You Cannot Guess Cloud Costs From Memory
Before understanding how cost estimation tools work, it helps to understand why cloud pricing is so difficult to reason about intuitively. Cloud providers publish pricing pages, but the actual cost of a workload depends on a combination of dimensions that interact in non-obvious ways.
Consider a single AWS EC2 instance. Its monthly cost depends on:
- Instance type (
t3.microcosts ~$8/month;m5.4xlargecosts ~$560/month) - Pricing model (On-Demand vs. Reserved vs. Spot — up to 90% difference)
- Region (
us-east-1is typically cheaper thanap-southeast-1) - Operating system (Linux vs. Windows adds a licensing surcharge)
- Tenancy (shared vs. dedicated host)
- EBS storage attached to it (type, size, and IOPS are billed separately)
- Data transfer out (first 100GB/month free, then tiered pricing)
That is seven pricing dimensions for one resource. A realistic production environment might have dozens of resource types, each with their own pricing matrix. The combinatorial complexity makes mental arithmetic impossible.
The Three Categories of Cloud Cost
Understanding cost estimation requires knowing what kinds of charges exist:
Compute costs are typically the largest line item. These are charges for running virtual machines, containers, or serverless functions. They are usually priced per hour (or per millisecond for Lambda) and scale directly with instance size and runtime.
Storage costs are charged per gigabyte per month. They seem small individually — S3 Standard storage is about $0.023/GB/month — but they accumulate silently. A team that stores 50TB of logs without a lifecycle policy is paying over $1,100/month just for storage, before any data retrieval costs.
Data transfer costs are the most commonly overlooked. Moving data into AWS is free. Moving data out of AWS (egress) is charged per gigabyte. Moving data between AWS regions or between Availability Zones also incurs charges. A microservices architecture with chatty cross-AZ communication can generate thousands of dollars per month in transfer fees that never appear in any resource's cost estimate.
A common interview question is: "What are the most commonly overlooked cloud costs?" The strong answer covers three areas: (1) data egress fees — often invisible until traffic scales; (2) NAT Gateway charges — both hourly and per-GB processed, which surprise teams running many private subnets; and (3) idle resources — load balancers, Elastic IPs, and RDS instances that run 24/7 even when unused. Demonstrating awareness of these shows you think about cost holistically, not just at provisioning time.