Azure SQL has evolved from a managed SQL Server into a family of managed database services with distinct performance profiles. Understanding when each option is appropriate is an architecture skill that the serverless and hyperscale tiers make more important.
Azure SQL Database tiers
The Azure SQL product family includes: SQL Database for general-purpose managed SQL Server, SQL Managed Instance for near-complete SQL Server compatibility in a managed environment, and SQL Edge for IoT scenarios. Within SQL Database, the serverless compute tier (auto-pause and resume based on activity) and the Hyperscale tier (up to 100TB storage with distributed storage architecture) address distinct scaling patterns that the standard tiers cannot.
The serverless compute tier
Azure SQL Database serverless auto-pauses after a configured period of inactivity (minimum 1 hour) and resumes on the next connection. The first connection after a pause has a cold start delay of 10-30 seconds. For databases that are used intermittently (development environments, infrequently accessed reporting databases, webhook processors), the serverless tier reduces cost to zero during idle periods.
Hyperscale for large databases
Azure SQL Hyperscale separates compute and storage, distributing page server storage across multiple nodes. The result: database files can grow to 100TB without performance degradation, read replicas are provisioned in minutes rather than hours, and backups are almost instantaneous. For databases that have outgrown Standard tier storage limits or need near-instant backup and restore, Hyperscale is the path. The cost is higher than Standard tier but the operational simplicity at scale justifies it.
Query performance analysis
The Azure SQL Query Performance Insight surfaces the top resource-consuming queries with execution plans and recommendations. The index recommendations from the Automatic Tuning feature are based on actual query patterns from the query store. The combination of Query Performance Insight and Automatic Tuning provides actionable performance data without needing to manually analyse execution plans and DMVs. For teams that do not have a dedicated DBA, these tools cover 80% of the performance optimisation work.