Chinas AI Empire Is Rising From Potato Fields

The New Power Grid Is Being Built on Grass

Ulanqab in China’s Inner Mongolia Autonomous Region is officially known as the potato capital. For years, that was enough to put it on the map — a dry steppe town where farmers harrowed fields and herded sheep across volcanic terrain. This year, something else has arrived.

Rows of vast rectangular data centers are going up alongside the potato fields. Hundreds of construction workers in yellow helmets move earth and weld steel. Cranes lift roof sections into place while crews balance on scaffolding. One worker joked that this is exactly what you’d expect from a country obsessed with infrastructure. Another described it as “step two” of a multi-phase buildout that will keep expanding.

The buildings are not for potato processing.

They are becoming the engine room of China’s artificial intelligence push. By 2030, Beijing wants AI embedded in 90 percent of industry and society. These centers are how it plans to get there.

Why Inner Mongolia

China isn’t building these data centers in Shanghai or Shenzhen, where land is expensive, power grids are strained, and cooling systems would guzzle water in a city already drinking itself dry. Instead, it is going north and west — to Ulanqab, Ningxia, Gansu, and Guizhou — places where the ground is cheap, the air is cool, and the land belongs to people who rarely make headlines.

These regions already produce surplus wind and solar power. The grid has room to absorb new loads. The climate cuts cooling costs. It is the same logic that drove Amazon and Microsoft to build data centers in rural Sweden and Oregon decades ago, just scaled up for an AI race that Beijing sees as existential.

Huawei, ByteDance, and DeepSeek are among the companies getting infrastructure in these regions. The state is directing it. The timeline is aggressive.

The Open-Source Gambit

While the US doubles down on export controls and keeps its largest models behind paywalls, China is pushing in the opposite direction. Xi Jinping has publicly encouraged domestic AI firms to develop cheaper, modifiable open-source models. The theory is straightforward: if everyone can build on top of everything else, innovation accelerates.

It also solves a problem that Washington loves to remind Beijing about — semiconductor shortages. China cannot easily buy the most advanced chips now that US export restrictions are tightening. Open-source models lower the barrier to entry. A company that cannot access H100 GPUs can still compete if it can run a well-tuned model on older hardware. That is the strategy, at least.

Anthropic CEO Dario Amodei warned last week that if China takes the lead in AI, it will pose “serious risks” to the US and the world. Trump agreed in principle, calling AI victory the only victory that matters. Xi, arriving in Washington this week, will hear those warnings and likely dismiss them as Cold War theater, as the Global Times already has.

The Hidden Costs

The data centers are not free. China still relies on coal for much of its electricity, and Inner Mongolia’s dry climate makes water a constraint for cooling systems. The government says most new centers will run on green energy by 2030, but that target is unverified.

Local communities bear a steeper price. Some villages have been relocated entirely. How many people moved, and what they received in compensation, is impossible to verify — the topic is too sensitive to discuss openly. One farmer spoke off-camera and asked simply: what does a giant data center do for someone like him?

Another farmer, checking his sheep near a construction site, said he barely uses the internet. His phone screen is cracked. He pays about 40 yuan a month for mobile data. AI is a foreign language to him. The future, as far as he is concerned, belongs to other people.

The Talent Shift

America’s immigration restrictions and the Trump administration’s visa policies have changed the flow of talent. Jack Zhang, who studied at Yale and Cambridge before returning to China through a state talent program, said the trend is clear: fewer Chinese students going to the US, more returning home. The American dream, he noted, has dimmed.

China produces roughly five million STEM graduates annually — ten times the US output. That pipeline is not drying up. It is converging. Young engineers are choosing domestic labs over Silicon Valley. A 27-year-old AI developer at a leading Chinese firm says campus culture is opening up, with researchers collaborating globally on everything from AI-assisted drug discovery to astrophysics.

Not everyone agrees. Liu Jun, a statistics professor at Tsinghua who has taught at Harvard and Stanford, says the system still rewards conformity over challenge. Students, he says, are told to question more — and then struggle to do it.

What Comes Next

The US-China AI summit in Washington this week will touch on safety and regulation. Beijing has called for a global governance framework that other countries could adopt. Washington has proposed stricter controls on chip sales and model exports. Neither side expects the other to slow down.

What is visible in Ulanqab suggests neither side is bluffing about scale. The data centers are rising from grassland faster than most infrastructure projects anywhere. The question is whether China’s open-source, distributed model can close the chip gap, or whether the next round of US restrictions will force a reckoning.

For the farmers still working the fields nearby, the answer may never matter. Their potatoes will keep growing. The machines will keep humming. And the world will argue about who wins the next phase of a race that none of them chose to run.