Every major cloud has a storage tier called Archive, or something close to it. The names line up neatly, the prices per GB look similar, and it's tempting to treat them as interchangeable. They aren't. The rules attached to each tier differ in ways that decide whether moving data there saves money or just moves the cost somewhere less visible.
Issue #2 included a small reference table that lined these tiers up by rough role. That was a simplification. This is the more careful version.
The two rules that actually matter.
Price per GB is the number everyone compares, and it's the least useful one. Two other rules decide the outcome.
Minimum storage duration. Every cold tier charges as though the data stayed for a minimum period, even if you delete it or move it sooner. AWS and Azure both document this explicitly, and Google applies minimum storage durations the same way. Move something to a cold tier and then delete it three weeks later, and you pay for the remainder of the minimum anyway.
How you get the data back. This is where "Archive" stops meaning one thing. Some tiers keep data online and readable in milliseconds, with a higher fee when you read it. Others take it offline, and getting it back is a separate operation measured in hours.
What the docs actually say.
Provider | Class | Minimum duration | Getting data back |
|---|---|---|---|
AWS | Glacier Instant Retrieval | 90 days | Milliseconds |
AWS | Glacier Flexible Retrieval | 90 days | 1-5 min (expedited), 3-5 hrs (standard), 5-12 hrs (bulk) |
AWS | Glacier Deep Archive | 180 days | Within 12 hrs (standard), within 48 hrs (bulk) |
Azure | Cool | 30 days | Online, per-GB access charge |
Azure | Cold | 90 days | Online, per-GB access charge |
Azure | Archive | 180 days | Offline; rehydrate first, up to 15 hrs |
GCP | Nearline | 30 days | Online, low latency |
GCP | Coldline | 90 days | Online, low latency |
GCP | Archive | 365 days | Online, milliseconds |
Two things jump out. Google's Archive has the longest minimum of the nine rows, 365 days, and also the fastest access. Azure Archive and AWS Deep Archive are the opposite: the data is effectively offline until you ask for it back.
So "Archive" on Google is a pricing decision. On Azure and AWS Deep Archive, it's also an operational one.
The details that bite later.
Azure's archived blobs can't be read or modified. They have to be rehydrated to hot, cool, or cold first, and lifecycle policies can't do the rehydration for you. If an incident or an audit needs that data today, the 15-hour wait is the plan, not a worst case.
Azure's minimums depend on the account type. The 30 and 90 day minimums for Cool and Cold apply to general-purpose v2 accounts. Legacy Blob Storage accounts don't have them. Check which type you actually run before assuming either way.
AWS adds per-object overhead on the deep tiers. Glacier Flexible Retrieval and Deep Archive add about 40 KB of metadata per object, part of it billed at Standard rates. That's trivial for large files and expensive for millions of tiny ones. Glacier Instant Retrieval has a 128 KB minimum billable object size, which has the same effect on small objects.
Early-deletion charges are quiet. They show up as storage cost on a bill, not as a labeled penalty, so a lifecycle rule that moves data too aggressively can look fine in a dashboard while quietly costing more than leaving it alone.
How to choose without getting burned.
Ignore the tier names and ask three questions about the data:
How long will it actually live? If it might be deleted or moved in under 90 days, a 90-day or 180-day tier is a bad bet no matter how cheap the per-GB price looks.
How fast might you need it back? If the honest answer is "within minutes, during an incident," avoid the offline tiers entirely and look at the online options instead.
How big are the objects? Millions of small files change the math on every deep tier.
If you can't answer the first two reliably, Google's Autoclass (automatic class transitions per bucket) and AWS's Intelligent-Tiering from Issue #2 exist for exactly that case, with their own tradeoffs.
The actual takeaway: Matching tier names across clouds will mislead you. Match on minimum duration and retrieval behavior instead, because those two rules decide the real cost. The cheapest tier on paper is often the most expensive one once data moves out early or has to come back fast.
Which cold tier has surprised you on a bill? Hit reply and tell me what happened. I read every one, and the good ones might show up (anonymized) in a future issue.