Calls to slow down the artificial intelligence race took center stage this week as U.S. President Trump and Chinese President Xi Jinping spent three days together in Washington, D.C., including a state dinner with the leaders of the largest American AI companies.
The U.S. and China have different problems when it comes to energy. The U.S. is short on capacity in the places it’s most needed, leaving the country’s hyperscalers scrambling to buy it up wherever it’s available (even including in outer space). At the same time, they’re battling rising public opposition, pegged in part to the belief that data centers will increase their already high electricity prices. Against this backdrop, utilities have increasingly invested in gas to power the build-out.
China, meanwhile, has added more than 1,000 gigawatts of wind and solar to the grid in recent years. But — like in the U.S. — much of that new supply is geographically mismatched with where it’s needed, so the country has largely opted to move computing to where the power is. China’s AI buildout is hampered instead by limited access to the most advanced AI chips as a result of U.S. export controls, even as it builds out its own domestic semiconductor supply chain.
Beijing also has a fundamentally different view of the AI boom’s power needs, explained David Fishman, who leads the mainland China team at energy advisory firm The Lantau Group.
In China, AI is one of many drivers of power consumption growth, but it’s neither the only nor the largest driver; sectors like high-tech manufacturing and EV charging are also drawing a lot of power. The government identified data centers as something that could derail the country’s plan to hit a 2030 carbon peak, unless they are “regulated quite aggressively.” Even with restrictions, Beijing is betting it can outplay the rest of the world in the long run, because it can build infrastructure incredibly quickly.
China’s state-planned approach to AI development may have some valuable lessons for the U.S., explained Cosimo Ries, an analyst at policy research firm Trivium China. “China is proceeding with a much more strategic, long-term approach to the data center buildout,” he told Latitude Media. “Once it lands on a system that works, it will “go full in on it and ramp up investments.”
China’s effort to geographically coordinate the data center buildout dates back to the early 2020s, when the government identified eight national computing hubs and ten data center clusters. That strategy was designed to move processing work to places with lower land cost, cooler temperatures, and plentiful electrons.
Each hub is set to serve a different type of demand, and the idea is to match the location and type of computing with local power, network, and latency constraints, Ries said. In other words, the Chinese government is “telling both the AI labs and data center developers that this is where stuff is going to be built, and doing a lot of the power system planning specifically to optimize for data center needs.”
Importantly, a more recent initiative requires the new data centers in those national hubs to source at least 80% of their electricity from renewable sources by 2030 — whether directly or through the country’s version of renewable energy credits.
Hubs in action
Those data centers are already going up, including in Inner Mongolia, where one of the designated data center hubs handles AI workloads that don’t need low latency, like model training, Ries explained. The region’s plentiful, cheap power from wind farms matters more for that work than proximity to users. Ries pointed to Envision, a Chinese wind developer that has pivoted to data centers and is building a gigawatt-scale campus in Inner Mongolia. The Galaxy Campus project will be primarily served by dedicated lines to nearby wind generation, with support from battery storage.
The U.S. has also experimented with linking AI growth geographically to energy assets and infrastructure planning. Last summer, the Department of Energy identified 16 federally controlled sites around the country that are “uniquely positioned for rapid data center construction, including in-place energy infrastructure with the ability to fast-track permitting for new energy generation such as nuclear.”
That helped tee up projects like SB Energy and AEP Ohio’s proposed 10 GW project at the former Portsmouth uranium enrichment site in Ohio, but hasn’t spurred a coordinated strategy for siting and connecting computing capacity around available power.
That’s not surprising, in Fishman’s view. “When you say, ‘Can we do a thing that China is doing?’ very often, the answer is no, as long as you have a different political economy,” Fishman said.
That’s the reality of comparing a decentralized democracy with a country with a centralized, single-party authoritarian government. In China, data centers are getting built with money from state banks, which are fulfilling state mandates and objectives, often by state companies, also fulfilling those mandates and objectives, Fishman explained, while in the U.S. the power-for-AI boom is driven by “private companies trying to make a buck.”
This creates an entirely different power dynamic between the private and public sectors. “If something they want to do ever runs against the geostrategic objectives of the state, then they will yield to the needs of the state,” Fishman added. “That is not the case in the United States…where very often the state yields to the needs of the corporations.”
A capability problem
Fishman, for his part, is skeptical that there’s anything about China’s approach — building abundant renewables and planning data center capacity for where the power is — that the U.S. could replicate on any kind of meaningful timescale.
For one thing, China doesn’t have the same challenges with public opposition to data centers. In China’s most sparsely populated regions, there’s “less of a feeling either that [data centers] are going to hurt or help,” Fishman said.
Developers aren’t pitching data centers as job creators, and there isn’t a risk that a new data center will raise power costs in a region. The government regulates retail electricity prices using a tiered system under which prices for residential ratepayers pay significantly less than commercial and industrial ratepayers. That means households are “totally protected from supply and demand fluctuations,” Fishman said.
Ultimately the U.S. government simply can’t dictate where data centers get built or what powers them to the same extent as China, which is targeting 60% of new data center capacity to be built in its eight hubs, Fishman said. Even if the U.S. government implemented sweeping development policies, there’s still the barrier that it takes years to get new assets online.
“ I’m not sure there’s anything obviously in place that would make the United States government agencies work faster,” he said. “Because that’s really what we’re asking for, right? We’re asking you to hire more regulators and overseers, asking you to improve your capacity to process interconnection requests and also construct grid and transmission infrastructure.”
