S3 storage classes compared: when each one actually saves money

S3 pricing looks simple until the first month where the archive tier costs more than the standard one. That usually happens because storage price is only one of four variables, and the other three do the damage.

Every class trades along the same four axes: price per gigabyte month, retrieval fee per gigabyte, retrieval latency, and a minimum billable duration per object. Pick a class by looking at all four against your actual access pattern.

The classes, in the order you will meet them

S3 Standard. Millisecond access, no retrieval fee, no minimum duration. The default, and the right answer for anything actively served.

S3 Intelligent-Tiering. Moves objects between access tiers automatically based on observed use, with no retrieval fee for the frequent and infrequent tiers. You pay a small monitoring charge per object per month. Ideal when the access pattern is genuinely unknown.

S3 Standard-IA. Lower storage price, a per gigabyte retrieval fee, a 30 day minimum duration and a 128 KB minimum billable object size. Good for backups and older logs you read occasionally.

S3 One Zone-IA. The same economics as Standard-IA at a lower price, stored in one availability zone. Appropriate only for reproducible data.

S3 Glacier Instant Retrieval. Archive pricing with millisecond access. A 90 day minimum and a higher retrieval fee. Suits archives you rarely touch but need immediately when you do, such as medical images or compliance records.

S3 Glacier Flexible Retrieval. Retrieval takes minutes to hours depending on the option you choose. A 90 day minimum. This is the classic archive tier.

S3 Glacier Deep Archive. The cheapest storage in the family, retrieval measured in hours, and a 180 day minimum. For data you keep because you must, not because you use it.

S3 Express One Zone. The opposite end. Single digit millisecond latency at a much higher storage price, for workloads where request latency dominates the bill.

The three traps

Minimum durations. An object written to Standard-IA and deleted after five days is billed for thirty. Pipelines that write and delete quickly should stay on Standard, no matter how attractive the archive price looks.

Minimum object size. The infrequent access and archive tiers bill a minimum of 128 KB per object. A bucket holding fifty million 4 KB files is billed as if each were 128 KB. At that scale, packing small files into archives before storing them is usually the bigger win.

Transition costs. Lifecycle rules charge per object transitioned. Moving ten million objects to a cheaper class has a real one off cost, and if the objects are small it may never pay back.

Lifecycle rules that behave

A pattern that works for most log and backup buckets:

1. Write to Standard.
2. Transition to Standard-IA after 30 days, but only for objects above roughly 128 KB.
3. Transition to Glacier Flexible Retrieval after 90 to 180 days.
4. Expire entirely at whatever your retention policy requires.
5. Add a rule to abort incomplete multipart uploads after seven days. Those fragments are invisible in the console and billed like everything else.

That last one is worth doing today. Abandoned multipart uploads are one of the most common sources of storage nobody can account for.

Where storage cost meets transfer cost

Storage class only governs what it costs to keep and read data inside S3. Moving that data out to the internet is billed separately as data transfer, and for content heavy workloads that line item usually dwarfs the storage one. We cover it in data transfer costs across cloud providers.

If your bill is compute rather than storage led, the EC2 instance families guide is the better starting point, and the free tier article covers what S3 usage sits inside the free allowance.