The US power market’s forecasting problem comes at a cost to ratepayers

Virtually every market on earth needs more power, more flexibility, and better transmission. But the United States has different and more high-pressure demands on its energy system than its peers. The specifics of the artificial intelligence boom are complicating the industry’s ability to predict what further changes are coming — and that uncertainty comes with a cost for ratepayers.

Globally, electrification is driving 87% of energy production, said Sarah Kapnick, managing director and global head of climate advisory at JP Morgan, onstage at the Aurora Energy Transition Forum in Manhattan last week. Stateside, however, about half of expected growth in energy production is due to data centers.

This makes the U.S. market fundamentally much harder to predict, as several speakers unpacked over the course of the day-long conference. Data center load forecasts vary widely, with the highest expecting 94 to 131 gigawatts of load growth by 2030 and the more conservative hovering around 22 GW. In Aurora’s projections, Texas, Virginia, and Pennsylvania will together account for about 45% of data center load growth by 2030 — Texas with 74 terawatt-hours, Virginia 28 TWh, and Pennsylvania 15 TWh.

Grid operators are expecting significantly more demand by 2030 than Aurora’s baseline projections — roughly 63 GW more. U.S. peak demand currently hovers just below 760 GW, so that forecasting gap represents about 8% of the entire existing system. If Aurora’s projections are correct, overbuilding could saddle ratepayers with steep, unnecessary costs. If they’re wrong, underbuilding could strain the grid and threaten reliability.

But there’s good reason to doubt those largest projections. Power market operators have historically struggled to accurately predict their future power needs; thus far in 2026, PJM’s actual demand has come in 21% lower than their 2025 wide load forecast.

According to Aurora’s North America managing director, Oliver Kerr, that forecasting gap could cost ratepayers anywhere between $79 billion to $624 billion in extra capital expenditure costs — and that’s before transmission and distribution costs, which are rising nationwide, especially in PJM, where 3.7 GW of ready-to-build projects have withdrawn specifically because of the price of transmission.

Quantifying the hurdles

However, there are a lot of variables. Roughly 58% of planned projects are being constructed by untested, first-time developers, said Julia Hoos, head of eastern power markets at Aurora. And historical precedent suggests that just 15% of those planned projects will ultimately materialize, she added.

The rest are likely to fall prey to obstacles like local and state moratoriums, turbine shortages, permitting challenges, and local pushback.

But there’s also a new policy hurdle: The bulk power system executive order, which blocks the importation and installation of foreign-made grid equipment from 24 countries — most notably China — will also likely drive prices up when implementing rules are released in late December, Hoos suggested. China provides roughly 65% of the U.S.’ transformer equipment, but accounts for only 8% of transformer equipment spending. U.S. developers are already grappling with wait times as long as five years for transformers, and demand is expected to double next year.

Jonathan Kaufman, senior managing director at the global investment bank Evercore, said that the industry will inevitably experience some form of consolidation as inexperienced developers unable to handle the logistics of engineering, permitting, and equipment shortages are forced from the field.

Hear from leading utilities about what it takes to build VPP programs that solve the overlapping challenges confronting today’s grid.

Although big banks have been relatively cautious when it comes to data center investment, private credit has had no such qualms. A recent report from Moody’s forecasted $1 trillion in combined capital spending from the five largest hyperscales and Coreweave — but also warned that many loans will start coming due between 2027 and 2029, potentially driving some players out of the game.

The state of the behind-the-meter workaround

Behind-the-meter generation, which some developers are pursuing as a work-around given grid interconnection challenges, comes with hurdles of its own. “Running your own microgrid isn’t easy,” said Kerr, even if colocating generation with data centers eliminates some risks inherent to trying to connect to the overtaxed transmission system.

Shoring up the existing grid, Sarah Orban Salati, chief commercial officer at National Grid Ventures, said, appeals to developers because “nobody wants to be isolated, because, in the end, [the grid] is the biggest battery in the world.” That said, speakers broadly acknowledged building behind-the-meter as a necessary evil, at least for now, given the collective assumption that the AI arms race will be won or lost in the next 18 months — far faster than a new project can get interconnected.

All of that assumes that behind-the-meter power is allowed to progress unfettered. A post-conference report from Ross Fowler and Rinny Singh at Bank of America predicted that as it becomes more widespread, building behind the meter will attract more government intervention “limiting [hyperscalers’] ability to simply bypass the regulatory, reliability, and cost-allocation issues.” The work-around, in other words, may be short-lived.