Both promise up to 72% off your AWS bill. Both demand a one-to-three year commitment. But they work very differently — and picking the wrong one can lock real money into capacity you never use.
Use our Cloud Cost Calculator to compare real pricing across AWS, Azure, GCP, Backblaze, Wasabi and more — side by side, in seconds.
Try the Free Calculator →If you have been running workloads on AWS for any meaningful stretch of time, you have probably noticed that the default on-demand pricing adds up fast. A single EC2 instance running around the clock can quietly eat through budget that nobody planned for. AWS offers two main ways to lock in lower rates: Reserved Instances and Savings Plans.
On the surface, both look like they solve the same problem. You make a commitment, you get a discount. But the details matter a lot here, especially when you are deciding where to put real money on the line for one to three years. This article gets into what each option actually does, where the numbers land, and which one makes more sense depending on what your cloud environment looks like.
AWS bills everything at on-demand rates by default. That pricing model is designed for unpredictability. You spin something up, you pay for it, you shut it down. Fine for occasional or experimental workloads. Not fine for production systems that run 24/7.
AWS offers discounts of up to 72% compared to on-demand rates through both Savings Plans and Reserved Instances. The catch is that you have to commit upfront to a term of either one or three years. In exchange for that commitment, AWS gives you a substantially lower hourly rate.
The question is not whether to commit. For most teams running stable infrastructure, it is almost always financially sensible to do so. The real question is which commitment model fits your situation better.
You commit to a specific instance type, size, and region. AWS bills you for that reservation whether it is running or not.
Up to 72% off Instance-locked Capacity guaranteeYou commit to spending a minimum of $X per hour on compute. AWS automatically applies discounts to your most expensive usage first.
Up to 66% off Flexible coverage Auto-appliesThe core structural difference: RIs lock you to a configuration; Savings Plans lock you to a spending level.
AWS Reserved Instances are a billing discount that lets you commit to a specific EC2 instance type in a particular AWS Region for a one-year or three-year term. In exchange for that commitment, AWS offers discounts of up to 72% compared to standard on-demand pricing. Reserved Instances are not pre-provisioned physical servers. They are a pricing construct applied to eligible on-demand usage in your account.
Think of it like reserving a parking spot in a garage. You pay for the spot whether your car is there or not. If your car is always there, you win. If it is not, you are still paying.
There are two main classes: Standard and Convertible. They sit at opposite ends of a flexibility-versus-discount spectrum.
Standard Reserved Instances average a 40% discount on one-year commitments and 60% on three-year commitments. Convertible Reserved Instances offer 31% on one-year and 54% on three-year terms, but allow you to change instance families, operating systems, tenancy, and payment options during the term.
Standard RIs give you the deepest discount, but you are locked into a specific configuration. If you want to move from an m5.large to a c6g.large six months in, you cannot do that without going to the Reserved Instance Marketplace to sell it — which is cumbersome and often results in financial loss.
When you buy a Reserved Instance, you can tie it to a specific Availability Zone or let it apply regionally. Zonal RIs provide a capacity reservation, meaning AWS guarantees it can spin up your instance in that zone even during high-demand periods. Regional RIs do not offer that guarantee, but the discount applies automatically across all AZs in the region without you having to manage placement.
For most teams, the regional option is more practical. But if you are running something where guaranteed capacity is mission-critical, like a financial system that needs to scale on demand during peak hours, the zonal reservation is worth considering.
Savings Plans are a commitment to a dollar amount of compute usage per hour, regardless of instance type. Instead of saying "I will run this specific instance," you are saying "I will spend at least $X per hour on compute for the next one to three years."
This is a fundamentally different structure. Instead of locking into a machine configuration, you are locking into a spending level. AWS then automatically applies the discounted rate to your usage, pulling from your most expensive instances first to maximize your savings.
Compute Savings Plans are the most flexible. They cover Amazon EC2, AWS Fargate, and AWS Lambda across all instance families, operating systems, and regions with discounts around 66% off on-demand. You can change instance types, move regions, shift from EC2 to Lambda or Fargate, and the discount follows automatically.
EC2 Instance Savings Plans are narrower. You commit to a specific instance family in a specific region (for example, M-series in us-east-1), but within that family you can vary the size and operating system. The discount reaches up to 72%, closer to what Standard RIs offer.
SageMaker Savings Plans are purpose-built for machine learning workloads and apply specifically to SageMaker instance usage.
AWS processes your hourly usage and applies the Savings Plan rate to the most expensive usage first. Any usage beyond your committed hourly amount is billed at standard on-demand rates. So if you commit to $8 per hour and your actual usage is $12 per hour, the first $8 gets the discounted rate and the remaining $4 is on-demand.
