Here’s a startling fact: the rise of artificial intelligence could force us to build billions of dollars’ worth of new power plants—unless we rethink how data centers use electricity. But here’s where it gets controversial: a groundbreaking Duke University study suggests that simply shifting when data centers operate could save the U.S. up to $150 billion over the next decade while curbing reliance on natural gas. Could this be the game-changing solution we’ve been overlooking?
As AI demand skyrockets, companies like Microsoft are racing to build data centers, prompting utilities to plan new power plants—often fueled by natural gas. But the Duke study, led by Jackson Ewing of the Nicholas Institute, challenges this approach. Ewing explains, ‘The real challenge isn’t the total electricity used—it’s those rare peak moments that strain the grid.’ By embracing load flexibility—allowing data centers to shift computing tasks to off-peak hours or locations with surplus power—we could avoid massive infrastructure investments.
And this is the part most people miss: electricity demand had been stagnant for years until AI, advanced manufacturing, and electrification reignited growth. Utilities, which plan years in advance, risk overbuilding if they misjudge future needs. ‘We could end up paying for underused assets,’ warns Ewing. Meanwhile, former Federal Energy Regulatory Commissioner Allison Clements notes, ‘Data centers are piling onto existing pressures like aging infrastructure and extreme weather, making quick solutions critical.’
In North Carolina, Duke Energy is already experimenting with large-load contracts that incentivize data centers to reduce usage during peak times. While these centers currently account for less than 1% of peak demand, they’re projected to make up 10% of electric sales by 2030. A recent 1-gigawatt agreement promises $1 billion in customer savings over 15 years—but even with flexibility, Duke still plans to build new natural gas plants.
Here’s the bold question: Can we truly balance affordability, resilience, and sustainability by integrating data centers smarter into the grid? Or will the AI boom outpace our ability to adapt? Ewing believes, ‘If done right, we could create a more balanced energy system.’ But what do you think? Is load flexibility the answer, or are we underestimating the scale of the challenge? Let’s debate in the comments!
