Skip to content

IaaS, PaaS, SaaS, and serverless

Service models describe how much the provider manages for you.

Model Customer focus Azure/provider focus Example
IaaS OS, applications, configuration, data Physical infrastructure and virtualization Azure Virtual Machines
PaaS Application code, data, access Infrastructure, OS, runtime, managed platform Azure App Service, Azure SQL Database
SaaS Users, data, access settings Application, platform, infrastructure Microsoft 365

The managed stack

Layer                  IaaS          PaaS          SaaS
───────────────────────────────────────────────────────
Data                   Customer      Customer      Customer
Application            Customer      Customer      Provider
Runtime/middleware     Customer      Provider      Provider
Operating system       Customer      Provider      Provider
Virtualization         Provider      Provider      Provider
Physical infrastructure Provider     Provider      Provider

Identity, access, data classification, and endpoint security remain customer concerns across all three models.

IaaS: configure the server

Choose Infrastructure as a Service when the scenario requires:

  • control of the operating system;
  • installation of custom or legacy software;
  • a lift-and-shift server migration;
  • custom server configuration.

IaaS offers the most control of the three models and the most customer administration.

IaaS operational checklist

The customer normally handles:

  • guest OS hardening and patching;
  • installed software and runtime updates;
  • backup and recovery design;
  • malware protection and host configuration;
  • VM availability and scaling architecture;
  • network rules and exposed ports;
  • data, identities, and permissions.

Microsoft operates the datacenter, physical hosts, physical network, and hypervisor.

IaaS scenario

A vendor application requires a specific Windows Server version, a custom driver, and administrator access to the OS.

Best fit: Azure Virtual Machines as IaaS.

PaaS: publish the application

Choose Platform as a Service when the team wants to:

  • focus on code and data;
  • host a web app or API without patching an OS;
  • use a managed database;
  • reduce platform maintenance.

PaaS trade-offs

Benefits:

  • less patching and platform administration;
  • built-in scaling and availability options;
  • faster application delivery;
  • integrated deployment, identity, and monitoring features.

Trade-offs:

  • less control over the operating system and runtime;
  • supported languages, versions, networking, or extensions can be constrained;
  • migration might require application changes.

PaaS scenario

A team wants to deploy a REST API, configure autoscale, and use deployment slots without maintaining Windows or Linux servers.

Best fit: Azure App Service as PaaS.

SaaS: use the software

Choose Software as a Service when users need a ready-made application and the organization does not want to build or operate the application platform.

SaaS customer responsibilities

The provider runs the application, but the customer still manages:

  • user lifecycle and access;
  • data classification and sharing;
  • tenant-level settings;
  • retention and compliance choices;
  • endpoint security and user behavior;
  • integration configuration.

SaaS scenario

Employees need email, calendars, document collaboration, and meetings without the company building those applications.

Best fit: Microsoft 365 as SaaS.

Serverless

Serverless means the customer does not provision or manage the underlying servers. It does not mean no servers exist.

Typical characteristics include:

  • event-driven execution;
  • automatic scaling;
  • highly abstracted infrastructure;
  • consumption-based billing on eligible plans.
Blob uploaded → event → Azure Function runs → execution completes

Azure Functions is the common AZ-900 example. Billing and always-on behavior depend on the selected hosting plan, so avoid absolute statements such as “serverless always charges only per execution.”

Serverless execution model

Serverless platforms commonly combine:

  • triggers: the event that starts execution;
  • bindings/integrations: connections to queues, storage, databases, or HTTP;
  • automatic allocation: platform adds or removes execution capacity;
  • stateless design preference: durable state is stored in external services;
  • managed runtime: platform operates the underlying hosts and runtime infrastructure.

Long-running workflows, warm instances, networking requirements, and predictable performance can require different hosting plans or related services. "Serverless" describes the infrastructure abstraction, not a guarantee that every workload has zero idle cost or unlimited execution time.

PaaS versus serverless

Serverless is often treated as a highly managed form of PaaS.

PaaS application hosting Serverless function model
Usually hosts an application continuously Commonly invokes code in response to events
Scaling follows the hosting plan Scaling often follows trigger demand
App is the deployment unit Functions or workflows are smaller execution units
May reserve always-on instances Consumption or always-ready plans can be selected

Database example

The same database engine can appear in different service models:

  • SQL Server installed on an Azure VM → IaaS; customer patches OS and SQL Server.
  • Azure SQL Database → PaaS; Microsoft operates OS and database platform.
  • A finished online business application that stores data internally → SaaS to its user.

The model depends on what the customer consumes and manages, not only on the product category.

Decision rule

  • Need OS control → IaaS.
  • Need to deploy code without managing the OS → PaaS.
  • Need finished software → SaaS.
  • Need event-triggered code with no server administration → serverless.

Exam elimination method

  1. Does the customer need guest OS control? → IaaS.
  2. Does the customer supply code but not manage OS/runtime? → PaaS.
  3. Does the customer consume a complete application? → SaaS.
  4. Is small event-driven code emphasized? → serverless/Functions.
  5. Does an answer claim SaaS removes responsibility for data and identities? → reject it.

Official reference

Describe cloud service types — Microsoft Learn