Frontrow Technology
← All insights & guides
Guide

Azure Cost Optimisation

Azure Reserved Instances vs Savings Plans: Which Saves More (Australia 2026)

A plain-English Australian guide to Azure Reserved Instances vs Savings Plans in 2026 — real discount percentages, the flexibility trade-off, and what we'd actually commit.

Daniel Brown · 16 June 2026 · 7 min read

If your Azure bill has crept past a few thousand dollars a month, someone in finance has probably asked the obvious question: are we paying full retail for compute we run around the clock? For most Australian mid-market organisations the answer is yes — and the fix is one of two commitment-based discounts Microsoft offers, Reserved Instances or a Savings Plan for compute. They sound similar, they're often confused, and choosing the wrong one quietly leaves real money on the table.

This is a decision we work through with Brisbane and east-coast clients regularly, so here's the senior-practitioner version: what each one actually is, the real discount percentages, where the catch sits, and what we'd commit to if it were our own bill.

The three ways you can pay for Azure compute

Pay-as-you-go (PAYG) is the default. You're billed per second of compute at list price, you can switch a VM off at any time, and you owe nothing the moment it stops. Maximum flexibility, maximum price. It's the right setting for genuinely spiky or short-lived workloads — dev/test environments, seasonal batch jobs, anything you can't predict a month out.

The other two options trade some of that flexibility for a discount, in exchange for a one- or three-year commitment to a baseline level of usage. That's the whole game: you're telling Microsoft "we'll definitely use at least this much", and they reward the certainty.

  • Reserved Instances (Azure Reservations) — commit to a specific VM family in a specific region.
  • Savings Plan for compute — commit to a flat hourly dollar spend across most compute, any family, any region.
  • Pay-as-you-go — no commitment, full list price.

Reserved Instances: deepest discount, narrowest fit

A Reservation is a commitment to a particular VM series (say, the Dsv5 family) in a particular region (say, Australia East) for one or three years. In return Microsoft applies the steepest discount it offers — up to around 72% off pay-as-you-go on a three-year term for stable Linux workloads, with one-year terms landing lower (indicative figures from Microsoft and industry FinOps sources — confirm against your own SKUs at purchase).

Layer on Azure Hybrid Benefit — using Windows Server or SQL Server licences you already own with Software Assurance — and the effective saving on those eligible workloads can climb higher again. For a steady production estate that isn't going anywhere, RIs are the most aggressive lever available.

Savings Plans: a bit less off, far more flexible

A Savings Plan flips the model. Instead of committing to a SKU, you commit to a fixed hourly dollar amount of compute spend — for example, "we'll spend at least AUD-equivalent of a set rate per hour on compute" — for one or three years. Azure then automatically applies the discounted rate to whatever eligible compute you run, up to that hourly commitment, across VM families, regions, and even App Service, Container Instances and Functions.

The headline discount is lower — up to around 65% on a three-year term (indicative — confirm at purchase) — but you've bought yourself room to move. Resize a VM, shift a workload from Australia East to Australia Southeast, swap families during a re-platform, and the discount follows your spend rather than evaporating.

The trade-off versus RIs is roughly a 7-percentage-point gap at the top end. On a $500,000 annual compute spend that difference is in the order of $35,000 a year — not nothing, but the right question is whether your estate is stable enough to actually capture the deeper RI rate without stranding the commitment.

The commitment trap most teams don't read until it's too late

Before you commit a dollar, understand the exit. Azure Reservations can be cancelled for a prorated refund, but Microsoft caps refunds at roughly USD 50,000 per billing profile in a rolling 12-month window (indicative — verify current policy), and the exchange rules tightened in recent years. One-year reservations can generally be exchanged through their term; three-year exchanges are more constrained outside the grace period because they're processed as a cancel-and-repurchase.

Savings Plans are blunter still: there's no exchange and no cancellation once purchased — you ride out the term. The one useful escape hatch runs in one direction: you can trade an existing Reservation in for a Savings Plan at any time, but not the reverse. So a Savings Plan is the more flexible product to run, yet the harder one to walk away from. Plan the commitment level conservatively.

What we'd actually do

For most Australian mid-market estates we land on a blended approach rather than picking a single winner.

  1. 1Measure first. Pull 60–90 days of Azure Cost Management data and find your true compute floor — the spend that's on every hour of every day. Don't commit above it.
  2. 2Reserve the truly fixed core. For long-lived production VMs on a settled SKU and region — domain controllers, line-of-business servers, stable databases — buy Reserved Instances and stack Azure Hybrid Benefit where you hold the licences.
  3. 3Cover the flexible middle with a Savings Plan. For compute that's steady in volume but likely to move family or region as you modernise, a Savings Plan captures most of the discount without locking the SKU.
  4. 4Leave the spiky top on PAYG. Dev/test, batch, and anything seasonal stays pay-as-you-go so you're never paying for idle capacity.
  5. 5Start with one-year terms if you're unsure. The three-year discount is deeper, but a one-year commitment is a sensible way to validate your baseline before locking in for the longer haul.

The headline percentages matter less than getting the baseline right. A 72% discount on a reservation you've stranded is worse than a 65% Savings Plan that keeps applying. Match the commitment to how stable each slice of your estate genuinely is, review it every quarter as workloads shift, and treat the whole thing as an ongoing FinOps habit rather than a one-off purchase.

All figures here are indicative AUD list — confirm at purchase, and always price your own SKUs and regions in the Azure pricing calculator with currency set to AUD ex GST before you commit, because discount percentages vary by VM family, term length and region.

Common questions

Frequently asked

Is an Azure Savings Plan worth it over Reserved Instances?
It depends on how stable your workloads are. Reserved Instances deliver the deeper discount (up to ~72% on three-year terms, indicative) but lock you to a specific VM family and region. A Savings Plan gives a slightly lower discount (up to ~65%) but applies automatically across families and regions. If your estate is settled, RIs save more; if you're modernising or likely to resize and relocate workloads, a Savings Plan usually captures more real-world value because the discount follows your spend rather than stranding.
Can I combine Reserved Instances and a Savings Plan?
Yes, and we usually recommend it. Reserve the truly fixed core of your estate — long-lived production VMs on a stable SKU and region — with Reserved Instances, cover steadier-but-movable compute with a Savings Plan, and leave spiky or seasonal workloads on pay-as-you-go. Azure applies Reservations first, then the Savings Plan, then PAYG for anything above your commitments.
What happens if I commit and then need to cancel?
Reservations can be cancelled for a prorated refund, but Microsoft caps refunds at roughly USD 50,000 per billing profile in a rolling 12-month window (verify current policy), and three-year exchanges are constrained outside the grace period. Savings Plans can't be cancelled or exchanged at all once purchased — you run out the term. You can trade a Reservation in for a Savings Plan at any time, but not the other way. Size every commitment to your guaranteed usage floor to avoid being trapped.
Should I choose a one-year or three-year term?
Three-year terms carry the deeper discount, but a one-year term is the safer starting point if you haven't yet validated your true compute baseline. Commit one-year on anything you're less certain about, prove the workload is permanent, then step up to three-year at renewal for the bigger saving.

Want us to run this with your team?

30 minutes. No deck. We'll walk through your tenant, your priorities, and the next sensible move.