The US is fighting over data centers. China is building them. Fast

The area appears like a high-tech mirage. Block after block of hulking data centers – both new and under construction – rise up seemingly out of nowhere from rolling plains and fields of sunflowers and sorghum.

On their sides are some of the biggest names in Chinese tech – hardware giant Huawei, social media platform Kuaishou, service provider VNET – all packed into a zone thrumming with bustle and movement.

This is Ulanqab, Inner Mongolia, a Chinese city on the edge of the Gobi desert. It’s fast becoming one of the data hubs fueling a massive AI boom in China that’s made the country the only serious rival to the US’ ambitions to dominate the technology. Nearly 90 data centers are slated to be built in this city of fewer than two million people.

As of last year, the US had roughly 10 times more AI data centers than China, according to Stanford University research – and vastly more computing power, other estimates show . American tech firms have something China’s don’t: access to the most advanced chips, the single most important ingredient for developing AI.

But Ulanqab – with its growing number of AI and big-data centers – showcases the advantages on China’s side in the race for the technology that could define the 21st century.

The towering units here were constructed at breakneck speed, with most of the densely packed Bayin Big Data Industrial Park taking shape in just two years, testament to China’s capacity to build fast.

Heavy winds, beating sun and wide, open spaces give the area plentiful access to cheap green energy and land.

And, crucially, throughout the Communist Party-run country, public opposition to the mass construction of data centers – the kind that is making headlines in the US – is non-existent.

At Ulanqab’s Bayin industrial park on a recent weekday, a steady stream of construction workers had their faces scanned as they filed past guards to enter building sites, while trucks rumbled along, feeding supplies to the facilities taking shape within. Outside their perimeters, vendors hawked noodles and cigarettes from makeshift stalls.

Qilai Shen/Bloomberg/Getty Images

The din of construction didn’t appear unwelcome to locals living nearby. Some who spoke to CNN said they were benefiting from the projects – and the influx of migrant construction workers that came with them to an area whose economy was otherwise largely reliant on agriculture.

“Before, we’d just sit there and wait for customers. Now we’re constantly busy – picking up orders, doing this and that. We barely get a break,” said Ma Rong, 56, who has run a grocery store and guest house in the area for the past 30 years. “Our business has gotten much better than before,” he said.

Some have been uprooted by changes sweeping the region, like restaurant owner Liu Baoping, 58, who said he had to leave his original business location due to the demolition of a highway. But he describes his daily clientele as doubling alongside the data-center buildout. “The environment isn’t as clean as before, but people’s incomes and the economy – it’s way better,” he said.

Ulanqab is part of Beijing’s vision to rapidly scale up data infrastructure – a key priority for a government that sees AI as a crucial engine of growth for the country’s slowing economy with its shrinking population.

“China has embraced AI with open arms,” Chinese leader Xi Jinping told a global artificial intelligence conference in Shanghai this past July. He pointed to China’s initiative to inject AI into sectors from health to home goods and hailed its push to embed AI in manufacturing as a “shining hallmark of Chinese modernization.”

Already more than 700 million people use AI in China, according to official statistics. And the plan to weave AI across manufacturing, industry and daily life also requires massive amounts of compute.

Beijing last month announced a $560 billion plan to rapidly scale up information infrastructure, including computing facilities, over the next five years. Private tech companies are also ramping up investment in AI infrastructure, though they still trail their Silicon Valley peers by a longshot.

Such investment from listed Chinese companies will double this year to around $139 billion, according to research firm Rhodium Group. But that is still less than one-fifth of the $800 billion in AI data center investment from their US peers.

And that’s not the only bottleneck.

Chinese firms face another critical challenge: limits on the amount of high-end hardware available to fill these centers. Washington’s export bans restrict their access to the best AI chips, like certain models from Nvidia and AMD, while Beijing has restricted imports of some less powerful options. Domestic chipmakers are racing to catch up, but their tech still lags the US – and supplies of their cutting-edge models are tight.

“If you look at the overall production cost of building a data center, chips are the most important factor, and they just don’t have enough,” said Nick Corvino, a researcher at the Carnegie Endowment for International Peace.

But what China lacks in chips and capital, it’s trying make up for with an advantage: ample, affordable power, much of it green.

