Inside China’s Race To Build The World’s Data Centers

We're at a factory in the trading city of Yangzhou that could build data centers in just a few months using a method called prefab construction. And all of these steel boxes are meant to build one three story AI data center. We can cut down the speed of deployment by at least half, and we can also control the quality right at the factory. The AI boom is driving massive demand for data centers worldwide, but building them is a challenge, particularly in the U.S., where red tape protests, soaring labor and construction costs, steep tariffs and supply shortages can delay production. It takes, on average, 2 to 3 years to build one in the U.S., but Chinese suppliers could significantly shorten the time frame. China is definitely very important towards the U.S. supply chain for data centers. At the moment, we're in a situation where we don't have what we need domestically, and if we do, it's going to be several year long lead times. China can offer those at a much more favorable timing. As the AI race between the US and China intensifies, it's unclear how or if the two countries will seek ways to collaborate. American policymakers are concerned about overreliance on China. Officials have looked to limit or outright ban Chinese made products and technology. National security is also a concern, and more generally, a recent push for an artificial intelligence slowdown could curb development. So this is the behind the scenes of the data center. We are constructing all the different building blocks of the data center, right in a factory in a controlled environment. This company, PrefabDC, is Chinese and the sole manufacturer of Singapore registered BrightRay, which manages the construction of the build outs. The data centers are built in modules, each one made with elements like cables and equipment. A typical mid-sized 50 megawatt data center is broken into about 1000 modules. This material and equipment are sourced from different parts of the world, and a very big portion come from China. Because China's supply chain is very integrated and comprehensive, that's the reason why we have the factory here in China. The prefabricated and modular construction method is common in China, building houses and data centers like you make cars. This company not only uses the method to make modules for the insides of the data centers, but for the entire data center itself. The building is done in a controlled environment where things like weather won't pose an issue. Outside of the weather, are there other problems that builders face when they're building up a data center? The on site constraint and also the different type of coordination on site where you have different equipment coming to the site, you will you will experience very complex coordination. And that is the one that slows down the entire construction of the data center. Skid-mounted and micro-modular systems are widely used internationally, but few companies build fully modular data halls. Between 40 and 60% of a US data centers. Individual parts are made of modules, some even more. Western conglomerates like Schneider Electric, Vertiv and ABB are some of the biggest international players in the space. Here a steel structure is built first. Then cooling equipment, ventilation or universal power supply is assembled and later added. Inside. Each module is different. One will have space for server racks, another will house the cooling system and still another will have the power utility equipment installed. Each module is labeled with its own specific building block number. That makes it easier to put together on site, similar to building furniture. So then these modules are going to be next to each other. Once they get onto the site. You will see this the exact module in the factory and you will be seeing the same thing on site. Then each module is tightly packed and later lifted onto a transport truck. Once they arrive on site, the modules are assembled by crane and snapped together like Lego blocks. BrightRay told us that from beginning to end, the construction takes nine months in Southeast Asia. That includes design, manufacturing, transportation, assembly and testing. The company tries to be price competitive, but argues its selling point isn't cost. Rather speed, reliability and connections to China's extensive supply chain. It wouldn't share how much its data centers cost in the U.S., prices vary, but can be on average, $17.6 million per megawatt for the most modern facilities. BrightRay currently has data centers in China and Southeast Asia, but the company hopes to expand in the U.S.. We can see very big potential in the US market and the market is equally important, but we can see that there's stronger demand in the US. The US had 5427 data centers in 2025, while China had 449. That difference is mainly due to greater access to capital. Total spend on building data centers in the US has grown five fold since late 2022. An estimate shows that US hyperscale investors will outspend China roughly six times on AI and cloud infrastructure in 2026, and total US private AI investment figures surpassed China by more than 23 times. I think Chinese companies just are not generating as much revenues from their AI services, and that's one of the reasons why they're not able to invest in these enormous data center build outs that their competitors in the US are doing. But in early June, it was reported that the Chinese government is investing $295 billion into building data centers over the next five years. That's still far from what the US private sector is investing. Time will tell in terms of how much of this money is actually allocated and dispersed. I think the deeper issue here is in the past, the Chinese government has supported the build out of a lot of data centers, but they've gone idle and unused, in part because there's not as much demand for the AI services. So while Chinese manufacturers face weak demand at home, they want to tap the huge demand in the United States. China is the world's largest manufacturer. Not only do its factories make all kinds of electronics like in this multi-story marketplace for tech, it also dominates the supply of electrical equipment. Huaqiangbei is the world's largest wholesale electronics market, located in the manufacturing tech hub of Shenzhen. Here, vendors hawk every conceivable electronics part you can think of. These are some of the fiber optic cable sellers in the market. Some told me they've raised prices by as much as eight fold products are sold worldwide. The. Government. In Shanghai. I met American Jean Esquier, the CEO of NGEN Mission critical. The company has been around for about a year and has offices in both the U.S. and China. It sources equipment for U.S. data center developers. Combining really the manufacturing supply chain engineering capacity on the Asia side with all of the services that we provide to our customers on the US side, that's what really bridges that gap. The company says products like transformers, switchboards, backup power systems and batteries can be made in China 2 to 3 times faster than in the US, and depending on the product, can be 10 to 30% cheaper. The benefit for the customers really is scale and and competitive pricing. You know, things are are made in China with incredible optimization and cost and also incredible engineering ingenuity. While the AI boom faces growing questions about its long term risks, the US and China are grappling with their reliance on each other. Both governments are discouraging their domestic companies from investing in the other. The US is considering bans on Chinese open source AI models and new types of Chinese data center components, as well as curbs on the sale to China of U.S. tech. We also have to have safeguards guardrails to ensure that those technologies, whether it's data centers, other access to our best technology does not fall into the hands of our adversaries. Chinese businesses and industry analysts believe that if the US doesn't buy from China, the American AI rollout could be set back. Yet other people in the US worry about cybersecurity risks and America, depending on the supply chain of its main AI competitor. What would you say to those people? Well, I can say that if that is the case in the AI industry in the US might be affected, there will be a longer timeline and that will affect the way business is done in us. That would add to the challenges the US is facing building new data centers. Backlash is growing and there have been nationwide protests over fears the centers will strain resources like water and energy. The opposition, when it comes to data centers, is certainly something that has to be considered at this point, mainly as it's tied to the affordability of the industry. I think it's inevitable that it's going to hinder the data center pipeline, especially if this sort of opposition continues over the coming years. The issue has also become a factor in the US midterm elections in November. President Trump has called out communities that oppose new developments, saying AI fears are a hoax and a scam. States like New York, Texas and Pennsylvania have put limitations on new build outs in the first quarter of 2026. At least 75 projects, worth about $130 billion were blocked or delayed. A number of industry leaders have called for an AI slowdown due to safety concerns. We've seen too many developers who don't care about our communities, who are trying to run roughshod over our communities, and those are the kinds of data center projects we don't want here. We want to work with people that want to work with local communities. China has its own problems. Beyond the shortage of capital, data centers need advanced semiconductors in large numbers, something the US restricted in 2022. The Biden administration cut off China from acquiring America's best in class AI and Nvidia chips. And though President Trump has loosened those controls, access is still a challenge. China is reportedly requiring that new state funded data centers must use domestically made AI chips. Those aren't as powerful as the ones made in the U.S.. While the AI demand accelerates, the U.S. and China are likely better off partnering. But tensions are high and the future of the AI boom uncertain. It's inevitable that we're going to be requiring Chinese built infrastructure unless we see increased investment domestically to really enhance the level of supply that we have. Factors that demand is increasing faster than the supply is still. And we're going to continue to see the import dependance.