Study Azure AZ-900 Compute and Networking: key concepts, common traps, and exam decision cues.
On this page
Compute and networking questions on AZ-900 are service-family questions. You do not need engineer-level design depth. You do need to know what problem the service category solves and when Microsoft is testing a compute model versus a network boundary.
Compute chooser
Need
Strongest first fit
Why
Traditional VM control
Azure Virtual Machines
strongest IaaS compute lane
Auto-scaling group of VMs
Virtual Machine Scale Sets
coordinated scaling for similar VM instances
Event-driven serverless execution
Azure Functions
function-first serverless lane
Desktop virtualization
Azure Virtual Desktop
managed desktop delivery rather than app hosting
Managed web application hosting
Azure App Service
app platform instead of full VM management
VM-related terms the exam still expects
Term
Why it matters on AZ-900
availability set
basic VM fault and update-domain concept
VM scale set
scalable fleet of similar VMs
disk, NIC, and VNet attachment
reminds you that a VM needs networking and storage resources too
Networking chooser
Need
Strongest first fit
Why
Private network boundary
Azure Virtual Network
main Azure network container
Subdivision inside a VNet
Subnet
smaller address boundary inside the VNet
DNS resolution
Azure DNS
name-resolution lane
Private connectivity from on-premises
VPN Gateway or ExpressRoute
hybrid connectivity lane
Service exposure boundary
public or private endpoints
how a service is reached
Hosting and connectivity comparisons
If the stem emphasizes…
Strongest first reading
full control over the server and OS
VM
hosting a web app with less infrastructure work
App Service
code that reacts to events
Functions
connectivity over the internet
VPN Gateway
dedicated private connectivity
ExpressRoute
access over private IP to a platform service
Private Endpoint
Decision order that usually wins
Decide whether the requirement is full VM control, managed web hosting, event-driven execution, or networking connectivity.
Keep compute choice separate from network delivery choice.
Prefer the more managed option when low-level control is not required.
Read event-driven cues before defaulting to virtual machines.
Match the service to the workload behavior, not to what looks most familiar.