Neither is cheaper in general. Redshift Serverless lists $0.375 per RPU-hour and Databricks SQL Serverless lists $0.70 per DBU, so the winner depends on how many hours a day your warehouse runs and how big it is. If your team is SQL and BI on AWS, stay on Redshift and cut its bill in place first. Pick Databricks when the work is code-first pipelines and ML.
Which is cheaper for your workload?
Take a BI-and-ELT warehouse that's active 6 hours a day, 22 days a month: 132 hours. Assume it only bills while queries run.
Amazon's Redshift pricing page lists Serverless at $0.375 per RPU-hour in US East (N. Virginia). An 8 RPU warehouse costs $3.00 an hour, so 132 hours is $396 a month in compute. Databricks lists SQL Serverless at $0.70 per DBU, and that rate includes the cloud instance cost. Divide the Redshift hourly cost by $0.70 and you get the number of DBUs per hour at which the two bills match.
| Redshift Serverless size | Redshift cost per hour | Databricks is cheaper below |
|---|---|---|
| 4 RPU | $1.50 | 2.14 DBU per hour |
| 8 RPU | $3.00 | 4.29 DBU per hour |
| 32 RPU | $12.00 | 17.14 DBU per hour |
To use it, find your Redshift size in the first column, then compare your Databricks warehouse's DBU per hour with the last column: if yours is lower, Databricks costs less. Read it as a threshold, not a size match. An RPU and a DBU aren't the same unit, so an 8 RPU warehouse doesn't map to any particular Databricks size. What you need is your own DBU burn: the Databricks pricing calculator gives it for a given warehouse, or you can run one warehouse for a day and read the bill.
$0.70 is the rate Databricks shows before you pick a plan, cloud and region. Set all three in the pricing calculator. Redshift managed storage is $0.024 per GB-month, so 2,000 GB adds $48.
What does each platform bill for?
Redshift Serverless scales from 4 to 1,024 RPUs and bills per second with a 60-second minimum. The 60-second floor matters if your dashboards fire many short queries after idle gaps.
Databricks bills DBUs per second too. The difference is what the rate covers. Serverless SQL includes the cloud instances. Classic and Pro SQL warehouses don't: you pay the DBU rate and your own cloud bill for the machines underneath. Databricks recommends serverless for most workloads, so the self-managed tiers are the exception.
Should you cut the Redshift bill instead of leaving?
Probably try it first. Redshift Serverless Reservations cut compute by up to 24% on a 1-year term and up to 45% on a 3-year term. On the $396 example, that's as low as about $301 or $218 a month. It also lowers every break-even figure in the table by the same percentage, which makes Databricks harder to justify on price alone.
Stay on Redshift
This is the pick for SQL analysts and BI dashboards that already live in your AWS account. You keep your SQL, your tooling and your team's habits, and you change the commercial terms instead of the platform. The catch is the one Databricks' own migration guide points at: costs rise with data volume and workloads, and tuning (sort and distribution keys, queues, cluster sizes) gets harder at size. That's the vendor's framing of its competitor, so weigh it accordingly, but anyone who has tuned a large Redshift cluster will recognise it.
Move to Databricks SQL Serverless
This fits teams whose work is pipelines in code, ML next to analytics, or a wish for open table formats (Delta, Iceberg) and one governance layer across both. You get a warehouse with no clusters to run or scale by hand. The catch is price predictability: you need the DBU number from the table above before you can say it's cheaper.
Move to Databricks SQL Classic or Pro
Databricks lists Classic at $0.22 per DBU and Pro at $0.55, before you choose plan, cloud and region, plus your own cloud bill. The low headline rate is the trap, because it isn't comparable to the $0.70 serverless figure. You also run the clusters yourself: one cluster per 10 concurrent queries, with minimum and maximum clusters set by hand. Pick these when you need control over the infrastructure, not to save money.
| Option | Best for | Catch |
|---|---|---|
| Stay on Redshift | SQL and BI teams on AWS | Tuning effort grows with data |
| Databricks SQL Serverless | Code-first pipelines and ML | Need your DBU per hour first |
| Databricks SQL Classic or Pro | Teams wanting infrastructure control | You size clusters and pay cloud separately |
What would you have to change to move?
Databricks' field guide to leaving Redshift points to four jobs, and none is a copy-paste.
- Translate your schemas rather than copying the DDL.
- Rewrite stored procedures from Redshift's PL/pgSQL into what Databricks supports.
- Redo how you manage concurrency: Redshift's workload management queues become Databricks warehouse sizing and scaling settings.
- Run both platforms in parallel and compare results before you cut over.
The same guide calls migration "straightforward since both platforms are cloud-based" and also calls code and logic conversion "often the most complex and critical part". Believe the second statement. If your Redshift estate is mostly tables and views, the work is modest. If it's a few hundred stored procedures, that's your migration. No timeline is published, so size it by counting your procedures and queues.
Performance has no neutral answer either. Take your ten slowest or most expensive queries, run them on both, and decide from that.
What this means for your platform decision
If you're the data engineer (or the backend engineer who inherited analytics) on AWS, the decision is a split by who works in the warehouse. SQL analysts and dashboards point to staying and buying a Reservation. Pipeline and ML engineers who are tired of tuning keys and queues point to a migration, and they should bring the DBU number with them.
Before any migration plan, do two things this week. Pull your warehouse hours and RPU size from last month's bill, and get the DBU per hour for the same workload from the calculator. If Databricks isn't under the break-even line, the case for moving rests on engineering reasons alone, and you should say so in the proposal.
Next, how Snowflake and Databricks differ covers the other pairing teams weigh, and there are more stack comparisons and decisions by tech stack in the same section.
Frequently asked questions
Is Redshift still relevant?
Yes. AWS still prices it per RPU-hour from 4 to 1,024 RPUs and sells Serverless Reservations that cut compute by up to 45%. For SQL and BI teams already in AWS, it's a live option, not a legacy one.
What are the disadvantages of Redshift?
Databricks' migration guide names rising cost with data volume, harder tuning at size, and, on the other side, the draw of open table formats and one governance layer across analytics and ML. That's a vendor's view. Your own bill and your own tuning hours are the better evidence.
Can you use both Redshift and Databricks?
Yes, and during a migration you should. Databricks recommends running both in parallel to validate results before cutting over.
