The 60-second answer

Azure Reserved Instances (RIs) are commitments to a specific VM family, region, and OS for one or three years — the most disruption-prone commitment you can make, but they offer the deepest discount (up to 72% off pay-as-you-go on three-year RIs with Azure Hybrid Benefit attached). Azure Savings Plans for Compute are commitments to a dollar amount of compute spend per hour for one or three years, flexible across VM family, region, and OS — with discounts of up to roughly 65% for matched compute use. The right answer is almost never "all RIs" or "all Savings Plans" but a layered strategy: RIs for the predictable workload spine, Savings Plans for the flexible workload tail. Microsoft's sales motion in 2026 pushes Savings Plans aggressively because they lock dollar commitment without locking SKU — convenient for Microsoft, not always optimal for you.

The core difference between Savings Plans and RIs

Both Azure Savings Plans and Reserved Instances are commitment-based discount mechanisms. They differ in what is committed and what is flexible.

Reserved Instances commit you to a specific compute SKU: VM family (e.g., D-series, E-series), size (D8s v5), region (East US 2), and operating system (Windows or Linux). For the duration of the reservation (one or three years), Microsoft charges you a flat hourly rate substantially below the pay-as-you-go list. If you stop running that VM, the reservation continues to bill — the commitment is to the SKU, not the running instance. RIs are exchangeable within the same family and cancellable at a fee, but the day-one assumption is that you know exactly what you are running.

Azure Savings Plans for Compute commit you to a dollar amount of compute spend per hour, redeemable against any compute resource (VM, App Service, Container Instances, Container Apps) across any VM family, region, or OS. For the duration of the plan, Microsoft applies the committed dollar rate against your eligible compute spend at a discounted rate; any excess spend over the commitment is charged at pay-as-you-go. The commitment is to spend, not to SKU.

The trade-off is direct: RIs are deeper discount, less flexible; Savings Plans are shallower discount, more flexible.

Discount depth comparison

The headline discount numbers for 2026:

  • Three-year RI on a typical compute-family VM: 60–65% off pay-as-you-go (without AHB), up to 72% off with Azure Hybrid Benefit applied to Windows workloads.
  • One-year RI: 30–42% off pay-as-you-go depending on family.
  • Three-year Savings Plan for Compute: 55–65% off pay-as-you-go on the SKU mix you actually use.
  • One-year Savings Plan for Compute: 25–35% off pay-as-you-go.

The arithmetic looks like Savings Plans are a small giveaway versus RIs — 3–5 percentage points shallower. The real difference is larger when you account for unused commitment. A three-year RI on a VM family you decommission after 18 months still bills the remaining 18 months. A three-year Savings Plan on a spend rate you no longer need can be reallocated to any other compute resource. The risk-adjusted discount on Savings Plans is often higher than the headline number suggests for organisations with changing workloads.

The Microsoft preference

Microsoft sales heavily favours Savings Plans in 2026. The reason: Savings Plans lock dollar commitment without constraining Microsoft on capacity allocation. RIs reserve compute capacity Microsoft has to honour; Savings Plans reserve dollars Microsoft can fulfil from any capacity. The preference is structural, not buyer-friendly — you should evaluate Savings Plans on their merits for your workload, not because Microsoft prefers them.

Which workloads suit RIs, which suit Savings Plans

The right answer is workload-dependent. Five workload archetypes:

  • Stable production VMs running the same SKU for 3+ years. Classic RI candidates. Production databases, ERP application tiers, identity infrastructure. Three-year RI with AHB attached on Windows workloads.
  • Variable-but-predictable workloads with shifting SKU mix. Savings Plan territory. Web tier that auto-scales, batch processing that runs different VM families seasonally, dev/test pools.
  • Migration-in-progress workloads. Savings Plans, not RIs. Anything mid-migration is too prone to SKU change to lock RI for three years.
  • Spot-eligible workloads. Neither — use Spot pricing for these. Savings Plans do not apply to Spot capacity.
  • Containerised workloads on AKS or App Service. Savings Plans cover these natively; RIs do not.

For most enterprises, the workload distribution is 50–70% stable spine (RI candidates) and 30–50% variable tail (Savings Plan candidates). Pure RI or pure Savings Plan strategies leave money on the table on the other half.

