AWS S3 Pricing: Replication and Intelligent Tiering

An S3 bill can double before anyone notices, especially when a bucket replicates data into another Region. Storage is only the first meter. Requests, inter-Region transfer, retrievals, and per-object automation can all alter the total.

In 2026, the S3 cost model needs a workload-level view of what each object costs across its full lifecycle. S3 object storage is only one meter, while replication creates another footprint. Intelligent-Tiering can lower charges when access patterns are uncertain, but AI and lakehouse workloads need a separate product-level estimate for S3 Vectors.

How AWS S3 pricing works before replication

Amazon S3 bills several independent usage types. A file can generate storage charges, S3 request costs, transfer charges, retrieval charges, and lifecycle charges in the same month. Request and retrieval activity can create different data access charges, and replication adds destination-side usage rather than replacing source-side usage.

Start with the official Amazon S3 price list for the Region pair and storage classes in your design. AWS can change rates, and prices differ across Regions, so a US East estimate is not a universal price card.

For a replicated object, model these cost paths separately:

• The source bucket continues to incur storage, request, retrieval, and lifecycle charges.
• The destination bucket incurs its own storage charges in the selected destination storage class.
• Cross-region replication, or CRR, adds inter-Region data transfer charges.
• Replication actions create request charges, while optional services such as Replication Time Control can add separate fees.

Storage is measured in GB-months and prorated over the month. Therefore, a 1 TB object stored for half a month does not cost the same as 1 TB held for a full month. This matters during migrations, backfills, and disaster recovery testing.

A replicated copy is a separately billed dataset. A lower-cost destination storage class can reduce the storage line item, but it does not remove CRR transfer or replication request charges.

Data classification should come before any storage-class decision. A copy used for active analytics needs low latency. A copy held only for regulatory recovery may tolerate a restore process that takes hours.

Storage classes and the price of access

Amazon S3 storage classes differ in storage pricing, availability, performance, and access costs. Storage pricing alone doesn’t describe the full cost of access, because data access charges can change the better choice.

The following list prices are planning figures for US East (N. Virginia), us-east-1, in USD, as of August 2026. They assume the first storage volume tier and exclude requests, retrievals, replication, and transfer. Prices vary by Region and can change.

S3 Standard storage is built for frequently read data, such as application assets, active media, and data lake inputs. Its $0.023 per GB-month rate is the baseline for many models in us-east-1.

S3 Standard-IA stores less frequently accessed objects at a lower rate, but data access charges can offset savings when reads are frequent. Archive tiers reduce storage rates further, but data retrieval fees and slower access affect the tradeoff. It also has a 30-day minimum storage duration, and AWS bills objects smaller than 128 KB as 128 KB in this class. Millions of small objects can therefore cost more than their logical data size suggests.

S3 One Zone-IA costs less than multi-AZ infrequent-access storage because it stores data in a single Availability Zone. It suits re-creatable backups and secondary copies, but not the only copy of critical data.

S3 Express One Zone targets high-performance workloads requiring single-digit millisecond access in a single Availability Zone. S3 Express One Zone has distinct pricing and directory bucket requirements, so don’t compare it directly with general-purpose classes without checking both.

S3 Glacier Deep Archive costs about 23 times less per GB-month than S3 Standard in this example. However, it suits preservation data rather than interactive workloads. Standard retrieval can take up to 12 hours, while bulk retrieval can take up to 48 hours. Deleting data before 180 days also creates an early deletion charge.

The S3 Intelligent-Tiering storage class starts at the S3 Standard rate for Frequent Access, then reduces the storage price as qualifying objects become inactive. Compared with S3 Standard storage, the tier’s economics depend on object count and access behavior, not storage volume alone.

Replication pricing is a second workload

S3 Replication requires versioning on both source and destination buckets. Same-Region Replication, or SRR, keeps copies in one AWS Region. Cross-region replication sends copies to another Region and adds inter-Region transfer charges.

With SRR, you still pay for destination storage and replication requests. However, there is no inter-Region data transfer because the copy remains in the same Region. SRR can fit account isolation, production-to-test data distribution, or compliance designs that require separate buckets without a second geographic location.

Multi-region access points route requests, but they don’t create a replica or eliminate CRR charges.

CRR adds more cost dimensions. AWS bills storage at the destination, replication PUT requests, requests needed to read source objects for replication, and data transfer across the Region boundary. The transfer price depends on the exact source and destination pair.

S3 Express One Zone has different directory-bucket and versioning considerations. Don’t assume it is a drop-in replication destination without verification.

Replication is asynchronous. If a recovery objective requires a predictable replication time, S3 Replication Time Control adds per-object and per-GB charges. Replication metrics, event notifications, and AWS Key Management Service activity can also add charges.

