Every database migration project starts with the same fork in the road: which managed migration service actually moves your data without weeks of downtime or a surprise bill. AWS Database Migration Service (DMS), Azure Database Migration Service, and Google Cloud Database Migration Service all promise the same thing – near-zero-downtime replication from your old database to a new one – but the pricing, supported engines, and even the survival of the tool itself differ sharply heading into the back half of 2026. Microsoft retired the classic Azure DMS tool for SQL Server scenarios on March 15, 2026, and killed off Azure Data Studio’s SQL migration extension two weeks before that, forcing thousands of teams to rethink their migration tooling mid-project.
Meanwhile AWS folded DMS into its new Database Savings Plans in December 2025 and keeps expanding DMS Serverless, and Google has spent 2026 pushing Oracle and SQL Server customers toward AlloyDB with an aggressive free tier on homogeneous migrations. This comparison breaks down the pricing models, the supported source-and-target combinations, what actually happened with the Azure retirement, and which service fits which migration scenario – using only the numbers each vendor has published, not the marketing copy layered on top.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
AWS DMS vs Azure DMS vs Google DMS: Quick Verdict
If you want the short answer before the full breakdown: choose AWS DMS if you need the broadest source-engine coverage (Oracle, SQL Server, PostgreSQL, MySQL, MongoDB, MariaDB, and more) or you’re already running Database Savings Plans on AWS. Choose Google Cloud Database Migration Service if your destination is Cloud SQL or AlloyDB for PostgreSQL – homogeneous migrations are free of any DMS charge, and even heterogeneous Oracle-to-AlloyDB jobs come with 500 GiB of free backfill every month. Choose Azure Database Migration Service only if you’re migrating MySQL, PostgreSQL, or MongoDB workloads – the classic SQL Server migration path through Azure DMS and Azure Data Studio no longer exists as of March 2026, and Microsoft has redirected that traffic to newer tooling built around Visual Studio Code.
The pricing gap is where this gets interesting. Google’s per-GiB heterogeneous pricing starts at $2.00/GiB for change data capture (CDC) on the first 2,500 GiB processed monthly, then steps down to $0.80/GiB above 10,000 GiB. AWS bills replication instances by the hour (standard EC2-equivalent rates) or by DCU-hour under DMS Serverless, with Schema Conversion thrown in free. Azure’s Standard tier is free outright, but its Premium tier displays no public hourly rate on the pricing page beyond the six-month free window – a level of pricing opacity the other two clouds don’t share.
What Are AWS DMS, Azure DMS, and Google Cloud DMS?
AWS Database Migration Service is Amazon’s managed replication engine for moving databases into or between AWS data stores. It supports both homogeneous migrations (same engine on both ends, like PostgreSQL to PostgreSQL) and heterogeneous migrations (different engines, like Oracle to Aurora PostgreSQL), using change data capture to keep the source and target in sync until cutover. AWS DMS ships in two billing modes: provisioned replication instances billed hourly, and DMS Serverless, which is metered in DMS Capacity Units (DCUs), where 1 DCU equals 2 GB of RAM and jobs can scale from 1 DCU up to 384 DCUs automatically.
Azure Database Migration Service is Microsoft’s equivalent tool, historically split into a Standard tier (free, 1/2/4 vCore) and a Premium tier (billed hourly after an initial free window). Azure DMS supports SQL Server, PostgreSQL, MySQL, and MongoDB migration scenarios – but the SQL Server-specific path through the classic DMS service and its Azure Data Studio extension officially retired in March 2026, a change that reshapes how Microsoft wants customers moving SQL Server workloads going forward.
Google Cloud Database Migration Service takes a different pricing philosophy entirely: it’s free for homogeneous migrations into Cloud SQL or AlloyDB for PostgreSQL, and only charges for heterogeneous jobs (Oracle or SQL Server moving into a different target engine), billed per GiB of data processed during backfill and ongoing CDC replication. Google has spent 2026 widening the heterogeneous path specifically to push Oracle and SQL Server customers onto AlloyDB for PostgreSQL, now supporting PostgreSQL 16 and 17 as migration targets.