Cover your baseline, not your peak. Over-committing is just as wasteful as not committing at all. A good rule of thumb is to commit to about 80% of your minimum daily compute spend and leave 20% on-demand as a buffer.
| Feature | Standard RI | Convertible RI | EC2 Instance SP | Compute SP |
|---|---|---|---|---|
| Max discount (3yr, all-upfront) | 72% | 66% | 72% | 66% |
| Max discount (1yr) | ~40% | ~31% | ~40% | ~33% |
| Flexibility | Very low | Medium | Medium | High |
| Covers Lambda & Fargate | No | No | No | Yes |
| Capacity reservation | Zonal only | Zonal only | No | No |
| Resale on marketplace | Yes | No | No | No |
| Management overhead | High | Medium | Low | Low |
If you can guarantee that your workload will run the same instance type in the same region for three years, a Standard Reserved Instance or an EC2 Instance Savings Plan will give you the deepest discount. If your architecture is evolving or you use a mix of compute services, a Compute Savings Plan gives you nearly equivalent savings with far less operational headache. For a deeper look at how these stack up against Spot Instances, see Reserved vs On-Demand vs Spot Instances: A Cost Breakdown.
Both Reserved Instances and Savings Plans offer three payment structures: All Upfront (pay everything now for the best discount), Partial Upfront (split between upfront and monthly), and No Upfront (pay monthly at the lowest discount tier).
The discount difference between all-upfront and no-upfront is typically 5 to 8 percentage points. If you have the capital and high confidence in your usage forecast, all-upfront is the most cost-efficient. If forecasting is uncertain, no-upfront reduces your risk at a modest cost. Understanding where hidden costs accumulate is just as important — Hidden Costs in Cloud Billing: What Your Provider Isn't Telling You covers this in depth.
Run your compute spend through our Cloud Cost Calculator to see projected savings across commitment tiers before locking in a multi-year term.
Try the Free Calculator →Savings Plans have been gaining ground, and AWS itself now recommends them over Reserved Instances for most use cases. But there are scenarios where RIs are genuinely the better choice.
Stable, long-running database workloads. RDS Reserved Instances are the only commitment-based pricing option for relational databases because Savings Plans do not cover RDS. RDS reserved instance pricing cuts on-demand rates by 29% on the lowest one-year commitment and up to 69% on a three-year all-upfront term. For a db.r8g.xlarge PostgreSQL running Multi-AZ, the difference between on-demand and three-year reserved is roughly $4,800 per year on a single instance.
Capacity guarantees matter to you. Zonal Reserved Instances are the only AWS discount mechanism that also reserves physical capacity. If your workload needs guaranteed spin-up capacity in a specific AZ during high-demand periods, a zonal RI delivers that. Savings Plans do not.
Large-scale, predictable EC2 fleets. At very large commitment volumes, Standard RIs unlock volume discount tiers that can add a few extra percentage points off once you exceed certain regional purchase thresholds. These tiers are quite high, so only large enterprises hit them, but they are worth knowing about if you are operating at that scale.
We covered this in depth: Ep 1 — Rewriting the Cloud Playbook with Backblaze CEO Gleb Budman — a candid conversation on egress fees, vendor lock-in, and why cloud cost strategy matters more than ever.
Listen to the Episode →If you run EC2, Lambda, and Fargate together, Compute Savings Plans are the only model covering all three with one purchase.
Startups and scale-ups change instance families as they optimize. A Savings Plan lets you do that without penalty or marketplace hassle.
Savings Plans apply automatically. No tracking which instances are covered, which have expired, or which purchases an architecture change might strand.
Compute Savings Plans apply across all regions automatically. Reserved Instances are region-specific and will not follow migrated workloads.
For the majority of teams, Savings Plans offer nearly identical savings to Reserved Instances while providing superior flexibility, easier management, and automatic discount application. The additional discount from Reserved Instances is at most around 3%, and that marginal difference rarely justifies the extra risk and operational complexity.
The most sophisticated cloud cost strategies do not pick one or the other. They use both, each where it fits.
Use Savings Plans for variable workloads that extend into Lambda or Fargate, and combine them with Reserved Instances for stable, EC2-centric baselines. In practical terms: Standard RIs for your production database fleet and core services that have not changed instance type in two years, plus a Compute Savings Plan sitting on top to catch everything else. Pairing this approach with smart auto-scaling strategies can reduce your effective cloud spend even further.
AWS is not the only place these decisions come up. In Azure, Savings Plans can cut eligible compute spend by up to 65%, while Reserved Instances can reach up to 72%, or up to 80% in some eligible Windows or SQL cases with Azure Hybrid Benefit. GCP takes a different approach with Committed Use Discounts, where you commit to vCPU and memory rather than specific instance types.
The core principle holds across all three clouds: predictable workloads benefit from reservation-style commitments, variable or evolving workloads benefit from flexible spending commitments.
Neither option is universally better. What matters is how closely the commitment structure matches your actual usage patterns.
Reserved Instances still hold the edge in raw discount depth for stable, specific workloads, especially where you need capacity guarantees or are covering RDS. Savings Plans win on flexibility, operational simplicity, and breadth of coverage across the modern AWS compute stack.
For most teams, the practical answer is to start with Savings Plans as your default commitment vehicle, layer in Standard RIs only for services you are genuinely confident about, and avoid over-committing in either direction. That combination tends to deliver the best effective savings rate without trapping budget in unused reservations.
The worst thing you can do is nothing. On-demand rates for predictable workloads are money handed directly back to the cloud provider with no reason for it.
Join 1,200+ CTOs, architects, and cloud professionals who get our weekly briefing on storage strategy, GPU compute, and cloud cost intelligence.
Subscribe Free →
Compare AWS, Google Cloud, Azure, and alternatives like Backblaze B2 Discover how much you could save in seconds
