“Just put it in Glacier” is common advice for data you don’t need often, but S3 actually offers multiple cold-storage tiers under that umbrella, and picking the wrong one is one of the easiest ways to accidentally overpay — or get burned by a restore that takes half a day longer than expected.
The storage cost is not the whole story
On storage price alone, Deep Archive looks like the obvious winner — it’s roughly a quarter of the cost of Glacier Instant Retrieval per GB/month. But storage is only one side of the ledger. The other side is retrieval cost and retrieval time, and that’s where the two tiers diverge sharply:
| Glacier Instant Retrieval | Glacier Deep Archive | |
|---|---|---|
| Storage cost | Higher (~$0.004/GB/month) | Lowest (~$0.00099/GB/month) |
| Retrieval time | Milliseconds | 12 hours (standard) |
| Retrieval cost | Low | Higher per-GB |
| Best for | Data accessed a few times a year | Data accessed almost never (compliance archives, disaster recovery) |
Why the “cheapest tier” isn’t always cheapest
If you have data you restore even occasionally, the math flips fast. Deep Archive’s per-GB retrieval fee, multiplied across a real restore event, can erase months of storage savings in a single incident. And the 12-hour standard retrieval window (or 48 hours for bulk retrieval) means Deep Archive is structurally unsuitable for anything you might need same-day — a compliance audit, an incident investigation, a customer data request with a response SLA.
Glacier Instant Retrieval exists specifically to close this gap: same rock-bottom storage tier positioning, but retrieval in milliseconds instead of hours, at a storage cost that’s still a fraction of S3 Standard. The trade-off is a higher storage rate than Deep Archive — you’re paying a premium for the option to retrieve instantly, whether or not you use it.
A simple way to decide
The real question isn’t “which tier is cheapest” — it’s “how often will I actually need this data back, and how fast”:
- Never, except for a rare compliance/legal event, and 12-48 hours is fine → Deep Archive.
- A handful of times a year, and you can’t predict when → Glacier Instant Retrieval.
- Regularly, just infrequently → you probably don’t want Glacier at all; S3 Standard-IA or Intelligent-Tiering is likely cheaper once retrieval frequency is factored in.
The break-even point depends entirely on your specific storage volume and expected retrieval frequency — there’s no universal answer, because a single unplanned bulk restore can cost more than a year of the storage savings that got you into Deep Archive in the first place.
The migration itself has a hidden cost
Once you’ve decided Deep Archive is the right tier, there’s a second trap waiting in the migration itself: AWS charges a per-object fee for every file transitioned into Glacier via a lifecycle rule. That fee is trivial for a handful of large files and brutal for millions of small ones — archiving ten million 500 KB log files costs vastly more in transition fees than archiving the same total volume as a thousand 5 GB backups, even though the storage savings afterward are identical. Teams that only model the monthly storage-rate delta routinely miss this and end up with a migration that takes over a year to pay for itself, or in extreme small-file cases, never does within their retention window.
Our S3 vs Glacier Cost Calculator models exactly this: given your data volume, average file size, and expected retention period, it computes the one-time transition fee and the month your cumulative Deep Archive cost actually drops below what you’d have paid staying on S3 Standard — so you can catch a bad small-files migration before you run it, not after the bill arrives.
If the data you’re archiving is coming from a large, ongoing upload pipeline, the S3 Multipart Upload Sizer is worth a look too — it calculates the chunk size and thread concurrency that gets large objects into S3 fastest before they age into a colder tier.