Billing, Pricing, and Support: Cost Models and Plan Selection
Pricing fundamentals
AWS pricing rests on three ideas: pay-as-you-go, save when you commit, and pay less by using more. Compute is billed by time and capacity, storage by amount stored, and data transfer out of AWS is charged while inbound is generally free. Trap: forgetting that data transfer between Availability Zones or out to the internet can cost money, while data transfer in is typically free.
Purchasing options
On-Demand is flexible with no commitment. Savings Plans and Reserved Instances trade a one- or three-year commitment for a discount; Savings Plans are more flexible across instance families and compute services. Spot Instances offer deep discounts for interruptible workloads. Dedicated Hosts suit licensing and compliance needs. Trap: using Spot for a database that cannot tolerate interruption, or Reserved Instances for a short-lived experiment.
Cost visibility and governance
Know the tools: AWS Cost Explorer visualizes and forecasts spend, Budgets sends alerts or triggers actions, the Cost and Usage Report gives the most granular billing data, and Cost Allocation Tags plus consolidated billing in AWS Organizations help attribute and combine spend. Trusted Advisor surfaces cost optimization, security, and fault tolerance checks.
Support plans
Basic is free with account and billing support. Developer adds technical email support during business hours. Business adds 24/7 phone, chat, and email with faster response and access to a technical account manager pool. Enterprise adds a designated technical account manager, concierge support, and the fastest response times. Trap: assuming a technical account manager is included below Enterprise (Business gets access to TAM pool guidance, not a designated TAM).
Worked example
A steady-state production workload running continuously for three years should use a Savings Plan or Reserved Instances; a batch job that can restart should use Spot; a two-week proof of concept should use On-Demand. Match commitment length to workload lifetime.