How to move S3 objects to Glacier without losing money

S3 Standard costs about $0.023 per GB-month in us-east-1; Glacier Deep Archive about $0.00099. The gap invites a lifecycle rule for everything old. The rule is right for large objects and wrong for small ones, and the bill is where you find out.

The costs that are not the storage price

Transition request fee: every object moved is one billed request, $0.01 per 1,000 into Standard-IA and Intelligent-Tiering, up to $0.05 per 1,000 into Deep Archive. A hundred million small objects is thousands of dollars before any saving starts.
Minimum billable size: Standard-IA and Glacier Instant Retrieval bill every object as at least 128 KB. A 10 KB object is billed as 128 KB, so its storage cost goes up by 12 times.
Glacier overhead: Glacier Flexible Retrieval and Deep Archive add about 40 KB per object (32 KB billed at the archive rate, 8 KB at the Standard rate).
Minimum storage duration: 30 days for Standard-IA, 90 for Glacier Instant and Flexible, 180 for Deep Archive. Delete or overwrite sooner and AWS bills the remaining days.
Intelligent-Tiering monitoring: $0.0025 per 1,000 objects per month for objects of 128 KB or more; smaller objects are not monitored or tiered and stay at the Standard rate.
Retrieval: per-GB fees on the infrequent and archive classes. Intelligent-Tiering has none, which is what you pay its monitoring fee for.

The size filter

A lifecycle rule can skip small objects. ObjectSizeGreaterThan is a strict comparison, so 131071 keeps every object of 128 KiB or more:

{"Rules": [{"ID": "archive-large-old-objects", "Status": "Enabled",
  "Filter": {"ObjectSizeGreaterThan": 131071},
  "Transitions": [{"Days": 180, "StorageClass": "DEEP_ARCHIVE"}]}]}

The analyzer prints exactly this for the action with the best net 12-month saving. It compares each class with and without the filter.

Get the numbers from your own bucket

Enable S3 Inventory with CSV output, then run the report locally. The file never leaves your machine:

node cli.mjs analyze inventory.csv --retrieval-pct 2 --fail-on warn --waste-threshold 250

The access pattern is the number an inventory cannot tell you. Set the share of bytes you read each month; if the best action changes when you raise it, you are close to a break-even and should measure access with S3 Storage Lens or server access logs first.

What this cannot see

List prices ignore your discounts, volume tiers, region and the free tier. It does not model Glacier retrieval request fees, data-transfer charges, versioned non-current objects (they are left out), or objects that will age past a threshold later. Treat the output as a reason to test a rule on one prefix, not as a forecast of the invoice.

Common questions

Does moving objects to a colder S3 class always save money?

No. Cold classes charge per object (a transition request, a 128 KB minimum billable size or about 40 KB of overhead, and a 30 to 180 day minimum duration), so objects under about 128 KB can cost more after the move. Filter them out with ObjectSizeGreaterThan 131071.

Which S3 class should old data go to?

It depends on how often you read it. Glacier Deep Archive is cheapest to store but slow and billed for 180 days; Standard-IA and Glacier Instant Retrieval keep millisecond access with a per-GB retrieval fee; Intelligent-Tiering has no retrieval fee but charges a monthly monitoring fee per object. Estimate the monthly percentage of bytes you read and compare the net figures.

Where do I get an S3 Inventory report?

Enable S3 Inventory on the bucket (Management, Inventory configurations) with CSV output and the Size, LastModifiedDate and Storage class fields. AWS delivers the report to a destination bucket daily or weekly; download the file, decompress it, and use it here.