Cloud Pricing Models Explained: On-Demand vs Reserved vs Spot vs Savings Plans
The Cloud Pricing Menu
Every major cloud offers multiple pricing tiers for the same compute. Choosing the right one for each workload can reduce costs by 50-70% without changing a line of application code.
On-Demand (Pay As You Go)
The baseline pricing model. You pay for exactly what you use, per second (AWS, GCP) or per minute (Azure). No commitment, no upfront cost. Most expensive option — but essential for truly variable workloads.
Use when: You need maximum flexibility, running experiments, workloads with highly unpredictable demand, new applications where sizing is uncertain.
Reserved Instances / Reserved Capacity
Commit to 1 or 3 years in exchange for 40-65% discount. Comes in variants:
| Variant | AWS | Azure | GCP | Discount |
|---|---|---|---|---|
| No Upfront, 1yr | Standard RI | Savings Plan | CUD 1yr | 30-41% |
| Partial Upfront, 1yr | Standard RI | Reservation | CUD 1yr | 37-44% |
| All Upfront, 1yr | Standard RI | Reservation | CUD 1yr | 40-45% |
| No Upfront, 3yr | Standard RI | Savings Plan 3yr | CUD 3yr | 50-58% |
| All Upfront, 3yr | Standard RI | Reservation 3yr | CUD 3yr | 55-63% |
Savings Plans (AWS and Azure)
More flexible than RIs — commit to a spend rate ($/hour) rather than a specific instance. AWS offers:
- Compute Savings Plans: Apply to any EC2, Lambda, Fargate. Maximum flexibility.
- EC2 Instance Savings Plans: Apply to specific instance family in one region. Maximum discount (up to 72%).
Azure Savings Plans: Apply to any compute regardless of VM size, series, or region.
Spot / Preemptible
Use spare provider capacity at 60-90% discount. Provider can reclaim with 2 minutes notice. No commitment required.
Use when: Batch processing, CI/CD, ML training, stateless web servers, any fault-tolerant workload.
Sustained Use Discounts (GCP Only)
GCP automatically applies discounts when instances run more than 25% of the month. No action required — discount applies automatically. Up to 30% for instances running 24/7. This is why GCP quotes often appear lower than AWS/Azure without reservations.
Choosing the Right Model — Decision Framework
| Workload Pattern | Recommended Model | Expected Saving |
|---|---|---|
| 24/7 stable (production DB, core services) | 3yr RI / Reservation | 55-65% |
| Business hours only (8×5) | On-Demand + scheduling | 60-70% via scheduling |
| Mostly stable with occasional peaks | 1yr RI + On-Demand for overflow | 35-45% |
| Batch processing | Spot Instances | 60-90% |
| Highly variable, unpredictable | Savings Plan (flexible) | 30-56% |
| Short-term experiments | On-Demand | 0% (full flexibility) |
Start with a 1-year Reserved Instance or Savings Plan commitment for your stable baseline workloads. Add Spot for variable batch workloads. Revisit Spot after building confidence in your interruption handling. This approach typically achieves 50-60% total compute cost reduction without architectural changes.
Ready to Calculate Your Cloud Costs?
Use TCOIQ's free comparison tool or build a full inventory across all 5 clouds.