SRC→CUTOVER→DESTIndependent · No vendor bias · Updated Apr 2026

Tool · AWS Database Migration Service

AWS Database Migration Service (DMS) pricing, 2026

What AWS DMS actually costs across instance classes and the Serverless option. Per-hour replication instance pricing, the six-month free tier for new accounts, storage and egress arithmetic, DMS Serverless economics, and a worked 5 TB SQL Server to RDS scenario.

AWS Database Migration Service (DMS) is the AWS-native database replication tool for cloud migrations. It supports both homogeneous migrations (same engine on source and target, e.g. SQL Server on-premise to RDS for SQL Server) and heterogeneous migrations (different engine on source and target, e.g. Oracle to Aurora PostgreSQL with AWS Schema Conversion Tool). The pricing model is per-replication-instance per-hour, plus storage and standard data transfer. This page is the 2026 pricing reference for DMS specifically.

The DMS pricing model

DMS provisioned pricing uses replication instance classes priced like EC2. Customers provision a replication instance (a specialised EC2 instance with the DMS replication engine installed), connect it to source and target endpoints, configure replication tasks, and pay for the instance plus its storage for the duration of the replication.

DMS provisioned replication instance pricing, US East (May 2026)

Instance class	vCPU	Memory	On-demand $/hour	Use case
dms.t3.micro	2	1 GB	$0.018	Test and small migrations; free tier eligible
dms.t3.small	2	2 GB	$0.036	Small databases under 100 GB
dms.t3.medium	2	4 GB	$0.072	Standard small workloads
dms.t3.large	2	8 GB	$0.145	Mid-size workloads
dms.r5.large	2	16 GB	$0.182	Standard production migrations
dms.r5.xlarge	4	32 GB	$0.365	Larger production migrations
dms.r5.4xlarge	16	128 GB	$1.459	High-throughput, large databases
dms.r5.24xlarge	96	768 GB	$8.757	Very large or high-concurrency
dms.c5.large	2	4 GB	$0.157	CPU-intensive transformations
dms.c5.xlarge	4	8 GB	$0.315	CPU-intensive, larger workloads

On top of the instance cost, DMS replication storage is charged at standard EBS rates (typically gp2 or gp3) for the replication state. Most production DMS migrations use 100 GB to 500 GB of replication storage, costing $11.50 to $57.50 per month at gp2 rates. Data transfer in to AWS is free; data transfer out (for example, replicating into a non-VPC target) is charged at standard egress rates documented on the AWS data transfer pricing page.

DMS Serverless, the alternative billing model

DMS Serverless was made generally available in 2023 and provides DMS replication as a pay-per-use service without instance provisioning. Pricing is per DCU (DMS Capacity Unit) per hour. DCUs scale automatically based on workload demand. The current Serverless rate is approximately $0.094 per DCU-hour. A typical DMS Serverless replication for a 5 TB SQL Server migration uses 4 to 8 DCUs on average, costing roughly $9 to $18 per day of active replication.

DMS Serverless vs provisioned economics. For continuous long-running replication (weeks to months), Serverless typically saves 15 to 35 percent versus equivalent provisioned capacity because DCU scaling tracks actual workload demand rather than provisioning for peak. For short-burst replication (full-load that completes in hours), Serverless typically costs marginally more because of the cold-start scaling behaviour. The crossover point is typically around 7 to 14 days of active replication; longer than that, Serverless wins.

DMS Serverless rule of thumb

For migrations expected to run more than two weeks of active replication (including any CDC parallel-running window), DMS Serverless typically saves 15 to 35 percent versus provisioned. For migrations that complete within a single weekend cutover window, provisioned dms.r5.large or larger is typically cheaper.

The free tier for new AWS accounts

AWS provides 750 hours per month of dms.t2.micro or dms.t3.micro replication instance for the first six months in a new AWS account. The free tier covers continuous use of a single small replication instance for six months, suitable for small databases or for testing larger migrations before production cutover.

