Cloud Commitment Strategies: Reserved Instances, Savings Plans, and CUDs

On-demand cloud pricing is the retail price — full markup, maximum flexibility. Commitment discounts are the wholesale price — lower cost, reduced flexibility. The typical enterprise saves 30-60% on compute by making the right commitments. The typical enterprise also wastes 15-25% of their commitment spend on resources that go unused.

The difference between these outcomes is analysis, not luck.

Commitment Types by Provider

AWS

Reserved Instances (RIs): Commit to a specific instance type in a specific region for 1 or 3 years.

On-demand:  m5.xlarge = $0.192/hr = $1,682/yr
1-yr RI:    m5.xlarge = $0.120/hr = $1,051/yr  (37% savings)
3-yr RI:    m5.xlarge = $0.075/hr = $657/yr    (61% savings)

Savings Plans: Commit to a dollar amount of compute per hour, applied across any instance type.

Compute Savings Plan: $10/hr commitment
  Applied to: EC2, Fargate, Lambda — any instance type, any region
  1-yr discount: ~30%
  3-yr discount: ~50%

Recommendation: Savings Plans for baseline compute. RIs only when you are certain about instance types for 3 years.

Azure Reservations

1-year reservation: 30-40% savings
3-year reservation: 50-60% savings

Applied to VMs, SQL Database, Cosmos DB, Storage, and more. Azure also offers Azure Hybrid Benefit for Windows/SQL Server licenses.

GCP Committed Use Discounts (CUDs)

1-year CUD: 37% discount
3-year CUD: 55% discount

Resource-based CUDs commit to vCPUs and memory, not specific machine types. Spend-based CUDs commit to a dollar amount.

Coverage Analysis

Before making commitments, analyze your usage patterns:

Step 1: Baseline Usage

Pull 90 days of usage data and identify your steady-state baseline:

Hour-by-hour compute usage (last 90 days):
  P10:  45 instances  (overnight minimum)
  P50:  68 instances  (typical workday)
  P90:  112 instances (peak periods)
  Max:  247 instances (Black Friday)

Step 2: Identify Commitment Layers

Layer 1: Always-on baseline (P10) → 3-year commitment
  45 instances × 3-year price = Maximum savings

Layer 2: Typical load (P10-P50) → 1-year commitment
  23 instances × 1-year price = Good savings, moderate risk

Layer 3: Peak load (P50-P90) → On-demand
  44 instances × on-demand price = Full flexibility

Layer 4: Burst (P90-Max) → Spot/Preemptible
  135 instances × spot price = Lowest cost for interruptible work

Step 3: Financial Modeling

                    On-Demand    Optimized     Savings
Layer 1 (45):       $75,686/yr   $30,274/yr   $45,412  (60%)
Layer 2 (23):       $38,684/yr   $27,079/yr   $11,605  (30%)
Layer 3 (44):       $73,997/yr   $73,997/yr   $0       (0%)
Layer 4 (burst):    Variable     Variable      Spot ~70%
─────────────────────────────────────────────────────────
Total:              $188,367     $131,350      $57,017  (30%)

Commitment Pitfalls

Over-Commitment

Committing to 100% of current usage ignores future changes:

Workloads migrate to containers or serverless
Business priorities shift, services are decommissioned
Instance types become outdated (new generations launch)

Rule: Commit to no more than 70-80% of your P10 baseline for 3-year terms.

Under-Utilization Monitoring

Track commitment utilization monthly:

Commitment: 50 × m5.xlarge reserved
Utilized:   38 × m5.xlarge
Waste:      12 × m5.xlarge × $0.075/hr = $7,884/yr

Set up alerts when utilization drops below 85%.

Ignoring Flexibility Options

AWS RIs offer flexibility tiers:

Standard: Cheapest, least flexible (cannot change instance type)
Convertible: Slightly more expensive, can exchange for different instance types
Savings Plans: Most flexible, applied automatically across services

For most organizations, Compute Savings Plans provide the best balance of discount and flexibility.

Organizational Process

Quarterly Review Cycle

Month 1:  Pull usage data, run coverage analysis
Month 2:  Model scenarios, get finance approval
Month 3:  Purchase commitments, update tracking dashboard

Governance

Who approves: Finance + Engineering jointly (not engineering alone)
Maximum term: 1-year default, 3-year requires VP approval
Break-even analysis: Every commitment includes a break-even calculation
Expiration tracking: Calendar reminders 90 days before each commitment expires

Tools

Tool	Strengths
AWS Cost Explorer RI/SP Recommendations	Native, free, decent for simple cases
Azure Advisor	Built-in reservation recommendations
GCP Committed Use Recommendations	Automated CUD suggestions
CloudHealth / Apptio	Multi-cloud, enterprise governance
Spot.io (NetApp)	Automated commitment management

Anti-Patterns

Anti-Pattern	Consequence	Fix
Committing 100% of usage	Wasted spend when usage drops	Cap at 70-80% of baseline
3-year terms everywhere	Locked in as tech evolves	1-year default, 3-year for proven stable
No utilization monitoring	Invisible waste	Monthly utilization reports
Engineering buys without finance	Budget surprises	Joint approval process
Set and forget	Expired commitments return to on-demand	90-day expiration alerts

Cloud commitments are financial instruments with engineering dependencies. Treat them with the rigor you would apply to any significant capital expenditure — and the flexibility you need in a rapidly changing technology landscape.