Why Cloud Pricing Models Matter
The pricing model you choose can affect your cloud bill by 60–90%. A workload costing $1,000/month on on-demand pricing might cost only $300/month with the right commitment discount. Yet many organizations default to on-demand pricing out of inertia or fear of commitment — leaving significant savings on the table.
Understanding when and how to use each pricing model is one of the highest-ROI activities in cloud financial management (FinOps). This guide will help you make informed decisions based on your specific workload characteristics.
On-Demand Pricing
On-demand is the default pricing model: you pay for compute capacity by the hour or second with no long-term commitments. All three major providers support per-second billing (with minimum charges typically of 1 minute).
When On-Demand Makes Sense
- Unpredictable workloads: New applications where usage patterns haven't stabilized
- Short-term projects: Proof of concepts, migrations, and testing
- Spiky demand: Auto-scaling groups that need capacity only during peaks
- First 3–6 months: Of any new workload, to gather utilization data before committing
Comparative On-Demand Pricing (4 vCPU, 16 GB RAM, Linux, US East)
A typical general-purpose instance with 4 vCPUs and 16 GB RAM costs approximately $120–150/month on-demand across all three providers. AWS (m7i.xlarge) and Azure (D4sv5) are typically within 5% of each other, while GCP (n2-standard-4) is often 5–10% cheaper after sustained use discounts.
Reserved Instances and Committed Use
All three providers offer commitment-based discounts, but the mechanics differ significantly:
AWS Reserved Instances
AWS RIs are tied to a specific instance type, region, and tenancy. You choose between three payment options: No Upfront (lowest commitment, ~30% savings), Partial Upfront (~40% savings), or All Upfront (~42% savings) for 1-year terms. 3-year terms amplify savings to 40–60%.
Limitation: Standard RIs lock you to an exact instance type. If you buy a reservation for m5.xlarge, it doesn't apply to m6i.xlarge or c5.xlarge. Convertible RIs offer more flexibility but less discount (~33% for 1-year).
AWS Savings Plans
Introduced as a more flexible alternative to RIs, Savings Plans commit you to a consistent hourly spend rather than a specific instance type. They come in two flavors:
- Compute Savings Plans: Apply across all instance families, regions, and even Fargate/Lambda — maximum flexibility with up to 66% savings
- EC2 Instance Savings Plans: Apply to a specific family in a region, with up to 72% savings — less flexible but deeper discounts
Azure Reserved VM Instances
Similar to AWS RIs but with better flexibility. Azure reservations can be scoped to a single subscription, resource group, or shared across all subscriptions in a billing account. Azure also supports exchanging reservations for different VM sizes within the same family.
GCP Committed Use Discounts
GCP CUDs are resource-based — you commit to vCPUs and memory in a region, not specific instance types. This provides inherent flexibility that AWS and Azure require special reservation types to achieve. 1-year CUDs offer up to 37% savings; 3-year CUDs offer up to 70%.
Spot and Preemptible Pricing
Spot/preemptible instances use spare cloud capacity at massive discounts (60–90%) with the trade-off that your instances can be reclaimed by the provider when capacity is needed.
Key Differences Between Providers
- AWS Spot: Market-driven pricing with dynamic fluctuation. 2-minute interruption notice. Instances can be reclaimed at any time.
- Azure Spot: Configurable maximum price with eviction policies (stop/deallocate or delete). Supports both capacity-only and price-or-capacity eviction types.
- GCP Spot VMs: More stable pricing than AWS. 30-second termination notice. No maximum runtime limit. Pricing is typically 60–91% below on-demand.
Workloads Suitable for Spot Pricing
- Batch processing and ETL jobs with checkpointing
- CI/CD pipelines and automated testing
- Distributed computing frameworks (Apache Spark, Hadoop)
- Stateless web application tiers behind load balancers
- Image and video rendering farms
- Machine learning training with checkpoint/resume support
Making the Decision: A Practical Framework
Use this framework to decide which pricing model to apply to each workload:
Step 1: Assess Workload Stability
Is this workload running 24/7 with predictable resource needs? If yes → consider commitments. If no → start with on-demand.
Step 2: Evaluate Interruption Tolerance
Can the workload handle sudden termination? If yes → consider spot/preemptible for 60–90% savings. If no → use on-demand or reserved.
Step 3: Determine Commitment Horizon
Will this workload run for 1+ years? High confidence → 3-year commitment (maximum savings). Moderate → 1-year. Uncertain → on-demand.
Step 4: Mix and Match
Most organizations use a blend: reserved for baseline capacity, on-demand for auto-scaling peaks, and spot for batch jobs. Target a 70/20/10 split.
Compare Real-Time Prices Across All Models
CloudMetrics lets you compare on-demand, reserved, and spot pricing across AWS, Azure, and GCP in one view. Find the optimal pricing model for your workload.
Compare Cloud Prices →