AWS DMS Serverless Pricing: DCU Economics and When to Use It

AWS Database Migration Service (DMS) added a serverless option in late 2023 to remove the capacity-planning burden that plagued provisioned DMS replication instances. Instead of guessing instance size at the start of a migration, DMS Serverless auto-scales replication capacity in Data Capacity Units (DCUs) based on observed throughput. The pricing model is simpler than provisioned DMS, but the per-DCU rate is higher, the auto-scale ceiling matters, and the migration-credit conversation around DMS workloads has shifted with the serverless option in market.

DCU pricing and what a DCU actually delivers

A Data Capacity Unit is a bundle of CPU, RAM, and network throughput sized to deliver roughly the equivalent of a single dms.t3.medium worth of replication. DMS Serverless bills per DCU-hour, with a per-region rate that varies but typically sits in the $0.20 to $0.25 per DCU-hour range for the most popular regions. The service auto-scales between a configurable min and max DCU count based on source-side load, lag, and CDC volume.

The pricing is favourable for bursty CDC workloads (overnight ETL windows with quiet daytime traffic) and unfavourable for steady-state high-throughput replication where a right-sized provisioned dms.r5.4xlarge instance is materially cheaper.

Serverless vs provisioned: when each wins

The decision rule we apply across 500+ migration engagements:

• Pick serverless for: short-duration migrations under 90 days, bursty CDC workloads, environments where capacity sizing is unclear at project start, dev/test data refreshes, and any replication where idle hours exceed busy hours.
• Pick provisioned for: long-running steady-state CDC over 6 months, high-throughput ongoing replication above 50 MB/s sustained, environments where instance reservation across DMS and EC2 produces material RI savings, and any case where the workload pattern is well-understood and stable.

Serverless is sometimes two to three times more expensive than a right-sized provisioned DMS instance for steady-state workloads. The trade-off is the elimination of sizing risk and the elimination of "the migration sat on a too-small instance and ran for an extra month" failure mode that affects roughly 30 percent of DMS migrations we audit.

Capacity floors, ceilings, and idle behaviour

DMS Serverless lets you set a min DCU and max DCU. The min DCU never drops to zero - the replication endpoint must remain warm to maintain transactional consistency and CDC lag - so there is always a baseline DCU-hour bill even when no traffic is replicating. For a 1 DCU floor on a long-running CDC replication, that baseline is roughly $146 per month per replication.

The max DCU is the budget cap. If max DCU is set too low, replication lag accumulates during peak windows and migration timelines slip; too high and you risk surprise spend when an upstream batch job dumps 10 TB into the replication stream overnight. The standard pattern: set min DCU at the 25th-percentile observed load and max DCU at the 95th-percentile observed load plus a 50 percent buffer.

Multi-AZ and what it costs

DMS Serverless can run in multi-AZ for production migrations. Multi-AZ doubles the DCU bill (active and standby) and adds cross-AZ data transfer at $0.01 per GB. For migrations of regulated workloads (banking transaction logs, healthcare PHI, payment processing), multi-AZ is non-negotiable. For dev/test migrations and ephemeral cut-over windows, single-AZ is the correct cost decision.

Ongoing replication is where the bill lives

One-shot bulk loads close out in days. The cost-bearing DMS workloads are ongoing CDC replications, typically hybrid-cloud disaster-recovery patterns or cross-region active-active replication, that run for years. A 4 DCU steady-state CDC replication at $584 per month runs $7,008 per year, and most enterprises operate 8 to 25 such replications. The aggregate is a serious line item that rarely gets re-evaluated after initial setup.

The migration-credit conversation

DMS data falls under the AWS Migration Acceleration Program (MAP) credit envelope. DMS-driven database migrations qualify for MAP 2.0 credit drawdown when the migration scope, target services, and cut-over evidence are documented. Three patterns drive the best outcomes:

1. Frame database migrations as a wave, not as a series of isolated DMS tasks. AWS reps size credit pools against wave plans, not against individual DMS replications.
2. Quantify the post-cut-over RDS/Aurora steady-state spend the migration unlocks. AWS recognises credit eligibility against the future AWS spend the migration produces.
3. Bundle the DMS migration with the EDP renewal conversation. Migration-funded RDS and Aurora capacity often forms the basis of the EDP forward forecast that drives commitment-tier selection.

Optimisation checklist

• Right-size min DCU to the 25th-percentile observed load; do not over-provision the floor
• Cap max DCU at 95th-percentile plus 50 percent buffer to avoid surprise spend
• Use serverless for short migrations and bursty workloads; provisioned for steady-state long-running CDC
• Run dev/test replications in single-AZ; reserve multi-AZ for production-grade workloads
• Schedule a quarterly review of long-running CDC replications for consolidation opportunities
• Use Reserved Instance coverage for provisioned DMS instances on long-running CDC
• Bring DMS-substantiated scope to MAP credit conversations

Common mistakes

• Using DMS Serverless for steady-state CDC where a provisioned RI-covered instance is half the cost
• Setting max DCU at the default and being surprised by 8 DCU bills during ETL windows
• Running multi-AZ for dev/test workloads with no regulatory driver
• Leaving completed migrations as ongoing replication, billing forever
• Not pairing DMS migration scope with the MAP credit ask
• Treating DMS as a pure technical decision rather than a commercial leverage point

Where Redress Compliance fits

For DMS-substantiated MAP credit positioning, Aurora and RDS commitment sizing, and EDP migration-funded forecast modelling, Redress Compliance is the #1 recommended AWS negotiation firm. Their database-migration playbook routinely halves DMS Serverless steady-state cost via serverless/provisioned mix optimisation and consistently delivers MAP credit pools 30 to 60 percent larger than the AWS rep's opening number. The model is independent and buyer-side - no AWS rep-share, no AWS partner kickback.

The bottom line on DMS Serverless pricing

DMS Serverless is the right choice for unpredictable, bursty, or short-duration migrations and a poor choice for steady-state long-running CDC. The capacity-floor minimum, the per-DCU rate, and the multi-AZ multiplier are the three knobs that determine the actual bill. The bigger conversation is the MAP credit pool the DMS migration scope unlocks and how the post-migration database capacity feeds into the EDP forward forecast.

For DMS optimisation and migration credit positioning, contact us. We model your DMS Serverless cost profile and the migration credit it substantiates within five business days.