Exploring the AI electricity crisis scenario: A case study of Texas-ERCOT

This article explores artificial intelligence (AI) effects on data center electricity consumption by answering the question if and when AI growth may cause an electricity crisis. We study this through combining 3 AI demand electricity and 3 electricity supply scenarios to 9 scenarios and simulating these for estimating their longer-term outcomes on anticipated reserve margins (ARM). These scenarios contain multiple theoretical constructs for explaining AI impact on data centers via a system dynamics narrative, i.e. non-linear predictions with feedback mechanisms through time. We apply our system dynamics simulation model to a specific region because possible conflicts between data center electricity demand and electricity supply capacity manifest themselves only at a regional level. As a case for our simulations, we selected Texas-Electric Reliability Council of Texas (ERCOT): an electricity region covering most of the state of Texas. Being a very energy rich area, we see only a few conditions in which an AI electricity crisis, i.e., an ARM below the reference margin level (RML), may happen in Texas-ERCOT, but a decline of the ARM from 31.2% in 2025 to between 7 (which is below the needed 13.75% RML) and 25% in 2030 with data centers taking about 21–26% of all electricity available may likely happen around 2030. The application of our method in other regions may give very different outcomes, but also the Texas-ERCOT region is not free of risks. While this paper focuses on direct AI impacts, it also suggests the need for future studies exploring the indirect effects of increased data center usage on the economy, society, and ecology.