Server-side encryption with AWS KMS deserves extra review. CRR of SSE-KMS data needs permissions for the destination KMS key and may generate KMS API request costs. Those charges can become visible when an application writes many small encrypted objects.

A new replication rule also doesn’t automatically backfill old versions. S3 Batch Replication can copy existing data through an S3 batch operations job. Its object-operation, request, storage, and transfer charges require a separate migration estimate. Lifecycle rules are configured independently on the destination bucket, so a retained DR copy can remain in S3 Standard long after the source transitions to archive storage.

An AI workload using S3 Vectors may replicate raw objects, rebuild its vector index, or protect both as separate decisions.

A 10 TB cross-region replication example

Assume 10 TB equals 10,000 GB for billing math, split across 10,000 objects. The source is S3 Standard in US East (N. Virginia), us-east-1, and the destination is S3 Standard in US West (Oregon), us-west-2. The data remains stored for a full month in both buckets. This example assumes a one-time initial CRR copy, no retrievals, no KMS, no Replication Time Control, and current list prices as of August 2026. Prices can change.

The ongoing storage cost is about $460 per full month before requests and future changes. The $200 transfer charge belongs to the initial copy in this example. Each newly created or changed replicated object will add new transfer and request charges.

Validate every Region pair, object count, and retention period in the AWS Pricing Calculator before approving a budget. A model that assumes 10 TB of large objects can be far too low for a workload with hundreds of millions of small files.

S3 Intelligent-Tiering costs and behavior

S3 Intelligent-Tiering is best for data with uncertain or changing access patterns. AWS places eligible objects in the Frequent Access tier first. After 30 consecutive days without access, an object can move to Infrequent Access. After 90 days without access, it can move to Archive Instant Access.

Those first three tiers provide millisecond access. AWS does not charge for automatic tier movement, and accessing a lower automatic tier returns the object to Frequent Access. Optional archive tiers have different data access charges because they use asynchronous restoration.

Teams can also enable two optional asynchronous archive tiers. Archive Access can begin after a configured inactive period of at least 90 days. Deep Archive Access can begin after at least 180 days. Deep Archive Access is comparable to S3 Glacier Deep Archive for preservation, but access behavior differs. Both tiers require restoration before an application can read the object, and data retrieval fees may apply.

AWS charges a monitoring and automation fee of $0.0025 per 1,000 eligible objects each month in the us-east-1 planning example. Objects with a billable object size below 128 KB stay in Frequent Access and do not incur that monitoring fee.

The AWS Intelligent-Tiering configuration guidance is useful when setting archive-tier activation periods. Choose those periods based on recovery requirements, not only the lowest storage rate.

A storage example with Intelligent-Tiering

This example uses US East (N. Virginia), us-east-1, August 2026 USD list prices, 10,000 GB of data, and 100,000 eligible objects. It assumes the month’s average distribution is 30% in Frequent Access, 50% in Infrequent Access, and 20% in Archive Instant Access. It reports monitoring separately from storage, while excluding request, retrieval, transfer, replication, and optional archive-tier costs. Prices can change.

The same 10,000 GB held entirely in S3 Standard would cost $230 per month under the same Region and list-price assumptions. Actual savings depend on how long objects remain inactive. New data does not arrive pre-qualified for an archive tier.

Object count can change the answer. At the same us-east-1 monitoring rate, 100 million eligible objects create a $250 monthly monitoring charge before storage. An independent 2026 Intelligent-Tiering cost analysis also highlights this per-object effect.

The class is a poor fit for objects that expire quickly or have fixed, predictable retention. A lifecycle policy can move known cold data directly to Standard-IA or Glacier classes without a recurring monitoring charge, but compare their minimum storage duration requirements. Use S3 Intelligent-Tiering when access history is unclear and keeping access fast matters.

Egress, requests, and lifecycle rules can change the bill

Internet egress can outweigh storage for data-heavy applications. For a planning example in US East (N. Virginia), us-east-1, assume 1 TB, or 1,024 GB, of data transfer to the public internet during one month. Assume the account has not used its aggregate 100 GB monthly AWS Free Tier allowance for internet egress and receives no private pricing discount. At the $0.09 per GB tier, the remaining 924 GB costs about $83.16. This allowance doesn’t cover all internet egress and is separate from destination, volume-tier, and AWS agreement-specific pricing, which can change.

Traffic between S3 and Amazon EC2 in the same Region usually avoids S3 internet egress because this data transfer stays on AWS’s regional path. However, an EC2 workload in a private subnet can still create AWS NAT Gateway data-processing charges if it reaches S3 through a NAT Gateway. A gateway VPC endpoint for S3 can remove that avoidable network path, but it doesn’t eliminate every possible network charge.

For long-distance uploads, S3 transfer acceleration has separate pricing and doesn’t eliminate ordinary download or egress costs.

