Cold cloud storage: retrieval time, retention, and total cost

Cold cloud storage is intended for data you expect to read infrequently. It can reduce the cost of keeping old project files, historical datasets, or long-term backups, but the storage rate is only part of the decision. Retrieval fees, minimum billable durations, and the time needed to restore usable data can change whether a cheaper tier fits the work.

Start with two questions: how often will someone need these files, and how long can they wait? A media archive that can wait until tomorrow has different requirements from an old customer document that must open during a support call. This guide compares those trade-offs and shows how to test them before moving a large archive.

Hot vs. cold storage: choose by access pattern

Hot data supports active work: current application content, frequently edited files, or datasets read repeatedly. Cold data is retained for later use but accessed less often. Age is a useful clue, not a complete rule. A five-year-old product image served every day is still active data.

Look at actual reads, updates, and deletions over a representative period. Include background activity such as indexing, antivirus scans, previews, analytics, and backup verification. An archive may look untouched to its owner while software reads much of it every week.

Good candidates for cold data storage include finished production assets, completed research data, and older backup copies with a suitable recovery deadline. Keep a faster copy when an urgent restore is required. If the access pattern is unknown, test a small group before applying a rule to everything.

Cold and archive storage do not always mean delayed access

Providers use different names for different retrieval models. Some low-cost classes remain online; others require a restore or rehydration operation before you can read the contents. Check the exact class, region, account configuration, and retrieval option. The examples below were checked against provider documentation on October 28, 2026.

Amazon S3 Glacier storage classes

The S3 Glacier comparison distinguishes three options:

• Instant Retrieval: millisecond access; a 90-day minimum storage duration.
• Flexible Retrieval: a restore step, with options ranging from minutes to hours; a 90-day minimum.
• Deep Archive: standard restores typically within 12 hours and bulk restores within 48 hours; a 180-day minimum.

These retrieval descriptions are not a promise that an entire application or large archive will be usable by that time.

Google Cloud Coldline and Archive

Google Cloud storage classes provide low-latency access without an offline retrieval step, including Coldline and Archive. Their minimum storage durations are 90 and 365 days respectively, and both have retrieval fees. The word “Archive” therefore does not imply the same restore process as S3 Deep Archive.

Azure Cold and Archive

Azure's Cold tier is online, with immediate access and a 90-day minimum. Archive is offline and has a 180-day minimum. Archived blobs must be rehydrated before their contents can be read. Our Azure Blob Storage guide explains the wider account, blob-type, and access choices.

Microsoft's rehydration documentation describes standard retrieval of objects under 10 GB as potentially taking up to 15 hours. High priority can be faster but costs more; size, demand, and account-level throughput limits affect completion. Do not apply a single-object estimate to a bulk recovery.

Measure time until the data is usable

Retrieval latency is one stage of recovery. Allow time to locate the right objects, obtain approval, start the request, wait for data to become available, transfer it, decrypt or unpack it, and check the result. Restoring an application may also require rebuilding its database or reconnecting files to records.

For an online cold tier, a large transfer can still take hours. Available bandwidth, concurrent requests, object sizes, destination capacity, and software behavior all matter. “Immediate access” means you can start reading without an archival restore step; it does not mean an unlimited amount of data arrives immediately.

Set a deadline for a concrete scenario, such as recovering one project or a full backup. Measure from the request for recovery to a successful use of the restored files. If the workflow cannot meet that deadline, keep the necessary data in a faster tier or retain an additional accessible copy.

Separate minimum billing duration from data retention

A minimum storage duration usually sets a billing obligation. Deleting or moving data sooner can trigger a charge for the remaining period. It does not, by itself, prevent deletion. For example, Google's pricing documentation explains early-deletion charges and exceptions for particular lifecycle and Autoclass operations. Check the rule for the operation you intend to use.

Enforced retention is a separate control. S3 Object Lock, for example, protects specified object versions through retention settings or legal holds. Its governance and compliance modes have different permissions and restrictions. Choosing a cold storage class does not enable those controls automatically.

Record three separate decisions: how long the organization needs the data, how long the service charges for it, and whether deletion must be restricted. Have the responsible owner approve any retention or hold settings. A low storage price does not establish compliance with a legal or contractual requirement.

Calculate the full cost of a cloud archive

Build the estimate for your chosen region and expected behavior. The Amazon S3 pricing page separates storage, requests, retrieval, and data transfer, among other charges. Different services bundle or waive some components, so use the current terms for the exact class.

