According to Reuters, China generated over 10,000 terawatt-hours of electricity in 2024, more than double the United States, with renewables expected to supply 5,500 TWh by 2030. This “electron gap” has led Nvidia CEO Jensen Huang to warn that China could win the AI race due to lower energy costs. However, the country is lagging in actually building data centers, with firms expected to spend just $147 billion on AI capex in 2027—less than Amazon’s projected total. Meanwhile, a government plan to relocate data centers west for cheaper power has created a glut of facilities with utilization as low as 20%, and AI startups like Zhipu are reporting massive losses despite a booming market forecast.
The Power Paradox
On paper, China‘s energy position is enviable. They’ve got the megawatts. The U.S. grid is aging and facing a projected 44-gigawatt shortfall for data centers, while China is doubling down on wind and solar. Industrial power is 30% cheaper there. That should be a massive tailwind for training energy-hungry AI models, maybe even letting them brute-force their way around U.S. chip restrictions by throwing more, less-efficient domestic chips at the problem.
But here’s the thing: having power and being able to use it where you need it are two completely different challenges. Most of China’s shiny new renewable capacity is out west, in remote areas. The AI workloads, the tech companies, the manufacturing—they’re in the east. Transmitting that power across the country is a huge bottleneck. In some regions like Tibet, they’re curtailing—basically wasting—over 30% of solar and wind power because the grid can’t handle it. So that theoretical advantage? It’s stuck in the desert.
The Data Center Glut
So, officials had a plan: if you can’t bring the power to the data, bring the data to the power. The “Eastern Data, Western Compute” initiative sparked a construction boom in places like Inner Mongolia. The idea was to build data centers where the juice is cheap and plentiful and beam the data back on fiber.
It hasn’t worked. The transfer speeds are too slow for the real-time processing that advanced AI needs. Now, Reuters reports there’s a surplus of basically useless data centers sitting with utilization rates as low as 20%. That’s a brutal misallocation of capital. And it points to a bigger issue: when you’re building critical industrial computing infrastructure, you can’t just chase the cheapest kilowatt-hour. You need the right location, the right connectivity, and the right hardware. It’s a systems problem, not just a power bill problem.
Involution On The Horizon
This feels familiar, doesn’t it? The data center surplus might just be the first sign of a broader overcapacity issue creeping through China’s AI stack. Bernstein analysts think the supply of domestic chips for AI inference will exceed demand by 2028. In the software layer, companies like Alibaba and ByteDance are already in a race to the bottom on model pricing.
Look at Zhipu AI. Their prospectus paints a rosy picture of a market growing to nearly $142 billion by 2030. Yet in the first half of 2025, they lost 12 times more money than they made. Now, officials are warning of a potential bubble in humanoid robotics, with over 150 manufacturers chasing unproven tech. This is “involution”—the destructive, deflationary competition that has plagued sectors from solar panels to EVs. It wins market share through sheer volume and low prices, but it crushes profitability and can stifle real innovation. Is that the future of Chinese AI?
Buying Time Or Building A Bubble?
The U.S. power crunch might buy Chinese firms some time to catch up on the chip front. But time alone doesn’t solve structural problems. The real risk is that China’s AI push follows the same script: massive state-directed investment, a frenzy of building, followed by a glut, price wars, and poor returns.
Jensen Huang might be right about the electron gap, but he’s missing the bigger picture. Winning in AI isn’t just about who has the most electricity or even the most data centers. It’s about efficiently connecting cutting-edge compute, sustainable power, and viable business models. Right now, China has a surplus of the first two ingredients and a dangerous shortage of the third. That’s a recipe for another boom-and-bust cycle, not a guaranteed path to leadership.
