AWS S3 Storage Classes Explained: Standard, Intelligent-Tiering, IA, Glacier & Express One Zone

Simple explanation

The key insight: cheaper storage classes are not worse quality. They are optimised for a different access pattern. A 7-year compliance archive does not need to live on the fastest storage tier. Putting it there is just paying for speed you never use.

Every class except Intelligent-Tiering requires you to decide upfront which tier is appropriate. Get it wrong and you either overpay on storage (too expensive a class for cold data) or overpay on retrieval (wrong class for data you access more than expected). Intelligent-Tiering removes that guesswork at the cost of a small per-object monitoring fee.

How S3 storage classes work

Every storage class involves the same three cost variables:

• Storage cost per GB: what you pay just to hold the data each month
• Retrieval cost per GB: what you pay when you read the data back
• Retrieval speed: how quickly the data is available after a request

Standard has the highest storage cost and no retrieval cost. Deep Archive has the lowest storage cost but the highest retrieval cost and a 12–48 hour wait. Most classes sit somewhere between.

Minimum storage durations add a wrinkle. Several classes charge you as if an object was stored for a minimum period, even if you delete it earlier. Standard-IA has a 30-day minimum; the Glacier classes have 90 or 180-day minimums. Transitioning objects before they have aged enough can make your bill higher, not lower.

Multi-AZ vs single-AZ is the other key variable. Most classes replicate across at least three Availability Zones, giving 99.999999999% (11 nines) durability. One Zone-IA and Express One Zone store data in a single AZ. They are cheaper, but an AZ failure can cause permanent data loss. Only use single-AZ classes for data you can regenerate.

Lifecycle policies are what make storage classes practical at scale. Rather than manually reclassifying objects as they age, you write rules once and S3 handles the transitions. See S3 Lifecycle Policies for how to configure them.

Quick decision guide

• Use S3 Standard when data is actively used: served to users, processed by pipelines, accessed multiple times per month
• Use Intelligent-Tiering when access patterns are genuinely unpredictable, object lifetimes are long, and managing manual lifecycle rules is impractical
• Use Standard-IA when data is accessed rarely but must be available instantly: DR snapshots, backup files, processed reports
• Use One Zone-IA when data is rarely accessed, can tolerate single-AZ risk, and can be regenerated if lost (thumbnails, derivative files)
• Use Glacier Instant Retrieval when data is archived for months or years but might be needed urgently: scanned documents, old media, compliance records that occasionally get pulled
• Use Glacier Flexible Retrieval when data is needed at most a few times per year and a wait of hours is acceptable: annual audits, project archives
• Use Glacier Deep Archive when data must be retained for years (legal, regulatory) and will almost never be accessed
• Use S3 Express One Zone when workloads need maximum throughput and single-digit millisecond latency: ML training data, real-time analytics, high-performance compute

Full storage class comparison

S3 Standard

S3 Standard is the default class. Data is stored across at least three Availability Zones with 99.999999999% durability. There are no retrieval fees and no minimum storage duration.

Use it when: data is actively in use: user uploads being served, files being processed by Lambda functions, datasets fed into analytics pipelines, assets read on every request.

Avoid it when: data sits unread for months. At that point you are paying Standard prices for cold data. A lifecycle rule to transition to Standard-IA after 30–60 days of inactivity can cut that cost substantially.

Real-world example: A SaaS application stores user-uploaded PDFs in S3 Standard. Users access the files frequently in the first few weeks after upload. Standard has no retrieval fees, so it stays cheap under frequent access. A lifecycle rule moves files to Standard-IA after 90 days of no access.

S3 Intelligent-Tiering

Intelligent-Tiering monitors each object’s access pattern and moves it between internal tiers automatically:

• Frequent Access tier: same cost as S3 Standard; used for recently or regularly accessed objects
• Infrequent Access tier: triggered after 30 consecutive days without access; meaningfully lower storage cost
• Archive Instant Access tier: triggered after 90 days without access; lower still, with millisecond retrieval
• Archive Access tier (optional): objects not accessed in 90–180+ days; Glacier-range pricing with asynchronous retrieval
• Deep Archive Access tier (optional): objects not accessed in 180+ days; deepest archive pricing

There are no retrieval fees within Intelligent-Tiering. Instead, AWS charges a small per-object monitoring fee each month. This makes Intelligent-Tiering uneconomical for objects smaller than around 128 KB, or for objects that will be deleted within days of creation, where the monitoring fee exceeds any savings.

