Optimizing long and short-duration storage for AI data centers

Kyri Baker – University of Colorado Boulder

In this talk, we will discuss an optimization framework we’ve developed for AI data centers that reduces capital and operational costs, while taking into account uptime and carbon goals. In particular, we look at the role that different on-site generation technologies and energy storage (both short and long duration) have on the operation and planning of the data center. We additionally analyze the impact of carbon-free energy (CFE) goals on the optimal resource selection, and how demand response and grid-interactivity of the data center can impact its costs and operational uptime. Lastly, we look at the impact of exposing the data center to time-varying wholesale prices versus fixed prices, and run a Monte Carlo analysis to perform a sensitivity analysis on our results.