Billing and cost basics

By the end of this lesson you will know which line items to watch on every AWS bill, how to match a workload to the right purchase option, and how to set up the alerts that catch a runaway egress charge before it lands in your inbox as an invoice.

Pay-per-use and the three cost dimensions

There is no flat AWS subscription. You pay for what you provision and consume, metered per service, and the total is the sum of three recurring dimensions that show up on almost every bill.

Compute time. An EC2 instance, a Lambda invocation, a Fargate task — you pay per second or per hour that capacity exists, whether or not it is doing useful work. This is the trap of idle provisioned capacity: a stopped EC2 instance costs nothing for compute, but a running one you forgot about bills 24/7. Provisioned-but-idle is the most common avoidable line item.

Storage (GB-month). A gigabyte stored for a month is the unit. S3 Standard runs roughly $0.023 per GB-month in common regions (and is designed for 11 nines — 99.999999999% — of durability, so the cost buys near-zero data-loss risk); EBS volumes, snapshots, and RDS storage are billed the same way. Storage also meters requests, not just bytes: S3 charges around $0.0004 per 1,000 GET and $0.005 per 1,000 PUT requests, so a workload making millions of tiny object reads can run up a request bill that rivals the byte storage itself. Storage is usually the predictable part of the bill — it grows slowly and you can see it coming. The catch is that storage you stop using keeps billing until you delete it: orphaned EBS volumes from terminated instances and forgotten snapshots accrue silently.

Data transfer (egress). This is the dimension that ambushes teams. The rule of thumb: ingress (data into AWS) is generally free; egress (data out to the internet) is not — commonly around $0.09 per GB for the first tier, varying by region and volume. Worse, transfer inside AWS is not free either: traffic crossing Availability Zones typically costs ~$0.01/GB each way (so ~$0.02/GB for a round trip between two services in different AZs), and cross-Region transfer runs ~$0.02/GB. A chatty microservice mesh sprawled across AZs, or a database replicating across Regions, can run up data-transfer charges that dwarf the compute generating them. Put numbers on it: a service pushing 50 TB/month of cross-AZ chatter at ~$0.02/GB round-trip is ~$1,000/month you can erase just by pinning the two services into the same AZ — pure architecture, zero feature change.

EC2 purchase options: the price/commitment/interruption tradeoff

The same EC2 instance can cost wildly different amounts depending on how you buy it. There are three buckets, trading price against commitment and reliability.

On-Demand is the default: no commitment, pay the full rate per second/hour, start and stop whenever. It is the most expensive per hour and the most flexible — right for spiky, unpredictable, or short-lived workloads where you can’t forecast usage.

Reserved Instances / Savings Plans are a commitment for a discount. You promise 1 or 3 years of usage (Savings Plans commit to a dollars-per-hour spend; Reserved Instances to a specific instance configuration) and AWS discounts the rate — often up to ~72% off On-Demand. Right for steady, baseline, always-on workloads whose floor you can predict a year out. The risk is the commitment itself: you pay even if usage drops.

Spot sells AWS’s spare capacity at up to ~90% off On-Demand — but AWS can reclaim it with a two-minute warning whenever it needs the capacity back. Right only for fault-tolerant, interruptible work: batch jobs, CI runners, stateless workers, big-data processing that checkpoints. Wrong for anything that can’t survive sudden termination.

Real fleets blend all three: Reserved/Savings Plans cover the predictable baseline, On-Demand absorbs the variable middle, and Spot soaks up bursty interruptible load at a steep discount.

The Free Tier and its traps

New accounts get three flavors of free, and conflating them is how the trap springs. 12-month free offerings (e.g., 750 hours/month of a t2.micro/t3.micro, 5 GB of S3) expire one year after signup — then bill at full rate, often unnoticed. Always-free offerings (e.g., 1M Lambda requests/month, a slice of DynamoDB) persist but have hard caps. Trials are short-term (30–90 days) per-service freebies.

The traps are structural. Free-tier limits are per category, not a blanket “free account” — exceed 750 instance-hours by running two instances, or spill past the S3 cap, and you’re billed on the overage immediately. The 12-month clock is the big one: teams stand up a “free” demo, forget it, and twelve months later it silently converts to a paying instance running 24/7. And the Free Tier covers the service but rarely the data transfer — your free instance can still generate billable egress. Treat the Free Tier as a metered allowance with an expiry, not as “AWS is free for a year.”

Cost controls: make the bill observable before it surprises you

When you see a spike in the bill mid-month, you want a tool that already told you it was coming — not a dashboard you had to remember to open. These four instruments are your early-warning system.

You cannot manage what you can’t see, and AWS bills after the fact — so the entire game is making spend observable early. Four tools form the baseline.

AWS Budgets lets you set a spend (or usage) threshold and get alerted — by email or SNS — when actual or forecasted cost crosses it. This is your tripwire; a forecasted-overage alert catches a runaway egress charge mid-month instead of in the next invoice. Cost Explorer is the analysis surface: it breaks spend down by service, region, and time so you can answer “what grew, and when.” Cost-allocation tags let you attribute spend to teams, projects, or environments — tag everything (team, env, service) and the bill becomes a per-owner ledger instead of one opaque number. The classic billing alarm (a CloudWatch alarm on EstimatedCharges) is the simplest first line of defense for a personal or small account.