Use it when: you have a large bucket of long-lived objects with genuinely unpredictable access patterns. User-uploaded document archives and media libraries are typical candidates. The fee is trivial at scale, and you avoid the manual work of writing lifecycle rules to predict access patterns.

Avoid it when: you already know the access pattern. If logs go cold after 30 days and are deleted after 90, a pair of lifecycle rules is cheaper and simpler than Intelligent-Tiering. See S3 Lifecycle Policies for how to set those up.

Real-world example: A legal firm stores tens of thousands of client documents. Some files get accessed daily during active matters; others are untouched for years. Intelligent-Tiering automatically keeps active files in the frequent tier and moves dormant ones to archive, without any manual effort.

S3 Standard-IA

Standard-IA (Infrequent Access) offers a lower storage cost than Standard in exchange for a per-GB retrieval fee and a 30-day minimum storage duration. Multi-AZ replication is unchanged: still at least three AZs.

Use it when: data is accessed rarely but must be immediately available when requested. Disaster recovery snapshots, monthly billing archives, compliance documents, and long-lived processed datasets are typical use cases. See S3 Security Best Practices for guidance on access controls for these sensitive files.

Avoid it when: data is accessed more than once or twice per month (retrieval fees add up quickly), or when objects are short-lived (the 30-day minimum kicks in).

Real-world example: A company keeps nightly database backups in Standard-IA. Restores happen maybe twice a year, but when needed the restore must start immediately. Standard-IA handles both requirements at lower cost than Standard.

S3 One Zone-IA

One Zone-IA has the same retrieval speed and fee structure as Standard-IA, but stores data in a single Availability Zone. This makes it roughly 20% cheaper than Standard-IA, but an AZ failure results in permanent data loss.

Use it when: the data is reproducible from another source. Thumbnails generated from originals stored in Standard, resized images derived from user uploads, or cached processing outputs are all appropriate.

Avoid it when: the data is irreplaceable. One Zone-IA is not appropriate for database backups, user-uploaded originals, compliance records, or anything else you cannot regenerate.

Real-world example: A media platform stores image thumbnails in One Zone-IA. The originals are in S3 Standard. If the AZ fails and thumbnails are lost, the platform regenerates them from the originals. No data is permanently lost and storage costs are lower.

S3 Glacier Instant Retrieval

Glacier Instant Retrieval delivers archival pricing with millisecond retrieval speed. It has a 90-day minimum storage duration, higher retrieval fees than Standard-IA, and multi-AZ replication.

Use it when: data is archived for months or years but might be pulled urgently. Scanned medical records, old customer documents, historical financial records, and compliance files that occasionally get requested during audits all fit this pattern.

Avoid it when: data is accessed more than once or twice per quarter (the retrieval fee becomes expensive), or when objects might be deleted before 90 days.

Real-world example: A hospital stores patient scans older than two years in Glacier Instant Retrieval. Most scans are never accessed again, but occasionally a doctor needs a scan from five years ago during an appointment. Millisecond retrieval means the workflow is not disrupted.

S3 Glacier Flexible Retrieval

Glacier Flexible Retrieval is cheaper than Glacier Instant Retrieval but adds retrieval delay. You choose from three retrieval modes:

• Expedited: 1–5 minutes (highest retrieval fee)
• Standard: 3–5 hours
• Bulk: 5–12 hours (lowest retrieval fee)

The 90-day minimum storage duration applies. For more detail on Glacier retrieval options, vault locks, and compliance use cases, see AWS Glacier Overview.

Use it when: data is needed at most a few times per year and you can plan the retrieval in advance. Annual audit records, project archives, and historical datasets that analysts occasionally need are typical candidates.

Avoid it when: there is any chance the data will be needed urgently.

Real-world example: A software company keeps previous-year project deliverables in Glacier Flexible Retrieval. Occasionally a client requests a file from a closed project. A standard retrieval is initiated and the file is ready within a few hours.

