AWS Reserved vs Spot vs On-Demand: Complete EC2 Pricing Comparison
AWS offers four distinct ways to pay for EC2 compute: On-Demand, Spot, Reserved Instances, and Savings Plans. Each trades off between cost, commitment, and flexibility. Choosing the right mix can cut your EC2 bill by 50–70% — choosing wrong means overpaying or getting interrupted at the worst possible time.
This guide breaks down every pricing model, compares them head-to-head with real numbers, and shows you the optimal strategy most teams use.
The 4 EC2 Pricing Models at a Glance
On-Demand is the default — pay by the second with no commitment. Spot gives you spare capacity at 60–90% off, but AWS can reclaim it with 2 minutes' notice. Reserved Instances lock you into a specific instance configuration for 1–3 years at up to 72% off. Savings Plans commit you to a $/hr spend level for 1–3 years, also up to 72% off, but with far more flexibility than RIs.
Most teams should use Savings Plans for their predictable baseline, Spot for interruptible and variable workloads, and On-Demand only for unpredictable spikes. Reserved Instances are largely superseded by Savings Plans for new purchases.
Master Comparison Table
| Feature | On-Demand | Spot | Reserved Instances | Savings Plans |
|---|---|---|---|---|
| Discount | 0% (baseline) | 60–90% | Up to 72% | Up to 72% |
| Commitment | None | None | 1 or 3 years | 1 or 3 years |
| Interruption Risk | None | Yes (2-min notice) | None | None |
| Flexibility | Full | Full | Locked to config | High (Compute SP) |
| Best For | Spikes, prototyping | Batch, CI/CD, ML | Steady baseline | Steady baseline |
| Payment | Per-second | Per-second | Upfront/monthly | $/hr commitment |
| Capacity Guarantee | Yes | No | Yes | Yes |
On-Demand: Maximum Flexibility, Full Price
On-Demand is the simplest model: launch an instance, pay by the second, stop it whenever you want. No commitments, no paperwork, no risk of interruption.
The price is the highest of all models, but the flexibility is unmatched. You're paying a premium for zero commitment.
When On-Demand is the right call
- Prototyping and experimentation — Spin up instances to test architectures, benchmark configurations, or run proof-of-concept workloads. No commitment needed.
- Unpredictable traffic spikes — Auto Scaling groups that handle sudden demand surges. You can't predict the spike, so you can't commit ahead of time.
- Short-term projects — A 2-week migration job or a one-off data processing task. The overhead of purchasing a commitment isn't worth it for days or weeks of usage.
- Workloads that can't tolerate Spot interruption and don't run long enough to justify a Savings Plan.
Many teams default to On-Demand for everything and never revisit. If you've been running the same EC2 instances On-Demand for more than 2–3 months, you're almost certainly overpaying. Even a 1-year No Upfront Savings Plan would save ~37% on that spend.
Spot: Massive Savings, Interruption Risk
Spot Instances use AWS's spare EC2 capacity at discounts of 60–90%. The tradeoff: AWS can reclaim them with a 2-minute warning when they need the capacity back.
Since 2017, Spot pricing uses a flat-rate model — prices change gradually rather than spiking. You pay the current market price, not an auction bid.
Best workloads for Spot
- Batch processing & data pipelines — Checkpoint to S3, restart from last checkpoint on interruption
- CI/CD build runners — Interrupted builds simply requeue; savings at scale are enormous
- ML training — Checkpoint model weights regularly; resume on new instance
- Stateless web tier — Behind a load balancer with ASG, users never see interruptions
- Dev/staging environments — Uptime isn't critical; a t3.micro Spot costs ~$2.23/month
Diversify across 5–10 instance types and multiple AZs. Use capacity-optimized allocation strategy. The more pools you draw from, the lower your interruption probability. See our Spot vs On-Demand deep dive for architecture patterns.
Reserved Instances: The Legacy Discount
Reserved Instances (RIs) were AWS's original commitment discount, introduced in 2009. You commit to a specific instance configuration (type, region, OS, tenancy) for 1 or 3 years and receive up to 72% off On-Demand pricing.
