The Rise Of AI-Ready Microgrids: What's Driving A New Wave Of Behind-The-Meter Innovation

George Sakellaris is the founder, chairman of the board, president and chief executive officer of Ameresco.

Over my career building energy infrastructure solutions, I’ve seen demand shift and grow in ways that reshape power delivery. AI is now driving that change, and meeting the demand will require more than what the grid alone can deliver.​

Data center power infrastructure is being pushed harder than ever. Facilities designed to train and run AI workloads are drawing more power, at higher density, than earlier data centers. Goldman Sachs Research forecasts that U.S. data center power demand will “more than double,” rising from 31 gigawatts in 2025 to 66 gigawatts by 2027, driven by AI infrastructure build-out.

Grid operators and developers are working to keep pace, but in many regions, connecting new capacity to the grid now takes years instead of months, and supply constraints on transformers, switchgear and other core equipment are adding further delay.

All of this is changing how the industry approaches infrastructure planning. Rather than relying solely on the grid, they are increasingly turning to on-site power and behind-the-meter solutions. Reliable power infrastructure has become as critical to an AI facility’s success as the compute itself. Speed-to-power has emerged as a key differentiator for developers balancing fast deployment with long-term resilience. Dedicated on-site energy infrastructure and microgrids are increasingly helping organizations secure both.

Understanding The Load Profile Behind AI Demand

That load profile brings several challenges to the grid at once. First, the amount of capacity required by a single campus can be measured in tens or hundreds of megawatts, and the largest developments are beginning to approach utility-scale demand.