Contact Us
Contact Us
Cloud Cost Savings

Savings Plans vs Reserved Instances: What to Commit To and What to Avoid

How AWS and Azure commitment discounts work, where the risk sits for finance and engineering, and a simple rule for how much to commit.

JRC Software Team
7 min read
Savings Plans vs Reserved Instances: What to Commit To and What to Avoid

The CFO has read that AWS offers discounts of up to 72% for a one or three-year commitment, and wants to know why you haven't bought any. The platform lead is halfway through moving services to containers and doesn't want to be tied to today's instance types. Both have a point. So the purchase sits in a backlog for another quarter, and you keep paying On-Demand rates.

Savings plans vs reserved instances is really a question of how much flexibility you need and how much risk you're willing to carry. Commit too little and you overpay every hour. Commit too much and you pay for capacity you're not using, with no way to cancel.

Below: how each option works on AWS and Azure, the traps, and a rule for sizing the commitment that finance and engineering can both sign.

More commitment options, faster-changing estates

The menu has grown. AWS now has four Savings Plans types, including Database Savings Plans that cover Aurora, RDS, DynamoDB, ElastiCache and other database services. Azure offers savings plans for both compute and databases.

Meanwhile, estates change faster than they used to. Teams move to containers and serverless, adopt newer processor families, and add AI workloads with very different compute patterns. A commitment that fitted perfectly when you bought it can be a poor fit long before the term ends. That makes flexibility at least as important as the headline percentage.

Savings plans vs reserved instances: what you're actually committing to

Both are billing discounts, not capacity. Your resources run exactly as before; the provider applies a lower rate to matching usage. The difference is the thing you promise.

  • Reserved instances (RIs) and Azure reservations commit you to a specific configuration, such as an instance type in a region.
  • Savings plans commit you to a fixed spend per hour (say, USD 10 an hour of eligible compute) that applies across eligible services and configurations.

AWS options compared

AWS's comparison of Savings Plans and RIs lays out the trade-off:

OptionMax discount vs On-DemandWhat it's tied toFlexibility
Compute Savings PlansUp to 66%Hourly spendAny instance family, size, OS, tenancy or Region; also Fargate and Lambda
EC2 Instance Savings PlansUp to 72%Hourly spend in one instance family and RegionAny size, OS or tenancy within that family
Convertible RIsUp to 66%Instance configurationCan be exchanged, but you do the exchanges manually
Standard RIsUp to 72%Instance configurationCan be modified, not exchanged; can be sold on the RI Marketplace

Database Savings Plans go up to 35% across eligible database services, and SageMaker AI Savings Plans up to 64%. AWS's own Reserved Instances documentation now opens with a plain statement: "We recommend Savings Plans over Reserved Instances."

Azure options compared

Microsoft's guidance on choosing between a savings plan and a reservation follows the same logic. Use reservations, at up to 72% off pay-as-you-go, for workloads that run continuously with no expected change to instance type, family or region. Use savings plans when workloads are changing, span several services or run in more than one region.

Our default choices

  • Compute that will change: Compute Savings Plans on AWS, compute savings plans on Azure. You give up a few points of discount for the freedom to modernise.
  • Stable, well-understood workloads: EC2 Instance Savings Plans or Azure reservations, for the deeper discount.
  • Databases: Database Savings Plans on AWS, or database savings plans or reservations on Azure, once the engine and size have settled.
  • Term: one year by default. Three years only where there's a genuine multi-year roadmap for that workload.
  • Payment: AWS offers All Upfront, Partial Upfront and No Upfront for Savings Plans, as it does for RIs, and its RI documentation notes that a higher upfront payment generally saves more. Let finance weigh that against cash flow. On Azure, upfront and monthly payment cost the same in total.