For most production migrations the free tier covers the planning and testing phase but not the production cutover. A typical pattern: use the free dms.t3.micro for source schema discovery and small-scale testing, then provision a larger dms.r5.large or dms.c5 for the production full-load and CDC. The free tier still meaningfully reduces the tooling cost for first-time AWS migrations.

Full-load vs CDC vs ongoing replication

DMS supports three replication modes that combine to handle different migration patterns. Full-load copies the source database to the target in a single bulk transfer; suitable for migrations that can accept downtime equal to the full-load duration. Full-load plus CDC starts with a bulk copy then maintains continuous change data capture replication until cutover; suitable for near-zero-downtime migrations. CDC-only assumes a bulk copy via a different tool then replicates ongoing changes; used for very large databases where DMS full-load is not the fastest path.

The cost shape varies by mode. Full-load is the cheapest pattern because the replication instance only runs for the bulk copy duration (typically days). Full-load plus CDC is more expensive because the replication instance runs continuously for the parallel-running window (typically weeks to months). CDC-only is similar to full-load plus CDC in replication instance time but eliminates the full-load DMS cost (the bulk copy is done via another tool, often AWS Snowball Edge for very large databases).

Worked 5 TB SQL Server to RDS scenario

A representative DMS-specific cost build for a 5 TB SQL Server database migration to RDS for SQL Server, using full-load plus CDC with two weeks of parallel-running, dms.r5.large provisioned instance.

Worked DMS cost build, 5 TB SQL Server to RDS, 2 weeks parallel

Cost line	Low estimate	Typical estimate	High estimate
DMS replication instance (dms.r5.large, 14 days)	$60	$62	$70
Replication storage (100 GB gp3, 14 days)	$3.50	$4.00	$5.00
EBS snapshots (replication state)	$0	$5	$15
Data transfer (source to VPC endpoint, free if Direct Connect)	$0	$150	$450
DMS task validation cost (one-off CPU spike)	$0	$5	$20
AWS SCT analysis (free for SQL to SQL; SCT not strictly required)	$0	$0	$0
Cutover-window DMS active (1-2 hours)	$1	$2	$5
DMS total (excluding SCT remediation labour)	$64.50	$228	$565

The typical-column number, $228, illustrates how small the DMS tooling cost is relative to the overall database migration cost. The total cost is dominated by the partner labour ($30K to $100K per database for typical mid-market migrations), the validation and cutover work, and the destination RDS instance cost. DMS is rarely the line that matters in cost modelling; the tool's value is in operational reliability and the elimination of third-party tooling cost.

Heterogeneous migrations and AWS Schema Conversion Tool

For heterogeneous migrations (different engine on source and target), DMS is paired with AWS Schema Conversion Tool (SCT). SCT analyses the source schema, generates the target schema, and produces a conversion report identifying objects that cannot be automatically converted. The remediation work for those objects is the largest cost line in heterogeneous database migrations.

Typical conversion rates by source-target pair (May 2026):

AWS SCT conversion rates by source-target pair

Source	Target	Auto-conversion rate	Typical remediation labour per database
Oracle	Aurora PostgreSQL	70 to 90%	$60K to $300K
Oracle	PostgreSQL	75 to 92%	$50K to $250K
Microsoft SQL Server	Aurora PostgreSQL (via Babelfish)	80 to 95%	$30K to $150K
Microsoft SQL Server	PostgreSQL	75 to 90%	$40K to $200K
Microsoft SQL Server	MySQL	70 to 85%	$40K to $200K
Sybase ASE	PostgreSQL or MySQL	60 to 80%	$80K to $400K
IBM Db2	PostgreSQL	55 to 75%	$100K to $500K
Teradata	Amazon Redshift	65 to 85%	$150K to $800K

