Hey @hboleto 

It’s difficult to accurately estimate the final cost of a Serverless cluster, as it is fully managed by Databricks. In contrast, Classic clusters allow for finer resource tuning since you can define spot instances and other instance types to optimize costs.

I understand that you’re considering the use of SQL Warehouse Serverless vs. Classic for running notebooks, with the assumption that it will help you spend less on your cloud provider. While this is partially true in terms of management, you should keep in mind that Serverless is typically 10-20% more expensive than Classic, based on my experience. The key advantage of Serverless is its immediacy and convenience for specific queries, but if your team is growing, the overall usage might become more intensive than expected, even with a 1-minute auto-termination policy.

If you’re using it solely for “small” and occasional queries, Serverless could be a cost-effective option in some cases. However, in a real-world development environment, it’s unlikely that a user would spin up a cluster just to run a single SELECT statement and not use it again. Additionally, for queries that take only a few seconds, Serverless’ seamless scaling can lead to unexpected costs, as you don’t directly control how the cluster operates. This could result in paying for additional instances without realizing it.

When it comes to job clusters, I wouldn’t recommend using Serverless under any circumstances. It’s much better to invest time in properly provisioning resources on a Classic cluster, as it allows for more granular cost control and predictable performance over time.

I also recall reading this post, where someone shared their experience with Serverless clusters in a very insightful way.

Pricing calculator 

Best regards 🙂