America Could Lose AI Race Against China Due To Power Crunch, Goldman Says

The future of artificial intelligence may depend less on chips from Nvidia Corp. (NASDAQ:NVDA) and Advanced Micro Devices Inc. (NASDAQ:AMD), and more on the raw watts needed to power data centers.

America's lead in the global artificial intelligence race is dangerously walking a tightrope, as a looming domestic power shortage threatens to stall progress and open the door for China to seize the upper hand, according to Goldman Sachs.

The bank’s analyst Hongcen Wei indicated the key ingredient to dominating the global AI race is having reliable and ample power supply—something the United States may struggle to maintain.

"The global AI race is heating up, with fierce competition centered on chips, rare earths access, energy supply, talent, and AI adoption," said Wei in a Thursday report.

But what could soon matter most, he said, is electricity.

"As AI demands massive power, reliable and ample power supply is likely to be a key factor shaping this race, especially because power infrastructure bottlenecks can be slow to solve."

US Data Center Growth May Hit A Wall

Currently, the U.S. leads the world in AI infrastructure, hosting 44% of global data center capacity, equivalent to over 50 gigawatts (GW). That's roughly equal to the combined capacity of China, the EU, Japan, South Korea and India.

Data centers already account for 6% of total U.S. electricity demand, and Goldman projects this to rise to 11% by 2030.

But that growing demand is colliding with a strained power grid.

Effective spare power capacity—a key measure of energy system flexibility—has dropped from 26% five years ago to 19% today, nearing the 15% threshold commonly viewed as critically tight.

Eight of 13 U.S. regional power markets have already hit or fallen below that level.

Goldman projects that, even under conservative assumptions, spare capacity in most U.S. regional markets will drop below 15% by 2030, tightening constraints on future data center development. In a scenario where AI infrastructure grows faster—as Goldman's equity analysts expect—spare capacity would dip even further below critical levels.

This power bottleneck is already impacting costs.

Real-time electricity prices surged last summer, and capacity prices in PJM—the regional grid that includes Virginia, the global data center capital—jumped sharply.

China Could Pull Ahead With 400 GW In Spare Capacity

While the U.S. grapples with shortages, China is gearing up to meet future AI demands head-on.

"China—the world's second-largest data center hub—already has major power spare capacity and is planning to boost power supply, across renewable, coal, natural gas, and nuclear sources," Wei said.

By 2030, Goldman projects, "China to have effective power spare capacity equivalent to over three times the world's expected data center power demand (~400 GW vs. ~120 GW), positioning it to fuel rapid data center expansion."

China's planned spare capacity would reach 25% of its peak summer demand by 2030.

Given that China's power market is already twice the size of the U.S., this provides Beijing with far greater flexibility to fuel AI-related infrastructure, alongside other energy-intensive industries like aluminum production.

Why The US Is Falling Behind

Goldman says the U.S. grid is tightening due to three issues: strong demand growth from data centers, insufficient renewable and natural gas capacity to replace retiring coal, and limited deployment of storage and emerging technologies.

These challenges are worsened by long lead times and permitting backlogs for natural gas turbines.

Goldman sees little additional nuclear or gas capacity coming online before 2030, citing construction and supply chain constraints.

AI's Future May Be Powered By Watts, Not Just Chips

While U.S. policy could boost renewables or delay coal retirements, the bank says those measures are "unlikely to reverse the tightening trend in the next few years."

The AI race is now about more than chips and talent—it's about energy. As Goldman put it, "A key ingredient to win the global AI race is having reliable and ample power supply to feed data centers."

And right now, that's what China is building—and what the U.S. may soon lack.

Read Next:

• Taiwan Semiconductor Hikes Dividend, Approves $15 Billion Spending Plan For New Chip Factories

Image created using artificial intelligence via Midjourney.