Cold and archive storage pays off when data is held for a long time and read almost never. The deep archive tiers, AWS S3 Glacier Deep Archive, Azure Archive and Google Cloud Archive, charge a tiny fraction of standard storage to hold a gigabyte, which makes them extraordinarily cheap for data you must keep but rarely touch. The catch is on the other side: they charge meaningful retrieval fees per gigabyte read, impose minimum storage durations of ninety days or more, and add retrieval latency that can range from minutes to hours. Archive pays off precisely when the storage saving over the data's lifetime exceeds the retrieval and minimum-duration costs you will actually incur, which for genuinely cold data is almost always, and for occasionally accessed data is often not.
This article is part of our complete guide to cloud storage and data cost optimization, the cluster pillar it links up to. It deepens the tier comparison in object storage tiers compared across AWS, Azure and GCP by focusing on the break-even for the coldest tiers specifically.
Archive trades a low storage price for a high access price and a minimum commitment. It pays off when you will hold the data long and read it almost never. The break-even is set by retrieval frequency, not by the storage price.
What you actually pay for archive
The archive bill has four components, and only the first one looks cheap. Storage per gigabyte is very low, which is the whole appeal. Retrieval fees are charged per gigabyte read back and are far higher than for warmer tiers; reading a large archived dataset can cost more than a month of standard storage for the same data. Minimum storage duration means an object deleted before the minimum, often ninety or one hundred eighty days, is billed for the full minimum regardless, so archive punishes short-lived data. And retrieval latency, from a few minutes for expedited options to many hours for the cheapest, is an operational cost rather than a dollar cost but matters if the data is ever needed urgently. The decision turns on weighing the low storage price against the other three.
| Data profile | Archive pays off? | Why |
|---|---|---|
| Compliance records kept 7 years, never read | Strongly yes | Long hold, near-zero retrieval |
| Old backups kept for DR, rarely restored | Usually yes | Retrieval is the exception, not the rule |
| Logs read a few times a year | Maybe | Depends on retrieval volume vs storage saving |
| Data read monthly | No | Retrieval fees overwhelm the storage saving |
| Data deleted within 90 days | No | Minimum-duration charge applies anyway |
The break-even rule
The math is simpler than it looks. Archive pays off when the monthly storage saving, the difference between what you pay in standard versus archive for the data, multiplied by the months you will hold it, exceeds the expected retrieval cost over that period plus any minimum-duration penalty. Because the storage saving is large and recurs every month, long-held data clears this bar easily even if you read it once or twice. The failure cases are short-held data, where the minimum-duration charge negates the saving, and frequently read data, where retrieval fees accumulate past the storage saving. A useful heuristic: if you expect to read the data more than a couple of times a year, model the retrieval cost explicitly before archiving, because that is where the surprise bills come from.
Want the archive break-even run on your data?
Our cloud cost audit models the storage, retrieval and minimum-duration costs for your datasets, archives what genuinely pays off, and proves the saving against a clean baseline on AWS, Azure, GCP and OCI. On the performance model, you pay only from realized savings. No savings, no fee.
Book a cloud cost audit →Where archive clearly wins
Several data types are nearly always good archive candidates because they are held long and read almost never. Compliance and regulatory records that you are required to retain for years but expect to read only in the rare event of an audit are the textbook case: the storage saving compounds for years and the retrieval, if it ever happens, is a single rare event. Old backups kept for disaster recovery fit too, since a restore is the exception rather than the routine, though the retrieval latency must be acceptable for your recovery objective. Completed project data, historical datasets past their active analysis window, and raw source files you keep only as a system of record all belong here. For these, archive is not a gamble, it is straightforwardly the cheapest correct choice, and the move is best automated through a lifecycle policy.
Where archive backfires
Archive backfires whenever the access or churn assumptions are wrong. Data that turns out to be read monthly racks up retrieval fees that quietly exceed what standard storage would have cost, the trap warned about in the tier comparison. Data deleted or overwritten before the minimum duration pays the full minimum anyway, so churny data placed in archive can cost more than leaving it warm. And data needed urgently in an incident, sitting in a tier with hours of retrieval latency, creates an operational problem that no storage saving justifies. The way to avoid all three is to base the archive decision on a measured access pattern rather than an assumption, which is why profiling comes first.
The Cloud Storage and Egress Cost Playbook includes the archive break-even calculator we use to decide, per dataset, whether the storage saving beats the retrieval and minimum-duration cost.
How to archive in practice
Once the break-even says yes, move data to archive through automation rather than by hand. A storage lifecycle policy transitions objects to archive on a schedule tied to age, so the move happens consistently and nothing is left warm by oversight. Confirm the data has cooled, using access analytics, before the transition fires, and respect the minimum-duration window so you are not transitioning data likely to be deleted sooner. Exact archive storage prices, retrieval fees, retrieval options, latency tiers and minimum durations differ across AWS, Azure and GCP and change over time, so verify the current figures in each provider's pricing documentation before committing data, as of May 2026. Placing the right data in archive and keeping it there is the Cut and Lock work of our See, Cut, Lock, Run method.
The short version
Cold and archive storage pays off when data is held long and read almost never, because the very low storage price compounds while the high retrieval fee is rarely triggered. It backfires on data read monthly, churned within the minimum duration, or needed urgently. Decide with the break-even rule, base it on a measured access pattern rather than an assumption, and automate the move with a lifecycle policy. Verify current rates per provider, since they change. When you want the break-even run on your data and the archiving done across the estate, that is part of what our rightsizing and waste elimination service delivers.