S3 Glacier Deep Archive

Deep Archive is the lowest-cost storage class AWS offers. Retrieval takes 12–48 hours (standard is 12 hours; bulk is 48 hours), retrieval fees are significant, and the minimum storage duration is 180 days.

Use it when: regulatory or legal requirements mandate long-term retention of data you are highly unlikely to ever retrieve. Seven-year GDPR retention archives, regulatory filing records, and long-term audit trails are common use cases.

Avoid it when: there is any realistic chance you will need the data within the next six months, or if you have not confirmed that the retention requirement actually needs this level of archiving.

Real-world example: A financial institution is required to retain transaction records for seven years. The records are almost never accessed after year one. Glacier Deep Archive keeps costs minimal over the retention period, and the retrieval delay is acceptable for a once-in-seven-years access pattern.

S3 Express One Zone

S3 Express One Zone is a distinct storage class designed for the highest-performance storage workloads. It uses a different bucket type called a directory bucket rather than a general-purpose bucket, and delivers consistent single-digit millisecond latency with higher throughput than standard S3.

It stores data in a single Availability Zone. There are no retrieval fees and no minimum storage duration. Storage cost per GB is higher than Standard, but request costs are lower, which makes it cost-effective for workloads with very high read/write rates.

Use it when: your workload is latency-sensitive and throughput-intensive. The primary use cases are ML training jobs reading large datasets repeatedly, real-time analytics pipelines, and interactive compute workloads where storage latency directly increases EC2 or SageMaker costs.

Avoid it when: multi-AZ resilience matters, you need S3 features like versioning or cross-region replication, or the workload is not actually latency-sensitive. For most standard object storage, S3 Standard is the better fit. Express One Zone is a specialist class for specific performance workloads, not a general upgrade.

Real-world example: A machine learning team reads a 500 GB training dataset hundreds of times per day during model development. Switching from Standard to Express One Zone reduces per-request latency and cuts training time, which directly reduces EC2 compute costs.

S3 Standard vs Intelligent-Tiering

These two classes are often compared because both serve active data with millisecond retrieval and no retrieval fees.

Choose Standard if:

• The data stays hot: accessed frequently and consistently
• Objects are small (under ~128 KB) and the per-object monitoring fee in Intelligent-Tiering would exceed savings
• Objects are short-lived (created and deleted within days)
• You already have lifecycle rules that handle the transition to IA automatically

Choose Intelligent-Tiering if:

• Access patterns are genuinely unpredictable: some objects go cold, others stay active, and you cannot reliably predict which
• Object lifetimes are long (months or years)
• The bucket is large and managing individual lifecycle rules per prefix is unwieldy
• You want zero retrieval fees across all access patterns

The monitoring fee in Intelligent-Tiering is small per object, but across millions of tiny, short-lived objects it can outweigh any storage savings.

When to use each class: real scenarios

For S3 versioning, it is common to leave current versions in Standard and use lifecycle rules to push noncurrent versions to Standard-IA after 30 days and Glacier after 90 days. This alone can cut versioning costs significantly.

For pre-signed URL workflows, keep the objects in Standard or Standard-IA. Pre-signed URLs still incur retrieval fees for IA classes.

Lifecycle policy and cost gotchas

Lifecycle policies can save significant money. Used carelessly, they create surprises on your bill.

Transition fees are sometimes overlooked. Transitioning objects between classes is not free. AWS charges a per-object transition request fee. For millions of tiny objects, these fees can surprise you. Consider setting minimum object size filters in your lifecycle rules to skip objects too small to benefit from transitioning.

Access patterns can fool you. A lifecycle rule that moves everything to IA after 30 days assumes objects go cold after 30 days. If some objects are still accessed at day 45, you will pay retrieval fees on those objects. Intelligent-Tiering handles this case better.

Glacier retrieval costs for unexpected access are real. Glacier is very cheap to store but meaningful to retrieve. If a team assumes they will never need certain archived data and then needs it urgently during an incident, an Expedited retrieval of hundreds of gigabytes can generate a notable charge. Budget for this if there is any chance of unplanned retrieval.

See S3 Lifecycle Policies for a full guide to writing rules, handling versioned buckets, and testing lifecycle configurations before production use.