Typically, the largest savings in a cloud environment are realized during a right sizing exercise for the compute resources. Since compute typically dominates costs, it is reasonable to expect that the biggest cost savings can be found here. Right sizing refers to the act of selecting the most adequately sized virtual machine for the given application (size here refers to the amount of CPU and memory allocated to the machine).

This problem is acutely present in environments that were just migrated from on-premises to a cloud environment. The resource sizing process for on-premises typically goes through many approvals and requisite teams, each layer adding extra cushion to CYA. This results in overpowered machines that are then migrated to comparable VMs in the cloud.

If your application can run in clusters and is stateless, it is advantageous to use more numerous, smaller instances. This optimizes for better cost (and performance), since you will be able to scale the cluster to better match demand. AWS and Azure have a rich set of VM sizes to choose from that can match any use case. GCP offers Custom Machine types where the user can set the sizing that's available. This offers even more control over matching machine resources to application demand.