Full Specs Comparison Table
Here’s how the three services stack up across the specs that actually determine which one fits a given migration project.
| Feature | AWS DMS | Azure Database Migration Service | Google Cloud DMS |
|---|---|---|---|
| Pricing model | Hourly replication instance or DCU-hour (Serverless) | Free (Standard tier); hourly vCore billing (Premium, rate not publicly listed) | Free (homogeneous); per-GiB backfill + CDC (heterogeneous) |
| Free tier | Free-plan credits for accounts created after July 15, 2025 | Standard tier free; Premium 4-vCore free for first 183 days | Homogeneous migrations always free; 500 GiB/month free backfill on heterogeneous |
| Homogeneous migrations | Supported (e.g., PostgreSQL to PostgreSQL, MySQL to MySQL) | Supported (SQL Server, PostgreSQL, MySQL, MongoDB) | Supported to Cloud SQL (MySQL/PostgreSQL/SQL Server) and AlloyDB for PostgreSQL |
| Heterogeneous migrations | Supported across Oracle, SQL Server, MySQL, PostgreSQL, MongoDB, MariaDB | Limited; classic SQL Server heterogeneous path retired March 15, 2026 | Oracle → Cloud SQL/AlloyDB for PostgreSQL; SQL Server → Cloud SQL/AlloyDB for PostgreSQL |
| Schema conversion tool | DMS Schema Conversion – free to use, pay only for storage | Delegated to separate tooling since Azure Data Studio’s SQL Migration extension retired | Built into the heterogeneous migration workflow |
| Change data capture (CDC) | Yes, continuous replication until cutover | Yes, on supported engine scenarios | Yes, metered separately from backfill on heterogeneous jobs |
| Serverless option | DMS Serverless (1-384 DCUs, autoscaling) | Not offered as a distinct serverless SKU | Fully managed, no instance provisioning required for either migration type |
| Newest 2025-2026 addition | DMS included in AWS Database Savings Plans (Dec. 2025) | Classic DMS SQL scenarios retired; Azure Data Studio retired Feb. 28, 2026 | PostgreSQL 16/17 support for heterogeneous targets (2026 release notes) |
| Primary target engines | RDS, Aurora, Redshift, DocumentDB, and self-managed EC2 databases | Azure SQL Database/Managed Instance, Azure Database for MySQL/PostgreSQL, Cosmos DB (Mongo API) | Cloud SQL (MySQL/PostgreSQL/SQL Server), AlloyDB for PostgreSQL |
| Recommended tooling path for SQL Server (2026) | DMS + Schema Conversion | Visual Studio Code with MSSQL extension (post-retirement) | SQL Server → AlloyDB for PostgreSQL heterogeneous path |
| Cloud ecosystem lock-in | Tightest coupling to AWS-native targets (Aurora, RDS) | Tightest coupling to Azure SQL family | Tightest coupling to Cloud SQL / AlloyDB |
| Cost predictability | Predictable hourly/DCU billing, no per-GB data fee | Free tier predictable; Premium pricing opaque beyond free window | Fully predictable – free or flat per-GiB tiers published on the pricing page |
Pricing Compared: Instance Hours, DCUs, and Per-GiB Billing
The three vendors don’t just charge different amounts – they charge for fundamentally different units, which makes an apples-to-apples cost comparison harder than it should be. AWS bills by compute time (either a provisioned replication instance at standard hourly rates, or DCU-hours under DMS Serverless). Azure bills by provisioned vCore-hour on its Premium tier, with the Standard tier free outright. Google bills by data volume processed, not compute time, for anything that isn’t a pure homogeneous migration.
