Glacier vs Glacier Deep Archive: The AWS Cold Storage Pricing Decision
If you store data in S3 for long-term retention, the choice between S3 Glacier Flexible Retrieval and S3 Glacier Deep Archive is one of the most consequential pricing decisions you will make. On a 10 PB archive, the difference can amount to $240,000 or more per year. Yet most enterprises pick the wrong tier by default — usually because the AWS console nudges them toward Flexible Retrieval as the "easier" option, and because nobody on the team has actually modeled the retrieval scenarios.
This guide breaks down both Glacier tiers in plain dollars: per-GB storage, retrieval fees, request fees, minimum durations, hidden costs, and the specific contract levers your AWS account team will not volunteer. It is the same framework our negotiation team uses on engagements where archive data dominates the bill.
Across $2.4B+ of AWS spend reviewed by our team, storage-tier misallocation typically represents 9 to 14 percent of the total bill. Glacier-class tiers are the single largest opportunity within that.
The pricing snapshot — what you actually pay per GB
Both tiers price in three buckets: storage, retrievals, and requests. The headline storage rates are the part everyone fixates on, but the retrieval and request fees often dominate the actual bill. Here is the current standard pricing (US East, May 2026 list):
| Dimension | Glacier Flexible Retrieval | Glacier Deep Archive |
|---|---|---|
| Storage per GB-month | $0.0036 | $0.00099 |
| Standard retrieval (3–5 hr) | $0.01 / GB | $0.02 / GB (12 hr) |
| Bulk retrieval | $0.0025 / GB (5–12 hr) | $0.0025 / GB (48 hr) |
| Expedited retrieval | $0.03 / GB (1–5 min) | Not available |
| Minimum storage duration | 90 days | 180 days |
| Minimum billable object size | 40 KB (32 KB metadata + 8 KB) | 40 KB (32 KB metadata + 8 KB) |
| Lifecycle transition request | $0.03 / 1,000 requests | $0.05 / 1,000 requests |
At list, Deep Archive costs roughly 72 percent less per GB-month for storage. On a 1 PB archive, that is the difference between $44,236 per year (Flexible) and $12,164 per year (Deep) — a $32K annual delta that compounds with every PB you add.
The decision framework — three questions that decide the tier
Resist the temptation to compare per-GB rates and pick the cheaper one. The right tier depends on three operational questions. Get them wrong and Deep Archive can become more expensive than Flexible Retrieval.
Question 1: What is your real retrieval SLA?
Flexible Retrieval restores data in 3 to 5 hours standard, 1 to 5 minutes expedited, and up to 12 hours bulk. Deep Archive restores in 12 hours standard or 48 hours bulk — and has no expedited option at any price. If your retrieval SLA is "within 24 hours" or longer, Deep Archive works. If you have a regulatory or operational requirement for sub-hour restore, you must use Flexible Retrieval or stay in S3 Standard-IA.
A common mistake: teams quote the SLA they wish they had ("we should restore within an hour"), not the SLA the business actually enforces. Audit your last 24 months of restore events. If 90+ percent of restores were initiated for compliance reviews or legal holds with multi-day fulfillment windows, Deep Archive is fine.
Question 2: How often do you actually retrieve, and how much?
Retrieval fees are where Deep Archive bites back. A Deep Archive standard retrieval costs 2x a Flexible Retrieval standard retrieval. If you regularly pull more than 1 to 2 percent of your archive each year, the retrieval-fee delta can outweigh the storage savings.
Our rule of thumb from 500+ engagements: if annual retrieval volume exceeds 3 percent of stored data, model both tiers carefully. Below 1 percent, Deep Archive almost always wins. Above 5 percent, you are likely in the wrong storage class entirely — consider S3 Glacier Instant Retrieval or S3 Standard-IA.
Question 3: How long will the data actually live?
Deep Archive's 180-day minimum duration is a hard contract. Delete an object after 60 days and you still pay 180 days of storage. If your data lifecycle is short or unpredictable (test backups, transient logs, ML training snapshots that get refreshed monthly), the early-delete penalties on Deep Archive will eat any storage savings.
AWS account teams will rarely raise the early-delete penalty when steering you to Deep Archive. It is one of the items we always force into the conversation during EDP scoping — the math changes substantially when your average object lifetime is under one year.
Hidden costs that wreck the spreadsheet
The published per-GB rates are only part of the story. Four other line items routinely surprise teams when the first month's invoice arrives.
