Pricing Models Compared: On-Demand, Reserved, Savings Plans, Spot, and Dedicated
A structured comparison of the five main AWS pricing models, when to use each, how they differ in commitment, flexibility, and cost.
Why pricing models matter
AWS offers several ways to pay for compute capacity. Each model trades off between commitment, flexibility, discount depth, and interruption risk. Choosing the right model for each workload is one of the most impactful cost decisions you can make.
The five pricing models at a glance
| Model | Commitment | Typical Savings | Can Be Interrupted? | Best For |
|---|---|---|---|---|
| On-Demand | None | Baseline (0%) | No | Variable or unpredictable workloads |
| Reserved Instances | 1 or 3 years | Up to 72% | No | Steady-state, predictable workloads |
| Savings Plans | 1 or 3 years ($/hr) | Up to 72% | No | Flexible long-term compute needs |
| Spot Instances | None | Up to 90% | Yes (2-min warning) | Fault-tolerant, flexible jobs |
| Dedicated Hosts | On-Demand or Reserved | Varies | No | Licensing compliance (BYOL) |
On-Demand Instances
On-Demand is the default pricing model. You pay by the second (or hour) with no upfront commitment.
Key characteristics:
- No long-term contract or upfront payment
- Highest per-unit cost of all models
- Full flexibility to start and stop anytime
- Billed only for running time
When to use: Development and testing, unpredictable workloads, short-term projects, or any workload where you cannot commit to a usage pattern yet.
Reserved Instances (RIs)
Reserved Instances provide a significant discount in exchange for committing to a specific instance type in a specific Region for 1 or 3 years.
Key characteristics:
- Up to 72% savings versus On-Demand
- Locked to a specific instance type, Region, and tenancy
- Three payment options: All Upfront (deepest discount), Partial Upfront, or No Upfront
- The discount applies automatically to matching running instances
When to use: Steady-state production workloads that run 24/7 with a known instance type, databases, application servers, or baseline capacity you are confident about.
Note
Reserved Instances do not prevent you from running On-Demand instances. If your usage exceeds the reservation, the extra capacity is billed at On-Demand rates.
Savings Plans
Savings Plans are a newer, more flexible commitment model. Instead of locking to a specific instance type, you commit to a consistent hourly spend (e.g., $10/hr) for 1 or 3 years.
Two types:
- Compute Savings Plans, the most flexible. Apply across EC2, Lambda, and Fargate regardless of instance family, size, Region, or OS. Up to 66% savings.
- EC2 Instance Savings Plans, locked to a specific instance family in one Region, but flexible on size and OS. Up to 72% savings.
Key characteristics:
- Commit to $/hr, not a specific instance
- Compute Savings Plans cover EC2 + Lambda + Fargate
- Discounts apply automatically to matching usage
- Unused commitment is still billed
When to use: Organizations that want commitment-based savings but need flexibility to change instance types, Regions, or shift between EC2 and serverless over time.
Savings Plans vs Reserved Instances
| Reserved Instances | Savings Plans | |
|---|---|---|
| Commitment unit | Specific instance type + Region | $/hr spend |
| Flexibility | Low, locked to instance type and Region | High, especially Compute Savings Plans |
| Services covered | EC2 (and RDS, etc.) | EC2, Lambda, Fargate (Compute plan) |
| Max discount | Up to 72% | Up to 72% (EC2 plan) / 66% (Compute plan) |
| Best for | Predictable, static workloads | Evolving workloads that may shift services |
Tip
If you know the exact instance type you will run for years, Reserved Instances may give the deepest discount. If you want flexibility to change instance families or shift between EC2 and serverless, Savings Plans are often the better choice.
Reserved Instance flexibility: Standard vs Convertible
Reserved Instances come in two offering classes with different trade-offs between discount depth and flexibility.
| Feature | Standard Reserved Instance | Convertible Reserved Instance |
|---|---|---|
| Discount depth | Up to 72% off On-Demand | Up to 66% off On-Demand (lower than Standard) |
| Exchange capability | Cannot be exchanged | Can exchange for different instance family, OS, or tenancy |
| Modification | Can change AZ, instance size, networking type | Can change AZ, instance size, networking type |
| Best for | Stable, predictable long-term workloads | Evolving needs where requirements may change |
Convertible RI key concept: You can exchange a Convertible RI for another Convertible RI with different attributes (instance family, operating system, or tenancy) during its term. The exchange must result in equal or greater value. This flexibility trades some discount depth for adaptability.
Note
AWS now recommends Savings Plans over Reserved Instances for most use cases. Savings Plans provide similar flexibility to Convertible RIs with automatic application across instance types and even Lambda and Fargate.
Spot Instances
Spot Instances let you use unused EC2 capacity at steep discounts, up to 90% off On-Demand pricing. The trade-off: AWS can reclaim (interrupt) Spot capacity with a 2-minute warning when it needs the capacity back.
Key characteristics:
- Up to 90% savings, the deepest discount available
- Can be interrupted at any time with 2 minutes' notice
- No commitment or upfront payment
- You request capacity; AWS fulfills when available
When to use: Batch processing, data analysis, CI/CD builds, image rendering, or any workload that can checkpoint, pause, and resume without data loss.
Warning
Never use Spot Instances for workloads that cannot tolerate interruption, such as a primary database, a single-instance web server, or any stateful service without redundancy.
Dedicated Hosts and Dedicated Instances
These two options provide physical isolation, but they serve different purposes.
Dedicated Hosts:
- You get an entire physical server dedicated to your account
- Full visibility into sockets, cores, and host ID
- Required for bring-your-own-license (BYOL) compliance with per-socket or per-core licenses (e.g., Windows Server, SQL Server)
- Most expensive option, priced per host
Dedicated Instances:
- Your instances run on hardware not shared with other AWS accounts
- You do not get visibility into the underlying host hardware
- Useful for compliance requirements that mandate single-tenant hardware but do not require socket/core visibility
| Dedicated Hosts | Dedicated Instances | |
|---|---|---|
| Physical isolation | Yes | Yes |
| Socket/core visibility | Yes | No |
| BYOL support | Yes | No |
| Pricing | Per host | Per instance + per-Region fee |
Licensing Strategies: Included vs BYOL
AWS offers two fundamental approaches to software licensing on EC2. The choice depends on whether you have existing license entitlements and what those license terms require.
Quick Reference
License Manager pairing
AWS License Manager automates license tracking and host administration for BYOL workloads. It monitors socket and core usage, enforces license limits, and reduces the operational overhead of managing Dedicated Hosts at scale.
Capacity planning pitfall
Dedicated Hosts are physical capacity allocations. You must plan host capacity upfront. Auto Scaling configurations must account for host availability, not just instance counts. Running out of dedicated hosts blocks instance launches even if your Auto Scaling group requests more capacity.
Decision tree: which model to pick
Ask these questions in order
- Can the workload be interrupted? → If yes, consider Spot Instances first for maximum savings.
- Do you have licensing requirements (BYOL)? → If yes, use Dedicated Hosts.
- Is the workload steady-state for 1+ years? → If yes, compare Reserved Instances vs Savings Plans.
- Do you need flexibility across services or Regions? → Prefer Savings Plans (Compute plan).
- Is the workload unpredictable or short-term? → Use On-Demand.
Micro-activity: Match the workload to the pricing model
Workload → Pricing Model
Connect each workload scenario to the most cost-effective pricing model.
Examples
Choose one, then match it on the right
Characteristics
Select an example first
0 of 5 matched so far.