Standard vs Convertible RIs
| Feature | Standard RI | Convertible RI |
|---|---|---|
| Max Discount | Up to 72% | Up to 66% |
| Change Instance Family | No | Yes |
| Change OS | No | Yes |
| Marketplace Resale | Yes | No |
The declining relevance of RIs
Since AWS launched Savings Plans in 2019, Reserved Instances have become increasingly niche. Savings Plans offer the same discount levels with dramatically more flexibility — you can change instance types, sizes, OS, and regions without re-purchasing.
The only scenario where RIs still make sense: you want to resell unused capacity on the AWS Marketplace. Savings Plans can't be resold.
If you have active RIs, let them expire naturally and replace them with Savings Plans. There's no penalty for running both simultaneously — AWS applies the best available discount automatically.
Savings Plans: The Modern Commitment Discount
Savings Plans flip the RI model: instead of committing to a specific instance, you commit to a dollar-per-hour spend level for 1 or 3 years. Usage up to your commitment is discounted; anything above bills at On-Demand.
Compute SP vs EC2 Instance SP
| Feature | Compute SP | EC2 Instance SP |
|---|---|---|
| Max Discount | Up to 66% | Up to 72% |
| Instance Family | Any | Locked |
| Region | Any | Locked |
| Covers | EC2, Lambda, Fargate | EC2 only |
| Best For | Dynamic workloads | Predictable baseline |
For most organizations, Compute Savings Plans are the safest bet. The slightly lower max discount is worth the flexibility to change instance types, move regions, or shift workloads to Lambda/Fargate. Read our full Savings Plans guide for a deep dive.
Real Cost Comparison: 10× m5.large for One Year
Let's put real numbers on this. Running 10 m5.large instances 24/7 in us-east-1 for a full year under each pricing model:
💰 On-Demand
🟡 Spot (~70% discount)
🔵 Reserved Instance (1-yr, No Upfront, Standard)
✅ Compute Savings Plan (1-yr, No Upfront)
Spot is cheapest but risky. RIs and Savings Plans save 37–72% with no interruption risk but require a commitment. On-Demand is the most expensive but the most flexible. The right answer is almost never just one model.
The Optimal Mix Strategy
The best-optimized AWS accounts don't pick one pricing model — they layer all of them. Here's the framework:
1. Cover your baseline with Savings Plans (or RIs)
Identify your minimum steady-state compute — the instances that run 24/7 no matter what. Purchase a Savings Plan commitment sized to this floor. Use Cost Explorer's recommendations to find your optimal commitment level.
2. Fill variable capacity with Spot
Workloads that scale up and down — batch processing, CI/CD, ML training, extra web capacity — should run on Spot. Architect for interruption (checkpointing, diversified instance pools, capacity-optimized allocation) and enjoy 60–90% savings on this layer.
3. Handle unpredictable peaks with On-Demand
Traffic spikes, emergency scaling, and workloads that can't tolerate interruption and exceed your Savings Plan commitment get billed at On-Demand. This should be the smallest portion of your bill if you've sized your Savings Plan correctly.
✅ Example: Optimal Mix for a Typical Web App
Frequently Asked Questions
Which AWS pricing model is cheapest?
Spot is the cheapest at 60–90% off, but it comes with interruption risk. For guaranteed capacity, Savings Plans and Reserved Instances both offer up to 72% off with a 1- or 3-year commitment. The cheapest overall strategy is combining all models — Savings Plans for baseline, Spot for variable workloads, On-Demand for spikes.
Can I combine pricing models?
Yes, and AWS encourages it. Discounts are applied automatically — your Savings Plan or RI rate covers eligible usage first, Spot pricing handles interruptible workloads, and any remaining usage bills at On-Demand. You don't need to assign specific instances to specific pricing models; AWS optimizes the billing for you.
Are Reserved Instances still worth it in 2026?
For most new purchases, Savings Plans are strictly better. They offer the same discounts with far more flexibility. The only reason to choose RIs is if you plan to resell unused capacity on the AWS Marketplace. If you have existing RIs, let them expire and replace with Savings Plans. See our Savings Plans guide for the full comparison.
Find Your Optimal Pricing Mix
CloudBench shows On-Demand, Spot, and Reserved pricing for 600+ EC2 instance types across all AWS regions. See exactly where each pricing model saves you the most.
Compare Instance Pricing →