| Service / Tier | Billing unit | Free allowance | Paid rate |
|---|---|---|---|
| AWS DMS (provisioned instance) | Per instance-hour | Free-plan credits for post-July 2025 accounts | Standard on-demand instance-hour rate (varies by instance class and region) |
| AWS DMS Serverless | Per DCU-hour | Included in free-plan credits where applicable | Scales 1-384 DCUs; 1 DCU = 2 GB RAM |
| AWS DMS Schema Conversion | Storage only | Tool itself is free | Standard S3/storage rates for conversion artifacts |
| Azure DMS Standard | Per vCore (1/2/4) | Always free | $0 – no charge at any usage level |
| Azure DMS Premium (4 vCore) | Per vCore-hour | Free for 183 days from service creation | Not publicly listed as a flat figure past the free window |
| Google DMS homogeneous | N/A – no DMS charge | Unlimited (only target instance costs apply) | $0 for the migration service itself |
| Google DMS heterogeneous – backfill | Per GiB | First 500 GiB/month free | $0.40/GiB above 500 GiB |
| Google DMS heterogeneous – CDC | Per GiB/month | None | $2.00/GiB (0-2,500 GiB), stepping down to $0.80/GiB above 10,000 GiB |
For a mid-sized heterogeneous migration – say, an Oracle database with 3 TB of initial backfill and ongoing CDC replication of a few hundred GB per month during a multi-week cutover window – Google’s model means the first 500 GiB of that backfill is free, with the remaining roughly 2.5 TB billed at $0.40/GiB. That’s a meaningfully different cost structure than AWS, where the same job is priced purely on how long a replication instance (or how many DCU-hours) runs, regardless of data volume. Teams moving small databases slowly will find AWS’s compute-hour model cheaper; teams moving large databases quickly will often find Google’s per-GiB model more predictable, since it doesn’t punish you for running a bigger instance to finish faster.
Azure’s Standard tier being free is the most generous headline number of the three, but it only covers vCore sizes up to 4, and Microsoft’s own pricing page for Database Migration Service (classic) lists Premium 4-vCore pricing as a placeholder rather than a number, which is unusual transparency-wise compared to the line-item pricing AWS and Google publish. For teams doing FinOps planning around a migration project, that gap matters – you can’t budget a number that isn’t published. For a deeper look at how the three clouds handle cost visibility more broadly, see our comparison of AWS Cost Explorer vs Azure Cost Management vs Google Cloud Billing.
Total Cost of Ownership: A Worked 5 TB Migration Example
Abstract pricing tables are useful for a quick scan, but they hide how differently these three billing models behave once you plug in a real workload. Take a common scenario: a company needs to move a 5 TB Oracle database into a PostgreSQL-compatible target, with an expected six-week cutover window and roughly 1 TB of ongoing change data generated during that period before the final switchover.
On Google Cloud DMS, that 5 TB (roughly 5,120 GiB) backfill gets its first 500 GiB free, leaving about 4,620 GiB billed at $0.40/GiB – a backfill cost in the range of $1,848. The 1 TB (1,024 GiB) of CDC traffic during the six-week window falls entirely inside the first pricing tier, at $2.00/GiB, putting CDC costs around $2,048. Total exposure for the migration service itself lands near $3,900, on top of whatever the destination AlloyDB or Cloud SQL instance costs to run during the same window.
On AWS DMS, the same job is priced entirely differently – by compute time, not data volume. A team running DMS Serverless at a mid-size DCU allocation for six weeks pays for the DCU-hours consumed regardless of whether the underlying database is 500 GB or 50 TB, which means the exact dollar cost depends heavily on how much compute capacity the migration actually needs to sustain the CDC replication lag, not on the raw byte count. That makes AWS DMS harder to budget precisely in advance for very large one-time jobs, but it also means a team that finishes early pays for fewer hours than one that badly underestimates the timeline.
On Azure DMS, this specific scenario (Oracle heterogeneous migration) doesn’t map cleanly onto the current tooling at all – the classic heterogeneous SQL Server paths that once ran through Azure DMS were the ones affected by the March 2026 retirement, and Oracle-source migrations were never Azure DMS’s core use case to begin with. Teams in this exact scenario on Azure would typically look outside DMS entirely, which is itself a data point worth weighing: Azure’s tool is the most cost-effective of the three for its supported scenarios, but it also covers the narrowest slice of migration scenarios overall.
Supported Database Engines and Migration Paths
Engine support is where AWS DMS still leads by a wide margin. AWS documents support for nearly all of today’s popular DBMS engines as source endpoints, covering Oracle, SQL Server, PostgreSQL, MySQL, MongoDB, and MariaDB, with target flexibility spanning RDS, Aurora, Redshift, DocumentDB, and self-managed databases on EC2. That breadth is why AWS DMS remains the default choice for enterprises running a mixed fleet of legacy engines they need consolidated onto fewer platforms.
