Aligning compute resources with the actual workload curve—scaling down during off-peak periods and scaling up during peak periods—can significantly reduce waste from idle compute.