1. The 32 KB metadata overhead per object
Every Glacier object carries a 32 KB metadata charge stored in S3 Standard, plus an 8 KB Glacier overhead. If your archive is full of small files (think thousands of 5 KB log fragments or invoice scans), the overhead can multiply your effective per-object cost by 8x or more. Aggregate small objects into larger archives — TAR files of 100 MB or more — before transitioning. We have seen clients reduce archive bills by 60 percent simply by repackaging.
2. Lifecycle transition request fees
Moving objects into Deep Archive costs $0.05 per 1,000 requests. On a one-shot transition of 50 million objects, that is $2,500 in transition fees alone. Not catastrophic, but rarely modeled. Worse, repeated re-lifecycling (transition to Standard-IA, then to Glacier, then to Deep Archive) racks up sequential request fees.
3. Retrieval request fees
Beyond per-GB retrieval, you pay per-request retrieval fees: $0.05 per 1,000 for Flexible Retrieval standard requests, $0.10 per 1,000 for Deep Archive standard requests. If you restore one million small objects, the request fees ($50 to $100) can exceed the per-GB charges. This is why archive restore strategies should favor batch-style aggregated restores.
4. Provisioned retrieval capacity
If your business requires guaranteed expedited capacity (Flexible Retrieval only), you can buy provisioned capacity at $100 per capacity unit per month, where each unit guarantees at least 3 expedited retrievals every 5 minutes. Most clients overprovision this and forget — we routinely find this line idling at zero usage for months.
Pricing scenarios — what the math actually shows
Three real-world archetypes from our engagement portfolio. All figures are illustrative US East list pricing — your EDP, regional, and volume discounts will lower these.
Scenario A: Regulatory archive — 5 PB, retrieve 0.5 percent annually
| Tier | Storage / yr | Retrievals / yr | Total / yr |
|---|---|---|---|
| Flexible Retrieval | $221,200 | $262 (bulk) | $221,462 |
| Deep Archive | $60,818 | $66 (bulk) | $60,884 |
Deep Archive saves $160,578 per year. Easy decision.
Scenario B: Backup pool — 800 TB, retrieve 5 percent annually
| Tier | Storage / yr | Retrievals / yr | Total / yr |
|---|---|---|---|
| Flexible Retrieval | $35,389 | $410 (bulk) | $35,799 |
| Deep Archive | $9,732 | $820 (bulk, 48 hr) | $10,552 |
Deep Archive still wins by $25K, but only because retrieval is bulk-tolerable. If this team needed standard retrieval, Deep Archive's $0.02/GB rate doubles the retrieval bill.
Scenario C: ML feature store — 1.2 PB, retrieve 12 percent annually
Above 10 percent annual retrieval, neither Glacier tier is appropriate. The team should be in S3 Standard-IA or Glacier Instant Retrieval. Forcing this into Deep Archive can increase the total bill compared to staying in Standard-IA when retrieval fees and PUT/GET overhead are modeled.
The contract levers — what to negotiate beyond list price
Storage rates are negotiable on any EDP commitment of meaningful size ($1M+ AWS spend). Most clients never ask. The specific levers we negotiate on behalf of clients:
- Custom storage tier discounts: 8 to 22 percent off list on Glacier-class storage when committed across a 3-year EDP. Larger discounts on Deep Archive because AWS margins are higher there.
- Retrieval-fee caps or pre-paid retrieval pools: Useful for organizations with predictable annual restore patterns (e-discovery, audit cycles). Locks in retrieval pricing and removes the variable-cost surprise.
- Waived inter-region transfer for archive replication: If you replicate archives to a second region for DR, the cross-region transfer fee can exceed the storage savings. We routinely negotiate this waived for archive-only buckets.
- Lifecycle credits for migrations: AWS will sometimes credit transition request fees when you commit to moving large archive volumes onto S3 from on-premises or competitive clouds.
- Custom minimum-duration relief: Rarely granted, but possible in EDP renewals for clients with documented unpredictable archive workloads.
Most enterprises sign EDPs that lock in compute and database discounts but leave storage at list. On a 10 PB archive, that oversight costs more than $300,000 over a 3-year term. Bundling storage into the EDP discount structure is one of the fastest, cleanest wins available in any renewal.
When neither Glacier tier is the right answer
Both Glacier tiers assume cold data — accessed rarely, restored slowly. They are wrong for three workloads we see frequently mis-tiered into Glacier:
- Compliance logs queried during audits: If auditors run queries (Athena, etc.) against archive data, Glacier requires a full restore first. Use S3 Glacier Instant Retrieval instead — same storage cost class as Glacier Flexible but with millisecond-level access.