Google’s supported paths are narrower but deliberately focused. Homogeneous migrations land in Cloud SQL for MySQL, Cloud SQL for PostgreSQL, Cloud SQL for SQL Server, or AlloyDB for PostgreSQL. Heterogeneous migrations – the ones that actually cost money – are limited to four specific combinations: Oracle to Cloud SQL for PostgreSQL, Oracle to AlloyDB for PostgreSQL, SQL Server to Cloud SQL for PostgreSQL, and SQL Server to AlloyDB for PostgreSQL. In 2026, Google’s release notes confirmed heterogeneous migrations now support PostgreSQL 16 and 17 as target versions, closing a gap that had left Google DMS a version or two behind community PostgreSQL releases.
Azure’s supported scenarios cover SQL Server, PostgreSQL, MySQL, and MongoDB – but the SQL Server heterogeneous and homogeneous paths that used to run through the classic DMS engine and the Azure SQL Migration extension in Azure Data Studio are the ones caught up in the March 2026 retirement. Microsoft has been explicit that MySQL, PostgreSQL, and MongoDB scenarios were out of scope for that retirement notice, meaning teams on those engines can keep using Azure DMS largely unaffected. SQL Server teams are the ones who need a new plan.
The Azure Retirement Shake-Up: What Actually Changed in 2026
This is the story that makes 2026 a genuinely different year for this comparison than any prior one. Microsoft retired Azure Data Studio entirely on February 28, 2026, and with it went the Azure SQL Migration extension that many teams relied on to drive SQL Server-to-Azure migrations through a GUI. Two weeks later, on March 15, 2026, Microsoft retired Database Migration Service (classic) for SQL scenarios specifically – the underlying service that extension talked to.
Microsoft’s guidance, published on its official documentation, points displaced users toward Visual Studio Code with the MSSQL extension and other dedicated SQL migration tooling rather than a like-for-like DMS replacement. That’s a meaningful workflow change for any team that had automated part of its migration pipeline around Azure Data Studio’s assessment and migration wizards – those wizards simply don’t exist anymore for SQL Server scenarios.
Two details temper how disruptive this actually is. First, the retirement is scoped narrowly: MySQL, PostgreSQL, and MongoDB migration scenarios through Azure DMS were explicitly called out as unaffected. Second, Azure Migrate – a broader migration and modernization service – still lists Database Migration Service as free for the first 180 days in its own pricing documentation, suggesting Microsoft is consolidating database migration functionality under the Azure Migrate umbrella rather than abandoning the capability outright. Teams mid-migration on SQL Server in 2026 should treat this as a forced tooling change, not a signal that Azure can no longer receive SQL Server workloads – just that the on-ramp changed.
AWS DMS Deep Dive: Serverless, Schema Conversion, and Savings Plans
AWS spent 2025 and 2026 quietly making DMS cheaper to operate rather than adding flashy new features. The biggest structural change was folding AWS DMS into Database Savings Plans, announced in December 2025 – a commitment-based discount model previously available for RDS and Aurora that now extends coverage to DMS usage as well. For any team running DMS continuously (as opposed to a one-time cutover job), that’s a real lever for cutting the compute-hour bill without changing architecture.
DMS Schema Conversion remains AWS’s answer to the hardest part of heterogeneous migrations: translating stored procedures, triggers, and schema objects from one SQL dialect to another. AWS keeps this tool free to run, charging only for the storage consumed by conversion artifacts – a pricing decision that removes one of the biggest cost-uncertainty factors in an Oracle-to-Aurora or SQL-Server-to-PostgreSQL migration, where schema conversion effort historically dwarfed the data-movement cost.
DMS Serverless continues to be the recommended default for new migrations, scaling automatically from 1 DCU up to 384 DCUs based on workload demand, with each DCU representing 2 GB of RAM. That autoscaling matters most during the initial backfill phase of a migration, when throughput demand is highest, and then again during CDC catch-up after a long cutover window – DMS Serverless avoids the manual instance-resizing that provisioned replication instances require. AWS also adjusted its free-tier structure in mid-2025: any account created after July 15, 2025 can choose between a Free Plan or Paid Plan, with the Free Plan’s credits explicitly usable against DMS and DMS Serverless charges.