Traps we see in commitment purchases

  1. Committing before cleaning up. Microsoft's recommended sequence starts with right-sizing, because "Discounts reduce rates, not waste." Do the quick wins first; our 10 quick wins guide covers them.
  2. Assuming you can cancel. AWS Savings Plans can't be cancelled during the term, and EC2 RIs can't be cancelled either. Azure savings plans can't be cancelled or refunded. Azure reservations can be refunded, but only up to USD 50,000 in a rolling 12-month window across your agreement.
  3. Forgetting it's per hour. On Azure, unused savings plan commitment for an hour expires and doesn't roll over. AWS works the same way in practice: you pay the hourly commitment whether or not usage turns up. A weekend dip below your commitment is money gone.
  4. Missing expiry dates. EC2 RIs don't renew automatically. When they expire, the instances keep running at On-Demand rates, and the first anyone hears of it is a bigger bill. Put expiry dates in a shared calendar, or turn on renewal where the provider offers it.
  5. Covering work that comes and goes. Savings Plans don't apply to Spot usage. Keep batch jobs, CI runners and short projects off your commitments.
  6. One team buying alone. Engineering buys without finance, or finance buys without the roadmap, and someone gets surprised. Make it a joint decision with both names on it.

Commit to the floor, not the average

This is the rule we use with clients:

Commit to the floor, not the average. Look at your hourly eligible spend over a recent period after clean-up. Find the level it rarely drops below, nights and weekends included, and commit to that. Everything above the floor stays on demand, Spot or pay-as-you-go.

Then layer it:

  1. Clean up first so the floor reflects real need, not waste.
  2. Buy in tranches, a portion now and more each quarter, rather than one big purchase. Staggered start dates also mean commitments expire at different times, which gives you regular chances to adjust.
  3. Make the first layer flexible. Compute Savings Plans or Azure compute savings plans for the base. Add instance-specific RIs or reservations only for workloads you're certain about.
  4. Check utilisation and coverage monthly. AWS Cost Explorer and Azure Advisor both give commitment recommendations and utilisation reports. If utilisation drops below the high 90s, stop buying and find out why before the next tranche.

This deliberately leaves some savings on the table. That's the point. Being slightly under-committed costs a little. Being over-committed costs you every hour for the rest of the term.

Our part in a commitment purchase

Through our cloud cost optimisation service, a savings assessment starts with an hour-by-hour export of your eligible usage across AWS and Azure for at least the last 30 days, and longer where you have it. We strip out waste we can see, find the floor, and put it next to your engineering roadmap: which services are being containerised, which instance families are on the way out, which regions you plan to leave.

You get a recommended mix of savings plans and reservations, term and payment option, a tranche schedule with dates, and the expected saving in dollars at each step. After purchase, we set up utilisation and coverage reports with alerts so you know when commitments are under-used and when it's safe to buy more.

What we'll often advise against: three-year terms on anything being modernised, and covering 100% of current usage. If you run AI workloads, the same discipline applies to provisioned model throughput, as described in our FinOps for AI guide.

Not sure how much to commit? Book a savings assessment and we'll model the options against your own usage.

Frequently asked questions

What is the main difference between savings plans and reserved instances?

Reserved instances commit you to a specific configuration, such as an instance type in a region, in exchange for a discount. Savings plans commit you to a fixed spend per hour and apply the discount automatically to eligible usage. On AWS, Compute Savings Plans also cover Fargate and Lambda and work across Regions. Savings plans trade a little discount for much more flexibility.

Are AWS Savings Plans better than Reserved Instances?

For most compute, yes, and AWS itself recommends Savings Plans over Reserved Instances. EC2 Instance Savings Plans match Standard RI discounts of up to 72%, and Compute Savings Plans match Convertible RIs at up to 66% while applying automatically across families, sizes and Regions. Standard RIs can still suit very stable workloads, partly because they can be sold on the RI Marketplace if needs change.

Can I cancel a savings plan or reservation if my usage drops?

Generally no. AWS Savings Plans and EC2 Reserved Instances can't be cancelled, although some RIs can be modified, exchanged or sold. Azure savings plans can't be cancelled or refunded. Azure reservations can be exchanged for another reservation of the same type, and refunded up to USD 50,000 in a rolling 12-month window. That's why we size commitments conservatively.

How much of my cloud spend should be covered by commitments?

There's no universal percentage. Our rule is to commit to the floor of your steady hourly spend after removing waste, not to your average. Buy in tranches over several quarters, start with flexible plans, and review utilisation monthly. That keeps unused commitment low while still capturing most of the available discount.

Cloud Cost Savings

Book a savings assessment

Tell us where you're stuck and we'll come back with practical next steps.

Get in touch

Sources