Click here to download a pdf of the memo.
Executive Summary
The unprecedented energy demands of artificial intelligence infrastructure—projected to be as high as 5 GW by 2030 with quintupling annual growth thereafter—are rapidly outpacing traditional data center power solutions. Grid constraints, multi-year interconnection queues, and extended lead times are forcing the industry to seek alternative power approaches that can scale quickly and reliably.
Behind-the-meter (BTM) gas generation, whereby power is generated on site and consumed directly by the facility, has emerged as a leading solution for several reasons. The technology offers proven reliability, established supply chains, and relatively straightforward permitting compared to alternatives. Recent deployments by companies such as xAI, which uses modular gas turbines in Tennessee, and Novva, which deploys on-site gas generators, demonstrate the viability of this approach.
The technical options for BTM gas generation fall into three categories:
1. Combined cycle plants, which offer the highest efficiency for large-scale deployment.
2. Industrial gas turbines, which can be deployed quickly and provide flexible operating profiles.
3. Traditional backup solutions like internal combustion engines, which remain important but face scaling limitations.
BTM generation still faces significant regulatory challenges. Federal environmental laws impose extensive permitting requirements. State power plant rules subject even BTM projects to lengthy environmental reviews. And growing local opposition to data centers and their supporting infrastructure threatens project approvals.
This memo considers how these regulatory frameworks interact across jurisdictions to affect BTM gas deployment for AI infrastructure. Understanding these barriers is crucial for site selection, technology choice, and development planning as the AI industry continues its rapid expansion.
While this memo emphasizes BTM natural gas generation as a current leading solution, the long-term power architecture for AI facilities remains uncertain. Many data centers may ultimately adopt hybrid approaches combining BTM generation with grid connections; such approaches would introduce additional regulatory complexities beyond this memo’s scope.
