Cost Comparison Between Databricks, ADLS (Archive/Cold), and Azure SQL Hyperscale

1. Databricks Serverless SQL Warehouse + ADLS Archive Tier

Databricks Serverless SQL Warehouse + ADLS Cold Tier

Azure SQL Database Hyperscale (Serverless)

How should we do a fair, apple-to-apple compute comparison between Databricks Serverless and Azure SQL Hyperscale — for example, when we are assuming 16 vCores worth of compute?

Can we consider the “X-Small” Databricks SQL Warehouse (6 DBU/hr) to be roughly equivalent to 16 vCores in Hyperscale?

  Or is there a better reference or guideline to ensure the CPU/memory/throughput are comparable across these two services?
  
  **What are the correct billing parameters we should consider for cost modeling across the three options?**
  
     For example, should we estimate:
     
           Compute cost in DBU/hour for Databricks
           
                 vCore-seconds for Hyperscale
                 
                       Read/rehydration charges for ADLS Archive or Cold?
                       
                          Should backup storage, HA replicas, or other infra components be included in these comparisons?
                          
                          **How is cost metering done in all three services — per second or per hour?**
                          
                             For instance:
                             
                                   Is Databricks Serverless billed per-second, per-minute, or per-hour?
                                   
                                         Is Hyperscale serverless billed strictly per second of usage?
                                         
                                               Are reads from Cold or Archive tier charged per GB read, or are there session-level charges?
                                               
                                               **Are there any hidden or indirect costs we should take into account?**
                                               
                                                  For example:
                                                  
                                                        ADLS Archive has a rehydration delay and early-deletion charges — what are the other hidden charges in Archive and Cold tiers?
                                                        
                                                              In Hyperscale, are there additional costs due to write-ahead logging, autoscaling behavior, or background tasks (e.g., checkpointing, backups)?
                                                              
                                                                    In Databricks, are there minimum active times, startup delays, or idle time billing patterns to be aware of?
                                                                    

We are not asking you to verify exact price numbers — but rather seeking guidance on how to do this cost modeling correctly, using appropriate billing units and assumptions. We want to ensure our client is making a sound cost decision with full transparency across all layers — storage and compute.