How to reduce DMS cost

Use the free tier for testing and small migrations. dms.t3.micro is sufficient for many proof-of-concept and small-database scenarios.
Use DMS Serverless for long-running parallel-replication scenarios. The DCU scaling saves cost on workloads with fluctuating throughput.
Provision the right instance class. Over-provisioning is the most common waste; start small and scale up if replication lag becomes an issue.
Minimise the parallel-running window. Every day of CDC adds replication instance cost plus the underlying destination RDS or Aurora cost.
Use Direct Connect or VPC endpoint for source-target connectivity to eliminate egress cost.
Stop DMS tasks immediately after cutover. Continued replication for already-cutover databases wastes instance time.
Apply MAP credits to DMS consumption. DMS is small relative to overall migration but counts toward MAP cost recovery.

AWS DMS is one of the cheapest tooling line items in any cloud migration. The pricing model favours short, focused replications and DMS Serverless extends the favour to long-running replications with variable load. The cost discipline for DMS is rarely about DMS itself; it is about minimising the parallel-running window (which drives both DMS and destination database cost) and choosing the right replication mode for the downtime tolerance of the workload.

Q&A

Frequently asked

Q. What does AWS DMS cost?

A. A typical DMS replication instance (dms.r5.large) costs roughly $0.18 per hour on-demand, or about $130 per month if continuously provisioned. For a 5 TB database migration with two weeks of replication, the DMS instance cost lands around $60 to $250 depending on instance class. Additional costs: storage on the replication instance (typically 100 GB for replication state at $0.115 per GB-month for gp2), data transfer out from AWS to a non-VPC source (standard egress rates), and any source-side network costs.

Q. Is the first DMS instance free?

A. Yes. AWS provides 750 hours per month free of dms.t2.micro or dms.t3.micro replication instance use for the first six months in a new AWS account. The free tier is intended for testing and small migrations; production migrations typically use larger instance classes (dms.r5.large or above) that fall outside the free tier. For the first migration in a new AWS account the free tier covers most of the tooling cost.

Q. What is DMS Serverless?

A. DMS Serverless was made generally available in 2023 and provides DMS replication on a per-DCU (DMS Capacity Unit) per-hour billing model without the need to provision an instance class. DCUs scale automatically based on workload demand. DMS Serverless typically saves 15 to 35 percent versus provisioned DMS for long-running migrations or migrations with fluctuating throughput. The trade-off is slightly higher peak billing during burst replication.

Q. Does DMS support all source and target databases?

A. DMS supports a broad source list: Oracle, SQL Server, MySQL, MariaDB, PostgreSQL, MongoDB, Db2 LUW, Db2 z/OS, SAP ASE, and Amazon S3. Targets include all major AWS database services (RDS, Aurora, DynamoDB, Neptune, ElasticSearch, S3, Redshift, Kinesis, Kafka). Heterogeneous migrations (Oracle to PostgreSQL) require AWS Schema Conversion Tool (SCT) in addition to DMS. The supported combinations are documented on the AWS DMS supported sources and targets page.

Q. How long does DMS take to migrate a typical database?

A. Initial full-load throughput for DMS varies by source and target combination. Typical full-load throughput is 100 to 500 GB per hour for homogeneous migrations (e.g. SQL Server to RDS for SQL Server) and 50 to 200 GB per hour for heterogeneous migrations (e.g. Oracle to PostgreSQL). A 5 TB database full-load typically completes in 10 to 50 hours depending on source IO capacity and target ingestion rate. Change Data Capture (CDC) replication after full-load runs continuously until cutover.

Q. What is AWS Schema Conversion Tool (SCT) and what does it cost?

A. AWS SCT is a free desktop tool that converts schema from one database engine to another. It is used in conjunction with DMS for heterogeneous migrations. SCT generates the target schema, identifies conversion gaps (objects that cannot be automatically converted), and produces a conversion report. SCT itself is free; the work of remediating conversion gaps is partner-led and typically costs $20K to $100K per database depending on schema complexity.