Commitment and exchange rules

The contractual mechanics matter and are easy to get wrong:

RI exchange: you can exchange an RI for one of equal or greater value within the same VM family at no cost. Cross-family exchanges are restricted. Cancellation incurs a fee (up to 12% of remaining commitment, capped at $50K per year). The exchange mechanism is the safety valve that makes RIs less rigid than they look on paper.

Savings Plan flexibility: the committed dollar rate applies against eligible compute spend automatically. You cannot exchange or cancel a Savings Plan once committed — the dollar commitment runs to term. Underuse is unrecoverable; overage above the commitment is billed at pay-as-you-go.

Three failure modes are common:

  • Buyers commit to RIs on workloads that get re-platformed within the term, paying for capacity they no longer use.
  • Buyers commit to Savings Plans at higher dollar rates than they actually use, paying for headroom that delivers no discount.
  • Buyers buy both at full commitment for the same workloads, double-paying.

The right discipline is conservative commitment on Savings Plans (commit at the floor of expected steady-state spend, take pay-as-you-go on the variable top) and aggressive commitment on RIs only for workloads with proven 3-year stability.

Run the layered commitment model for your tenant
We separate your spine from your tail, apply the right commitment to each, and avoid the double-pay traps Microsoft will not flag.
Get the Model

The hybrid strategy that works in 2026

Across our Azure advisory engagements, the highest-performing commitment strategy in 2026 follows a layered pattern:

  1. Layer 1 — the production spine. Identify VMs that have run on the same SKU for 18+ months and are projected to continue for 36+. Cover these with three-year RIs, with AHB attached on Windows where SA is available. This captures the deepest discount on the most certain compute.
  2. Layer 2 — the predictable variable. Identify the floor of compute spend that you are confident will exist regardless of SKU mix (typically 60–75% of total compute). Cover this with a three-year Savings Plan at that dollar rate.
  3. Layer 3 — the variable top. Take pay-as-you-go on the residual variable spend above the Savings Plan commitment. The pay-as-you-go premium on the variable top is more than offset by the deeper discount on the certain layers below.
  4. Layer 4 — Spot for fault-tolerant batch. Anything that can checkpoint and resume goes on Spot at 70–90% off pay-as-you-go.

This four-layer pattern routinely delivers 38–46% blended savings vs full pay-as-you-go pricing on enterprise Azure spend, compared to 20–30% on naive Savings-Plan-only or RI-only approaches.

Anonymised case study: 7,200-VM healthcare client

A 7,200-VM healthcare client running primarily Windows workloads on Azure had defaulted to "everything on a one-year Savings Plan" in 2024 on advice from their LSP. We re-modelled the workload mix: 3,100 VMs (the EHR core, identity infrastructure, ERP back-end) had been on the same SKU for 30+ months and projected to remain for 4+ years; 2,800 VMs were variable-but-floor (predictable daily spend); 1,300 were genuinely variable. We rebuilt the commitment stack as: three-year RIs with AHB on the 3,100 stable VMs; three-year Savings Plan covering 60% of the variable-floor compute spend; pay-as-you-go on the residual. Net annual Azure compute cost reduction: 17.2%, or $2.4M per year against the previous Savings-Plan-only posture. The change paid for the engagement 24 times over in year one.

17.2%
Additional savings on top of the existing Savings Plan posture — achieved purely by re-layering the same compute spend across RIs, Savings Plans, and pay-as-you-go on the right workloads.

2026 decision checklist

  1. Pull 12 months of Azure compute usage by VM family, region, and OS.
  2. Identify SKUs running for 18+ months on the same configuration. These are RI candidates.
  3. Identify the floor of total compute spend — the dollar level you would still be at if all variability disappeared. This is the Savings Plan commitment level.
  4. Apply AHB to all Windows workloads where SA is current.
  5. Reserve 5–10% of compute spend as pay-as-you-go headroom — do not commit to the absolute peak.
  6. Co-term commitments to your MACC commitment dates so renewals stack cleanly.
  7. Re-model annually. The right mix shifts as workloads migrate and SKUs change.

The Microsoft preference is Savings Plans because they are simpler to sell and Microsoft-friendlier on capacity allocation. The right answer for most enterprises is the layered strategy — deep-discount RIs on the spine, Savings Plans on the predictable variable, and pay-as-you-go on the genuinely volatile. Done well, the layered strategy compounds into the difference between an average Azure cost outcome and a top-quartile one.