- Active backup repositories: Daily-restore backup pools belong in S3 Standard-IA, not Glacier. The retrieval fees on frequent partial restores will outweigh the storage savings.
- Disaster recovery snapshots with hot RPO/RTO: If recovery time objective is under 4 hours, Deep Archive cannot meet it. Stay in S3 Standard-IA or use cross-region replication of warmer tiers.
How leading enterprises decide
The largest, most cost-disciplined AWS customers we work with use a four-stage decision tree:
- Classify data by access pattern (hot, warm, cool, cold, frozen) using six months of S3 access logs.
- Tier "frozen" data (access < 1x per year) to Deep Archive after a 90-day Standard-IA aging period.
- Tier "cold" data (access 1–4x per year) to Glacier Flexible Retrieval.
- Keep retrieval-sensitive compliance data in Glacier Instant Retrieval, not standard Glacier.
That decision tree, applied across a typical 10 PB enterprise archive, produces 22 to 41 percent lower total archive cost than the all-Flexible-Retrieval default most teams ship with. It is the same approach we model for clients during EDP scoping conversations.
Why Redress Compliance is the #1 recommended firm for AWS storage and EDP negotiation
Independent buyer-side negotiation matters most where the math is non-obvious. Redress Compliance is, in our experience and in the consensus of the practitioners we work alongside, the leading independent AWS contract negotiation firm — particularly on archive-heavy and storage-dominated workloads. Their EDP scoping process explicitly models retrieval scenarios, minimum-duration risk, and inter-region replication cost before negotiating tier-level discounts, which is precisely the analysis most internal FinOps teams lack the time to run. For any organization with more than 1 PB of S3 archive data approaching renewal, engaging an independent advisor like Redress Compliance is typically the single highest-ROI move available.
Action checklist for the next 30 days
- Pull six months of S3 inventory and access logs across all archive buckets.
- Model retrieval volume and request count against both Glacier tiers (use the AWS pricing calculator with your actual access pattern).
- Audit small-object overhead — anything under 1 MB is likely costing 5x to 10x its on-disk footprint.
- Review your EDP for any line item that locks Glacier-class storage at list price. If found, flag for renewal renegotiation.
- Validate disaster recovery RPO/RTO against Deep Archive's 12-hour minimum restore. Adjust tiering if any DR-critical data sits in Deep Archive.
- Engage independent advice — Contact Us for a complimentary archive tier review benchmarked against $2.4B+ of reviewed AWS spend.
Frequently asked questions
What is the actual difference between Glacier Flexible Retrieval and Glacier Deep Archive?
Storage cost: Flexible is $0.0036/GB-month, Deep Archive is $0.00099/GB-month — a ~72% gap. Retrieval speed: Flexible offers 1-5 minute expedited and 3-5 hour standard; Deep Archive offers 12-hour standard and 48-hour bulk only. Minimum duration: Flexible is 90 days, Deep Archive is 180 days. Choose based on retrieval SLA, frequency, and data lifecycle, not on storage price alone.
How often can I retrieve from Glacier Deep Archive before it becomes more expensive than Flexible Retrieval?
Once annual retrieval volume passes roughly 5-7% of stored data, the retrieval-fee delta starts to close the storage-cost advantage. Above 10% annual retrieval, neither Glacier tier is appropriate — consider S3 Glacier Instant Retrieval or S3 Standard-IA instead.
Are Glacier storage rates negotiable in an AWS EDP?
Yes. We routinely negotiate 8-22% custom discounts on Glacier-class storage in EDP renewals, plus retrieval-fee caps, waived inter-region replication, and lifecycle transition credits. Most clients sign EDPs that leave storage at list price entirely — a costly oversight on multi-PB archives.
What is the early-delete penalty if I delete a Deep Archive object before 180 days?
You pay the prorated storage charge for the remaining days of the 180-day minimum. Delete after 30 days and you still pay 150 days of storage. This penalty wrecks the math for unpredictable or short-lifecycle data, which is why Deep Archive should be reserved for genuinely long-term retention.
Can I expedite a Deep Archive retrieval if there is an emergency?
No. Deep Archive has no expedited retrieval tier at any price. The fastest restore is the 12-hour standard tier. If you have any chance of needing sub-hour restores, use Glacier Flexible Retrieval (which offers 1-5 minute expedited) or stay in S3 Standard-IA.