While energy-hungry data centers are straining an aging US grid that’s not expanded significantly in decades, China between 2022 and 2025 added capacity equivalent to that of industrial power sources across the entire US as of the end of 2025.

And it has more green power than it can use: the country wasted enough excess green energy in the first six months of 2026 to power the entire country of Mexico for a year, findings from independent research organizations the Centre for Research on Energy and Clean Air and Global Energy Monitor suggest .

Beijing wants data centers to sop up that excess – much of it in western, rural areas where sun, wind and land are plentiful – while boosting business for its green tech sector, under a plan known as “Eastern data, Western compute.”

Under that strategy, first launched in 2022, China has designated eight national computing hubs – including several in its west – and 10 national data center clusters. Its latest five-year national economic blueprint elevated the linkage further, calling for pairing the build-out of computing infrastructure and green energy.

“China’s trying to take advantage of the fact that it has this muscle memory for building energy (infrastructure) quickly,” said Kyle Chan, a fellow at the Brookings Institution. “Whereas for the US, we haven’t done this kind of grid buildout in a long time and so we are struggling on the power generation and grid connection front.”

Abundant power can also help China make up for its less powerful domestic chips. Chinese chip designers like Huawei are combining clusters of chips to create the computing power of superior tech. The fix can be effective for some tasks, analysts say, but uses significantly more electricity.

Observers say that even with ample power, Beijing’s vision has its limits.

Some computing hubs in the west are too far away to handle certain tasks at the speeds needed for urban hubs in the densely-populated east. There’s also concern about so-called “ghost data centers” or ill-conceived projects in rural areas, built to fit a topline government initiative, yet lacking the hardware or users to justify the investment.

But in hubs in Inner Mongolia, where optical cables can transmit data to Beijing – some 200 miles away – faster than the blink of an eye, and top players are pouring major resources, China’s vision appears to be taking shape.

Long one of China’s top coal producers, the wind-swept region bordering Russia and Mongolia has more recently become a hub for green energy.

Fields of solar panels blanket parts of Inner Mongolia, while the slender silhouettes of wind turbines spin above grazing sheep. Along some stretches of highway unused turbine blades line the roadsides like giant, fiberglass elephant tusks.

In Ulanqab, which falls within one of China’s 10 national data center clusters, nearly half the grid draws from wind and solar – and some data center providers are finding ways to rely even more heavily on clean sources.

Several miles from the thrumming construction of Ulanqab’s Bayin industrial park, the Envision “Galaxy Campus” is running one such model.

Spanning more than 20 football fields and slated to be among the world’s largest by area when completed, this data center will be home to servers and chips from tech companies, used to handle queries users input into AI apps in places like Beijing.

Upwards of 80% of the power for the facility will come directly from wind farms operated by the company, bypassing the grid completely, Envision says.

Battery systems then store wind power, while a cutting-edge solid-state transformer can rapidly adjust how much power is feeding the servers, based on demand, according to Ricky Zheng, the facility’s general manager, who showed CNN around the under-construction campus on a recent day.

Water is another key resource that’s sparked backlash against data centers in America, but Zheng says it’s not an issue here. The chilly climate in Ulanqab reduces the need for cooling overall, while the facility is running a so-called “zero-water” system that uses industrial alcohol as coolant, he says.

Zheng also seemed optimistic when it comes to demand from tech companies for their facilities – and those clients’ ability to kit out server racks with needed hardware.

The first phase of the planned total 2GW facility was already fully booked, he said, and the company was building additional phases, both for AI and cloud services.

AI chips and servers “are owned by our off-takers,” he said. “When they give us the order to build such a big data center … they were confident to have enough infrastructure to be installed in these (server) racks.”

And this company, like Ulanqab itself, is planning to be ready for an expected surge in AI demand in the coming years.

“We designed this for the next 10 years,” said Zheng. “You need to get prepared for the future.”

CNN’s Joyce Jiang contributed to this report.

