S3 Glacier Deep Archive: when it saves money and when it costs more

What Deep Archive is for

Deep Archive is an S3 storage class, DEEP_ARCHIVE, not a separate service. Objects stay in your bucket and keep their keys, but their contents cannot be read until you restore a temporary copy. AWS designs it for data accessed less than once a year, with the same eleven nines of durability as S3 Standard across at least three Availability Zones.

Typical fits are records kept for legal or tax retention, finished project files, raw footage after delivery, and backups of backups. The common thread is that nobody is waiting on them.

How to restore files from S3 Glacier Flexible Retrieval and Deep Archive

• Standard restore: within 12 hours.
• Bulk restore: within 48 hours, at a lower retrieval rate.
• There is no expedited tier; if you might need a file in minutes, Deep Archive is the wrong class.

The three costs that decide whether it pays off

The first is the 180-day minimum. Delete, overwrite or move an object out of Deep Archive before 180 days and you pay a prorated charge for the rest of the period. Glacier Flexible Retrieval and Glacier Instant Retrieval have 90-day minimums.

The second is per-object overhead. Every archived object carries 40 KB of extra billed storage: 8 KB for its name and metadata at the S3 Standard rate, and 32 KB of index data at the Deep Archive rate. On a 50 MB video that is noise. On a 20 KB invoice the overhead is larger than the file, and the 8 KB alone, billed at the Standard rate, can cost more than the archived bytes.

The third is transition requests. Each object a lifecycle rule moves into Deep Archive is a billed request. Moving a million small files costs the same per-request fee as moving a million large ones, and with small files the savings may never catch up.

Amazon S3 pricing for business file storageA cost-conscious stack for an S3 file workspace

• Large files, kept for years, rarely opened: Deep Archive almost always wins.
• Small files, or anything you might delete within six months: it often loses.
• If your files are small, zip them into larger archives before archiving.

Deep Archive compared with the other Glacier classes

The three Glacier classes trade storage price against how fast and how cheaply you can get data back. Instant Retrieval reads in milliseconds like Standard-IA but charges a higher retrieval fee. Flexible Retrieval restores in minutes to hours. Deep Archive is cheapest to keep and slowest to read.

• Glacier Instant Retrieval: 90-day minimum, 128 KB minimum billable size, millisecond access. For files opened about once a quarter.
• Glacier Flexible Retrieval: 90-day minimum, 40 KB overhead, restores in 1 to 5 minutes (expedited), 3 to 5 hours (standard) or 5 to 12 hours (bulk, free).
• Glacier Deep Archive: 180-day minimum, 40 KB overhead, restores within 12 hours (standard) or 48 hours (bulk).

A lifecycle rule that archives only what is worth archiving

Since September 2024, S3 lifecycle rules skip objects smaller than 128 KB by default, precisely because transition fees on small objects outweigh the savings. Keep that default unless you have a reason not to.

The rule below moves objects under a prefix to Deep Archive 365 days after creation, and adds an explicit size filter so the intent is clear to whoever reads it next. Transitions into Deep Archive are one way: to bring an object back to another class you restore it and copy it over itself.

{
  "Rules": [
    {
      "ID": "archive-closed-projects",
      "Status": "Enabled",
      "Filter": {
        "And": {
          "Prefix": "projects/closed/",
          "ObjectSizeGreaterThan": 1048576
        }
      },
      "Transitions": [{ "Days": 365, "StorageClass": "DEEP_ARCHIVE" }]
    }
  ]
}

S3 Object Lock: governance vs compliance modeRecover deleted or overwritten files with S3 versioning

• Billing at the Deep Archive rate, the 180-day clock and the overhead all start the day the rule is satisfied, even if the physical move happens later.
• In a versioned bucket, add a NoncurrentVersionTransitions rule too, or old versions stay in Standard.

Intelligent-Tiering as the alternative

If you cannot predict which files will go cold, S3 Intelligent-Tiering can make the call per object. Its optional Deep Archive Access tier moves objects untouched for 180 days to archive pricing, with no retrieval fee, in exchange for a small monthly monitoring fee per object. Objects under 128 KB are never monitored or moved.

The catch is the same as Deep Archive itself: once an object reaches that tier, it must be restored before anyone can open it.

What an archived file looks like to your team

An archived object still shows up in a bucket listing, which is where the confusion starts. Someone clicks a contract from 2021, the download fails, and a ticket lands with whoever runs the AWS account.

BucketDesk shows the storage class next to each file and, on the Business plan, lets an operator request a restore for a single object, pick the retrieval tier and confirm the estimated cost before AWS starts the job. AWS bills the retrieval to your account as usual.

The same applies to AI. Document AI, BucketDesk's single-document chat on Amazon Bedrock, can only read a file whose bytes are available, so a contract in Deep Archive needs a restore before Claude can answer questions about it. Plan the wait into any workflow that relies on old documents.

Ask questions about S3 documents with Claude on Amazon BedrockPreview files in Amazon S3 without downloading