Small edge data centers must manage cooling and power despite changing workloads, local conditions and limited human oversight. Machine learning models offer a route to data-driven thermal control, while transfer learning may reduce the effort needed to adapt those models to different modular deployments.