Project Title: “Dynamic Modeling and Stability Assessment of Large Data Center Loads for Utility Transmission Planning“

Project Summary:

Data centers differ fundamentally from traditional loads. Switch-mode power supplies with active power factor correction exhibit constant-power behavior (negative incremental impedance). UPS systems decouple IT loads from the grid during disturbances, with facility-specific ride-through behavior typically undisclosed to planners. Variable frequency drive (VFD)-coupled cooling systems contribute zero rotational inertia. In addition, AI training workloads exhibit power fluctuations up to 15 times greater than conventional cloud computing, with GPU synchronization events causing tens-of-megawatt swings at sub-second to minute timescales. Such new load characteristics are new problems for grid planning.

The rapid proliferation of hyperscale data centers is reshaping the load landscape of the US bulk power system. PJM projects 30 GW of data center load growth by 2030, representing 94% of its total load increase [1]. Dominion Energy’s contracted capacity queue has reached approximately 47 GW, nearly triple its historical peak. Duke Energy reports that data centers now account for half of its contracted pipeline for 2029. The impact of such rapid load growth on the power grid cannot be understated.

The inadequacy of existing planning tools and models was demonstrated in the July 2024 Northern Virginia incident, where a 230 kV transmission fault caused approximately 1,500 MW of data center load to simultaneously disconnect [2]. This event was not predicted by any planning model. NERC subsequently issued a Level 2 Alert in September 2025 identifying inadequate dynamic modeling as a systemic reliability gap [3], and the DOE has directed FERC to finalize large load interconnection rulemaking by April 2026 [4]. Those activities necessitate better models for dynamic studies.

The WECC Composite Load Model (CMPLDW) [5], the most advanced standard model in PSS/E, was designed for mixed residential, commercial, and industrial loads. It does not capture UPS battery dynamics, AI workload pulsing, cooling load, or coordinated simultaneous disconnection of collocated data centers. To the best knowledge of the investigators, no validated, PSS/E-compatible data center dynamic load model exists in any commercial tool library.

Objectives: This project addresses two interconnected research needs with a two-year work plan:

1. Dynamic Data Center Load Model (Year 1): A phasor-domain positive-sequence dynamic model for large data center loads is to be developed and validated in ANDES (open-source) [6-7], capturing UPS ride-through, AI workload pulsing, and VFD cooling dynamics.
2. Eigenvalue-Based Stability Assessment and PSS/E Deliverable (Year 2): An eigenvalue analysis framework is to be developed for characterizing oscillation risks from data center loads, including
   Rev. 6 Page 3 of 8
   electromechanical modal interactions and torsional screening. The model will be ported to PSS/E User Defined Model (UDM), and a transmission impact study will be delivered.

 

University Team Members:
Lead PI: Dr. Hantao Cui – Clemson University

Co-PI’s: Dr. Kevin Tomsovic – Clemson University and Dr. G. Kumar Venayagamoorthy – NC State University

Industry Advisors:

Drew Clarke – Duke Energy

Rebecca Rye – Dominion Energy