A skeptic’s take on AI electricity load growth

This Catalyst podcast episode interviews Jon Koomey, an expert on data center energy consumption, to discuss the impact of AI on electricity demand. The interview begins with a historical overview of data center energy use, highlighting periods of rapid growth followed by efficiency improvements. Koomey challenges the prevailing assumption of insatiable AI compute demand, arguing that factors like accuracy limitations and infrastructure constraints could limit electricity load growth. He suggests that while near-term growth is likely, the long-term trajectory may mirror past trends, with a period of rapid growth followed by a leveling off due to efficiency gains. Koomey cites the example of the International Energy Agency's varying projections for 2022 data center electricity use to illustrate the uncertainty in forecasting.