Your team wants to cut an S3 bill with a lifecycle rule that transitions every object in a bucket to S3 Standard-IA after 30 days. For which objects does that raise the bill instead of lowering it, and why?

## Why cheaper per GB is not cheaper S3's infrequent-access classes sell a lower per-GB storage rate in exchange for constraints on how you use the data. The interview question is always the same underneath: do you understand that the sticker price is one of four numbers? For S3 Standard-IA and S3 One Zone-IA the other three are the minimum billable object size, the minimum storage duration, and the retrieval fee. ## Minimum billable object size: 128 KB IA classes bill each object as if it were at least 128 KB. Store a 10 KB thumbnail in Standard-IA and you pay for 128 KB — roughly thirteen times the bytes you actually keep. At the published rates that makes a bucket of tiny objects strictly more expensive in IA than in Standard. AWS added a guard rail for exactly this: as of 2025, S3 Lifecycle does not transition objects smaller than 128 KB from S3 Standard into `STANDARD_IA` or `ONEZONE_IA` (nor into `INTELLIGENT_TIERING` or `GLACIER_IR`). So the naive "transition everything" rule often quietly skips the objects it would have hurt — but you cannot rely on that as a design: it makes the rule a no-op for that population, so your projected savings never appear, and the same objects are still sitting in Standard. The correct move is to look at the size distribution first and, if small objects dominate, aggregate them rather than tier them. ## Minimum storage duration: 30 days An object in Standard-IA is billed for at least 30 days of storage in that class. Delete it, overwrite it, or transition it onward after 10 days and you are still charged for the remaining 20. A pipeline that writes files, transitions them, and then reprocesses and replaces them a week later pays twice. There is a matching rule on the way in: an object must have been stored at least 30 days before lifecycle will transition it from S3 Standard to Standard-IA or One Zone-IA. That is why 30 days is the conventional first step in every staircase rule you see — it is the earliest permitted step, not a considered choice. ## Retrieval fees and request pricing IA classes charge per GB retrieved, on top of the request charges, and IA GET and PUT requests are priced higher than Standard's. This is the charge that surprises people, because it scales with how the application behaves rather than with what you store. Data read a few times a month across the whole dataset can easily burn more in retrieval than the per-GB discount saves. The break-even is a genuine calculation: roughly, the monthly retrieval volume as a fraction of stored volume, priced against the storage delta. ```text monthly cost (IA) = GB * ia_storage_rate + GB_retrieved * retrieval_rate + requests monthly cost (Std) = GB * std_storage_rate + requests ``` When the middle term dominates, IA loses. ## One Zone-IA adds a durability tradeoff, not just a price S3 One Zone-IA is cheaper still because the data lives in a single Availability Zone. The 128 KB minimum size and 30-day minimum duration still apply, and now an AZ loss can destroy the data outright. It is correct for reproducible derivatives — thumbnails you can regenerate, a secondary copy of something replicated elsewhere — and wrong for anything you cannot rebuild. ## How to decide instead of guessing S3 Storage Class Analysis is the AWS-native answer to "is this data actually cold?" You enable it on a bucket, prefix or tag filter, and after an observation period it reports how much of the storage is being retrieved by age band and recommends a transition age. It needs about a month of observation before its output means anything, and it deliberately only advises on the Standard-to-IA decision. S3 Inventory gives you the other half: a scheduled report listing every object with its size, storage class and last-modified date, which you query to get the size distribution the analysis does not show. ## The answer an interviewer wants Name the three charges, then apply them: small objects lose on the size minimum, churny objects lose on the duration minimum, hot objects lose on retrieval. IA is for data that is large, cold, and kept a long time — and you prove it is cold with Storage Class Analysis rather than assuming the number 30.