The headline rate is the cheap part
S3 Glacier Deep Archive holds a terabyte for about $1.01 a month, or $0.00099 per GB. That number is real, and it's why deep archive shows up in nearly every backup diagram. But the storage rate is the part AWS wants you to read. The part that bills you is retrieval, and pulling back data you pay pennies to store can land a four-figure invoice in a single afternoon. The trap isn't buried in the fine print. It's just on a different line of the pricing page than the one everyone screenshots.
Three meters run on a restore, and storage is the smallest
Pull data out of Deep Archive and several meters run at once. People who budgeted around the storage rate usually only know about the first.
- Retrieval fee — you pay per GB to lift data out of cold storage and stage it. Standard retrieval (≈12 hours) runs
$0.02/GB, i.e.$20/TB. Bulk (≈48 hours) drops to$0.0025/GB, roughly$2.50/TB. There is no expedited tier on Deep Archive; that option only exists on the warmer Glacier Flexible Retrieval class. "I'll just pay for fast" is not on the menu here. Your only lever is how long you can wait. - Data transfer out — once it's staged, getting the bytes off AWS to the internet costs
$0.09/GBfor the first 10 TB each month, sliding to$0.085and then$0.07at higher volumes. That's$90/TB, and it has nothing to do with which storage class you picked. On a real restore it's usually the biggest single number. - Request fees — small per-object charges. Negligible across a few large files. Painful across millions of tiny ones, because every object is a separate billable retrieval request.
Stack those and the storage line, the one you optimized for, becomes a rounding error the moment you need the data back.
A worked example: 50 TB you stored cheap, and the day you need it
Say you've parked 50 TB of backups in Deep Archive. The monthly bill is genuine and lovely:
| Line item | Rate | 50 TB cost |
|---|---|---|
| Storage (monthly) | $0.00099/GB-mo | ~$49.50/month |
Fifty dollars a month to sit on 50 TB. Defensible. Now ransomware hits your primary and you need all 50 TB back, fast, restored to on-prem hardware. Here's the one-time restore bill:
| Line item | Rate | 50 TB cost |
|---|---|---|
| Standard retrieval (~12 hr) | $0.02/GB | $1,000 |
| Data transfer out to internet | $0.09/GB first 10 TB, then tiered | ~$4,400 |
| Request fees | per object | tens to hundreds |
| One urgent restore, total | ~$5,500+ |
That's roughly 112 times your monthly storage bill, spent in one day, to get back data that cost fifty bucks a month to hold. Drop to bulk retrieval and the retrieval fee shaves to about $125, but now you've committed to a 48-hour wait with production down, and egress still owns the bill at ~$4,400. The math doesn't care that the storage was cheap. It never did.
Why does deep archive retrieval cost so much more than storage?
Because the price reflects the physics and the bet AWS is making. Deep archive sits on high-density media tuned for write-once, read-approximately-never. Spinning that data up, staging it, and shipping it across the internet is genuinely expensive work, and AWS prices retrieval to discourage you from doing it often. The whole class is a wager that you'll almost never read what you wrote. When the wager holds — the seven-year compliance copy nobody ever touches — it's one of the best deals in cloud storage. When it breaks, you pay the premium at the worst possible moment: an outage, or an audit due tomorrow, where every hour of downtime carries its own cost.
When deep archive genuinely wins, and when it's a false saving
Deep archive is the right tool when retrieval is truly rare. The rule of thumb is under ~1% of stored data read per month. Regulatory retention you keep because a law says so. Raw footage you'll likely never reopen but can't delete. The final backup of a backup. In those cases the cheap storage rate is the whole story, and you'll basically never touch the retrieval side.
It's a false saving when archive is quietly doing the job of a backup you actually expect to restore. Disaster-recovery copies live here. So does the "we might need this next quarter" dataset, and anything where a restore is a when, not an if. Slot active recovery data into deep archive to win on the storage line and you've simply moved the cost to a future invoice, one that arrives on the day you can least afford a surprise and least want to wait 12 to 48 hours.
The cheapest place to store data is not the cheapest place to store data you intend to read.
A note on source variance
Be honest about the numbers, because they move. Retrieval and egress rates vary by region, AWS reprices periodically, the egress tiers step down as volume climbs, and request fees swing wildly with object count. The figures here are list prices for a common US region as of mid-2026, and your console is the only authority for your account. The shape of the trap is stable even when the digits aren't: storage is cheap, retrieval plus egress is not, and the gap is widest exactly when you're under pressure. Before you commit anything to a deep-archive tier, run your own worst-case restore through a calculator — full volume, urgent timeline, egress included — and look at that total, not the monthly storage line.
The model we'd rather you reason about
Tiered cold storage exists because, at hyperscaler scale, storage and retrieval really are different cost centers, and pricing them apart is rational engineering. The failure is on the buyer side: optimizing the one rate that's visible and forgetting the two that aren't. We took a different stance with beebeeb. Storage is flat and predictable — up to 99 TB self-serve at €10.99 per extra TB, with a custom quote beyond that, no separate retrieval meter, no egress toll for reading your own files back. There's no tier where getting your data out is the expensive part, because charging you to access your own data at the moment you need it most was a business model we didn't want.
It's also end-to-end encrypted and zero-knowledge on every tier, no exceptions: AES-256-GCM on file contents, Argon2id for key derivation, keys derived and zeroized client-side, servers in Falkenstein, Germany. That's a different design center than deep archive. Deep archive optimizes for the cost of bytes you'll never read. We optimize for the cost, and the privacy, of bytes you'll actually need back. If you only ever read 1% of your data a month and never on a deadline, Glacier Deep Archive is a fine and cheap home for the other 99%. Just price the restore before you trust the storage rate.