When AI demand shifts to the distribution grid

Utilities have spent the last several years preparing for AI-driven load growth by focusing on hyperscale data centers, transmission infrastructure, generation adequacy, and large-load interconnections. That response was necessary. But it assumes future AI demand continues arriving in large, visible increments.

This paper explores a different possibility: a growing share of AI demand arrives through commercial buildings, branch offices, local inference infrastructure, and AI-enabled devices. As models become more efficient and enterprises increasingly prioritize cost, latency, and data privacy, AI processing moves closer to users. What begins as a technology trend quickly becomes a utility planning challenge.

The result is a forecast asymmetry. Utilities may be forecasting the right quantity of demand while planning for the wrong distribution of demand. A gigawatt spread across thousands of commercial buildings and millions of customer endpoints creates a fundamentally different planning problem than a gigawatt concentrated in a handful of hyperscale facilities.