On the Existence of AI Equilibrium in Electricity Markets

Vladimir Dvorkin – University of Michigan

AI models inform critical decisions in power system operations and markets. Traditionally, AI models are developed in silos, optimized exclusively for prediction accuracy or decision quality in specific applications. This perspective is now inadequate: As AI adoption proliferates, model performance depends not only on individual design and data but on others’, meaning AI models couple through the grid or market optimization problems they inform. Demand forecast errors influence electricity prices affecting all participants; generation forecast errors may amplify or offset these impacts. These interactions between coupled AI models in operational and market contexts require an equilibrium perspective: treating AI models as strategic agents with interdependent performance and system-wide impact. This talk explores AI impacts in grid operations and electricity markets through game-theoretic equilibrium analysis. We establish feedback loops, including Grid-AI (models interact with the system) and AI-Grid-AI (models interact through the system), then formulate equilibrium problems on these loops, and develop tractable equilibrium-seeking algorithms. Main result: AI equilibria exist in competitive two-stage wholesale electricity markets and improve cost efficiency by 13% relative to agnostic AI adoption.