S3 Storage Classes

Most data gets cold fast — an invoice PDF is read constantly for a week, then almost never. Paying S3 Standard prices for it forever is waste. S3 Storage Classes let the same object, through the same API, sit on progressively cheaper storage as it ages. Durability stays at 11 nines across every class (except One Zone). What changes is availability, minimum storage duration, per-GB price, and how fast/expensively you can read it back.

• Every class stores data across ≥ 3 AZs in the bucket’s Region — except S3 One Zone-IA, which uses a single AZ and therefore drops to ~99.5% durability against AZ loss (the 11-nines figure assumes the AZ survives).
• Storage class is an object-level property. Different objects in one bucket can be in different classes, and Lifecycle rules transition them automatically.
• Glacier classes are not a separate service here — they’re storage classes inside S3, addressed by normal S3 keys.

graph LR
HOT["S3 Standard (frequent access, ms latency)"] -->|"not accessed 30d"| IA["S3 Standard-IA (infrequent, retrieval fee)"]
IA -->|"90d"| GIR["Glacier Instant Retrieval (ms, archive price)"]
GIR -->|"180d"| GFR["Glacier Flexible Retrieval (mins to 12h)"]
GFR -->|"years"| DEEP["Glacier Deep Archive (12h to 48h, cheapest)"]
HOT -.->|"unknown / changing pattern"| IT["S3 Intelligent-Tiering (auto-moves, no retrieval fee)"]
HOT -.->|"reproducible data, cost over durability"| OZ["S3 One Zone-IA (single AZ)"]
classDef hot fill:#eaf7ec,stroke:#2e8b3d,stroke-width:2px,color:#1a3d20
classDef warm fill:#e8f0fe,stroke:#3b82f6,stroke-width:2px,color:#173a70
classDef cold fill:#f3ecfb,stroke:#8b3de0,stroke-width:2px,color:#3a1a5c
classDef special fill:#fdf0e6,stroke:#e0842e,stroke-width:2px,color:#6b3b0f
class HOT hot
class IA,GIR warm
class GFR,DEEP cold
class IT,OZ special

Class	Retrieval	Min duration	Use it for
Standard	Instant, ms	none	Active data, websites, analytics inputs
Intelligent-Tiering	Instant (auto-tiers)	none	Data with unknown or shifting access patterns — set and forget
Standard-IA	Instant, ms + per-GB fee	30 days	Backups, older files read a few times a month
One Zone-IA	Instant, ms + per-GB fee	30 days	Re-creatable data where a single-AZ risk is acceptable
Glacier Instant Retrieval	Instant, ms + fee	90 days	Archives you still occasionally need immediately (medical images)
Glacier Flexible Retrieval	Minutes–12 h	90 days	Backups, DR — expedited/standard/bulk retrieval tiers
Glacier Deep Archive	12–48 h	180 days	Compliance/regulatory retention, tape replacement — cheapest storage on AWS

Per-GB storage falls roughly 10× from Standard to Deep Archive, but colder classes add retrieval charges and minimum storage durations — delete or transition an IA object before 30 days and you still pay for 30. Intelligent-Tiering charges a small per-object monitoring fee, so it’s not economical for millions of tiny objects. The cost mistake the exam loves: putting frequently-read data in IA/Glacier and getting crushed by retrieval fees.

• “Access pattern is unknown / unpredictable” → Intelligent-Tiering. It’s the default correct answer whenever the question refuses to tell you how often data is read.
• “Data can be regenerated” / “thumbnails” / “logs we could rebuild” + cost focus → One Zone-IA.
• “Retrieve within milliseconds” but “rarely accessed” and “archive” → Glacier Instant Retrieval. If minutes-to-hours is fine → Flexible Retrieval. If “7–10 years, regulatory, lowest cost” → Deep Archive.
• Transitions must go warm → cold; you can’t lifecycle an object from Glacier back to Standard automatically — that requires a restore (temporary copy) then a manual copy.
• Lifecycle policies act on age or on noncurrent version age (with versioning) and can also expire objects and abort incomplete multipart uploads — mention that last one for cost-cleanup questions.
• Glacier Flexible Retrieval retrieval tiers: Expedited (1–5 min), Standard (3–5 h), Bulk (5–12 h, cheapest).