Whatu2019s the key difference between Reserved, Spot, and Onu2011Demand instances?

Reserved Instances offer lower pricing in exchange for a 1u2011 or 3u2011year commitment to a specific instance configuration; Spot Instances offer the deepest discounts (up to ~90%) but can be interrupted by AWS with a 2u2011minute warning; Onu2011Demand Instances require no commitment and guarantee availability, but at higher cost.

How much can I save with Reserved Instances versus Onu2011Demand?

Savings vary by term and payment optionu2014typical figures: 1u2011year Reserved (No Upfront) saves ~35%, 1u2011year (All Upfront) ~41%, 3u2011year (All Upfront) ~60% or more compared to Onu2011Demand.

When should I use Spot Instances?

Use Spot for workloads that are faultu2011tolerant and can handle interruptions, such as batch jobs, analytics, CI/CD pipelines, and stateless web tiersu2014because Spot gives up to ~90% savings but comes with interruption risk.

Can I mix all three instance types in my architecture?

Absolutely. A common FinOps strategy is to use Reserved Instances for baseline production, Spot for flexible or batch workloads, and Onu2011Demand to cover overflow or unpredictable demandu2014this yields significant cost savings while maintaining capacity.

How do I manage the risk of overu2011committing with Reserved Instances?

Monitor utilization via AWS Cost Explorer, favor Convertible RIs if you expect changes, and resell unused RIs via the AWS Reserved Instance Marketplace to mitigate risk.

Reserved vs Spot vs On‑Demand Instances: AWS Pricing Model Comparison

Spread the love

Use a blend of On‑Demand, Reserved, and Spot instances to minimize AWS cost without compromising reliability—knowing when to pick each model saves thousands annually.

This post may contain affiliate/partner links. If you buy or sign up through them, we may earn a commission at no extra cost to you.

Last updated: September 12, 2026

Table of Contents

What are Reserved, Spot, and On‑Demand Instances?

Reserved, Spot, and On‑Demand are three AWS EC2 instance pricing models. Reserved Instances (RIs) are billing discounts in exchange for a 1‑ or 3‑year term commitment to a specific instance configuration. Spot Instances are unused capacity available at deep discounts but are interruptible with a short notice. On‑Demand requires no commitment, and you pay per second/hour of usage with full flexibility and availability.

Reserved Instances: billing discount with commitment for specific instance type, region, purchase term; applies automatically when matching instances run ([docs.aws.amazon.com](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-reserved-instances.html?utm_source=openai)).
Spot Instances: spare EC2 capacity offered at up to ~90% off On‑Demand; may be terminated with 2‑minute warning ([aws.amazon.com](https://aws.amazon.com/ec2/faqs/?utm_source=openai)).
On‑Demand Instances: pay‑as‑you‑go, no long‑term commitment, guaranteed availability (unless AZ capacity exhausted) ([aws.amazon.com](https://aws.amazon.com/compare/the-difference-between-on-demand-instances-and-reserved-instances/?trk=faq_card&utm_source=openai)).

How much can you save with each model?

Here’s a real‑world example for a m7i.xlarge (4 vCPU, 16 GB RAM) in us‑east‑1 running 24×7 for one year (from CloudMetrics, Jan 2026):

Model	Annual Cost	Savings vs On‑Demand
On‑Demand	~$1,752	baseline
1‑Year Reserved (No Upfront)	~$1,138	35%
1‑Year Reserved (All Upfront)	~$1,042	41%
3‑Year Reserved (All Upfront)	~$666	62%
Spot (average)	~$526	70%

The difference between Spot and On‑Demand is over $1,200/year for just one instance; multiply that by hundreds or thousands—cost optimization compounds fast ([cloudmetrics.dev](https://www.cloudmetrics.dev/blog/reserved-vs-spot-vs-ondemand?utm_source=openai)).

When should you use each pricing model?

