AI data-center loads are rewriting utility planning, forcing grid operators to rethink dispatch and capacity economics.
For nearly two decades, electric utilities planned around modest load growth, incremental upgrades, and long development timelines. AI has ended that era.
Executives from utilities, grid reliability organizations, engineering firms, and commercial real estate said that hyperscale data centers are forcing the electric industry to redesign how it studies, prices, and connects large customers. The challenge extends well beyond adding generation. Transmission, substations, transformers, interconnection studies, and reliability standards are being rewritten for a class of customers that can demand hundreds of megawatts—or even gigawatts—at a single site.
"The data centers are ubiquitous. They are coming," said Brian Thiry, director of strategic engagement and reliability at ReliabilityFirst, a North American Electric Reliability Corporation regional entity. "We're going to need all the megawatts from all the resources to power the future."
Thiry and other panelists stressed that AI is accelerating pressures already building across the grid rather than creating an entirely new problem. Utilities spent years planning for little or no growth in electricity demand. That changed almost overnight as utilities began incorporating hyperscale AI campuses, broad electrification, and new manufacturing loads into their long-range plans. PJM Interconnection forecasts have risen sharply, driven partly by resource adequacy concerns tied to power plant retirements and electrification that predate AI projects. "Directionally, even if we get a portion of that load forecast, it's still much more than we've seen for the last couple of decades," Thiry said.
The issue is no longer simply finding generation. "It's not just about the generation and the load," Thiry said. "It's about the deliverability and the infrastructure." That means new transmission lines, substations, transformers, and switching equipment—assets that often require years to permit, manufacture, and install.
Rachel Lindesmith, director of national accounts at FirstEnergy, urged developers to begin conversations with utilities before projects are finalized. "The earlier the better," she said. FirstEnergy recently introduced a two-stage load study process that provides early-stage cost and schedule estimates while screening speculative projects before they consume engineering resources. Equipment shortages are reshaping timelines: "Some of these transformers and breakers and things that are required take four or five years," Lindesmith said. The utility is also investing in larger transmission projects, including a joint venture with American Electric Power to construct more than 300 miles of new 765-kV transmission across Ohio.
Industrial development conversations have flipped, said commercial real estate broker Terry Coyne. Developers once asked about highways, acreage, and workforce. Today, they ask about megawatts. "It's all about gas, and it's all about power," Coyne said. Manufacturers are increasingly competing with hyperscale operators for electrical capacity. Robotics, defense manufacturing, and advanced industrial projects are now requesting 20 MW or more—levels once reserved for only the largest facilities. Those projects compete for the same substations, transformers, and transmission capacity serving AI campuses. "The gold rush is kind of over," Coyne said of Ohio's data center land market. "The power is tapped out."
Panelists discussed behind-the-meter generation, small modular reactors, and recent federal efforts to speed large-load interconnections. While several viewed those developments as necessary, none described them as complete solutions. "Bringing your own generation doesn't necessarily solve all of our reliability questions," Thiry said. Even privately developed power plants must be integrated into the broader grid. Utilities still need models, operating studies, and transmission infrastructure to manage changing power flows during outages, maintenance, or system disturbances.
ReliabilityFirst is working with NERC on proposed standards for computational load entities—a recognition that AI campuses have become important operating elements of the bulk power system rather than simply large customers. "You are now part of the grid," Thiry said.
The discussion came a week after the Federal Energy Regulatory Commission ordered regional transmission organizations to justify or reform their rules for large-load interconnection, reflecting mounting concern that existing processes cannot keep pace with AI development. Panelists welcomed the move but cautioned that policy alone cannot shorten transformer lead times, build transmission corridors, or manufacture additional equipment—constraints governed by engineering, permitting, and supply chains.
AI may have accelerated the problem, but it is no longer the only source of demand. Manufacturing, defense production, and electrification are competing for the same generation, transmission capacity, and skilled labor. For utilities, the work extends beyond adding power plants: they must expand transmission, redesign interconnection processes, and build a grid that can accommodate a new class of customers measured in hundreds of megawatts rather than dozens.