Who will win the AI arms race? The U.S. is clearly miles ahead of China when it comes to the sophistication of its chips. But an Nvidia processor is only part of what's needed to adopt AI.
AI relies on real-world infrastructure: servers, data centers and stable power. That needs energy, land, and other resources, not just GPUs.
And that can be very costly. About a month ago, a consortium of investors including Microsoft and BlackRock pledged to mobilize as much as $100 billion to build data centers in the U.S. and partner countries. “The capital spending needed for AI infrastructure and the new energy to power it goes beyond what any single company or government can finance,” Microsoft president Brad Smith explained at the time.
The U.S. will need an incalculable amount of time to build up its physical infrastructure in an economy with more expensive land, labor and power. U.S. democracy also has several bottlenecks and veto points—like local legislation, regulations, and community hearings—that will delay, if not outright halt, new data center construction. And that may mean a huge advantage for China and its state-driven "command economy."
The media and investors tend to focus on computing power, and thus miss much of the real-world work that supports AI. Beijing's industrial policy and its ability to funnel public and private resources into growing sectors swiftly will make the country a major power in the world of AI, much like it already has proved to be in clean energy and electric vehicles.
Being at the leading edge of a new technology means little if a country can't figure out how to make it work at scale.
Expensive infrastructure investments today are needed to support tomorrow’s AI demand. Meta CEO Mark Zuckerberg recognizes this: In Meta’s most recent earnings call, Zuckerberg defended heavy investments in data centers by arguing the risk of underinvestment, and missing out on “the most critical technology for the next decade and a half” was far greater.
“I’d rather risk building capacity before it is needed, rather than too late, given the long lead times for spinning up new infra projects,” he told analysts.
The long lead time isn’t due to bottlenecks at Nvidia, but instead something far more mundane. The lead time on new power transformers and generators, needed to power the increased electricity demands from data centers, can be as long as four years. Even if the power equipment is in place, data center operators still need to wait for them to be connected to the grid. Local zoning laws and land permits would also take time, often months or years for approval.
Northern Virginia currently has the U.S.’s highest concentration of data centers. According to local power utility Dominion Energy, any data center that consume over 100 megawatts of power will now need to wait as much as three additional years, on top of the current three to four years wait, to get connected to the grid.
Big tech firms thus need to spend big on infrastructure today long before they can enjoy a meaningful return on their investments. And if they don’t invest now, they risk falling behind their competitors in the future, or worse, not being able to provide the raw power needed to run faster and more powerful algorithms when the technology advances further.
China may be able to ramp up its AI infrastructure faster and more cheaply than the U.S. can. That may not be enough to overcome China’s weakness in producing advanced chips. But if and when China does figure out how to make powerful semiconductors, it can catch up quickly to the U.S., if not potentially overtake it.
China’s state-driven system can put resources and energy into infrastructure at a speed and scale unmatched by other countries. Since 2014, the province of Guizhou has been a pilot zone for China’s data center industry. By 2022, the province already had 23 key data centers in operation or under construction, including eight very large data centers.
Guizhou is now one of the highest concentrations of large data centers in the world, hosing data for both local giants like Huawei, JD.com and Alibaba, and global companies like Apple and Microsoft.
Officials could replicate Guizhou's success in another province with ample land and little urbanization. Local governments would welcome data center investment, as it leads to new jobs and tax revenue, as well as finding a productive use for land that serves little agricultural purpose.
Experts in the industry are starting to realize that China’s advantage in energy and land could help ameliorate China’s lag in chip development. Wang Jian, the founder of Alibaba Cloud, said at a recent conference that China's ready access to renewable energy gives it a better foundation for AI development compared to the U.S.
Zhang Ping’an, CEO of Huawei Cloud, has emphasized China’s strengths in bandwidth, network and energy, and suggested that architectural innovation, rather than reliance on advanced AI chips, could help Chinese cloud computing come out on top.
China has previously leveraged scale to build new industries. The country’s command economy rallied clean energy companies, helping grow national and international champions in green energy. Just a decade ago, China had minimal renewable energy. But thanks to government policy and strategic investments in renewable technologies and supply chains, China is now adding more than 200 gigawatts of renewable energy capacity annually. The U.S. adds just 40.
Chips—and China’s lack of access to them—are what gets attention today. But infrastructure and data centers could decide the final winner of the AI arms race.
