AWS Backup cold tier: when "cheaper" storage costs you more

Cold storage. The word itself sounds like a bargain. You hear "cold tier" and your brain immediately goes: less accessible, therefore cheaper. That's how it works with S3 Glacier. That's how it works with most archive tiers across cloud providers. So when AWS Backup offers a cold storage option at 75% lower per-GB pricing, the instinct is obvious. Move everything there and save money.

I had exactly that instinct. And I should have known better.

Before starting CloudYali, I spent years at Druva, one of the leading SaaS backup companies. I worked specifically on their cold backup feature designed to lower retention costs for long-term data. I even share a patent on it. So when I looked at AWS Backup's cold tier, I thought I knew how this worked. Cold storage = cheaper for infrequently accessed backups. That's how I built it at Druva. That's how it should work everywhere.

But when I actually dug into how AWS Backup cold tier works, not the marketing page but the actual service architecture, what I found genuinely surprised me. For EBS backups specifically, cold storage can cost up to 2× more than warm. Not a little more. Double.

Think of it like switching your gym membership from a pay-per-visit plan to an unlimited plan because the per-visit cost is lower. Sounds smart until you realize the unlimited plan charges you for 24 hours a day whether you show up or not. The unit economics look better. The total spend looks worse. I'll show you exactly why in the numbers below.

If someone with a patent in cold backup storage can get this wrong, it's worth writing about. This post is me walking through the math that changed my mind, so you don't have to learn this lesson the expensive way.

The pricing table everyone sees (and the details they miss)

Here's the headline pricing for US East (N. Virginia), pulled from AWS Backup and service-specific pricing pages:

Look at that EBS cold price. $0.0125 vs $0.05. That's 75% cheaper per GB. Looks like a no-brainer, right?

Here's what the table doesn't tell you.

First, EBS warm restores are completely free. Zero. Nothing. Cold restores cost $0.03/GB, so every restore from cold is pure incremental cost you wouldn't have paid in warm.

Second, and this is the big one, EBS cold storage converts every incremental snapshot into a full snapshot. It uses the EBS Snapshots Archive Tier under the hood, and that tier doesn't support incrementals. Your 50 GB of changed blocks? They become a full 1 TB snapshot in cold.

Third, DynamoDB cold storage requires an irreversible opt-in to "advanced features," which actually changes your warm pricing from $0.10 to $0.05/GB-mo. Read that again. Irreversible.

The EBS cost paradox: how I discovered cold storage doubles your bill

This is the part that caught me off guard. Let me walk through the exact scenario that opened my eyes.

The setup: A 1 TB EBS volume, backed up monthly, 1-year retention, 5% monthly block change rate (typical for a database volume), one restore per year.

Warm tier: what you'd expect

EBS snapshots in warm storage are block-level incremental. Only changed blocks get stored after the initial full snapshot. At steady state with 12 monthly backups:

Cold tier: where the math breaks

When snapshots transition to cold, AWS converts each one into a full snapshot. Every single monthly backup stores the entire 1 TB volume:

Side by side, the picture is clear

Cold tier costs 97% more than warm. The 75% per-GB discount gets completely overwhelmed by the 8× increase in stored data from the incremental-to-full conversion.

And this isn't some contrived edge case. Any EBS volume with a change rate below ~18% will be cheaper in warm storage. Most production volumes like databases, app servers, and file systems sit at 1–10%.

So where's the actual break-even?

This is the question I kept coming back to: at what change rate does cold actually start saving money?

Let c = monthly change rate (fraction), V = volume size in GB, and n = number of retained monthly backups (12 for annual retention):

Warm storage cost (annual):

Total stored = V + (n − 1) × c × V = V × [1 + (n − 1) × c]

Annual cost = V × [1 + (n − 1) × c] × $0.05 × 12

Cold storage cost (annual, storage only):

Total stored = n × V

Annual cost = n × V × $0.0125 × 12

Setting warm = cold and solving for c:

V × [1 + 11c] × 0.60 = 12 × V × 0.0125 × 12

[1 + 11c] × 0.60 = 1.80

1 + 11c = 3.0

11c = 2.0

c = 0.182 (18.2%)

So you need an 18.2% monthly block change rate before cold tier even breaks even. Here's how it plays out:

Most production EBS volumes exhibit 2–10% monthly change rates. Cold storage is the wrong choice for the vast majority of EBS backup workloads.

Don't take my word for it. Try it with your own numbers:

EFS and DynamoDB: where cold tier actually makes sense

Now, I don't want to give the impression that cold storage is always a trap. For EFS and DynamoDB, the story is completely different because incremental behavior is preserved. No full-snapshot conversion happens.

