Should we use Azure Reservations, Savings Plans, or a mix for changing workloads?

I wouldn’t commit to the entire environment yet. First identify the boring baseline that runs consistently, such as VMs that stay online around the clock, and cover only that portion. Keep the rest flexible until the container migration and VM changes settle down.

Before buying anything, review several months of usage by VM family, region, uptime, and expected migration timeline. If the workload is mostly steady within one family and region, a reservation may be worthwhile. If it regularly changes shape or is moving to containers, flexibility is probably worth more than the maximum discount. For interruptible development or test jobs, Spot VMs may also reduce costs substantially if interruptions are acceptable.

A mixed strategy is often the most practical: use Reservations for the stable baseline and a Savings Plan for workloads that move between eligible compute options. You can also add Savings Plan commitments gradually instead of covering everything at once—for example, start with a conservative amount, watch usage for a month or two, and then add another layer if the baseline holds.

Reservations usually provide the better discount, but they’re more specific. They can work well when you stay in the same region and VM family, especially if you’re resizing within that family through instance-size flexibility. If VM types and regions change frequently, though, it becomes harder to use the reservation fully. Savings Plans are more flexible because they commit you to a spending level rather than a particular VM configuration, but the discount is generally lower and the commitment can’t simply be canceled if usage drops.