In this article:
- Why AI data centers create power grid challenges beyond their total electricity use
- How rapid growth is changing grid planning and connection rules
- Why successful integration will require both data centers and the grid to adapt
Across the United States, the electric grid is under growing pressure from an old problem and a new one. Much of the system is aging, while electricity demand is rising quickly, driven in part by the rapid expansion of artificial intelligence and the data centers that support it.
A U.S. Department of Energy report estimated that data centers consumed approximately 4.4% of U.S. electricity in 2023 and could account for between 6.7% and 12% by 2028. Those estimates include conventional data centers, AI facilities, and cryptocurrency operations. The wide range also demonstrates how uncertain the pace of growth remains.
Integrating these facilities is not simply a question of supplying enough electricity. Some large AI workloads can change their power use quickly, and data centers may disconnect from the grid during a disturbance. Utilities and grid operators must plan for both the scale of the new demand and the way these facilities interact with the system.
Why AI data centers create new grid reliability risks
Unlike many traditional industrial users, AI data centers can create highly variable patterns of electricity demand. During large-scale model training, power use may rise and fall rapidly as computing tasks change. How much of that variation reaches the grid depends on how the facility is designed and whether it uses energy storage, on-site generation, or systems that help smooth those changes.
There is also a physical dimension to this problem. Many large power plants and grid components were designed to operate steadily, not to repeatedly increase and decrease their power output. When generating units are cycled more frequently to match shifts in electricity demand, those changes place additional stress on rotating components, including the long shafts and couplings that connect the turbine to the generator. Over time, this mechanical fatigue can increase maintenance needs and shorten the life of some components.
Data centers rely on sensitive electronic equipment and protective systems that may disconnect equipment when voltage changes suddenly. If protective systems at multiple facilities respond to the same grid disturbance, a large amount of electricity demand can disappear almost simultaneously.
That happened in Northern Virginia in July 2024. After a fault on a high-voltage transmission line, protective systems at data centers across the region disconnected approximately 1,500 megawatts of electricity demand from the grid almost at once, according to the North American Electric Reliability Corporation.
When demand drops by that much so quickly, power plants may still be producing electricity at the previous level. The resulting imbalance can push grid frequency or voltage outside its normal range until system controls restore balance. Protective systems remain essential, but data center operators and grid operators need to coordinate how those systems respond.
For the public, this does not mean that every data center threatens a blackout. It does mean that large AI data centers must be modeled and operated as significant parts of the power system rather than treated like ordinary commercial customers. Greater uncertainty and volatility can complicate planning, increase operating costs, and require new infrastructure.
Planning for rapid and uncertain electricity demand
The scale and speed of development create another challenge. Major grid upgrades typically take years to plan, approve, and construct, while data center development decisions can move much faster. This mismatch in timelines creates tension and risk.
Utilities need credible demand forecasts to plan investments effectively. Without better coordination and information sharing between data center operators and grid planners, it becomes harder to determine how much infrastructure will be needed, where it will be needed, and how quickly it must be available.
Federal and state policymakers are beginning to respond. In June 2026, FERC directed the six regional grid operators it oversees, including PJM, to justify or revise the rules governing how data centers and other large electricity users connect to the grid. The commission identified questions involving how projects are studied, who pays for grid upgrades, what information developers must disclose, and whether data centers can reduce or shift their electricity use when the grid is strained.
In Pennsylvania, Gov. Josh Shapiro signed an executive order on Aug. 18 establishing new requirements for data center projects seeking state permits and benefits. Among other provisions, the order calls for developers to cover the cost of the new energy infrastructure their projects require, disclose energy and water consumption, and meet requirements involving environmental protection, workforce development, and community engagement.
These actions do not resolve the underlying engineering problems. They demonstrate, however, that the technical questions surrounding large-load integration are increasingly influencing grid rules and public policy.
How data centers and the grid can adapt
A central question is whether data centers should adapt to the grid or whether the grid should adapt to them. Should large computing facilities be required to better manage their demand, for example, by smoothing out spikes through energy storage or on-site generation? Or should utilities invest in new infrastructure to meet these new loads? In practice, successful integration will probably require both.
Another question is which energy sources are best suited to balance rapid shifts in electricity demand. Some generating facilities, including natural gas plants, can increase or decrease their output relatively quickly. Energy storage and demand-side management can also provide flexibility. In some locations, new power plants and grid infrastructure will still be necessary. Effective integration will probably require both more flexible data center operations and targeted investments in the grid.
What effective grid integration will require
Major uncertainties remain. Researchers and utilities still lack reliable data on how AI-driven electricity demand will grow, how large it will become, and how variable it will be over time.
Better coordination between data center operators and utilities could help align expectations and reduce surprises. Grid modernization — including advanced monitoring, control systems, and expanded energy storage — could improve flexibility and resilience. Demand-side management may also play an important role.
The Penn State workshop examined these needs across three broad areas: grid integration, data center architecture and design, and legislation, policy, and electricity markets. One concern emerged across the discussions: The shortage of power system engineers is becoming an important constraint. Utilities, grid operators, manufacturers, and research organizations need people who can perform increasingly complex studies, analyze how new facilities will affect the grid, and develop the systems needed to connect them safely and reliably.
Addressing that workforce need will require sustained investment in power engineering education, research, and collaboration between universities and industry. Penn State is exploring the possibility of developing an industry-informed research effort focused on AI data center integration.
For consumers, the stakes are high. The electric grid has long operated out of sight, but the rise of AI is bringing it into sharper public view. The decisions made now will shape whether the grid continues to support modern life quietly and reliably or becomes a growing point of strain.
Nilanjan Ray Chaudhuri is a leading expert in power system dynamics and control. His research addresses critical challenges in maintaining the reliability and resilience of modern electric power grids, including the impacts of rapid demand shifts from data centers, renewable integration, and phenomena leading to cascading blackouts and/or infrastructure damage. Through advanced lab facilities and simulation platforms, his team replicates real-world grid behavior to develop innovative solutions.