On‑Demand: When flexibility and availability matter most

Short‑lived, unpredictable, or spiky workloads that don’t justify commitment. Ideal for test, development, sandbox environments or benchmarking ([docs.aws.amazon.com](https://docs.aws.amazon.com/wellarchitected/latest/cost-optimization-pillar/select-the-best-pricing-model.html?utm_source=openai)).
No upfront cost, full control—instances can spawn anytime.
Guaranteed unless an AZ is out of capacity—can use On‑Demand Capacity Reservations for critical runs.

Reserved Instances: Cost‑efficient base for stable workloads

Best for long‑term, steady usage—such as persistent backend, production, or databases.
Choose between Standard RIs (deepest discount up to ~72% vs On‑Demand) and Convertible RIs (flexibility to change attributes at some cost) ([docs.aws.amazon.com](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-reserved-instances.html?utm_source=openai)).
Commit for 1 or 3 years; payment options: All Upfront, Partial, No Upfront.
Fixed pricing over term—makes cost predictable; capacity reservations possible by AZ.
Risks: over‑commitment if workload drops, limited flexibility (unless Convertible), cannot cancel but RIs can be traded/resold via AWS RI Marketplace ([docs.aws.amazon.com](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-reserved-instances.html?utm_source=openai)).

Spot Instances: Deepest discounts for fault‑tolerant workloads

Massive savings—often 70–90% off On‑Demand; no commitment required ([docs.aws.amazon.com](https://docs.aws.amazon.com/pdfs/whitepapers/latest/cost-optimization-leveraging-ec2-spot-instances/cost-optimization-leveraging-ec2-spot-instances.pdf?did=wp_card&utm_source=openai)).
Interruptible—AWS can reclaim with 2-minute notice; average interruption <5% ([d1.awsstatic.com](https://d1.awsstatic.com/whitepapers/architecture/AWS-Cost-Optimization-Pillar.pdf?utm_source=openai)).
Best for batch processing, analytics, CI/CD, distributed workloads, stateless tiers where interruption is acceptable ([aws.amazon.com](https://aws.amazon.com/ec2/faqs/?utm_source=openai)).
Requires automation: auto‑scaling, graceful termination, checkpointing, fallback to On‑Demand when needed ([en.wikipedia.org](https://en.wikipedia.org/wiki/Autoscaling?utm_source=openai)).

Comparison: Quick feature matrix

Dimension	On‑Demand	Reserved	Spot
Cost	Highest	~35–72% off	~70–90% off
Commitment	None	1 or 3 years	None
Reliability	High	High	Low–medium (interruptible)
Use case	Unpredictable, dev/test	Baseline production	Fault‑tolerant, batch

Steps to choose the right model

1. Classify workloads: steady vs bursty vs fault‑tolerant.
2. Use AWS Cost Explorer to analyze historical usage and identify steady-state baselines to convert to RIs or Savings Plans—and spot opportunities in flexible workloads ([docs.aws.amazon.com](https://docs.aws.amazon.com/wellarchitected/latest/cost-optimization-pillar/select-the-best-pricing-model.html?utm_source=openai)).
3. Apply multi-model strategy: use Reserved for baselines, Spot for batch or scalable parts, On‑Demand for overflow and unknowns.
4. Automate orchestration: build auto‑scaling groups using mixed instances with Spot fallback, reserve capacity where needed.
5. Review quarterly: monitor RI utilization, Spot interruption rates, adjust as workload shifts.

Concrete example

Production web app has steady backend (24×7), periodic batch jobs (nightly ETL), and unpredictable testing workloads.

1. - Backend: purchase 3‑Year Standard Reserved for baseline capacity → save ~60%.
  - Nightly ETL: use Spot + checkpointing + fallback to On‑Demand via Spot Fleet → save ~80% while maintaining completion.
  - Testing: On‑Demand or Spot as load allows, using budget limits.
  - Use Cost Explorer to monitor and adjust RI purchases, turnover idle capacity via RI Marketplace.

This mix can cut compute spend by ~50–70% vs On‑Demand only, depending on scale.

Where to go next

For computing cost optimization frameworks and tooling integrations, see our pillar guide AWS cost optimization hub.