If you want to know how to lower your Azure bill, the two biggest discounts Microsoft offers — Reserved Instances vs. Spot pricing — look tempting at first glance. Reserved Instances promise up to 72% off pay-as-you-go rates for a one- or three-year commitment, while Spot VMs can slash prices by up to 90%. The catch is that they serve opposite use cases: one rewards steady, predictable workloads, and the other rewards workloads that can be interrupted at any moment. Buy the wrong one and you either overpay or get evicted mid-task. This guide walks through the real math and tells you exactly which discount fits which workload.
How much can Azure Reserved Instances actually save you?
An Azure Reserved Virtual Machine Instance is a simple trade: you commit to using a specific VM size in a specific region for one or three years, and Microsoft drops the hourly rate. With a one-year term, the discount typically lands around 30–45% versus pay-as-you-go; with a three-year term, it can reach up to 72%.
Here is the break-even math on a typical general-purpose VM — a D2s v3 running full-time in East US (prices illustrative; check the current Azure pricing calculator before committing):
| Billing model | Illustrative monthly cost | Savings vs. pay-as-you-go |
|---|---|---|
| Pay-as-you-go | ~$70 | — |
| 1-year Reserved Instance | ~$45 | ~36% |
| 3-year Reserved Instance | ~$25 | ~64% |
| Spot VM (if eligible) | ~$12 | ~83% |
The critical rule for Reserved Instances: only reserve what runs 24/7. A VM that runs eight hours a day, five days a week is on for roughly 176 hours out of 730 each month. A one-year reservation billed for 730 hours costs more than just paying the on-demand rate for 176. Reserved pricing wins when your usage curve is flat — production web servers, always-on databases, steady batch processing — not when it spikes and dips.
Microsoft offers two scopes to soften the commitment risk. A shared scope reservation floats across subscriptions in your billing account, so if you shut down one VM, the discount automatically applies to another matching VM. If your footprint is small and shifting, choose shared scope instead of locking a reservation to a single resource group. You can also exchange reservations when your VM needs change, so a wrong guess is not a three-year prison sentence — but exchanges have limits, so start with a one-year term the first time you reserve.
How much do Azure Spot VMs save — and what’s the catch?
Azure Spot Virtual Machines use spare datacenter capacity that Microsoft is not selling at full price right now. Because that capacity can be reclaimed at any moment, the discount is dramatic — up to 90% off pay-as-you-go. But the catch is real and non-negotiable: Azure can evict your Spot VM with as little as 30 seconds’ notice whenever the capacity is needed elsewhere.
That single fact sorts every workload into eligible or not:
- Good Spot candidates: stateless batch jobs, rendering farms, CI/CD build agents, big data processing, dev/test environments, machine-learning training checkpoints, containerized microservices that reschedule themselves.
- Bad Spot candidates: production databases, single-instance web servers, anything stateful that cannot checkpoint, workloads with strict SLAs.
Think of Spot VMs the way a budget airline sells standby seats: the ticket is cheap precisely because your seat is not guaranteed. If your job can survive being paused and restarted — a nightly ETL pipeline that retries automatically, for example — Spot is the cheapest compute Microsoft sells. If it cannot, the “savings” will cost you an outage.
Reserved Instances vs. Spot Pricing: which one fits your workload?
Stop comparing the two discounts as rivals and start comparing your workloads instead. Run this three-step decision for every VM in your subscription:
Step 1: Is the workload interruptible? If no — a customer-facing app, a database, a VM that must answer in seconds — Spot is out. Reserved Instances or a savings plan are your only real levers.
Step 2: Does it run most of the month? If the VM idles at night and on weekends, neither discount may beat simply shutting it down. A D2s v3 running 176 hours a month at pay-as-you-go costs about $17 — less than a one-year reservation at ~$45. Auto-shutdown and start/stop automation beat discounts for part-time workloads, and they are free to implement.
Step 3: Match the winner to the pattern. Steady 24/7 workload → one-year Reserved Instance (start here; upgrade to three years once the pattern proves stable for six months). Bursty but interruptible workload → Spot VMs with checkpointing. Bursty and non-interruptible → pay-as-you-go plus auto-shutdown. Predictable but fluctuating compute across many VM families → look at Azure savings plans, which commit to an hourly spend amount rather than a specific VM size and apply more flexibly than reservations.
One more practical note: you buy Reserved Instances in the Azure portal under Reservations, and the discount applies automatically to matching running VMs. You do not tag anything. If you buy a reservation and your bill does not drop within the hour, you bought the wrong VM size or region — check the reservation’s utilization report before buying a second one.
Can you combine Reserved Instances and Spot VMs in the same subscription?
Yes — and mature cloud teams usually do. The textbook pattern looks like this: a base layer of Reserved Instances covers your minimum always-on compute (the database, the core API tier), while Spot VMs absorb the elastic overflow (batch jobs, build queues, scale-out workers). The two discounts never fight over the same VM, so there is no double-counting risk.
A concrete mini-example from a real small-team pattern: a team runs a production API on two D2s v3 VMs 24/7 (one-year Reserved Instances, ~$45 each instead of ~$70) plus a nightly data pipeline on four Spot VMs that runs two hours a day (~$0.40 per night instead of ~$4). Monthly compute drops from roughly $160 to under $105 — about a third off — with zero changes to the application code. The only engineering work was making the pipeline retry-safe, which took an afternoon.
If you want to know how to lower your Azure bill further, the next levers after these two discounts are right-sizing (most over-provisioned VMs are one size too big) and cleaning up unattached disks and idle load balancers — the silent line items that discount programs never touch.
Frequently asked questions
Q: What is the difference between Azure Reserved Instances and Azure savings plans?
A: A Reserved Instance commits to a specific VM size and region for one or three years. A savings plan commits to a fixed hourly spend on compute (for example, $2/hour) for one or three years, and the discount floats across VM families and regions. Savings plans are more flexible; Reserved Instances are simpler to reason about for a fixed fleet.
Q: Can I cancel an Azure Reserved Instance if my needs change?
A: Microsoft allows exchanges and, within limits, cancellations of reservations. You can swap a reservation for a different VM size or region, and early termination is possible with a cancellation fee that varies by terms. Because of this, first-time buyers should prefer one-year terms until their usage pattern is proven.
Q: How do I know if my workload is safe to run on Spot VMs?
A: Ask two questions: can the job checkpoint its progress, and can it tolerate a restart within minutes? If yes to both — batch processing, renders, builds, training runs — Spot is usually safe. If the workload holds in-memory state that cannot be rebuilt or serves live users, keep it off Spot.
Q: Do Reserved Instance discounts apply automatically?
A: Yes. Once you purchase a reservation, Azure applies the discount to any matching running VM in the reservation’s scope — no tagging or configuration needed. Check the reservation utilization blade in the portal to confirm your VMs are actually consuming the discount you bought.
