AI data centers threaten grid stability with correlated power loads

A new paper from Chaudhary, Abdelkader, Pei, Benidris, and Mitra, accepted for the 2026 IEEE Power & Energy Society General Meeting, tackles a growing blind spot in grid planning: the spatial load correlation introduced by AI data centers. Traditionally, power system analysis assumes that loads fluctuate independently, providing natural diversity that stabilizes frequency and voltage. But as data centers — often clustered in the same regions, running similar AI workloads — all ramp up or down together, their demand profiles become highly correlated. The authors derive analytically how these correlated fluctuations amplify aggregate stochastic disturbances, weaken reactive power stiffness (reducing voltage stability margins), and erode frequency stability by destroying load diversity.

Using real-time digital simulations, the team confirms that moderate spatial correlation among distributed data centers produces simultaneous frequency deviations and voltage fluctuations across multiple buses — phenomena that conventional planning tools would miss. The paper urges transmission system operators to adopt a physics-based perspective, designing stability criteria grounded in measurable load-correlation structures rather than outdated diversity assumptions. This is a critical wake-up call for utilities and hyperscalers alike as AI compute continues to scale.