The explosive growth of AI data centers is forcing electric utilities to confront a fundamental question: which proposed projects are real enough to justify billions of dollars in grid infrastructure? A review of integrated resource plans and transmission appendices from three major U.S. utilities shows they are each answering that question differently, but all are moving away from treating announced load as actual forecasted load.
Georgia Power: Risk-Adjust the Pipeline
Georgia Power’s 2025 Integrated Resource Plan lists a 22.8 GW pipeline of large-load economic development prospects. The utility is explicit that it does not expect all of that demand to materialize, and it does not feed the raw pipeline figure into its planning forecast.
Instead, Georgia Power applies a probabilistic, risk-adjusted forecasting process. Hundreds of thousands of simulations account for variables including state selection, electric provider selection, project delays, and outright cancellations. The output is a probability distribution, and the planning forecast that emerges from it is 8.2 GW through the winter of 2030-31.
That 8.2 GW figure then drives physical infrastructure decisions. The companion transmission plan calls for 1,142 miles of new transmission lines, 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. The projects are designed to increase transfer capability, relieve existing constraints, and accommodate changing generation patterns.
Duke Energy: Stress-Test Multiple Futures
Duke Energy takes a scenario-based approach rather than building around a single demand forecast. Appendix K of Duke’s Carolinas planning materials opens with a Local Economic Study that models 16 hypothetical large-load customers ranging from 100 MW to 500 MW, evaluated under both summer and winter planning conditions. The exercise is designed to identify where transmission upgrades may be needed before specific projects are even selected.
Duke’s Multi-Value Strategic Transmission process then prioritizes investments that remain justified across multiple futures. Engineers map transmission-constrained areas, referred to as red zones, to help developers understand where new generation is most likely to require network upgrades. The appendix also covers 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.
Dominion Energy: Separate Commercial Risk From Grid Risk
Dominion Energy draws a clear line between two distinct problems. On the commercial side, the utility categorizes prospective data center customers by the type of agreement in place, sorting them into engineering agreements, construction agreements, and binding electric service agreements, rather than treating all inquiries as equally likely to result in actual load.
On the technical side, Dominion’s transmission appendix examines grid reliability independently of the commercial pipeline. Engineers run separate analyses of import capability, system inertia, frequency response, short-circuit strength, and black-start capability using PJM planning models. One analysis finds that planned transmission upgrades will increase the amount of power the Dominion zone can import from the broader PJM market.
The appendix also cautions that transmission capacity does not guarantee energy availability if dispatchable generation elsewhere in PJM continues to decline during extreme weather conditions. Dominion’s planners are evaluating technologies including virtual inertia, grid-forming inverters, and synchronous condensers to assess how a shifting generation mix could affect system stability over time.
A Common Thread
Despite their different methodologies, all three utilities share a core principle: announced AI data center demand is not the same as reliable load, and transmission investments must be evaluated against a range of possible futures rather than a single optimistic projection. For hosting operators and data center developers, the practical takeaway is that the level and type of contractual commitment made with a utility will influence how, and whether, that project factors into long-range grid planning.