They are sold as the same thing and they are not. A backup protects what you are using. An archive holds what you have finished with. Cold storage is a price tier, not a purpose.
A backup protects against deletion, corruption, ransomware and mistakes. It is judged on the restore.
An archive protects against a different problem: a working system clogged with things nobody touches. Full cloud quotas, slow search, a file server kept alive for five requests a year. The data is not at risk; it is in the way, and expensive where it sits.
Cold storage makes storing something cheaper in exchange for making it harder to get back. That suits a backup of a backup. It does not suit an archive someone has to search next week.
Using a backup as an archive. Backups keep several versions of everything, so pointing one at thirty terabytes of finished work means paying to hold it several times over, for ever, while it stays in the live system taking up its quota.
The reverse mistake: moving finished work into an archive and assuming it is now backed up. An archive with one copy in one place has no protection.
Cold storage sits underneath either, where retrieval does not matter. Data described as never needed is often needed in a tender, a dispute or an insurance claim.
Cold storage is sold on the price of keeping data. The costs that matter are the fee to get something back and the time it takes.
| Where it sits | To keep, per TB a month | To get 1 TB back, including egress | How long for 1 TB | Minimum stay |
|---|---|---|---|---|
| S3 Glacier Deep Archive | about US$1 | about US$110: US$20 retrieval plus about US$90 egress | about 12 hours standard, about 48 hours bulk | 180 days |
| Azure Archive | about US$1 | about US$20 plus egress | about 4 days, capped at 10 GiB an hour | 180 days |
| SharePoint extra storage | £180 | nothing | straight away | none on a monthly term |
| Shed | £90 | nothing | when you want it | none |
Microsoft caps Azure Archive rehydration at 10 GiB per storage account per hour, so a terabyte takes about four days and ten terabytes about six weeks, at any price.
Retrieval and egress are two different charges and you pay both. Retrieval moves a file out of the cold tier; egress moves it across the internet to you, at about US$0.09 per GB on AWS after the first 100 GB a month.
Pulling a ten terabyte archive back out of Glacier Deep Archive, at October 2026 list prices:
About US$1,100 and a day, for data you already paid to store. The storage was a dollar a terabyte a month, which is why it got chosen.
Not every provider charges this way. Backblaze B2 includes egress up to three times your average monthly storage, and Wasabi includes it provided downloads stay under what you store. Both are far cheaper per terabyte than anything Microsoft sells, and for plain object storage they are a sensible answer.
They are not an archive. There is no verification report, nothing is searchable inside a document, nobody checks that the other copy is still good, and a folder does not go back where it came from. You supply the archive yourself.
Both AWS and Azure charge a 180 day minimum. Data removed in month two is billed for six.
Glacier and Azure Archive suit data that will not be read, held for a regulator who gives notice. They do not suit finished creative work, which gets asked for at short notice.
Competitor figures checked October 2026 for US East, and they move: confirm on the AWS S3 pricing page and the Azure Blob Storage pricing page before budgeting. Retrieval rates vary by region. The rehydration cap and the 180 day minimums come from AWS's and Microsoft's own documentation.
Shed is an archive with a backup's checks. Every file is compared with the original by hash before anything is removed from your cloud, snapshotted nightly and replicated across our nodes, and it stays open: searchable inside documents, previewable in a browser, and a folder goes back where it came from in one step.
It is also used as a backup on its own, with nothing moved or removed, by companies that want a checked copy somewhere their cloud provider is not.
Tell us which system is full, or read what happens when SharePoint storage is full.