledger — nothing else.

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

Builder

For builders the usable signal is workload placement economics: month-long training runs tolerate distance, so remote high-desert sites became viable in 2022, while two dedicated fiber cables from 2017 and 2019 pulled average latency under five milliseconds - the threshold that lets the same campus serve inference to eastern Chinese users. That is the architectural argument behind the migration from backup storage to frontier training and serving. What the single source does not give builders is cooling design, rack density, accelerator supply, or any operational capacity number, so the piece supports reasoning about siting logic rather than about system design.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence54: 0100
One outlet, decent primary anchors
Every claim traces to a single Wired article. That article does cite identifiable underlying evidence - a Goldman Sachs research note with a capacity tally, named experts Andrew Stokols and Damien Ma, local government weather data, a dated municipal water action and dated fiber builds - which lifts it above unsourced reporting. But there is no second publisher, no primary document is quoted directly, the flagship DeepSeek project is only 'reportedly' underway, and no operational capacity or water volume figure is provided to test the pledge total.
Adoption71: 0100
Physical buildout already at scale
Adoption here is measurable in concrete, dated facts rather than intent: nearly 100 facilities opened or under construction since 2016, anchor builds by Huawei in 2016 and Apple in 2019, a newly announced 2 GW Envision campus with dedicated clean power, and named AI firms building owned capacity. The score is held below the top band because the 12.5 GW headline is pledged rather than energized and no operational megawatt figure is disclosed.
Hype gap+24: −100 understated0 aligned+100 overstated
Pledges framed against a delivery target
Mildly overstated. The framing compares 12.5 GW of pledged, estimated capacity in one city against Stargate's 10 GW completion target, an announcement-versus-endpoint comparison, and over 70 percent of those pledges are under a year old with no disclosed energized capacity, chip supply or financing to back them. The gap is kept modest because the reporting itself supplies the deflators - the water rationing, the 37 percent coal share and the 'reportedly' hedge on DeepSeek - rather than only the boom narrative.
Incentives63: 0100
Sell-side tally plus announcement-driven actors
The headline capacity number originates in a Goldman Sachs research note, a sell-side product with an interest in AI infrastructure narratives, and the underlying pledges come from companies and a turbine manufacturer whose valuations benefit from announced scale. A national program, 'Eastern Data, Western Compute', and a stated government aim of absorbing excess renewable capacity add policy incentive to inflate committed capacity. Partly offset by two academic commentators with no disclosed stake and by municipal water data that works against the boom framing.
Confidence52: 0100
Directionally solid, numerically soft
Confidence is moderate. The direction of the story - a large, fast, commercially driven compute buildout in a water-scarce Inner Mongolian city - is well supported by dated construction facts and named experts. The specific magnitudes are single-sourced through one research note, mix pledged with built capacity, and lack any water-volume or operational-capacity disclosure that would let the central tension be sized.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

China AI InfrastructureFollow
Latency and Workload PlacementFollow
AI Data Center Buildout Follow
Compute Energy and SitingFollow
Data Center Water Use Follow

Entities

UlanqabFollow
Goldman Sachs Follow
Stargate Project Follow
OpenAI Follow
DeepSeek Follow
ByteDance Follow
Alibaba Follow
XiaohongshuFollow
Huawei Follow
Apple Follow
EnvisionFollow
Eastern Data, Western ComputeFollow
Moonshot AI Follow
Zhipu AI Follow
Andrew StokolsFollow
Singapore Management UniversityFollow
Damien MaFollow
Carnegie ChinaFollow