The 3-2-1 backup guide on this site makes the case for an off-site copy without picking a specific home for it. That’s deliberate, because “off-site” covers everything from a hard drive at your parents’ house to a proper cloud object storage tier, and which one makes sense depends entirely on how you actually plan to use it. This article is the follow-up: assuming you’ve decided cloud is part of your off-site story, which service, and why.

Three names come up constantly in data hoarding circles: Backblaze B2, Wasabi, and Amazon S3 Glacier (in its various tiers). They look similar on a pricing page and behave very differently once you factor in retrieval cost, retrieval speed, and the fine print around minimum storage duration. Getting this wrong doesn’t usually cost you data, it costs you money, sometimes a shocking amount of it, the one time you actually need to pull something back down.

The three services, and what they’re actually built for

Backblaze B2 is the closest thing to a straightforward object storage bucket built for exactly this use case. It speaks the S3 API (mostly, with some documented gaps), storage pricing has sat in the same low, flat range for years, and it has no minimum storage duration on individual objects. Egress (downloading your data back out) is the line item that matters most, and Backblaze’s model gives you a meaningful amount of free egress each month tied to how much you’re storing, plus free transfer to a short list of CDN/compute partners under the Bandwidth Alliance. Past that free allowance, egress is billed per GB.

Wasabi positions itself as “S3-compatible storage without the egress fees,” and largely delivers on that, with two catches worth knowing before you commit. First, storage is billed with a minimum retention period per object (commonly discussed as roughly 90 days), so uploading something and deleting it a week later still bills you for the minimum. Second, “no egress fees” comes with a fair-use style policy that caps free egress relative to how much you’re storing and how long you’ve stored it, meant to stop people using it as a free CDN rather than as backup. For a data hoarder who uploads once and rarely restores, that policy is background noise. For someone doing frequent restores or using it as anything more than cold storage, read the actual policy, not the marketing page.

AWS S3 Glacier isn’t one thing, it’s a family of storage classes (Glacier Instant Retrieval, Glacier Flexible Retrieval, and Glacier Deep Archive being the ones people actually reach for) sitting underneath regular S3. Storage cost drops as you move down that list, and Deep Archive in particular is dramatically cheaper per TB per month than either B2 or Wasabi, cheap enough that it’s genuinely the cheapest place to park data you expect to never touch. The tradeoff is retrieval: Deep Archive retrieval isn’t instant, it’s a job you submit that can take many hours to complete depending on the retrieval tier you pick, and both retrieval requests and egress are billed separately from storage. Glacier also carries its own minimum storage duration, longer than Wasabi’s, commonly discussed in the 90-180 day range depending on the tier.

None of these prices are quoted here as exact numbers on purpose. Cloud storage pricing shifts, tiers get renamed, and a number that’s accurate today reads as wrong in two years on a page meant to stay useful. Check each provider’s current pricing page before committing real money, but the relative shape described above (B2 as the balanced default, Wasabi as flat-rate with a minimum-duration catch, Glacier Deep Archive as the cheapest-but-slowest true archive tier) has held steady for years and is the part actually worth internalizing.

The number that actually decides this: retrieval, not storage

Storage cost gets all the attention on pricing pages because it’s the number you see every month. It’s also almost never the number that determines whether a service was the right choice. That’s decided by what happens the one time you actually need the data back.

Ask yourself honestly: if the primary array and the local backup both fail in the same bad week, how fast do you need this data back, and how much of it at once? A photo library or a handful of critical documents you’d want back same-day points toward B2 or Wasabi, where retrieval is effectively instant and the egress bill for a one-time full restore, while real, is a bounded and predictable cost. A 40TB media archive you’re keeping because deleting it feels wrong, not because you’ll ever restore it in bulk, points toward Glacier Deep Archive, where the low storage cost compounds in your favor every single month you don’t touch it, and the slow multi-hour retrieval and higher restore cost are a one-time toll you accept in exchange.

Mixing the two tiers by actual access pattern, rather than picking one service for everything, is usually the right call: the stuff you might genuinely need back in a hurry goes somewhere with instant retrieval, and the stuff that’s pure insurance against total loss goes somewhere optimized for storage cost instead.

What this replaces, and what it doesn’t

Cloud cold storage is not a substitute for the local backup layer covered in the 3-2-1 guide, and it’s not a substitute for the RAID redundancy or bit rot detection on your primary array. It’s specifically the off-site leg: the copy that survives a house fire, a burglary, or a ransomware attack that reaches every device physically present on your network. If you already have a second local copy on a NAS at a family member’s house or via something like the tape strategy in the LTO guide, cloud cold storage is a second off-site leg, not a replacement for the first, and layering both is reasonable at real scale rather than redundant.

It’s also worth being honest that none of these three services are meant to be browsed like a NAS share. They’re backup targets, not a place you mount and stream media from day to day, and treating Glacier especially as anything other than a write-once, restore-rarely target will produce a painful retrieval bill.

Getting data in and out without babysitting it

All three services speak the S3 API closely enough that the same tools work across all of them, which is the real practical win here: you’re not locked into a vendor-specific client.

rclone is the tool most homelabbers reach for first. It supports B2’s native API and S3-compatible endpoints for both Wasabi and Glacier, handles multi-threaded uploads well for large libraries, and can be scripted into a cron job or systemd timer for scheduled, incremental sync rather than a manual upload every time something changes.

restic and Duplicati both add encryption and deduplication on top of a cloud backend, which matters more than it sounds like it does. Uploading raw files to a bucket means anyone with the right credentials (or a provider-side breach) can read your data in plain form. Encrypting client-side, before it ever leaves your network, means the cloud provider is storing ciphertext they can’t read even if they wanted to, and it’s the same principle covered from the local-disk angle in any full-disk-encryption discussion, just applied to the off-site copy instead.

Whichever tool you pick, the same discipline that matters for local backups matters here too: automate the upload so it isn’t a thing you remember to do, and actually test a restore periodically. A cloud backup nobody has ever restored from is a hope, not a backup.

Picking one

If you’re setting this up for the first time and don’t have a specific reason to do otherwise, Backblaze B2 is the reasonable default for most home data hoarders: S3-compatible, no minimum storage duration to trip over, straightforward pricing, and fast enough retrieval that a real restore doesn’t turn into a multi-day ordeal. Add Wasabi if flat, predictable billing matters more to you than B2’s usage-based egress, and you’re comfortable with the minimum-duration tradeoff. Reach for Glacier Deep Archive specifically once you’re storing tens of terabytes of data you’re genuinely confident you won’t need back in a hurry, where the storage-cost savings at that scale outweigh the retrieval friction.

None of the three are wrong choices. They’re built for different points on the same tradeoff between storage cost and retrieval convenience, and the right one is whichever matches how you’d actually behave the day you need the data back, not which pricing page looks best sitting still.