Who would have thought that Ulanqab would also benefit from the AI dividend.
Recently, I saw that Goldman Sachs, Bloomberg, and Wired all mentioned Ulanqab when discussing China's AI.
They say this small city in Inner Mongolia is becoming a major hub for computing power in China's AI infrastructure landscape.
Because AI competition ultimately boils down to competition over energy.
You see, Musk has been talking about space data centers all year, one very important reason being that the electricity required by future AI will be enormous, and by moving to space, it can directly utilize a large amount of solar energy.
OpenAI is no exception, with Sam Altman directly stating that the essence of their company is to convert electricity into useful intelligence.
Why Ulanqab?
Because Ulanqab meets several conditions: 1) the weather is cold, 2) electricity is cheap, 3) there is an abundance of green energy, and 4) it is close to Beijing.
I'm very familiar with Ulanqab, and I always visit Huanghuagou there during summer breaks, so I must write about it in detail.
To put it this way, Ulanqab has basically two seasons all year round, one is winter and the other is summer.
Moreover, the winters are extremely long. I checked the data, and the average annual temperature is only 4.3°C.
I recall going to the grasslands of Ulanqab in mid-September one year, and being frozen to the point of shivering at night.
I asked a friend, and he said that Ulanqab usually starts preparing for heating soon after the National Day holiday, which continues until mid-April the following year, lasting almost half a year.
The weather is truly hot for only two months, July and August.
"I top of a top of a top of a " ( it's a just only minimal, ":(." ( " " ( "."."s."."."."."."."."."."."."."."."."."."."."."." "."."."."."."."."."."."."."."."."."."." this "." "." "."."."."."."."."." that."."."."."."."."."."."."."."."."."."."."."."."."."."."." "." "." " that." "." "." "." "." "."." "." "."."."."."."."."."."."."." "." "." "." " not only "." "." " the aori,"." "." " "." " " " "." " "." " "." " some only place,"." "." " "." " " " "." " "." " "." " " " "." " " "." " a only minor, "." " "." " "." "."." "." "." " a only minor, "." "." " a only minor, " this a of a
The cold and windy conditions may not be comfortable for the local residents living in Ulanqab, but for data centers, these are all advantages. The cold weather allows them to utilize natural cooling sources for most of the year, reducing heat dissipation costs.
The wind is strong, which also means that the area has extremely good wind power conditions.
Similarly, a blast of cold air can help cool down data centers while also generating electricity. Hahaha.
Moreover, the electricity here is also cheap.
Inner Mongolia is originally a major energy-producing province, with abundant wind power, photovoltaics, and sufficient coal-fired power.
Like Ulanqab, the combined power generation from wind and photovoltaic energy is expected to account for nearly half of the city's total power generation by 2025, and has continued to increase in recent years.
When it comes to the scale of AI data centers, just how important is the price of electricity? In our daily lives, we might think that a difference of a few cents in the cost of a unit of electricity is negligible. However, for GW-level data centers, this is not the case at all.
Let me do a rough calculation for everyone.
Assuming an industrial park has an average load of 1GW, with an average load factor of 90%, its annual power consumption would be approximately 78.84 billion kWh.
In August 2026, the agency purchase price of electricity for Mengxi Power Grid was approximately 0.3234 yuan per kilowatt-hour, while that of Jibei was around 0.3857 yuan per kilowatt-hour, with a difference of just over 6 cents.
But it's just this 6 cents or so. Calculated at 7.884 billion kWh, that's approximately 491 million yuan per year. In other words, at the GW level, saving a few cents per kWh translates to hundreds of millions of yuan per year.
Moreover, as mentioned earlier, Ulanqab has a significant advantage, which is an abundance of green electricity.
Green electricity, in simple terms, refers to the power generated from renewable energy sources such as wind power and photovoltaics. Why is this matter becoming increasingly important for AI data centers? It's the same calculation as before - a 1GW park consumes nearly 8 billion degrees of electricity per year.
If a large amount of electricity relies on coal-fired power, the resulting carbon emissions will be extremely high.
Nowadays, major global companies are paying close attention to carbon emissions, and when companies like Google and Microsoft build data centers, they are also very mindful of the proportion of electricity that comes from clean energy sources such as wind power and photovoltaics.
It can even be predicted that in the future, when discussing the cost of a model, in addition to the cost of tokens, people may also start to pay attention to the energy consumption and carbon emissions behind these tokens.
At this point, some people may ask, what about places like Ningxia Zhongwei and Xinjiang, where electricity is also very cheap? The wind and photovoltaic resources in those areas may even be more abundant than in Ulanqab.
This is related to another advantage of Ulanqab. For those who often take the train, they may have heard Ulanqab's city slogan: one step west of Beijing is Ulanqab.
It is only about 350 kilometers away from Beijing, and can be reached by high-speed rail in approximately two hours.
Beijing itself is one of the most densely concentrated areas for domestic AI companies, research institutions, and internet companies. If you continue to move computing power to Ningxia and Xinjiang, electricity may be even cheaper and land more abundant, but network latency will also increase.
When working on cloud computing in the past, western data centers mainly undertook tasks such as storage backup and offline computing, which were not time-sensitive, and a delay of several dozen milliseconds had virtually no impact.
Training large models is similar, once the training task is underway, it takes several days or weeks, and is not sensitive to the slight latency from Beijing to the data center.
However, AI is now using increasingly more computing power for inference. When an agent executes a task, it constantly needs to call models and various services in the middle. These calculations, which directly face users, are extremely sensitive to latency.
This is also the most significant advantage Ulanqab has over Ningxia and Xinjiang, as electricity has become sufficiently cheap and it is only a few hundred kilometers away from Beijing.
It was found that Ulanqab has been building a direct optical cable to Beijing since 2017, and later constructed a second one.
According to publicly available data, the network latency between Ulanqab and Beijing can be as low as 5 milliseconds, with one route achieving approximately 4.2 milliseconds.
Ulanqab's advantages are thus particularly timely.
Going west, the electricity in Ningxia and Xinjiang may be even cheaper, but it is farther away from Beijing. Going east, the distance to Beijing is shorter, but the costs of electricity and land are higher.
I had just excitedly told my friend from Ulanqab about these things, and he only asked me one question:
How many jobs can this bring to Ulanqab?
To be honest, the job opportunities that a computing power center can bring are extremely limited.