这片区域宛如一个高科技海市蜃楼。一栋栋庞大的数据中心——既有新建的，也有正在建设中的——仿佛凭空从绵延的平原和向日葵、高粱田中拔地而起。

它们两侧聚集着中国科技界的一些巨头——硬件巨头华为、社交媒体平台快手、服务提供商VNET——所有这些公司都挤在一个熙熙攘攘、充满活力的区域。

这里是内蒙古乌兰察布，一座位于戈壁沙漠边缘的中国城市。它正迅速成为推动中国人工智能蓬勃发展的数据中心之一，这场热潮使中国成为唯一能与美国在人工智能技术领域的雄心壮志相抗衡的国家。在这座人口不足两百万的城市里，计划兴建近90座数据中心。

根据斯坦福大学的研究，截至去年，美国的AI数据中心数量大约是中国的10倍；其他估算也显示，美国的计算能力远超中国。美国科技公司拥有中国公司所不具备的优势：获得最先进的芯片，而芯片是开发人工智能最重要的要素。

但乌兰察布凭借其不断增长的人工智能和大数据中心数量，展现了中国在可能定义 21 世纪的技术竞赛中的优势。

这里高耸的建筑群以惊人的速度建成，巴音大数据产业园的大部分密集建筑在短短两年内就建成，这证明了中国的快速建设能力。

强风、烈日和广阔的开放空间使该地区能够获得充足的廉价绿色能源和土地。

而且，至关重要的是，在整个共产党执政的国家，公众对大规模建设数据中心（这种数据中心在美国引起了广泛关注）的反对声音并不存在。

最近一个工作日，在乌兰察布巴因工业园区，络绎不绝的建筑工人在进入工地前都要接受人脸扫描，卡车隆隆驶过，为正在建设中的设施运送物资。园区外，小贩们在临时搭建的摊位上兜售面条和香烟。

沈启来/彭博社/盖蒂图片社

施工的喧嚣似乎并未让附近的居民感到不快。一些接受CNN采访的居民表示，他们正从这些项目以及随之而来的大量外来建筑工人中受益，而该地区的经济原本主要依赖农业。

“以前我们只是坐在那里等顾客。现在我们忙得不可开交——取货、做这做那，几乎没有休息时间，”56岁的马荣说道。过去30年来，他一直在该地区经营一家杂货店和旅馆。“我们的生意比以前好多了，”他补充道。

一些居民因席卷该地区的巨变而被迫搬迁，例如58岁的餐馆老板刘宝平。他说，由于高速公路拆除，他不得不离开原来的店面。但他表示，随着数据中心的建设，他的日均客流量翻了一番。“环境不如以前干净了，但人们的收入和经济状况——比以前好多了。”他说。

乌兰察布是北京快速扩大数据基础设施规模愿景的一部分——对于一个将人工智能视为国家经济增长关键引擎，以应对人口减少和经济增长放缓的国家政府而言，这是一项关键优先事项。

“中国张开双臂拥抱人工智能，”中国领导人习近平今年7月在上海举行的全球人工智能大会上表示。他指出，中国已将人工智能应用于从医疗健康到家居用品等各个领域，并称赞中国推动人工智能融入制造业是“中国现代化的闪亮标志”。

官方数据显示，中国已有超过7亿人使用人工智能。而将人工智能融入制造业、工业和日常生活的计划也需要海量的计算能力。

北京上个月宣布了一项5600亿美元的计划，旨在未来五年内快速扩大信息基础设施规模，包括计算设施。私营科技公司也在加大对人工智能基础设施的投资，尽管它们与硅谷同行相比仍有相当大的差距。

据市场研究公司荣鼎集团（Rhodium Group）预测，今年中国上市公司的此类投资将翻一番，达到约1390亿美元。但这仍然不到美国同行在人工智能数据中心领域8000亿美元投资额的五分之一。

但这并非唯一的瓶颈。

中国企业还面临另一项严峻挑战：用于填充这些中心的高端硬件数量有限。华盛顿的出口禁令限制了它们获取最先进的人工智能芯片，例如英伟达和AMD的某些型号，而北京也限制进口一些性能较低的芯片。国内芯片制造商正竞相追赶，但它们的技术仍然落后于美国——而且其尖端芯片的供应也十分紧张。

“如果你看一下建造数据中心的整体生产成本，芯片是最重要的因素，而他们却没有足够的芯片，”卡内基国际和平基金会的研究员尼克·科尔维诺说。

但中国在芯片和资金方面的不足，正试图用一项优势来弥补：充足且价格低廉的电力，其中大部分是绿色电力。

虽然耗能巨大的数据中心给几十年来没有大幅扩容的老旧美国电网带来了巨大压力，但中国在 2022 年至 2025 年间新增的装机容量相当于到 2025 年底美国所有工业用电设备的装机容量。

