Microsoft has confirmed that Azure reservation exchanges for savings plan-eligible services end on February 1, 2027. Until that date, a reservation purchased for one virtual machine series can be exchanged for another as workloads change. After it, that flexibility is gone for the affected services.
Reservation exchange has quietly underwritten a great deal of Azure commitment purchasing. The ability to correct a sizing mistake made over-commitment low-risk, and many organizations bought reservations more aggressively than their forecast confidence justified because the exit was cheap. That assumption needs retiring along with the policy.
Reservations and savings plans are now genuinely different instruments
The distinction has always existed but has been easy to blur. A reservation commits to a specific configuration — instance series, region, term — and delivers the deepest discount in exchange for that specificity. A savings plan commits to an hourly spend figure and applies automatically across eligible services and regions, delivering a smaller discount in exchange for flexibility.
With exchange available, a reservation behaved like a savings plan with a better rate and some administrative friction. Without it, the choice is real: reservations for workloads whose shape you are confident about over the full term, savings plans for everything else.
How to prepare
The first task is a portfolio review. Pull every active reservation with its expiry date, utilisation percentage and the workload it covers. Reservations running below roughly ninety percent utilisation are already leaking value; those are the candidates to exchange while exchange still exists. Between now and February 2027 there is a window to correct historical sizing mistakes, and after it there is not.
The second task is a confidence assessment on each workload. For each reservation approaching renewal, ask whether the underlying workload will be running on the same instance family in the same region in three years. For a stable production database, the answer is probably yes and a reservation remains correct. For an application tier that is a candidate for containerisation, a migration to a managed service, or a rearchitecture, the answer is no and a savings plan is the safer instrument even at a lower discount rate.
The third task is to rebalance the portfolio deliberately toward a mix rather than a default. Most estates we review are heavily weighted to reservations because that is what existed first and what the tooling recommends most readily. A blended position — reservations on the predictable core, savings plans on the variable layer — usually produces a better risk-adjusted outcome than either extreme, and it is now the only position that remains adjustable after February 2027.
The forecasting discipline this forces
The withdrawal of exchange makes commitment purchasing a forecasting exercise rather than an optimisation exercise. That is uncomfortable but not unreasonable, and it rewards organizations that maintain an actual view of their workload roadmap. If your platform team can tell you which workloads will still exist in their current form in three years, you can buy reservations confidently. If it cannot, savings plans are the correct instrument and the lower discount is the price of that uncertainty.
Worth noting alongside this: Microsoft has extended the Sentinel fifty gigabyte commitment tier promotional pricing through December 31, 2026, allowing customers to lock discounted pricing through March 31, 2027. Sentinel commitment tiers follow similar logic — commit to a data volume you are confident about, not one you hope to reach.
How Lorexus engages
We audit the reservation portfolio for utilisation and fit, execute the exchanges worth making inside the remaining window, and rebuild the commitment mix against a workload roadmap rather than against last quarter's consumption. For estates with significant Azure spend, this is a bounded exercise with a hard deadline. Book a free 15-minute call with our senior engineers.