Mastering AWS S3 Storage Costs

Amazon S3 pricing rests on four pillars: storage volume, request volume, data transfer, and retrieval. Storage is charged per gigabyte per month and varies enormously by class, from Standard at roughly $0.023 per GB down to Glacier Deep Archive at less than $0.001 per GB. Requests are billed per thousand operations, with PUTs around $0.005 and GETs around $0.0004 per thousand. Data transfer out to the internet is free up to a monthly allowance, then runs roughly $0.09 per GB. Retrieval charges apply only to the colder classes and are the line item teams most often forget when modelling a migration to Glacier.

Choosing the Right Storage Class

S3 Standard suits frequently accessed data, offering immediate availability with no retrieval fee and no minimum duration. When data is touched less than once a month, Standard-IA or One Zone-IA roughly halves the storage rate while adding a retrieval fee and a 30-day minimum. Glacier Instant Retrieval serves archives accessed a few times a year with millisecond access at a much lower storage price, subject to a 90-day minimum. Deep Archive is the floor of the market for data touched once or twice annually, with a 180-day minimum and restore times measured in hours. Intelligent-Tiering shifts objects between tiers automatically based on real access patterns, trading a small per-object monitoring fee for the elimination of lifecycle guesswork.

The Charges That Surprise People

Three costs account for most unexplained S3 bills. First, minimum storage durations mean that writing an object to Deep Archive and deleting it a week later still bills 180 days. Second, lifecycle transitions are charged per thousand objects rather than per gigabyte, so moving ten million small files to Glacier incurs roughly $500 in request fees before any savings begin. Third, versioning silently doubles or triples the storage line, because deleting a versioned object only writes a delete marker while every prior version keeps billing. Incomplete multipart uploads behave the same way and persist until a lifecycle rule aborts them.

Optimizing Your Monthly Bill

Start with the configuration-only wins: add a lifecycle rule to abort incomplete multipart uploads, expire non-current versions after a defined retention window, and transition genuinely cold data to IA or Glacier once you have confirmed the objects are large enough for transition fees to pay off. For read-heavy public workloads, placing CloudFront in front of the bucket usually lowers egress cost while improving latency, because CDN transfer is priced below direct S3 transfer. Model each scenario here before changing anything in production, paying particular attention to the average object size shown in Professional Mode, since that single number determines whether a cold-tier migration saves money or quietly costs more than it saves.