Hi @Sam500,

Before concluding that serverless is expensive, I would first want to understand what that assessment is based on. Have you done a like-for-like benchmarking exercise against classic or pro compute to arrive at that conclusion?

In many cases, people compare only the visible unit cost and stop there. But if you look at the total cost of ownership, serverless is generally lower. Databricks explicitly positions SQL warehouses as the most cost-efficient engine for interactive SQL workloads, and recommends serverless SQL warehouses for most workloads because they start in seconds, scale elastically, and scale down faster when demand drops.

A big reason for this is operational behaviour. Non-serverless warehouses take minutes to start, so in practice, many teams leave them running longer than necessary. Serverless, on the other hand, provides instant availability with automatic scale-up and early scale-down, which often leads to lower overall costs despite the perception that it is the more expensive option.

It is also important to correctly evaluate the full economics. For serverless services, the DBU price already includes the underlying virtual machine cost, so the comparison should be made at the end-to-end workload level, not only on sticker price.

I have personally done this comparison and benchmarking for a number of customers, and the conclusion has consistently been the same.... when measured properly, serverless is usually cheaper than classic compute from a total cost of ownership perspective.

That said, if costs are still higher than expected, the focus should usually be on workload optimisation rather than moving away from serverless. Right-size the warehouse, use aggressive auto-stop, limit max clusters, reduce unnecessary DirectQuery traffic from Power BI, and add proper tagging and monitoring so you can see exactly which reports, users, or models are driving spend.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.