aws dms create-replication-task \
--replication-task-identifier oracle-to-aurora-cutover \
--source-endpoint-arn arn:aws:dms:us-east-1:123456789012:endpoint:ORACLE-SRC \
--target-endpoint-arn arn:aws:dms:us-east-1:123456789012:endpoint:AURORA-TGT \
--replication-instance-arn arn:aws:dms:us-east-1:123456789012:rep:REPLICATION-INSTANCE \
--migration-type full-load-and-cdc \
--table-mappings file://table-mappings.jsonGoogle Cloud DMS Deep Dive: AlloyDB and the Oracle Exodus
Google’s DMS strategy in 2026 is unmistakably built around funneling Oracle and SQL Server customers toward AlloyDB for PostgreSQL, its PostgreSQL-compatible engine aimed at heavier transactional and mixed analytical workloads. The free-for-homogeneous, metered-for-heterogeneous pricing split isn’t neutral – it’s a deliberate incentive: if you’re already on MySQL or PostgreSQL and moving to Cloud SQL or AlloyDB, Google charges nothing for the migration service itself. If you’re moving off Oracle or SQL Server (both commercially licensed, both expensive to keep running), Google charges a metered fee, but one designed to look small next to the licensing costs you’re escaping.
The 500 GiB/month free backfill allowance on heterogeneous migrations is generous enough to cover a genuinely large first pass at moving an Oracle schema, and the CDC pricing tiers step down as volume increases, rewarding larger, more urgent migrations with a lower effective per-GiB rate. Google’s 2026 documentation update expanding heterogeneous support to PostgreSQL 16 and 17 as migration targets closed a real gap – until that update, Oracle and SQL Server customers migrating to AlloyDB or Cloud SQL for PostgreSQL were often landing on an older PostgreSQL major version than they’d get from a native PostgreSQL install, which complicated extension compatibility for anyone relying on newer PostgreSQL features.
Google’s documentation lists the source options for the Oracle heterogeneous path as including both self-managed Oracle deployments and Oracle running on Amazon RDS – meaning Google explicitly supports pulling an Oracle workload out of AWS and into AlloyDB, not just migrating on-prem Oracle. That’s a notable cross-cloud migration path built directly into the tool, rather than something you’d need a third-party ETL pipeline to accomplish.
Downtime, Cutover Time, and Throughput: What the Data Shows
Here’s where this comparison has to be honest about a gap in the public record: there is no independent, third-party benchmark published in 2025 or 2026 that directly measures downtime, cutover time, or throughput across AWS DMS, Azure DMS, and Google Cloud DMS side by side. Each vendor’s documentation describes its own tool as delivering “near-zero downtime” or “minimal downtime” migrations through continuous CDC replication, but none of the three has published a controlled, apples-to-apples comparison against the other two – and no credible third party appears to have run one either as of this writing.
What can be said with confidence is architectural, not empirical: all three services use the same fundamental pattern – an initial full-load backfill followed by continuous CDC replication until a cutover point, at which application traffic switches to the new target. The actual downtime experienced during cutover depends far more on how the migration is orchestrated (DNS switchover time, connection draining, application retry logic) than on which vendor’s replication engine moved the bytes. Teams evaluating these tools for a downtime-sensitive migration should budget time for their own proof-of-concept test against production-representative data volumes rather than relying on any vendor’s marketing claim about “zero downtime,” since none of the three publishes a verifiable SLA number specifically for migration cutover windows.
One place where the vendors do differ measurably is elasticity during the backfill phase. AWS DMS Serverless’s ability to scale from 1 to 384 DCUs automatically means a large initial backfill can complete faster without manual instance resizing – a structural advantage for one-time, time-boxed migrations where finishing the backfill quickly directly shortens the CDC catch-up window. Google’s per-GiB heterogeneous pricing has no equivalent throughput lever exposed to the customer; the service manages its own scaling internally, and you pay per gigabyte regardless of how long the job took to complete.
Real-World Use Cases: Which Tool Fits Which Migration
The right tool depends heavily on the specific migration scenario. Here are five common situations and how the pricing and engine-support differences play out in practice.
- A fintech company decommissioning an on-prem Oracle license: Google Cloud DMS is the obvious fit if the target is AlloyDB for PostgreSQL – the 500 GiB/month free backfill covers a substantial first pass, and escaping Oracle licensing costs typically dwarfs the per-GiB CDC fee for ongoing replication during a phased cutover.