该国拥有的绿色能源比它能使用的还要多：独立研究机构能源与清洁空气研究中心和全球能源监测机构的调查结果表明，该国在 2026 年上半年浪费的过剩绿色能源足以供整个墨西哥国使用一年。

北京希望数据中心能够吸收这些过剩的数据——其中大部分位于阳光、风能和土地资源丰富的西部农村地区——同时促进其绿色科技产业的发展，这项计划被称为“东方数据，西方计算”。

根据这项于2022年首次推出的战略，中国已指定了八个国家级计算中心（包括西部地区的几个中心）和十个国家级数据中心集群。其最新的五年国家经济发展规划进一步提升了这一联系，要求将计算基础设施建设与绿色能源建设相结合。

布鲁金斯学会研究员陈凯尔表示：“中国正试图利用其在快速建设能源（基础设施）方面积累的‘肌肉记忆’。而美国已经很久没有进行过这种规模的电网建设了，因此我们在发电和电网连接方面举步维艰。”

充足的电力也能帮助中国弥补国产芯片性能不足的缺陷。像华为这样的中国芯片设计公司正在将多个芯片组合在一起，以达到媲美先进技术的计算能力。分析人士指出，这种方法在某些任务上可能有效，但会显著增加电力消耗。

观察人士表示，即使拥有强大的实力，北京的视野也是有限的。

西部的一些计算中心距离东部人口稠密的城市中心太远，无法以所需的速度处理某些任务。此外，人们还担心所谓的“幽灵数据中心”，即位于农村地区、为迎合政府总体规划而建设的、但缺乏硬件或用户支持、因而无法证明投资合理性的欠妥项目。

但在内蒙古的枢纽地带，光缆可以将数据传输到约 200 英里外的北京，速度快得惊人，而且顶尖企业正在投入大量资源，中国的愿景似乎正在逐步实现。

这个与俄罗斯和蒙古接壤、风力强劲的地区长期以来一直是中国的煤炭主要生产地之一，近年来已成为绿色能源中心。

内蒙古部分地区遍布太阳能电池板，纤细的风力涡轮机在吃草的羊群上方旋转。在一些高速公路路段，废弃的涡轮机叶片像巨大的玻璃纤维象牙一样排列在路边。

乌兰察布位于中国十大国家数据中心集群之一，该市近一半的电网来自风能和太阳能——一些数据中心提供商正在寻找方法，以更加依赖清洁能源。

距离乌兰察布巴因工业园熙熙攘攘的建设工地几英里远的地方，远景集团的“银河园区”正在运行着这样一个模式。

该数据中心占地超过 20 个足球场，建成后预计将成为世界上面积最大的数据中心之一。它将容纳来自科技公司的服务器和芯片，用于处理用户在北京等地的 AI 应用程序中输入的查询。

Envision公司表示，该设施超过80%的电力将直接来自该公司运营的风力发电场，完全绕过电网。

据该设施总经理郑瑞奇（Ricky Zheng）介绍，电池系统可以存储风能，而尖端的固态变压器可以根据需求快速调节向服务器供电的电量。郑瑞奇最近带领 CNN 参观了正在建设中的园区。

水资源是美国数据中心引发强烈反对的另一个关键因素，但郑先生表示，在乌兰察布这不是问题。他说，乌兰察布寒冷的气候降低了整体冷却需求，而且该设施运行的是所谓的“零水”系统，使用工业酒精作为冷却剂。

郑先生对科技公司对其设施的需求，以及这些客户为服务器机架配备所需硬件的能力，似乎也持乐观态度。

他表示，计划总装机容量为 2GW 的设施的第一阶段已经全部预订完毕，公司正在建设其他阶段，用于人工智能和云服务。

他说：“人工智能芯片和服务器的所有权归我们的客户所有。当他们下令建造如此庞大的数据中心时……他们相信我们拥有足够的基础设施来安装在这些（服务器）机架中。”

这家公司和 Ulanqab 本身一样，正计划为未来几年人工智能需求的预期激增做好准备。

郑先生说：“我们设计的这个方案着眼于未来十年。你们需要为未来做好准备。”

CNN的Joyce Jiang对本报道亦有贡献。