Skip to content

Reserved Instances & Savings Plans — Long-Term Commitment#

"A team running prod on on-demand could have bought 3-year Reserved for 50-70% less. The 'lock-in' excuse costs $X. Correct forecast + commitment ladder = guaranteed savings."

This guide covers practical strategy for AWS Reserved Instances (RI), Savings Plans (SP), GCP CUDs, and Azure Reservations, and how to avoid over-committing.

📊 Cloud Commitment Types#

Type AWS GCP Azure
Compute discount EC2 RI / Compute SP CUDs Reservations
DB discount RDS RI CUDs SQL Reservations
Storage S3 reserved - -

AWS-specific#

Type Discount Flexibility
EC2 Standard RI 40-50% Low (instance type fixed)
EC2 Convertible RI 30-40% Medium (type can change)
Compute Savings Plan 40-66% High (any region/family)
EC2 Instance Savings Plan 50-72% Low (specific region)

🔑 2026 recommendation: Compute Savings Plan (flexibility) > Standard RI.

🎯 Commitment Strategy#

Step 1: Determine the baseline workload#

  • "What's the minimum capacity that will definitely run over 12 months?"
  • Production services, critical workloads
  • Excluding workloads that can run on spot/preemptible

Step 2: Forecast (12-36 months)#

  • Current growth rate
  • New feature roadmap
  • Customer-base change

Step 3: Commitment ladder#

Year 1: 50% baseline → 1-year SP (medium upfront)
Year 1+1: 70% baseline → 1-year SP renewal + additional
Year 1+2: 80% baseline → 3-year SP (highest discount)

🔑 Phased commitment → reduces over-commit risk.

Step 4: Spot + RI hybrid#

Workload composition:
  Stateful DB → On-demand or RI (1-year)
  Stateless prod replica → SP (60% baseline) + on-demand (peak)
  Background batch → Spot
  Dev/staging → Spot

💰 Commitment vs Pay-As-You-Go#

m5.large baseline (1 instance, 24/7):
  On-demand:       $0.096 × 720 = $69/mo
  1-yr no upfront: $0.060 × 720 = $43/mo  (38% savings)
  3-yr no upfront: $0.040 × 720 = $29/mo  (58% savings)
  3-yr all upfront: $24/mo equivalent     (65% savings)

Upfront options#

Type Cash flow Discount
No upfront Equal monthly Low
Partial upfront 40% upfront + monthly Medium
All upfront 100% upfront Highest

💸 All upfront: best when cash flow allows, 3-5% extra discount.

🛡️ Over-Commit Risk#

Scenario#

Year 0: 3-yr SP $1M commitment, 100 instances
Year 1: Workload migration → 50 instances suffice
Year 2: $500K idle commitment (unused)

Mitigation#

  1. Phased commitment ladder (above)
  2. Convertible RI (sell on AWS Marketplace)
  3. Partial commitment: 50-70% baseline (on-demand for peak)
  4. Quarterly review: actual usage vs commit

📊 AWS Cost Explorer — Recommendation#

# Compute Savings Plan recommendation
aws ce get-savings-plans-purchase-recommendation \
  --savings-plans-type COMPUTE_SP \
  --term-in-years ONE_YEAR \
  --payment-option NO_UPFRONT \
  --lookback-period-in-days SIXTY_DAYS

→ AWS recommends: "$Y savings with a $X yearly commitment".

Continuous monitoring#

  • AWS Cost Explorer → coverage % (commitment utilization rate)
  • < 85% coverage → over-committed (idle money)
  • < 70% → buy more SP

🌍 GCP CUDs (Committed Use Discounts)#

gcloud compute commitments create my-commitment \
  --project=<PROJECT> \
  --region=<REGION> \
  --plan=THIRTY_SIX_MONTH \
  --resources=vcpu=10,memory=40
Term Discount
1 year 25-37%
3 years 52-70%

GCP CUDs are resource-based (vCPU + memory), a bit different from AWS RI/SP.

☁️ Azure Reservations#

Type Discount
1-year Reservation 35-40%
3-year Reservation 55-65% 
az reservations reservation-order list

📈 Coverage Dashboard#

# Custom: AWS Cost Explorer API → Prometheus
aws_cost_explorer_savings_plans_coverage_percentage

Targets#

  • Coverage %: 70-85% (not over, not under)
  • Utilization %: > 95% (the used portion of the commitment)
  • On-demand %: < 30% (for peak buffer)

Quarterly review#

  1. Coverage: below 70% → buy more SP
  2. Coverage: 95%+ → over-committed, you may be losing peak
  3. New SP need: 30+ days of steady on-demand pattern

🚫 Anti-Pattern Table#

Anti-pattern Why it's bad Do this instead
All prod on-demand Pay 50-70% too much SP + spot mix
100% commit (no peak buffer) Capacity falls short on a spike 70-80% commit
Standard RI instead of Convertible Lock-in Compute SP / Convertible
3-yr commit on greenfield Workload keeps changing 1-yr ladder
No coverage tracking Over/under-commit Quarterly review
All-upfront without the cash flow Liquidity problem No-upfront or partial
No commitment, "we already have spot" Spot interruption and commit are separate niches Use them together
Committing per account Global pool lost Organization-wide SP (consolidated)
Not knowing about RI marketplace selling Over-commit loss Convertible + sell

📋 Commitment Strategy Checklist#

[ ] Baseline workload analysis (12 months)
[ ] Forecast: growth + feature roadmap
[ ] Commitment ladder plan (1-yr → 3-yr phased)
[ ] 70-80% baseline commit (peak buffer)
[ ] Compute SP > Standard RI (flexibility)
[ ] Coverage dashboard
[ ] Quarterly review (under/over-commit)
[ ] Convertible RI (for flexibility)
[ ] Marketplace sell strategy (over-commit fallback)
[ ] Hybrid: SP (steady) + spot (batch) + on-demand (peak)
[ ] Multi-account: org-wide SP pool
[ ] Budget alarm (commit + actual)
[ ] Annual: commitment review to leadership

📚 References#

"A Reserved/Savings Plan isn't 'lock-in' — it's a forecasted commitment. The right-size + spot + commit trio can cut the cloud bill by 40-60%. A team that does none of them is overpaying at on-demand price."