S3 request costs matter most in high-object-count designs. Image rendering, log ingestion, data lake manifests, and ML training jobs can generate large volumes of LIST, GET, PUT, COPY, and HEAD requests. Track request activity and egress beside GB-months when estimating data access charges at the workload level.

Lifecycle rules deserve the same scrutiny. Each transition can generate a request charge, and transitions into infrequent-access or archive classes can create minimum storage duration obligations. Moving a 10-day-old object into Standard-IA and deleting it soon afterward may save little or nothing.

AI storage adds meters beyond S3 object storage

AI and lakehouse workloads need a separate cost model. S3 Tables pricing, S3 Vectors, and S3 Metadata introduce usage categories that do not map neatly to standard object storage or replication estimates.

S3 Tables can generate charges for table storage, object monitoring, requests, and compaction data processed. S3 Tables pricing needs separate forecast lines for each meter. Compaction reduces small-file overhead in Apache Iceberg tables, yet frequent writes can increase the bytes AWS processes for maintenance. Treat compaction as a workload cost, not a background detail.

S3 Vectors adds costs for vector storage, vector writes, and vector query activity. With S3 Vectors, a retrieval-augmented generation system may store modest source documents while issuing far more vector queries. Raw-object storage savings won’t offset an unbounded query pattern.

S3 Vectors cost model

• S3 Vectors storage: Forecast the vector index separately from source objects. Growth follows embedding volume and retained versions.
• S3 Vectors writes and index updates: Count ingestion, upserts, deletes, and maintenance. Frequent changes can make index processing exceed initial load estimates.
• S3 Vectors queries: Model requests by query count and access pattern. Interactive applications can create far more activity than batch workloads.

S3 Metadata adds its own object monitoring, journal table storage, and journal update charges. High-churn buckets can create frequent journal activity even when retained data volume remains flat.

Keep raw source data, derived embeddings, table files, and metadata journals as separate forecast lines. If a second Region is required, decide whether S3 Vectors needs raw-data replication, a rebuilt table, a second vector index, or all three. Those choices have different storage, data transfer, and processing costs. Forecast S3 Vectors separately, including write, query, and index-maintenance assumptions.

A practical FinOps model for replication and tiering

Finance and engineering teams should review source and destination buckets as separate products, with tags for workload and retention. A bucket tagged “backup” may be a low-cost archive copy, a hot replica for analytics, or an expensive CRR target that never transitions.

Use S3 Inventory to measure object size, object count, age, version count, encryption status, and storage class. Classify high-performance, single-AZ workloads in S3 Express One Zone separately, and forecast S3 Vectors indexes as separate products. S3 Storage Lens adds estate-level visibility. Review S3 Storage Lens trends for growing object counts and stale versions.

Then compare that inventory with Cost and Usage Report data for requests, retrievals, data transfer, and data access charges. For S3 Vectors, compare query and write activity with raw-object storage before assigning shared costs.

Apply S3 lifecycle policies to known retention patterns, checking the minimum storage duration before transitions or early deletion. Use S3 Glacier Deep Archive for long-retention copies, and put uncertain datasets into Intelligent-Tiering. Monitor savings and monthly object fees; configure destination lifecycle rules separately because the source policy does not manage the replica.

Finally, test the model against a limited bucket or prefix before changing a petabyte-scale estate. Use S3 Storage Lens to validate object and version counts, storage growth, and the forecast in AWS Pricing Calculator. The S3 cost model improves when a forecast includes storage volume, access behavior, and copy location, supporting cloud cost optimization.

Frequently Asked Questions

Does S3 replication double storage costs?

A replicated object is billed as separate storage in the source and destination buckets, so storage costs can effectively double. Cross-Region Replication also adds inter-Region data transfer and replication request charges.

Is S3 Intelligent-Tiering always cheaper than S3 Standard?

No. Intelligent-Tiering can reduce storage costs for objects that become inactive, but it adds a per-object monitoring and automation fee. It may be a poor fit for short-lived objects, very small objects, or data with predictable retention.

What charges should be included in an S3 replication estimate?

Include source and destination storage, replication requests, source read requests, and inter-Region data transfer for CRR. Also review retrievals, KMS requests, Replication Time Control, lifecycle transitions, and S3 Batch Replication when those features apply.

How can teams reduce unexpected S3 costs?

Use S3 Inventory and Cost and Usage Report data to compare object count, versions, storage classes, requests, retrievals, and transfer. Apply lifecycle rules to known retention patterns, configure destination policies separately, and validate the forecast with a limited bucket or prefix before making large changes.

Final cost perspective

Replication buys a second copy, so its budget must include destination storage and data transfer. Intelligent-Tiering may reduce storage costs as access changes, but it adds monitoring costs and makes access patterns part of the decision.

The strongest cost model treats every replicated dataset as its own lifecycle. With clear retention rules, AWS S3 pricing stays predictable, and the same discipline extends to AI data products such as S3 Vectors.