• Stored capacity: include retained versions, metadata, and any minimum billable object sizes.
• Ingest and transitions: count uploads, copy operations, and lifecycle requests.
• Reads and restores: include request charges and the amount retrieved.
• Network transfer: price the route to the actual destination, including another region or an on-premises system.
• Restored copies: budget for temporary online storage and processing where needed.
• Early removal: check charges for deleting, overwriting, or moving data before the minimum period ends.

Model a quiet month, a routine retrieval, and a large recovery. Include the people and tools needed to operate the archive. Predictable costs come from measured behavior and clear limits, rather than the headline price alone.

A worked example with hypothetical rates

Suppose you keep 10,000 GB unchanged for 12 months. For illustration only, assume cold storage costs $0.004 per GB-month and an online comparison tier costs $0.020. Assume retrieval costs $0.05 per GB for the cold option and nothing for the comparison tier. Network transfer costs $0.08 per GB for both. These are invented rates, not provider quotes.

With no reads, storage costs $40 per month in the cold option and $200 in the comparison tier. In a month when you retrieve and transfer 4,000 GB, the calculation changes:

• Cold option: $40 storage + $200 retrieval + $320 transfer = $560.
• Online comparison: $200 storage + $0 retrieval + $320 transfer = $520.

This simplified example excludes requests, temporary copies, taxes, discounts, and early-deletion charges. It shows why a lower storage rate can lose its advantage when access increases. Substitute actual rates, billing units, and observed reads before making the decision.

Use lifecycle rules carefully

A lifecycle policy can move eligible objects to another class as they age. Before enabling one broadly, identify what the rule will match, which versions it affects, when charges begin, and how restored data will be treated. Keep an accessible inventory with object names, owners, dates, formats, and retention decisions.

Test the policy on a small, clearly labeled group. Check the resulting tier and bill, then retrieve representative objects. Make sure automated readers do not turn a storage saving into recurring retrieval charges. Keep an exception process for data whose access pattern changes.

Decide how to group small files. Bundling can reduce the number of operations, but may force a larger download to recover one item. Test the archive format, indexing, and tools you will need years later. A folder name alone is not a reliable recovery catalog.

Keep archiving and backup responsibilities clear

An archive preserves material for later use. A backup provides a recoverable copy from an earlier state. A cold tier can hold either, but putting the only copy there does not create another copy or a recovery history. Our cloud backup guide covers the broader recovery decisions.

Restrict access to sensitive files and separate routine archive access from administrative deletion rights where the service supports it. Preserve access to encryption keys and the software needed to open the data. Keep references to those dependencies in the runbook, without putting secret values in the inventory.

Low access frequency and delayed retrieval are not proof of isolation from ransomware. Review permissions, versioning, deletion protection, independent copies, and recovery tests as separate choices.

Run a restore test before moving the full dataset

Use a representative sample of file sizes, formats, versions, and permissions. Include a difficult case, such as a project with many small files or an encrypted backup. Our guide to recoverability testing explains why successful storage alone is insufficient.

1. Find the data: confirm another authorized person can identify the required objects from the catalog.
2. Request access: exercise the real approval, identity, and retrieval process.
3. Measure the complete recovery: record waiting, transfer, unpacking, and validation time.
4. Check data integrity and usability: compare suitable checksums or manifests, then open the files in the intended application.
5. Inspect the cost: record retrieval, request, transfer, and staging charges when billing data becomes available.
6. Record the result: save the procedure, limitations, owners, and the next review date.

Repeat after changes to permissions, encryption, archive software, or storage rules. A successful sample is useful evidence, but it does not prove that a full recovery will meet the same deadline. Test at the scale that matters for the recovery commitment.

Cold cloud storage questions

When is cold storage worth using?

When infrequent access, expected retention, and measured recovery time fit the chosen class, and the complete cost is lower for that workload. Stable completed projects can fit well; frequently scanned or repeatedly restored datasets need closer cost analysis.

Is cold cloud storage an offline backup?

Not automatically. Some cold classes allow immediate reads, while archival classes may require a restore step. Neither description establishes that the backup is isolated from compromised accounts or protected from deletion.

Can you delete archived files whenever you want?

That depends on permissions, retention settings, and holds. If deletion is allowed, a minimum-duration charge can still apply. Check both access controls and billing rules before relying on deletion to reduce the bill.