Show Me the Power!
Here’s a surprise. Your boring old electrical utility, you know that stock no hip tech investor like you would ever let into his portfolio, could become the high-tech growth company of the future—if it makes the right adjustments swiftly.
As AI model training accelerates and workloads scale, U.S. electricity demand, which had been flattish since 2010, has begun to rise rapidly. As other sources of demand moderated with efficiencies, the demand from digital computation and communication kept rising. with the net result that demand stayed flat.
As energy guru Mark Mills pointed out in a recent issue of the Dakota Digital Review, rising digital energy demand was a powerful force even before AI became a practical reality early this decade.
“The physics of transporting bits leads to a surprising fact: An hour of video using the Cloud infrastructure uses more energy than a single person’s share of fuel consumed on a 10-mile bus ride,” according to the Dakota Digital Review.
AI is accelerating this trend.
“In the U.S., the square footage of data centers—under construction or planned for the next few years—is greater than the entire existing inventory… And each square foot of data center inhales 100 times more electricity than a square foot of a skyscraper,” the Dakota Digital Review reported.
The result is that computational demand is finally pushing a net increase in total demand.
According to the Information Network’s report Hot ICs: A Market Analysis of Artificial Intelligence (AI), 5G, Automotive, and Memory Chips, U.S. data centers will triple their energy consumption from 196 TWh in 2023 to nearly 600 TWh by 2030. And the phenomenon is global. Across China, both government- and enterprise-led compute centers are being built at massive scale, with some individual sites exceeding 200 PFlops of AI-specific capacity.
Grid Strain and Structural Shortfalls
Where will we get all that power? Coal and natural gas currently generate 56% of U.S. electricity, but pending EPA carbon rules would phase out most coal and force natural gas plants to adopt carbon capture by 2032. These rules may ease up somewhat as green hysteria recedes, but a tripling of power from fossil fuels seems unlikely.
Meanwhile, utility-scale wind and solar deployment remains behind pace. Many completed projects are being revealed as embarrassing, even dangerous failures. Just ask a German friend about the Dunkelflaute or any Spainard where he was when the lights went out. A North American Electric Reliability Corporation report warned that two-thirds of the U.S. grid may be at risk of outages during peak weather by 2027.
This and several other factors critical to powering the digital world argue strongly for a revival in nuclear power, a revival that is already underway.
The digital future favors nukes in several ways. One traditional argument against nukes has been that they don’t handle peaks and troughs of demand very well. That’s improving a bit with new technology. More important though, data centers do not operate on the old daily demand cycles that peak in the afternoon and taper off at night. Their energy needs remain largely constant, requiring uninterrupted electricity day and night. AI is accelerating this shift, effectively turning 24/7 baseload generation into the most prized commodity in the energy sector.
Computation is not only increasing total energy demand—it is reshaping the structure of the electricity market. The shift away from peak power cycles is one way this is happening. Another is geography. Power demand used to vary mostly with population, even including commercial demand because factories were located near workers. Today, demand is being localized by the data centers themselves.
Northern Virginia is the world’s largest data center hub, hosting some 150 data centers, most of them in Loudoun County, Virginia. NVA alone accounts for more than a quarter of U.S. data center load. Largely driven by those data centers, between 2019 and 2023, Virginia’s electricity sales jumped by circa 14 billion kWh, the largest increase of any state.
PJM, the regional grid operator, now projects Virginia’s Dominion zone demand will grow roughly 5% per year, far above the average. Some analysts project monthly electricity use by Virginia data centers could nearly triple from 11,000 GWh in 2024 to ~30,000 GWh by 2040.
Rivaling Northern Virginia is Texas (ERCOT). Texas has become the second-largest data center market. From 2019 to 2023, Texas electricity consumption rose by 13 billion kWh, similar to Viginia’s increase. The Dallas–Fort Worth area had 591 MW of data center capacity leased in 2022, the second-largest concentration in the country. The influx of data center projects is so massive that ERCOT’s planners have published estimates that place Texas peak demand in 2031 at 218 GW (up from ~85 GW peak in 2023). Other estimates are somewhat lower, but the lowest are still breathtaking.
Other large and growing demand regions include the Pacific Northwest, where demand could grow by 10X through the end of this decade. However, growth there may be slowed by needed improvements to the grid.
As for the grid, it is important to note that hyperscale data centers do not simply draw from the general power pool. Nearly all negotiate direct interconnections and establish power purchase agreements (PPAs) with electricity providers, securing dedicated capacity and reliability guarantees.
Are there companies out there that fit the bill? Over at the Gilder Technology Report we are tracking several we believe are excellent investments.