Many organizations landed on Azure Databricks by circumstance rather than by decision: it was the path of least resistance in a Microsoft estate. A few years on, the workloads are bigger, the bill is bigger, and the question deserves to be asked on the merits rather than by default. Today we are making our answer official: Buoyant Data now offers Azure to AWS migration services for Databricks platforms.
First, the risk assessment, because that is what a decision like this actually hinges on. The migration risk is low: Delta Lake, Unity Catalog, MLflow, Databricks SQL, and the notebook experience are identical across clouds, and a PySpark job written for Azure Databricks runs on AWS with configuration changes only. Your code is not the obstacle. The real question is which cloud gives the platform its best foundation, and on cost and capability the evidence increasingly points to AWS.
The cost case
Compute is the dominant line item in any Databricks deployment, and AWS offers more levers to pull:
- EC2 Spot instances at up to 90% off On-Demand prices. Spark workloads tolerate interruption gracefully, making Spot practical for a large share of data engineering and CI workloads.
- AWS Graviton processors, which delivered up to 3x better price performance for Databricks at launch, and typically around 20% compute savings on everyday Spark workloads. ARM-based compute is not an option on Azure at all.
- AWS Marketplace procurement, so Databricks consumption draws down existing AWS committed-spend agreements instead of sitting outside them.
- Pay-per-token AI inference through Amazon Bedrock. Always-on GPU endpoints are the most common way to overspend on inference; on AWS, Databricks notebooks, SQL, and agents call Bedrock directly at per-token prices, with no standing infrastructure to keep warm.
These are not projections. Organizations that have replatformed analytics from Azure to Databricks on AWS have reported infrastructure cost reductions of around 40%.
Best-in-class data tools
Cost gets the attention; capability compounds over time. AWS was the first cloud Databricks launched on, and it remains the most battle-tested deployment: new platform capabilities reach general availability on AWS first, and the platform is operated directly by Databricks rather than delivered through a third party.
Around the lakehouse, Databricks on AWS sits inside the industry's richest data ecosystem:
- Amazon S3, the de-facto lakehouse storage standard, with 11 nines of durability and the broadest compatible tooling. If your data already lands in S3, moving Databricks to AWS means moving compute to the data instead of paying to move data to the compute.
- Streaming and interoperability with Kinesis, MSK, Glue, Athena, and Redshift for the workloads around the lakehouse.
- Foundation models through Amazon Bedrock, including providers such as Anthropic, available to Databricks workloads through external-model support.
- The widest selection of compute profiles: memory-optimized instances for joins, storage-optimized for shuffles, GPUs for training, and custom silicon like AWS Trainium for machine-learning price-performance.
How the migration works
A cloud-to-cloud Databricks migration is achievable, but it is not a lift-and-shift, and the schedule matters as much as the destination. Our engagements run in four phases with clear exit criteria: an honest assessment of workspaces, jobs, data volumes, and cost models; a pilot workload migrated end-to-end to prove the runbook; prioritized migration waves with validation gates between them; and a final handoff that leaves your team operating the new platform with documentation and transferred knowledge, not a permanent dependency on us. Production keeps running throughout; the waves are structured so no team loses access to the platform it depends on.
If your Databricks bill is climbing, your data already lives in S3, or your AI roadmap needs Bedrock-class inference economics, it is time to evaluate whether the platform is on the right cloud. Send us a note and we will walk through the numbers with you.