- A retailer replicating SQL Server during a change freeze: Teams that built their migration runbook around Azure Data Studio’s SQL Migration extension need a new plan in 2026 – Microsoft’s guidance points to Visual Studio Code with the MSSQL extension, which changes the tooling but not the underlying CDC mechanics.
- A SaaS startup consolidating a MongoDB fleet: AWS DMS’s broad engine support (including MongoDB as both source and via DocumentDB compatibility as target) makes it a natural fit for teams already running compute on AWS and wanting to keep the migration inside one console.
- An enterprise IT team running committed-use AWS spend: Since AWS folded DMS into Database Savings Plans in December 2025, teams that already carry RDS or Aurora savings commitments can extend that same discount structure to ongoing DMS usage, making AWS the lower-friction choice purely from a procurement standpoint.
- A healthcare provider moving PostgreSQL for data-residency reasons: A same-engine, cross-region PostgreSQL-to-PostgreSQL migration is homogeneous by definition, which means it’s free of any DMS service charge on Google Cloud – only the destination Cloud SQL or AlloyDB instance costs apply, a straightforward calculation for a compliance-driven move.
Notice the pattern: engine choice and target platform drive the decision far more than any single feature. Teams that have already standardized on RDS or Aurora as a target should default to AWS DMS regardless of source engine, and teams already committed to AlloyDB should let Google’s free homogeneous tier and metered heterogeneous tier do the math for them. For teams still deciding where the data should land in the first place, our AWS RDS vs Azure Database vs Google Cloud SQL comparison is the logical companion piece to this one.
Migration Guide: Planning a Cross-Cloud Database Migration in 2026
Regardless of which of the three services you pick, the planning sequence looks similar. Here’s the process most teams follow, adjusted for what changed in 2026.
- Classify the migration as homogeneous or heterogeneous. This single decision determines your cost exposure on Google Cloud and your tooling path on Azure – get it wrong and you’ll re-scope the whole project.
- Inventory schema objects that need conversion. Stored procedures, triggers, and proprietary SQL dialect features are the actual bottleneck in most heterogeneous migrations, not raw data volume. Run AWS DMS Schema Conversion (free) or the equivalent assessment step early, before committing to a cutover date.
- Confirm your tooling is still supported. If you’re migrating SQL Server on Azure, verify you’re not relying on the retired Azure Data Studio SQL Migration extension – pivot to Visual Studio Code with the MSSQL extension per Microsoft’s current guidance before building automation around anything deprecated.
- Provision the smallest viable replication capacity first. Start with AWS DMS Serverless at low DCU counts, or a Standard-tier Azure DMS instance, or Google’s managed heterogeneous job, and validate correctness on a subset of tables before scaling up for the full backfill.
- Run the initial full-load backfill during a low-traffic window. This is the highest-throughput phase and the one most likely to benefit from AWS DMS Serverless’s autoscaling or a larger provisioned instance class.
- Enable continuous CDC replication and monitor replication lag. Keep the source and target in sync for days or weeks if needed – all three services support this, and there’s no cost penalty on any of them for running CDC longer than expected, aside from ongoing compute or per-GiB charges.
- Validate data integrity before cutover. Row counts, checksums, and application-level smoke tests against the target database catch replication gaps that CDC monitoring alone won’t surface.
- Cut over during a scheduled maintenance window and decommission the source. Budget your actual downtime based on DNS propagation and connection-draining time in your application tier – not on vendor claims about the replication engine itself being “zero downtime.”
Pros and Cons of Each Migration Service
AWS DMS
- Pros: Widest engine support of the three; free Schema Conversion tool; DMS Serverless autoscaling from 1-384 DCUs; now eligible for Database Savings Plans discounts.
- Cons: Instance-hour and DCU-hour pricing isn’t published as flat dollar figures for every instance class, making cost forecasting harder without running the AWS pricing calculator directly; no per-GiB flat-rate option for teams that prefer volume-based billing.
Azure Database Migration Service
- Pros: Standard tier is free with no usage cap up to 4 vCore; tight integration with Azure Active Directory and the broader Azure Migrate program; MySQL, PostgreSQL, and MongoDB scenarios remain fully supported post-retirement.
- Cons: Classic SQL Server migration path retired March 15, 2026, alongside Azure Data Studio’s SQL Migration extension (retired February 28, 2026); Premium-tier pricing beyond the six-month free window isn’t published as an exact figure, the only one of the three vendors with that gap.
