S3 storage class comparison: when each tier wins

S3 has six storage classes plus a high-performance Express One Zone option, each with different storage cost, retrieval fees, request fees, and minimum storage duration. Picking the right class for each workload can cut S3 costs by 50-95%. Lifecycle policies that automatically transition objects between classes amplify the savings further.

This post walks through each class with concrete numbers, retrieval economics, and the typical workloads each one wins for.

The full pricing table

Storage cost per GB-month in us-east-1 (other regions are within 10-30%):

• Standard: $0.023/GB-month
• Intelligent-Tiering Frequent Access: $0.023/GB-month
• Intelligent-Tiering Infrequent Access: $0.0125/GB-month
• Intelligent-Tiering Archive Instant Access: $0.004/GB-month
• Standard-IA: $0.0125/GB-month
• One Zone-IA: $0.01/GB-month
• Glacier Instant Retrieval: $0.004/GB-month
• Glacier Flexible Retrieval: $0.0036/GB-month
• Glacier Deep Archive: $0.00099/GB-month

The cheapest (Deep Archive) is roughly 99x cheaper than the most expensive (Standard) per GB. But retrieval fees, minimum storage durations, and request costs change the picture significantly.

Standard

The default. Right for data accessed multiple times per month. Web assets, transaction logs, hot data lake tables, application state, user-generated content with active access.

No retrieval fees. Higher request fees than archive classes but lower than other warm tiers. No minimum storage duration. Reads and writes are both cheap relatively.

Intelligent-Tiering

Auto-tiering. AWS monitors access patterns and moves objects between tiers. Standard pricing for actively accessed objects, cheaper for inactive.

Costs: $0.0025 per 1,000 objects per month for monitoring (rarely meaningful even for buckets with millions of objects), plus the per-tier storage rate. No retrieval fees from the Frequent and Infrequent tiers. Archive Instant Access tier has a small retrieval fee.

Right when: access patterns are unpredictable or unknown. For well-understood workloads, manually picking the tier saves a little vs Intelligent-Tiering. For everything else, Intelligent- Tiering removes the decision burden.

Standard-IA

Single-region replicated, infrequent access. 45% cheaper storage than Standard. $0.01/GB read fee. 30-day minimum storage.

Right when: you know the access pattern is <1 access per month per object, retrieval times need to be immediate (no warm-up), and you don't want Intelligent-Tiering's monitoring fee.

Common use cases: long-tail user content, backup files accessed rarely for compliance, audit logs.

One Zone-IA

Single-AZ replicated only. 20% cheaper than Standard-IA. Same retrieval fees and minimum.

Right when: you have multi-AZ durability handled at the application layer (e.g., processing pipelines that can regenerate the data) or the data isn't critical enough to need 99.99% cross-AZ durability.

Wrong when: data is irreplaceable. The AZ-level risk is real if AWS has a regional incident.

Glacier Instant Retrieval

Cold storage with millisecond retrieval. About 82% cheaper than Standard. 90-day minimum storage. $0.03/GB retrieval.

Right when: you have data you'll access less than once a quarter but need fast access when you do (compliance queries, intermittent analytical queries).

Glacier Flexible Retrieval

Cold storage with minutes-to-hours retrieval. About 85% cheaper than Standard. 90-day minimum storage. $0.01-$0.10/GB retrieval depending on Expedited (1-5 minutes), Standard (3-5 hours), or Bulk (5-12 hours).

Right when: data is rarely accessed and you can tolerate retrieval delays. Backups, archived application data, training datasets that occasionally need to be rerun.

Glacier Deep Archive

Cheapest tier. About 96% cheaper than Standard. 180-day minimum storage. $0.02-$0.025/GB retrieval. 12-hour retrieval time.

Right when: data is for long-term compliance, regulatory retention, or genuinely never-touched archives. Don't put data here that might need to be retrieved this week.

S3 Express One Zone

High-performance single-AZ class. Higher storage cost than Standard ($0.16/GB-month) but dramatically cheaper requests and single-digit-millisecond latency.

Right when: data is hot and request volume is high. Machine learning training datasets, real-time analytics scratch space, request-heavy workloads where Standard's request fees dominate.

The lifecycle policy pattern

Most production workloads benefit from automatic tier transitions via lifecycle policies. A common pattern for log/backup data:

resource "aws_s3_bucket_lifecycle_configuration" "logs" {
  bucket = aws_s3_bucket.logs.id

  rule {
    id     = "tier-down-old-logs"
    status = "Enabled"

    transition {
      days          = 30
      storage_class = "STANDARD_IA"
    }

    transition {
      days          = 90
      storage_class = "GLACIER"
    }

    transition {
      days          = 365
      storage_class = "DEEP_ARCHIVE"
    }

    expiration {
      days = 2555  # 7 years
    }
  }
}

Effect for 100 TB of cumulative log data after 5 years:

• Without lifecycle: 100 TB × $0.023 × 12 = $27,600/year
• With lifecycle (steady state): about 10 TB in Standard, 20 TB in Standard-IA, 50 TB in Glacier, 20 TB in Deep Archive
• Effective monthly: 10×$0.023 + 20×$0.0125 + 50×$0.0036 + 20×$0.00099 = $0.69/TB-month avg
• Annual cost: ~$830/year

Savings: 97% on storage. Implementation cost: 20 lines of Terraform.

What c3x estimates

c3x reads lifecycle policy configurations and applies expected storage tier distribution based on age. Specify expected average object age in c3x-usage.yml for the most accurate estimate:

# c3x-usage.yml
resource_usage:
  aws_s3_bucket.logs:
    standard_storage_gb: 10240   # 10 TB Standard
    standard_ia_storage_gb: 20480
    glacier_storage_gb: 51200
    deep_archive_storage_gb: 20480

c3x then estimates each tier's cost separately. For the full S3 pricing reference, see the aws_s3_bucket catalog page.

FAQ

The decision framework

For most workloads, the picking algorithm is:

1. If access pattern is unknown or unpredictable: Intelligent-Tiering.
2. If access is multiple times per month: Standard.
3. If access is <1/month but needs instant retrieval: Standard-IA (or One Zone-IA if multi-AZ durability isn't required).
4. If access is <1/quarter but needs ms retrieval: Glacier Instant Retrieval.
5. If access is rare and minutes-to-hours retrieval is fine: Glacier Flexible Retrieval.
6. If access is essentially never (compliance archive): Glacier Deep Archive.

Combine with a lifecycle policy that ages data through these tiers automatically. Don't manually manage transitions; let AWS do it on a schedule.

For per-resource pricing reference including all storage classes, see the aws_s3_bucket catalog page. For the broader S3 cost picture including request fees and cross-region replication, see our post on NAT alternatives which covers VPC endpoints (the free S3 Gateway endpoint is one of the highest-ROI changes available).