On September 20, 2024, Constellation Energy announced that it signed a power purchase agreement with Microsoft to reopen Three Mile Island nuclear plant to power data centers, including those that operate artificial intelligence (AI) programs. On June 3, 2025, Meta signed a similar deal with Constellation Energy to buy about 1.1 gigawatts of energy from the facility’s nuclear plant in Illinois. These are only two examples of the recent boom in planned development of data centers built to power AI, and consequently the need for the development of new power generation and transmission capacity. Since 2010, the number of companies that have taken concrete steps to develop data centers nationwide has nearly quadrupled.
Data centers present unique challenges to the grid. First, AI requires a significant amount of energy to operate. Second, even if there is enough electricity generation to power the data centers, but the generation is not co-located with the data center, transmission capacity must also be expanded to ensure that the electricity can reach from the generation site to the data centers. The growing demand for data centers, generation and infrastructure begs the question, can the grid handle the emerging loads associated with such demand?
The Federal Energy Regulatory Commission (FERC) recently initiated a review of the issues associated with large loads, such as AI data centers. The North American Electric Reliability Corporation (NERC) submitted a comment in the FERC proceeding illustrating just one aspect of how data centers can increase the complexity of grid management. NERC described a large load loss event in the Eastern Interconnection in July of 2024 where a lightening arrestor failed on a 230 kV transmission line and resulted in a permanent fault that “locked out” the transmission line. The fault led to a loss of approximately 1,500 MW of voltage-sensitive load, consequently causing a data center loss of power from the grid. In this incident, the sudden loss of large load caused frequency and voltage to rise, which were, thankfully, successfully managed by the operators with the result that the event did not cause significant risk to the reliability of the grid. NERC observed that data centers are sensitive to voltage disturbances and may use uninterruptible power supply (UPS) to take over power supply until the loads are manually reconnected to the grid. While such programs protect the data center, NERC emphasized that reconnection of the large loads to the grid after use of UPS is an intricate exercise that must be performed in a controlled manner by grid operators. Ultimately, the increased installation of data centers, the need for more power to meet their needs, and grid reliability in light of unforeseeable or unpreventable events may continue to cause issues in the future.
The question is how to address this?
First and foremost, if the need for data centers to be built continues, then new generation will need to be built to meet the demand - a demand that typically outpaces the forecasted growth of the local utility. All forms of generation are suitable, but realistically the generation - gas, nuclear, wind, solar, etc. - needs to be either readily available or can be built and operational in a time frame that meets the needs of the data center without compromising the utility infrastructure. Often co-location with the data centers is sought out but in several markets where real estate square footage is at a premium this is not always possible.
NERC’s comment included several ideas for improvements to reliability associated with co-located large loads. These include increasing proximity between large loads and power generation sources to reduce energy loss, improve transmission reliability, and foster coordination between the utility (or seller of the power) and the end user to manage the load better and reduce the strain on the Bulk Power System (BPS). Proximity also improves grid stability by creating flexibility to adjust demand during critical conditions. NERC noted that nuclear energy, of which Microsoft and Meta recently purchased to power their data centers, is particularly advantageous for co-located large loads because of its inherent high-capacity factor, load-carrying ability, and consistent performance even during extreme weather conditions, a benefit for data centers which typically run 24 hours a day. However, due to the complexity of siting and building both traditional and small nuclear generating facilities, forecasts predict that only 10% of the necessary nuclear capacity needed to meet the demands of data centers will be available by 2030. Accordingly, it appears that while nuclear energy may have a prominent role in powering the AI boom, it will necessarily be only one of multiple energy sources used. Thus, no one solution will solve the pressure to the grid presented by the AI boom, and the complexity of the challenges invites the opportunity for nuanced and thoughtful approaches.
For more information on data centers, the impact on grid reliability and how data centers and utilities can plan successfully to meet the growing needs for generation while not compromising reliability, or how DWGP may be of assistance, please reach out to Sean Neal, Lisa Gast or Keith Gordon.
Article By DWGP Summer Associate Anne Vicari - The George Washington University School of Law, May 2026
