In 2023, Grid Strategies declared the end of “the era of flat power.” That report, which forecasted a “shocking” peak demand growth of 38 gigawatts through 2028, caused a ripple of alarm through the energy sector. It was a year since ChatGPT’s launch, and the reality of what artificial intelligence could mean for data center energy demand — especially in combination with electrification and onshoring of manufacturing in the U.S. — was just beginning to become clear.
John Wilson, who specializes in planning, rate, and other utility regulation issues at Grid Strategies, remembers thinking that the 2023 projection was “huge.”
But two years later, the organization’s latest estimate — which puts U.S. peak load growth at 166 GW over the next five years — makes that number seem trifling in comparison. “Now that we’re at 166 GW, 38 just looks like, ‘Oh, that was no big deal,’” Wilson told Latitude Media. “The scale of the growth in the forecasts is phenomenal.”
The new estimate, which is included in a report co-authored by Wilson and published today, is more than four times the original 2023 forecasts, and over double last year’s 64 GW estimate, implying that peak demand will grow by 3.7% annually.
Electricity use is projected to rise even faster, up 32% over the next five years, with an average annual increase of 5.7%.
The biggest driver of this surge, perhaps predictably, is data centers. Of the 166 GW in projected peak load growth, about 90 GW, or roughly 55%, is tied to data center development. Manufacturing accounts for another 30 GW, or about 20%, with the remainder coming from building and transportation electrification, plus mining and other loads.
It’s the first time in the sector’s history that data centers have so thoroughly dominated utility forecasts, overshadowing both manufacturing and electrification in terms of energy demand. This is consistent with estimates published last week by the International Energy Agency, which found that while globally data centers account for less than 10% of electricity demand growth by 2030, in the U.S., they make up over 50%.
In the U.S., “if the load forecasts are correct,” the Grid Strategies report anticipates that AI and data centers will go from consuming less than 2% of total electricity in 2024 to over 15% by 2030.
Too high?
However, even Grid Strategies itself thinks these numbers might be overblown. “We think the data center number is too high,” Wilson said.
Grid Strategies estimated the 90 GW of peak load growth by aggregating the forecasts that utilities and regional planners submitted to the Federal Energy Regulatory Commission, with CAISO, ERCOT, PJM, SPP, MISO, and Georgia Power together making up the bulk of the total. But the number is inconsistent with external reports tracking data center load growth, some of which estimate between 60 and 65 GW of data center load coming online by 2029-2030.
The discrepancy suggests both that the truth of what we can expect at the end of the decade is somewhere in the middle — approximately 25 GW less than what utilities are declaring, Wilson estimates — and, more broadly, that the industry is still struggling to accurately forecast large new loads.
“The utility industry still has work to do to figure out how to estimate not just how much load is going to come up, but when it’s going to come on,” Wilson said, explaining that the vast majority of utilities are grappling with outdated forecasting models and a lack of historical data. “Three or four years ago, your utility load forecast was an econometric projection, with the main things affecting growth being the number of new customers… and an economic growth indicator. Now, most of the growth is in the large load forecast, and they’re figuring out how to model it.”
It will likely take a couple of years, he added, for utilities to figure the new system out, “before load forecasting becomes boring again.”
This forecasting uncertainty creates risks of utilities overbuilding or underbuilding the infrastructure required to meet load growth, forcing unnecessary costs onto customers already facing steep electricity bills — or else prompting utilities to delay delivering power to large load customers.
That said, even if the reality is 20 GW lower than today’s estimate — or in the neighborhood of 140 to 150 GW — the updated load growth estimate is still staggering: “We’re in big power range,” Wilson said.
It begs the question of whether the U.S. system will be able to build enough new generation and transmission capacity to meet the peak load that’s coming, considering it would have to do so “at more than six times the rates seen in recent years,” as the report notes.
“That is a lot of power to build in five years,” Wilson said, stressing that the country’s capacity to build, buy, or import necessary materials such as power electronics, transformers, circuit breakers, and gas turbines is limited. “There’s going to be a lot of constraints on the ability to build out the system nationwide to meet this demand… and [understanding] the scale of the power industry’s response is what’s important about these numbers.”
