AWS Reserved Instances vs Savings Plans — What to Actually Buy in 2026

AWS gives you two ways to save on compute: Reserved Instances (RIs) and Savings Plans. Both offer up to 72% discount over On-Demand pricing in exchange for a 1-year or 3-year commitment. The problem? Most teams buy the wrong one, over-commit, or buy before right-sizing — and end up wasting money instead of saving it. This guide gives you a clear decision framework.

What Are Reserved Instances?

Reserved Instances are a billing discount, not actual instances. When you purchase an RI, AWS automatically applies the discount to any running instance that matches the RI’s attributes (instance type, region, OS, tenancy). There are two types:

• Standard RI — locked to a specific instance family, size, region, and OS. Highest discount (up to 72%). Can be sold on the RI Marketplace.
• Convertible RI — can change instance family, size, OS, and tenancy during the term. Lower discount (up to 66%). Cannot be sold.

What Are Savings Plans?

Savings Plans are a commitment to a dollar amount per hour of compute usage. Instead of committing to a specific instance, you commit to spending (e.g., $10/hour). Two types:

• Compute Savings Plan — most flexible. Applies to any EC2 instance (any family, any region, any OS), plus Fargate and Lambda. Discount up to 66%.
• EC2 Instance Savings Plan — locked to a specific instance family in a specific region (e.g., m5 in us-east-1). Discount up to 72%. More restrictive but cheaper.

Comparison Table

Feature	Standard RI	Convertible RI	Compute Savings Plan	EC2 Instance Savings Plan
Max discount	72%	66%	66%	72%
Change instance family	No	Yes	Yes	No (locked to family)
Change region	No	No	Yes	No
Applies to Fargate/Lambda	No	No	Yes	No
Commitment type	Instance attributes	Instance attributes	$/hour spend	$/hour spend
Can sell unused	Yes (Marketplace)	No	No	No
Size flexibility	Within family (Linux)	Yes	Yes	Within family

When to Use Standard RIs

Use Standard RIs only when:

• You are 100% certain of the instance type (e.g., you’ve been running the same m5.xlarge for 6+ months)
• The workload is steady 24/7 (databases, core API servers)
• You want the absolute maximum discount
• You’re comfortable with the risk — if you need to change, you can try selling on the Marketplace, but it’s not guaranteed

When to Use Convertible RIs

Honestly? Almost never in 2026. Compute Savings Plans offer the same discount with more flexibility. The only edge case is if you want to lock to a specific instance and later convert to a more expensive one (Convertible RIs allow “upgrading” mid-term).

When to Use Compute Savings Plans

This is the default choice for most teams:

• You have steady compute spend but might change instance types
• You use a mix of EC2, Fargate, and Lambda
• You operate across multiple regions
• You want flexibility without much analysis

When to Use EC2 Instance Savings Plans

Use this when:

• You’re committed to a specific instance family (e.g., m5 or c6i) in a specific region
• You want higher discounts than Compute Savings Plans
• You’re OK with size flexibility within the family (m5.large and m5.2xlarge both covered)

How to Calculate What to Buy

Do NOT guess. Use data.

1. Open AWS Cost Explorer
2. Go to Reports → EC2 Usage
3. Set time range to last 3 months
4. Group by Instance Type
5. Look for the steady-state baseline — the minimum number of instances running 24/7
6. Only commit to that baseline, not peaks

# Quick check: find your On-Demand spend baseline
# AWS CLI — get last 30 days EC2 spend
aws ce get-cost-and-usage \
  --time-period Start=2026-03-01,End=2026-04-01 \
  --granularity MONTHLY \
  --metrics "UnblendedCost" \
  --filter '{
    "Dimensions": {
      "Key": "SERVICE",
      "Values": ["Amazon Elastic Compute Cloud - Compute"]
    }
  }' \
  --group-by Type=DIMENSION,Key=INSTANCE_TYPE

AWS also provides Savings Plans recommendations in Cost Explorer under Recommendations → Savings Plans. Use it as a starting point, but verify against your actual usage patterns.

