Ulanqab: The 'Token Capital' Worth Investigating

I spent the weekend digging into the data chain for Ulanqab, the "Token Capital," and hit quite a few pitfalls.

This article from TMTPost describes how a small northern city became a computing hotspot thanks to data centers. Details like food trucks, workers, and chicken with tofu skin really paint a picture. But when picking topics, you can't just stop at the visuals. Last week I wrote an article breaking down computing power news into a fact checklist; this time I wanted to see if applying the same method to Ulanqab would be faster.

I started by doing it manually. I opened government plans, press releases, and corporate contract reports, separating signed contracts, projects under construction, operational ones, and planning targets. It took about two hours and only yielded around twenty usable data points, with lots of repetitive news. The hassle wasn't in searching, but in confirming timelines and entities.

Later I switched to a search agent. I've been using this tool for two weeks; it's good for scraping public web pages first. I asked it to grab sources based on Ulanqab's three-year action plan for data center signing and production, then fed the results to Codex for entity extraction. I've used Codex for a month; it breaks down projects, investors, and target numbers into short sentences, which is indeed convenient. My tests showed that candidate information was significantly more abundant than manual collection, but it mixed in plenty of marketing-speak like "$500 billion investment" or "comparable to a small nuclear power plant."

The key difference is that manual work is slow but better at spotting contradictions in metrics; tools are fast but tend to compress plans, signings, and targets into the same layer of "facts." For example, distance to Beijing: some say ~350 km, others say 300 km. Revenue targets exceeding 8 billion yuan and total computing scale surpassing 200,000 P are planning figures, not achieved realities. Electricity prices at 0.33 yuan, 10 months of natural air cooling per year, and 67% green electricity are cost advantage conditions; you can't directly infer that all big tech companies save 30% on electricity.

This news story is worth running through the data chain once. More specifically, AI infrastructure competition has shifted from model parameters to electricity, land, climate, and scheduling costs. Ulanqab's opportunity didn't appear out of nowhere; low temperatures, electricity prices, green energy, and location have long been there. They were ignored in the past but are now being repriced by computing demand.

This approach suits industry researchers, tech editors, and those doing pre-investment due diligence. Using a search agent plus Codex to build a draft saves a lot of time flipping through web pages. However, don't use it as direct evidence of successful urban transformation. Especially, don't interpret DeepSeek's 1GW computing center or Huawei/Alibaba/Apple/Kuaishou signings as fully implemented.

Let the tools run on public sources first, then manually label each piece of info as implemented, under construction, planned, or rumor. Any sentence containing "target," "breakthrough," or "plan" should go in a separate column. Without this step, tools will just organize the hype faster.