Cloud Computing Economics
One of the most compelling reasons organizations migrate to the cloud is the economic model shift from capital expenditure to operational expenditure. Understanding cloud pricing models enables cost optimization and informed architectural decisions.
CapEx vs OpEx
CapEx (Capital Expenditure) — Traditional on-premises model:
- Large upfront investment in servers, networking gear, storage arrays, and datacenter facilities
- Assets depreciated over 3–5 years on balance sheet
- Requires capacity planning 12–18 months in advance
- Idle capacity is wasted spend; insufficient capacity causes outages
OpEx (Operational Expenditure) — Cloud model:
- Pay only for resources consumed — no upfront hardware purchase
- Expense recognized in the period consumed (simpler accounting for many organizations)
- Scale up in minutes; scale down to zero when not needed
- Shift financial risk to the provider; no hardware obsolescence
AWS Pricing Models
On-Demand Instances
Pay by the hour (or second) with no commitment. Maximum flexibility, highest per-unit price. Ideal for unpredictable or short-lived workloads.
Reserved Instances (RIs)
Commit to 1 or 3 years of usage in exchange for up to 72% discount vs On-Demand. Two payment options: All Upfront, Partial Upfront, or No Upfront. Ideal for steady-state production workloads.
Spot Instances
Bid on unused AWS capacity. Up to 90% discount vs On-Demand, but instances can be interrupted with 2-minute notice. Ideal for batch processing, ML training, rendering, and fault-tolerant workloads.
Savings Plans
Flexible alternative to RIs: commit to a dollar/hour spend (e.g., $10/hr) across any instance family, region, or OS. AWS provides the discount automatically. Covers EC2, Lambda, and Fargate.
Pricing Model Decision Tree
Worked Example: On-Premises vs Cloud Over 3 Years
Workload: Web application running 2 × 8-core servers, 64 GB RAM each, 10 TB storage
On-Premises (CapEx):
| Item | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Server hardware (2×) | $24,000 | $0 | $0 |
| Storage array | $8,000 | $0 | $0 |
| Networking | $3,000 | $0 | $0 |
| Power + cooling (annual) | $4,800 | $4,800 | $4,800 |
| IT staff overhead (20% time) | $15,000 | $15,000 | $15,000 |
| Total | $54,800 | $19,800 | $19,800 |
| 3-Year Total | $94,400 |
AWS (Reserved Instances, 1-year term):
| Item | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| 2× r6i.2xlarge (8vCPU/64GB) 1-yr RI | $5,200 | $5,200 | $5,200 |
| 10 TB S3 storage | $2,760 | $2,760 | $2,760 |
| Data transfer + misc | $1,200 | $1,200 | $1,200 |
| Total | $9,160 | $9,160 | $9,160 |
| 3-Year Total | $27,480 |
Cloud saves approximately 71% over 3 years in this scenario — and includes zero hardware refresh risk, built-in HA, and instant scalability.
Additional Economics Considerations
- Economies of scale: AWS buys hardware at massive volume discounts and passes savings to customers
- Elasticity value: Ability to handle 10× traffic spikes without pre-provisioning saves significant overprovisioning costs
- Opportunity cost: Engineers spend time on product, not datacenter maintenance
- Hidden costs to watch: Egress (data transfer out), idle Reserved Instances, unattached EBS volumes, NAT Gateway charges