Real Example

Here’s what I did for a production environment running a multi-tenant SaaS platform:

• 3x m5.xlarge running 24/7 for core API — bought EC2 Instance Savings Plan for m5 family (72% discount)
• 2x t3.medium running 24/7 for background workers — covered by the same Compute Savings Plan
• Variable Fargate tasks for batch processing — partially covered by Compute Savings Plan
• Peak hours had 2 additional m5.xlarge — left these On-Demand (don’t commit to peaks)

Result: $1,400/month saved on a ~$4,000/month compute bill. 35% reduction.

1-Year vs 3-Year Term

Factor	1-Year	3-Year
Discount	~40% (varies)	~60-72% (varies)
Risk	Lower — easier to adjust	Higher — locked for 3 years
Best for	First-time buyers, evolving workloads	Stable workloads you’ve run 1+ years
Recommendation	Start here	Graduate to this once confident

My rule: Buy 1-year commitments first. After your first renewal, if the workload is still the same, switch to 3-year for the bigger discount.

Payment Options

No Upfront

Pay monthly. Smallest discount but no capital required. Best for most teams.

Partial Upfront

Pay ~50% upfront, rest monthly. Middle ground. Better discount than No Upfront.

All Upfront

Pay 100% upfront. Maximum discount. Only if you have the cash and certainty.

Recommendation: Start with No Upfront. The difference between No Upfront and All Upfront is usually only 3-5% additional savings — not worth the cash outlay risk for most teams.

Common Mistakes That Waste Money

• Buying before right-sizing — if your m5.xlarge is only using 20% CPU, right-size to m5.large first, then commit. Otherwise you’re locking in waste. See our EC2 right-sizing guide.
• Over-committing — only commit to your steady-state baseline, never your peak usage. Peaks should stay On-Demand or use Spot.
• Forgetting region locks — Standard RIs and EC2 Instance Savings Plans are region-locked. If you move to a different region, the commitment doesn’t follow.
• Not checking utilization — go to Cost Explorer → Reservations → Utilization to see if your existing RIs are actually being used.
• Buying Convertible RIs in 2026 — Compute Savings Plans are almost always the better choice now.
• Letting commitments auto-renew without review — set a calendar reminder 30 days before expiry. Check if your workload still matches.

Decision Flowchart

1. Have you right-sized your instances? No → Do that first
2. Do you have 3+ months of steady usage data? No → Wait and collect data
3. Are you certain of instance family + region? Yes → EC2 Instance Savings Plan
4. Do you use Fargate/Lambda/multiple regions? Yes → Compute Savings Plan
5. Do you need absolute maximum discount on a single type? Yes → Standard RI
6. Not sure? → Compute Savings Plan, 1-year, No Upfront

Frequently Asked Questions

Can I sell unused Reserved Instances?

Yes, you can list Standard RIs on the AWS Reserved Instance Marketplace. Convertible RIs and Savings Plans cannot be sold. There is a 12% service fee on sales.

What happens if I stop using the instance I reserved?

You still pay the commitment. RIs and Savings Plans are billing constructs — the discount applies automatically to matching usage, but if nothing matches, you pay anyway. This is why you should only commit to your baseline.

Savings Plans vs Reserved Instances — which is better overall?

For most teams in 2026, Compute Savings Plans are the best starting point. Same discount as Convertible RIs but with more flexibility (cross-region, cross-family, covers Fargate and Lambda). Only choose Standard RIs if you want maximum discount and are certain of your instance type.

Is there a minimum commitment for Savings Plans?

The minimum is $0.001/hour (essentially nothing). Start small, see how it works, then scale up.

Can I combine Reserved Instances and Savings Plans?

Yes. AWS applies RIs first, then Savings Plans to remaining eligible usage. A valid strategy: use Standard RIs for known database instances, Compute Savings Plans for everything else.