- Utilities are ramping infrastructure investments as they struggle to keep up with AI demand
- They're looking to insure these investments through upfront payments for expected usage
- Uncertainty about the scale and longevity of demand mean there's still a risk of overdevelopment of electric infrastructure
You’ve probably heard of the trickle-down theory in economics, which holds that regulatory and tax advantages for businesses spark investment that eventually benefits citizens. But what about trickle-down capex?
The latter, more of a survival instinct than a supposed public good, seems to be happening among utilities as they scramble to keep the lights on in North America despite a wave of electricity demand from data centers looming over the grid.
“Investor-owned utilities in the U.S. have already invested more than $1.3 trillion in the past decade to enhance energy infrastructure,” Morningstar DBRS wrote in a note to investors. “Growth expenditures by these utilities are expected to exceed $1.1 trillion over the five-year period ending December 31, 2029.”
Yes, you read that right. Utilities are expected to spend as much over the next five years as they have over the last 10 to beef up their electrical infrastructure. And for good reason.
In its most recent long-term forecast, the North American Electric Reliability Corporation (NERC) predicted 120% data center load growth in parts of Texas (served by ERCOT), California (served by CAISO) and the South eastern U.S. near Georgia and South Carolina by 2027. Additionally, 130% data center load growth was predicted in the Southwest Power Pool covering the middle of the country and 70% growth in the Mid-Atlantic region (served by PJM).
In Texas, which is home to one of OpenAI’s Stargate projects, ERCOT is expected to be swamped with over 20 GW of newly contracted large loads by 2028 (driven in large part by data centers) on top of its organic load growth.
All told, NERC warned that about half of utility regions in the U.S. and Canada would be an elevated risk for power shortfalls.
AI bubble risks
Of course, the money for these investments has to come from somewhere. We’ve already covered the battle over whether or not consumer rate payers should foot the bill for infrastructure investments spurred by data center load growth.
But Morningstar noted that even these “traditional funding sources are inadequate to meet future investment needs,” and thus private equity and debt providers have stepped in. By Morningstar’s count, investments from these sources jumped from $17.6 billion in 2016 to $37.4 billion in 2024.
Of course, with all the talk of an AI bubble floating around, utilities are looking to protect these investments.
In Ohio, for instance, a new rule was passed in July requiring data centers to pay for at least 85% of their contracted energy usage upfront each month even if they don’t use it all. New data center customers must also pay a hefty exit fee if their project is canceled.
Even with this insurance against the AI bubble bursting, there remains uncertainty around just how much data center demand there will be and how long it will last. And that means utilities are at risk of overdeveloping their infrastructure.
“The commercialization of AI is still at an early stage and there is neither convergence on the number of data centers needed before market saturation nor consensus that the planned electricity demand by data centers will be sustained for 15 to 30 years following infrastructure investment,” Morningstar noted. “Should data center usage fail to materialize in the scale and magnitude planned, investment recovery could be delayed, leaving utilities without adequate demand to recoup investments efficiently.”