AWS bills you at on-demand rates until you commit. Reserved Instances (RI) and Savings Plans (SP) are the mechanisms for that commitment. With proper analysis, savings on EC2, RDS, and Lambda range from 40% to 72% compared to on-demand. Get it wrong, and you lock up money in reservations that go unused. A typical company's monthly cloud bill contains hidden waste—you are likely overpaying by 30–40% due to suboptimal reservation use. For a client with a $50,000 monthly bill, we cut costs to $25,000 after implementing Compute SP. Our AWS Certified Solutions Architects have over 5 years of experience and have delivered 50+ projects with an average savings of 45%. We guarantee a minimum of 20% savings on eligible spend or we'll refine the strategy at no extra cost. Contact us to analyze your account—we will estimate the project in one day.
Therefore, the goal is not "buy RI" but to identify the stable baseline load and cover it with reservations, leaving the variable part for Spot or on-demand.
Reserved Instances and Savings Plans: Key Differences
Reserved Instances are tied to a specific instance type, region, and optionally Availability Zone. Standard RI offer the maximum discount but are non-convertible. Convertible RI allow you to change instance family, OS, or tenancy during the term—useful if you plan a migration to Graviton.
Savings Plans are more flexible. Compute Savings Plans cover EC2 (any instance family, region, OS), Fargate, and Lambda with a single hourly dollar commitment. EC2 Instance Savings Plans are tied to a specific instance family in a region but offer a slightly larger discount.
For most companies, we now recommend Compute SP: flexibility is more important, especially if you plan to move to Graviton or change instance types. The discount difference between Compute and EC2 Instance SP is typically 5–8%. Compute SP is 2x more flexible than EC2 Instance SP due to covering different services. Also, No Upfront payment is 5% less risky than All Upfront because you avoid capital lock-in.
How to Analyze Workload for Savings Plans?
The main tool is AWS Cost Explorer with Savings Plans Recommendations enabled. But AWS's default recommendations look at the last 30 days. If the last month was atypical (holidays, marketing campaign launch, load testing) — the recommendations will be inaccurate.
The correct approach: analyze 3–6 months of history. Build percentile distribution of consumption — p50 (median) is roughly what should be covered by reservations. The difference between p50 and p90 is a candidate for Spot. Everything above p90 — on-demand or automatic scaling.
Baseline load (p50) → Savings Plans / RI (1-year No Upfront)
Variable load (p50-p90) → Spot Instances + Spot Fleet
Spikes (>p90) → On-demand (Auto Scaling)In Cost Explorer, there is ec2_instance_savings_plan_eligible_spend — look at this metric over time, not just daily snapshots.
More about percentiles
p50 is the value below which 50% of observations lie. For example, if one hour consumes 10 EC2 units and the next hour 20, then p50 = 15. Use p50 as the baseline load for reservation. p90 is the value below which 90% of observations lie — cover this with Spot.Why Compute Savings Plans Are Often Better?
They cover not only EC2 but also Fargate and Lambda. If your architecture includes containers or serverless, this gives a unified optimization approach. According to the AWS Compute Savings Plans documentation, one commitment covers all three services, simplifying management. For example, a company with 30% load on ECS and 10% on Lambda can save on all services with one commitment, rather than buying separate RI for different instance types.
Which Payment Option to Choose?
| Payment Type | Extra Discount vs No Upfront | When It's Worthwhile |
|---|---|---|
| All Upfront | ~5% | If you have excess cash and stable 3-year load |
| Partial Upfront | ~2–3% | Compromise: small savings without full upfront |
| No Upfront | baseline | Best for most — flexibility and no capital outlay |
No Upfront is optimal for most cases: no capital lock-in, minimal loss if you close the account or change architecture.
Comparison of RI and SP
| Characteristic | Reserved Instances | Savings Plans |
|---|---|---|
| Flexibility | Low (tied to instance type) | High (Compute SP covers multiple services) |
| Maximum Discount | Up to 72% | Up to 72% (depends on term and payment) |
| Coverage | EC2, RDS, ElastiCache | EC2, Fargate, Lambda |
| Sell on Marketplace | Yes | No |
Term: 1 Year vs 3 Years
1 year is the standard choice. 3 years offers an even larger discount (~15% extra), but a three-year planning horizon in cloud infrastructure is rare. In three years, instance families can become outdated, regions change, and architecture may shift toward serverless.
Exception: RDS RI for production databases. If PostgreSQL or MySQL on RDS runs stable for several years and there are no migration plans — a 3-year RI is justified.
Step-by-Step: Setting Up Savings Plans
- Collect data: export Cost and Usage Reports for 3–6 months.
- Determine baseline load: calculate p50 of EC2, Fargate, Lambda consumption.
- Choose SP type: Compute or EC2 Instance, based on load stability.
- Choose term and payment: 1 year, No Upfront for most.
- Purchase SP via Console or AWS CLI.
- Set up alerts in Cost Explorer: notification when coverage < 80%.
What's Included in This Service
- Export and analysis of Cost and Usage Reports for 3–6 months
- Recommendations based on actual utilization rate
- Assessment of unused reservations for sale on Marketplace
- Purchase execution via Console or CLI
- Alert configuration for coverage monitoring
- Strategy documentation and monitoring plan
- Team training on Cost Explorer
- Support for strategy review after 6 months
- AWS Cost Explorer access setup and optimization
- Full documentation of the strategy and ongoing support
Estimated Timelines
Analysis and preparation of recommendations — 1 day. Approval and purchase — half a day. Total: 2–3 business days, including iterations for strategy discussion with your team. Pricing is determined after initial assessment.
Want to cut your AWS costs by 40–72%? Order an analysis of your account — we will evaluate the project in one day. Get a consultation on AWS cost optimization now.