Archive storage is the lowest-cost tier of cloud storage, designed for data that must be kept for years but read rarely, typically less than once a year. It trades retrieval speed and retrieval cost for a monthly storage price that is a small fraction of standard object storage, and it is often called by the name of its best-known service, Glacier.
In cloud storage, archive tiers are usually storage classes of an object storage service rather than separate products: S3 Glacier Flexible Retrieval and S3 Glacier Deep Archive on AWS, the Archive access tier of Azure Blob Storage, and the Archive storage class of Google Cloud Storage. Data keeps the same high durability as other classes; what changes is how quickly and how cheaply it can be read back.
How archive storage achieves its low price
Providers keep archive data on media and systems optimized for density and low energy use, and they schedule reads in batches instead of serving them instantly. The exact hardware is not disclosed, but the economic model is clear: the provider can promise eleven nines of durability while keeping data offline or on slow media, because customers accept a delay before first byte.
In several services, archived objects are not directly readable. A client first issues a restore or rehydration request, the provider copies the object to a readable tier, and only then can it be downloaded. On S3 the restored copy is temporary and expires after a chosen number of days; on Azure the blob is moved or copied to the Hot or Cool tier.
Pricing has four components, and the balance between them is what distinguishes archive from cold storage: storage per gigabyte-month, retrieval per gigabyte, request charges and a minimum storage duration that bills early deletion as if the object had stayed for the full period.
Archive tiers across major providers
The major platforms implement archive storage with different access models and commitments. The figures below are the published service characteristics and change only rarely; prices vary by region and are therefore omitted.
| Service class | Access model | Typical time to first byte | Minimum duration |
|---|---|---|---|
| S3 Glacier Instant Retrieval | Direct read | Milliseconds | 90 days |
| S3 Glacier Flexible Retrieval | Restore first | Minutes to 12 hours | 90 days |
| S3 Glacier Deep Archive | Restore first | Within 12 to 48 hours | 180 days |
| Azure Blob Archive tier | Rehydrate first | Under 1 hour to 15 hours | 180 days |
| Google Cloud Storage Archive | Direct read | Milliseconds | 365 days |
Google Cloud Storage and S3 Glacier Instant Retrieval show that archive does not always mean slow. In those classes the penalty is purely economic: reads are immediate but carry high per-gigabyte retrieval fees, and early deletion is billed against the minimum duration.
Restore-based classes offer retrieval speed options. S3 Glacier Flexible Retrieval provides expedited retrieval in minutes, standard retrieval in hours and bulk retrieval within about 5 to 12 hours, each at a different price. Choosing bulk retrieval for large planned restores reduces cost substantially.
Archive storage compared with neighboring storage classes
Archive storage is the last step of a tiering ladder that starts with frequently accessed data. Understanding where each step ends is essential for a correct storage class choice.
| Class | Intended access frequency | Storage price | Retrieval cost |
|---|---|---|---|
| Standard or hot | Daily or continuous | Highest | None or minimal |
| Infrequent access or cool | About once a month | Lower | Moderate per GB |
| Cold storage | About once a quarter | Low | Higher per GB |
| Archive storage | Less than once a year | Lowest | Highest, often with delay |
Archive storage also differs from backup. Backup exists to restore a recent working state quickly; archive exists to retain records long term, often for legal or historical reasons. Many organizations still send older backup generations to archive tiers once fast restore is no longer expected.
Moving data into archive with lifecycle policies
Data rarely gets written directly to archive. The usual path is a storage lifecycle policy that transitions objects automatically after a defined age, for example from standard to infrequent access after 30 days and to archive after 180 days.
- Filter by prefix or tag so only eligible data, such as closed financial records, moves to archive.
- Avoid archiving tiny objects: per-object overhead and request charges can exceed storage savings, so small files are often bundled into larger archives first.
- Combine with object versioning carefully, because noncurrent versions also consume archive storage and need their own expiration rules.
- Use object lock or immutability policies when regulations require write-once, read-many retention.
- Record restore procedures and test them, since restores are rare and easily forgotten.
The decisive planning question is how fast data must come back when it is needed. A legal hold that tolerates a two-day wait suits deep archive; an audit that must answer within an hour does not.
Pitfalls that erase the savings
Archive storage saves money only when access patterns match its assumptions. When they do not, retrieval and early deletion fees can make it more expensive than a warmer tier.
| Mistake | Consequence | Better approach |
|---|---|---|
| Archiving data that is read monthly | Retrieval fees exceed storage savings | Keep it in cool or cold storage |
| Deleting data soon after archiving | Early deletion charges | Respect the minimum storage duration |
| Archiving millions of small files | Request and metadata overhead dominate | Aggregate into larger objects first |
| Planning a fast mass restore | Expedited fees and throttling | Use bulk retrieval and stage it over time |
| Assuming archive means immutable | Data can still be deleted | Enable object lock or legal holds |
The safest rule is to model the total cost of a realistic restore before archiving a dataset, not only the monthly storage bill.
A non-technical view
Archive storage works like a long-term storage warehouse outside the city. Keeping boxes there costs very little each month, far less than keeping them in your office.
The catch is that you cannot grab a box instantly. You have to order it, wait for it to be delivered, and pay a delivery fee. That is perfect for old tax records you may never open again, but a poor choice for files you need every week. The key point is simple: cheap to keep, slow and costly to fetch.
A concrete example
A regional hospital must retain medical imaging studies for at least ten years, and its picture archiving system produces several hundred terabytes of scans. Studies older than six months are almost never opened, but they must be available if a patient returns or a legal question arises.
The IT team configures a lifecycle policy that keeps new studies in standard storage, moves them to infrequent access after 90 days and to a deep archive class after 12 months, with object lock enforcing the retention period. When an old study is requested, the system issues a standard restore and notifies the clinician when it is ready, usually the next day. The hospital cuts its storage bill for older studies by more than an order of magnitude while still meeting its retention obligations.