How Do Utilities Determine Which AI Data Centers Get Grid Access?. 6 Min Read. Getty Images.
The rise of AI data centers is transforming the energy landscape, creating unprecedented demand for electricity and forcing utilities to rethink how they plan for grid capacity. With billions of dollars in transmission investments at stake, utilities must decide which AI projects are real enough to build the grid around.
But their approaches to this challenge vary widely.. A review of planning documents from Georgia Power, Duke Energy, and Dominion Energy shows each utility responding differently. Rather than treating announced load as forecasted load, they filter prospective customers, test multiple futures, and evaluate transmission investments that remain valuable under a range of outcomes..
While one utility discounts a 22.8 GW pipeline of prospective customers before building its load forecast, another models 16 hypothetical 100 MW to 500 MW customers to test where transmission may be needed. A third sorts prospective data center projects by contractual commitment before adding them to its long-range plans..
Related:How NASA’s Artemis Lessons Apply to AI Infrastructure Planning. Regardless of the approach, the same question lingers: Which AI projects are real enough to build the grid around?. Model the Pipeline, Not the Hype.
Georgia Power’s 2025 Integrated Resource Plan reports a 22.8 GW Large Load Economic Development pipeline.The company says it does not expect all of that demand – or even all committed projects – to materialize.. Importantly, the utility does not assume all 22.8 GW will materialize.
Instead, Georgia Power uses what it describes as a risk-adjusted load forecasting process. The company explicitly says it does not expect every economic development project – or even the full load from committed projects – to materialize.. Its probabilistic model evaluates uncertainties including state selection, electric provider selection, project delays and whether projects ultimately materialize, running hundreds of thousands of simulations to produce the probability distributions underlying its planning forecast..
The result is an 8.2 GW forecast through the winter of 2030–31, and that forecast becomes the basis for transmission planning.. Georgia’s companion transmission plan proposes 1,142 miles of new transmission, including 543 miles of new 500 kV lines, 530 miles of 230 kV construction, 69 miles of 115 kV facilities, 982 miles of rebuilds or reconductoring and 21 new high-voltage transformers..
Several of the largest projects establish high-capacity corridors designed to increase transfer capability, relieve transmission constraints and accommodate changing generation patterns and forecasted load growth. Maps in the appendix show those corridors reinforcing Georgia’s high-voltage backbone..
Related:Texas’ 765 kV Decision: Build the Wires, the AI Will Follow. The transmission appendix rarely mentions AI, but the resource plan does. Together, the documents show how a risk-adjusted forecast becomes physical infrastructure..
Stress-Test the Future. Duke Energy starts somewhere else. Rather than relying on one demand forecast, the company develops multiple scenarios and asks which transmission investments still make sense if the future changes..
Appendix K opens with a Local Economic Study examining 16 hypothetical large-load customers ranging from 100 MW to 500 MW under summer and winter planning conditions. The study is part of the Carolinas Transmission Planning Collaborative’s effort to identify future transmission needs before specific projects are selected..
The appendix then lays out Duke’s Multi-Value Strategic Transmission process. Engineers identify transmission needs, develop solutions, evaluate alternatives and quantify multiple benefits before prioritizing projects. Figures in the appendix map transmission needs identified across multiple planning scenarios..
Duke’s transmission planning materials also identify transmission-constrained “red zones” that help developers understand where new generation is more likely to require network upgrades. As transmission projects are completed, those constraints can be relieved, opening additional areas for development..
Related:NERC Flags AI Data Center Grid Risks in Report. The appendix also evaluates Grid Enhancing Technologies, including dynamic line ratings, advanced power-flow control devices, transmission switching and advanced conductors, as part of Duke’s long-term planning and FERC Order 1920 compliance work..
Separate Commercial Risk From Grid Risk. Dominion Energy separates two different questions: the first is commercial, and the second is technical. The company categorizes prospective data center customers by engineering agreements, construction agreements, and binding electric service agreements rather than treating every proposed project as equally likely to materialize..
Its transmission appendix examines whether the future grid will remain reliable once those customers arrive.. Engineers perform separate studies of import capability, system inertia, frequency response, short-circuit strength and black-start capability using PJM planning models. One analysis concludes that transmission upgrades will increase the amount of power the Dominion zone can import.
It also warns that transmission capability does not guarantee energy will be available if dispatchable generation elsewhere in PJM continues to decline during extreme conditions.. The appendix also evaluates technologies including virtual inertia, grid-forming inverters and synchronous condensers as planners assess how a changing generation mix could affect future system reliability..
Planning for Uncertainties. Integrated resource plans (IRPs) have always been tools for managing uncertainty, helping utilities model variables like fuel prices, weather, economic growth, and regulatory outcomes. However, AI data centers introduce a new kind of challenge: individual customers requesting hundreds of megawatts – or even a gigawatt – of power years before construction begins..
“IRPs have always been a tool to plan for uncertainties, but that has largely been about regulatory uncertainties, climate and weather uncertainties, and to some extent growth uncertainties,” said Elizabeth Whitney, managing principal of Meguire Whitney. “Now the stakes are higher because the swings are more extreme.”.
The consequences extend far beyond planning departments. Underestimating demand could leave data center projects waiting years for transmission upgrades, while overestimating it risks building infrastructure for projects that never materialize – raising questions from regulators and consumer advocates about whether ratepayers should bear the cost.
These competing risks are forcing utilities to rethink how proposed AI campuses translate into billion-dollar transmission investments.. “Planning is changing, but it is not a wholesale abandonment of traditional methods,” said John Moura, director of Reliability Assessment and Performance Analysis at the North American Electric Reliability Corporation.
“The pace, size, and uncertainty of large-load requests, especially from data centers and computational-related demand, are forcing utilities and planners to use a broader range of tools.”. Utilities are adapting by expanding traditional methods with probabilistic models, sensitivity analyses, and scenario planning.
Georgia Power filters a large-load pipeline before it enters the forecast, Duke Energy tests transmission plans against multiple possible futures, and Dominion Energy separates commercial commitment from engineering analysis before committing infrastructure.. “The big question for utilities is what happens if the data center load vanishes (like a lot of crypto load did),” Whitney said.
“Getting upfront payments from data centers is a big fix for that, but removing the risk from investor-owned utilities may also remove their ability to recover the costs and profit from those builds.”. Moura emphasized that deterministic planning remains essential but must now be complemented by these expanded approaches.
“The risk is not just whether enough generation is built, but whether generation, transmission, fuel infrastructure, and operating capability are developed in time and in the right locations,” he said.. The filings from Georgia Power, Duke Energy, and Dominion Energy reveal a common goal: distinguishing announced load from expected load and determining which AI projects are substantial enough to justify billion-dollar transmission investments..
About the Author