EFS: the sweet spot for cold storage

That's a 76% savings. And the break-even on restores?

$600 + R × $20 = $120 + R × $30

$480 = R × $10

R = 48 restores per year

You'd need to restore nearly once a week before cold tier loses its advantage. For compliance backups that sit untouched for years, this is exactly what cold storage was designed for.

DynamoDB: decent savings, tighter margins

Break-even restores: ($600 − $360) / ($200 − $150) = 4.8 restores/year. Just 5 full restores per year and cold tier loses its advantage. The storage discount is smaller (40% vs 80%) and that $0.20/GB restore fee adds up fast.

The 90-day trap that nobody warns you about

Here's another thing that bit me. AWS Backup cold storage has a non-negotiable 90-day minimum retention. Delete before 90 days? You still pay for the full 90 days. Your retention policy must satisfy: DeleteAfterDays ≥ MoveToColdStorageAfterDays + 90.

This creates three traps that I see teams fall into:

Trap 1: Short retention policies can't use cold storage at all. If your retention is 60 days, it's architecturally impossible. The minimum total retention with cold is 91 days (1 day warm + 90 days cold).

Trap 2: Accidental cold transitions on short-cycle backups. Weekly backups with 30-day retention that accidentally transition to cold will incur 60 extra days of charges per backup. I've seen this happen.

Trap 3: Transitioning to cold too early. If you move backups to cold before 8 days in warm, AWS Backup creates an entirely new full warm backup. You end up paying for both. AWS themselves recommend at least 8 days in warm before transitioning.

The sweet spot for cold storage is retention periods of 6 months to several years, where the 90-day minimum becomes irrelevant and per-GB savings compound over time. This pattern of quiet penalties hiding in cloud services isn't unique to AWS either. Azure's extended support has its own version of this problem, and rising AI infrastructure costs are creating similar surprises across all three major providers.

The decision tree I wish I had from the start

After working through all of this, I built a decision tree that I now use for every backup cost review. It covers the three services where cold tier is available (EBS, EFS, DynamoDB) and walks through the key questions: retention length, change rate, backup frequency, restore tolerance, and restore frequency.

And here's the same information as a quick-reference table:

If cold tier is wrong for EBS, what should you actually do?

This was my next question after realizing cold tier was a trap for most EBS workloads. I found six alternatives that work, and I've written them up in detail in Part 2: Six ways to cut EBS backup costs without touching cold tier. Here's the short version:

1. AWS Data Lifecycle Manager (DLM) keeps snapshots incremental in warm tier. It's free. This is the highest-return switch for EBS-only workloads.
2. Reduce backup frequency in warm tier. Quarterly warm consistently beats monthly cold on total cost.
3. Cross-account snapshot copies give you account isolation without the conversion penalty.
4. S3 Glacier Direct is 68–92% cheaper than cold tier for long-dated compliance archives.
5. Third-party tools (N2WS, Veeam, Druva) archive incrementally to S3/Glacier without the full-snapshot conversion.
6. Manual Glacier archiving of individual checkpoint snapshots for surgical, long-term retention.

Each has exact pricing, worked examples, and honest limitations in Part 2.

What I learned

I spent years building backup features at Druva. I have a patent in this space. And I still got tripped up by AWS Backup's cold tier because I assumed it would work the way cold storage should work. The way I helped build it. It doesn't.

The lesson: always read the fine print and understand the service architecture before making cost decisions. The per-GB price on a pricing page is not your cost. It's a single input into a formula that includes data volume, retention behavior, restore patterns, and architectural constraints you didn't know existed. For EBS cold storage, that formula produces a genuinely counterintuitive result. A 75% per-GB discount that leads to a 97% cost increase.

Any EBS backup plan with cold storage enabled is worth reviewing. Switching to warm will almost certainly reduce costs. If you want a systematic way to catch issues like this across your AWS environment, the AWS cost optimization guide is a good place to start. For EFS and DynamoDB, cold storage delivers real savings of 25–76%, but only when retention exceeds 6 months and restores stay below the break-even thresholds (48/year for EFS, 5/year for DynamoDB).

The November 2025 launch of [S3 low-cost warm tier](https://aws.amazon.com/about-aws/whats-new/2025/11/aws-backup-low-cost-warm-storage-s3-backups/) with roughly 30% savings and no cold-tier penalties further narrows cold storage's value proposition. It offers a middle path that avoids the 90-day minimum, retrieval fees, and restore delays entirely.

The cheapest-looking option isn't always the cheapest. Sometimes the fine print is where the real cost lives. I learned my lesson.