Azure SQL deployment: SQL Database vs Managed Instance vs SQL Server on VMs vs Fabric vs Arc

Verdict: Azure SQL Database for greenfield cloud-native apps. Managed Instance for lift-and-shift needing SQL Agent and cross-database queries. SQL Server on VMs for OS-level control. Fabric for OneLake analytics; Azure Arc to manage instances outside Azure.

Criterion	Azure SQL Database	Azure SQL Managed Instance	SQL Server on Azure VMs	SQL in Microsoft Fabric	SQL Server enabled by Azure Arc
Model	PaaS single/pooled database	PaaS instance	IaaS, full OS control	PaaS mirrored to OneLake	Management plane for external SQL
Patching	Automatic	Automatic	Customer, or IaaS Agent extension	Automatic	Not applicable
Instance features	No SQL Agent or cross-db queries	SQL Agent, cross-db, Database Mail	Full engine	No	Runs on the source engine
Max storage	Hyperscale 128 TB	16 TB	256 TB	n/a	n/a
Choose when	Greenfield cloud-native app	Lift-and-shift needing instance features	OS-level control, CLR, third-party agents	Analytics in Fabric/OneLake	Inventory SQL instances outside Azure

Rules

Azure SQL Database is the PaaS default for new cloud-native apps: no OS access, automatic patching, elastic pay-as-you-go scale.
Managed Instance offers near-100% engine compatibility including SQL Agent, cross-database queries and Database Mail as a fully managed service.
SQL Server on Azure VMs is IaaS: full OS and engine control for CLR, third-party agents and custom recovery models.
Azure SQL Database and Managed Instance are auto-patched by Microsoft; VM SQL Server is not, unless the IaaS Agent extension is enabled.
Max single-database storage: SQL Server on VM 256 TB, Hyperscale 128 TB, Managed Instance 16 TB.
Resource Governor works only on SQL Server on VM and Managed Instance, not single Azure SQL Database or elastic pools.

Traps

Managed Instance has the smallest storage ceiling of the three at 16 TB, not the largest.
PaaS tiers carry a higher guaranteed SLA than IaaS; the VM SLA covers only infrastructure, not database availability.