Google Cloud Database Migration Service
- Pros: Homogeneous migrations are entirely free; 500 GiB/month free backfill on heterogeneous jobs; transparent, tiered per-GiB pricing published directly on the pricing page; newest PostgreSQL 16/17 target support for heterogeneous migrations.
- Cons: Heterogeneous support limited to four specific source/target combinations (Oracle and SQL Server only, landing in Cloud SQL or AlloyDB for PostgreSQL); no MySQL-as-source heterogeneous path publicly documented; per-GiB billing can get expensive fast on very large one-time migrations above the free tier.
Security and Compliance Considerations
All three services encrypt data in transit and at rest by default, and all three integrate with their respective cloud’s native identity system – AWS IAM, Azure Active Directory, and Google Cloud IAM – for controlling who can create, modify, or monitor a migration task. That said, a database migration is inherently a moment of elevated risk: source credentials, target credentials, and often a full copy of production data sit inside the migration service’s configuration and logs for the duration of the job.
For regulated workloads – healthcare data under HIPAA, financial data under PCI DSS, or anything covered by data-residency requirements – the cross-region and cross-cloud paths matter as much as the encryption defaults. Google’s documented ability to migrate Oracle running on Amazon RDS directly into AlloyDB, for instance, means the migration job itself briefly touches both AWS and Google infrastructure, which compliance teams need to account for in their data-flow documentation even though the underlying replication is encrypted end-to-end. Teams in regulated industries should treat the migration window itself as a temporary expansion of their compliance boundary and scope their access reviews accordingly, rather than assuming the migration tool’s default security posture covers every regulatory requirement automatically.
Disaster-recovery planning during a migration is a related but distinct concern – if the migration itself fails partway through, you need a rollback path that doesn’t depend on the very replication job that broke. Our AWS DRS vs Azure Site Recovery vs Google Cloud comparison covers the failover tooling that complements, rather than replaces, the migration services covered here.
Common Migration Pitfalls Across All Three Services
A handful of mistakes show up repeatedly regardless of which vendor’s tool a team picks, and most of them trace back to treating the migration service as a fire-and-forget appliance rather than something that needs active monitoring for the full duration of the CDC window.
- Underestimating schema conversion effort. Teams consistently budget for data movement time and forget that translating Oracle PL/SQL or SQL Server T-SQL into PostgreSQL-compatible syntax is usually the longer pole in the tent. AWS’s free Schema Conversion tool and Google’s built-in heterogeneous workflow both help, but neither eliminates the manual review pass a DBA needs to do on converted stored procedures.
- Building automation around deprecated tooling. The Azure Data Studio retirement on February 28, 2026 caught teams that had scripted parts of their migration pipeline around that extension’s CLI hooks – a reminder to check a vendor’s own deprecation notices before automating anything beyond a single migration project.
- Ignoring replication lag during long CDC windows. A six-week cutover window means six weeks of monitoring replication lag, not six weeks of “set it and forget it.” Lag that creeps up unnoticed can turn a planned weekend cutover into an extended outage when the final catch-up takes longer than expected.
- Assuming per-GiB pricing and per-hour pricing are directly comparable. Because Google bills by data volume and AWS bills by compute time, a straight side-by-side dollar comparison only works once you’ve modeled your specific data volume against your specific timeline – the worked example above shows why a generic “which one is cheaper” answer doesn’t exist.
- Skipping a rollback plan. None of the three services guarantees a clean rollback if a cutover goes wrong mid-flight. Teams need a tested plan to revert traffic to the source database that doesn’t depend on the same replication pipeline that just failed.
Verdict: Which Database Migration Service Should You Use in 2026?
There’s no single winner here because the three tools aren’t really competing for the same job anymore. If your migration involves Oracle or SQL Server and your target is PostgreSQL-compatible, Google Cloud DMS offers the most transparent and often the cheapest path, thanks to the 500 GiB free backfill and clearly tiered CDC pricing – provided your target fits within its four supported heterogeneous combinations. If you need to move a genuinely mixed fleet of engines – MongoDB, MariaDB, Oracle, SQL Server, PostgreSQL, and MySQL all in one program – AWS DMS remains the only one of the three with that breadth, and the new Database Savings Plans eligibility makes it cheaper to run continuously than it was in 2024 or early 2025.
