Nvidia CEO Jensen Huang has sounded an unusual warning about the AI race. According to Huang, China is moving faster not because of better chips or engineering breakthroughs, but because it can physically construct major infrastructure at speeds the United States cannot match. As reported by Fortune, Huang told Center for Strategic and International Studies President John Hamre that a data center built in the U.S. can take years, while similar mega-projects in China go up almost overnight.
“If you want to build a data center here in the United States from breaking ground to standing up an AI supercomputer is probably about three years,” Huang said. “They can build a hospital in a weekend.”
This speed gap matters because training AI systems requires massive clusters of GPUs housed inside energy-intensive data facilities. While the U.S. still holds a clear edge in chip design, Huang says China already holds a lead in the two resources needed to scale AI infrastructure: construction velocity and energy.
Huang highlighted that China generates roughly twice the total energy that America does, even though the U.S. economy is larger. He noted that China continues to add capacity rapidly while America has flattened out.
That imbalance has real consequences. With more electricity and faster construction timelines, Chinese cloud providers can deploy AI supercomputing facilities significantly sooner. More compute capacity also translates into faster training cycles, more model experimentation, and larger production environments.
Despite these warnings, Huang maintains that Nvidia’s technology remains ahead by several generations. China buys Nvidia chips today, but alternatives are emerging, and Huang pushed back on any assumption that China cannot eventually manufacture competitive systems.
“Anybody who thinks China cannot manufacture is missing a big idea,” he said.
Still, the U.S. is investing aggressively. Nvidia, Amazon, Google, Microsoft, and Meta are collectively expected to push between fifty and one hundred billion dollars into new data centers over the next year. Industry analysts estimate that a typical facility costs roughly ten to fifteen million dollars per megawatt, and even standard deployments require forty megawatts or more.
The challenge is timing. The permitting process alone can stretch years. Power grid approvals often lag behind demand. Some regions cannot guarantee immediate energy access, forcing companies to wait for upgrades.
Huang said he is optimistic that support from the federal government, including incentives to expand manufacturing capacity, will narrow the gap. But he made it clear that the AI race will not be won only in labs and chip fabs. It will also be won by whichever country can pour concrete faster, wire more facilities, and supply the energy to run them.
As Huang bluntly suggested, speed is a strategic differentiator. And right now, China is moving fast.