Azure Database Migration Service is the right call specifically for MySQL, PostgreSQL, or MongoDB migrations within the Azure ecosystem, where the Standard tier’s zero cost is hard to beat. But any team planning a SQL Server migration on Azure in 2026 needs to build around the post-retirement tooling from day one – assuming Azure Data Studio’s migration wizard still exists is the single most common planning mistake this year, given it stopped working on February 28, 2026. Whichever tool you pick, the data backs one consistent conclusion: engine support and target-platform lock-in should drive the decision far more than any single pricing line item, because all three vendors have converged on the same backfill-plus-CDC architecture underneath very different billing models.
Frequently Asked Questions
Is AWS DMS free to use?
AWS DMS itself isn’t free for ongoing use, but AWS DMS Schema Conversion is free (you pay only for storage), and accounts created after July 15, 2025 receive free-plan credits that can offset DMS and DMS Serverless charges. Provisioned replication instances and DCU-hours under Serverless are billed at standard rates beyond any applicable credits.
What happened to Azure Database Migration Service in 2026?
Microsoft retired Database Migration Service (classic) for SQL Server scenarios on March 15, 2026, and retired Azure Data Studio – along with its Azure SQL Migration extension – on February 28, 2026. MySQL, PostgreSQL, and MongoDB scenarios were explicitly not affected by this retirement and remain supported.
Is Google Cloud Database Migration Service really free?
It’s free for homogeneous migrations (same source and target engine) into Cloud SQL or AlloyDB for PostgreSQL. Heterogeneous migrations – different source and target engines, such as Oracle to AlloyDB – are metered: the first 500 GiB of backfill per month is free, with CDC billed per GiB on a tiered schedule starting at $2.00/GiB.
Which service supports the most database engines?
AWS DMS has the broadest documented engine support among the three, covering Oracle, SQL Server, PostgreSQL, MySQL, MongoDB, and MariaDB as source or target combinations across both homogeneous and heterogeneous migration types.
Can I migrate an Oracle database running on AWS into Google Cloud?
Yes. Google’s documentation explicitly lists Oracle running on Amazon RDS as a supported source for its heterogeneous migration path into AlloyDB for PostgreSQL or Cloud SQL for PostgreSQL, making a direct AWS-to-Google migration possible without a separate ETL pipeline.
What replaced Azure Data Studio for SQL Server migrations?
Microsoft’s official guidance points teams to Visual Studio Code with the MSSQL extension, along with other dedicated SQL migration tooling, following Azure Data Studio’s retirement on February 28, 2026.
Do any of these services publish independent benchmark data on migration downtime?
No. As of late 2026, none of the three vendors, nor any known independent third party, has published a controlled, side-by-side benchmark measuring downtime, cutover time, or throughput across AWS DMS, Azure DMS, and Google Cloud DMS. Vendor documentation describes each tool’s own architecture as low-downtime, but there’s no comparative data to verify relative performance.
Which tool is cheapest for a large one-time Oracle migration?
It depends on data volume and timeline. Google’s per-GiB model gets cheaper per gigabyte as volume increases (down to $0.80/GiB above 10,000 GiB of CDC), which can favor very large migrations. AWS’s compute-hour model can be cheaper for migrations that complete quickly on a smaller DMS Serverless footprint, and now benefits from Database Savings Plans if you’re already committed to AWS spend. Running the numbers against your actual data volume and expected migration duration is the only reliable way to compare the two for a specific project.
Does AWS DMS support MongoDB migrations?
Yes. AWS documents MongoDB alongside Oracle, SQL Server, PostgreSQL, MySQL, and MariaDB as supported source engines, with Amazon DocumentDB commonly used as a MongoDB-compatible target for teams consolidating a MongoDB fleet onto AWS-managed infrastructure.
Can I use more than one of these services in the same migration project?
Nothing technically prevents it, and some multi-cloud migrations do combine tools – for example, using Google’s Oracle-on-Amazon-RDS support to pull data out of AWS while the destination and ongoing operations move to Google Cloud. In practice, most teams pick a single service per migration project to keep monitoring, credentials, and rollback planning simple, reserving a mixed approach for genuinely cross-cloud consolidation efforts.
