AZ-900 cloud concepts: models, pricing and benefits
This page covers the first skill area of the AZ-900 exam, Describe cloud concepts, which is worth 25-30% of the score in the skills outline that took effect on July 20, 2026. It is vendor-neutral on purpose: the exam checks that you understand what the cloud is and why organizations use it before it asks about specific Azure services.
The examples follow one invented company, the team behind OrdersPortal (ordersportal.example.test), an order-management web app that today runs on two servers in a rented rack and is moving to Azure.
Topics on this page
Section titled “Topics on this page”| Topic | Skills outline item |
|---|---|
| Cloud computing | Define cloud computing |
| Shared responsibility model | Describe the shared responsibility model |
| Cloud models | Define public, private and hybrid cloud and their use cases |
| Consumption-based model | Describe the consumption-based model |
| Cloud pricing models | Compare cloud pricing models |
| Serverless | Describe serverless |
| Benefits of cloud services | High availability, scalability, reliability, predictability, security, governance, manageability |
| Cloud service types | IaaS, PaaS, SaaS and their use cases |
Cloud computing
Section titled “Cloud computing”Cloud computing is the delivery of computing services (servers, storage, databases, networking, software and analytics) over the internet, so that you rent capacity from a provider instead of buying and running your own datacenter. Microsoft Azure is Microsoft’s cloud platform; providers such as Microsoft are often called cloud service providers (CSPs).
The shift is easiest to see by comparing who does what. On-premises, OrdersPortal’s team pays for the rack space, owns the physical servers, replaces failed disks and carries all the risk. In the cloud, the provider owns the buildings, hardware and physical network and employs the people who run them; the customer configures services and owns its code and data.
| Responsibility | On-premises | Cloud |
|---|---|---|
| Real estate, power, cooling | You | Provider |
| Physical servers and network | You | Provider |
| Capacity planning months ahead | You | Mostly unnecessary: add or remove resources on demand |
| Configuration, code and data | You | You |
Computing moved from dedicated physical servers to virtual machines (several isolated operating systems sharing one host through a hypervisor), then to containers and fully managed services. Each step hands more of the undifferentiated work to someone else, which is the thread that runs through the rest of this page.
Pitfalls
- Treating “the cloud” as “someone else runs everything”. What you still own depends on the service type, covered next.
- Assuming the cloud is always cheaper. It removes up-front hardware spend; an idle, oversized deployment still costs money every hour.
Shared responsibility model
Section titled “Shared responsibility model”In the shared responsibility model, security and operational duties are split between the cloud provider and the customer, and the split moves with the service type.
The provider is always responsible for the physical datacenter, the physical network and the physical hosts. The customer is always responsible for the information and data stored in the cloud, the devices allowed to connect (phones, laptops) and the accounts and identities of people, services and devices. Everything in between, such as the operating system, network controls, applications and identity infrastructure, depends on the service type:
| Layer | On-premises | IaaS | PaaS | SaaS |
|---|---|---|---|---|
| Data, devices, accounts and identities | Customer | Customer | Customer | Customer |
| Identity and directory infrastructure | Customer | Customer | Shared | Shared |
| Applications | Customer | Customer | Shared | Provider |
| Network controls | Customer | Customer | Shared | Provider |
| Operating system | Customer | Customer | Provider | Provider |
| Physical hosts, network and datacenter | Customer | Provider | Provider | Provider |
A concrete OrdersPortal example: if the team installs SQL Server on an Azure virtual machine (IaaS), it patches the operating system and the database engine. If it moves the database to Azure SQL Database (PaaS), Microsoft patches the engine, but the team is still responsible for the data, who can sign in, and how the database firewall is configured.
Pitfalls
- “Microsoft secures my data.” Microsoft secures the platform; classifying, protecting and controlling access to your data stays with you in every model.
- Forgetting that IaaS puts the most responsibility on the customer and SaaS the least.
Cloud models
Section titled “Cloud models”A cloud model describes where resources are deployed and who can use them.
- Public cloud: built, controlled and maintained by a third-party provider; anyone can buy services. No capital expenditure to scale, resources can be provisioned and released quickly, and you pay for what you use, but you don’t have complete control over the hardware.
- Private cloud: used by a single organization, hosted in its own datacenter or a dedicated off-site one. It gives the most control over resources and security, at the cost of buying, maintaining and updating the hardware yourself.
- Hybrid cloud: public and private clouds connected so workloads can move between them. It offers the most flexibility: for example, a private cloud can “surge” into the public cloud for a seasonal peak, and an organization can keep regulated data on-premises while running the web tier in Azure.
- Multicloud: using two or more public cloud providers at once, for example during a migration or to use a feature that only one provider offers.
Azure has two services that come up in this area. Azure Arc manages servers, Kubernetes clusters, databases and SQL Server outside Azure (on-premises or in other clouds) as if they were Azure resources. Azure VMware Solution runs existing VMware workloads in Azure.
| Use case | Model |
|---|---|
| Startup with no datacenter and unpredictable traffic | Public |
| Strict regulation requires dedicated hardware the organization controls | Private |
| Keep an on-premises database, add cloud capacity for peak season | Hybrid |
| Analytics on one provider, line-of-business apps on another | Multicloud |
Pitfall: hybrid is about connected environments. Two unrelated environments that never exchange workloads or data are not a hybrid cloud.
Consumption-based model
Section titled “Consumption-based model”Cloud computing uses a consumption-based model: you pay for the resources you use and nothing more, and you can release them when you no longer need them.
This changes the type of spending. Capital expenditure (CapEx) is up-front spending on physical infrastructure such as servers, storage, network equipment and datacenter space, which is then depreciated over years. Operational expenditure (OpEx) is ongoing spending on services or products as you consume them. Because you pay for cloud services as you use them, cloud computing is an operational expense.
The practical effect is on capacity planning. OrdersPortal’s on-premises servers had to be sized for the Black Friday peak, so they sat mostly idle the rest of the year; undersizing them would have meant slow checkouts on the busiest day. In the cloud the team adds instances for the peak and removes them afterwards, paying only for the hours used.
| CapEx (on-premises) | OpEx (cloud) | |
|---|---|---|
| When you pay | Up front | As you consume |
| Capacity | Fixed, planned in advance | Matches demand |
| Risk of idle capacity | High | Low, if you scale in |
Pitfall: OpEx does not mean “unpredictable”. Budgets, alerts and the pricing calculator (see cost management) give you forecasting and control.
Cloud pricing models
Section titled “Cloud pricing models”The default cloud pricing model is pay-as-you-go: no commitment, billed for actual use. Azure adds three options that trade flexibility for a lower price; they are listed in the cost-management module on Microsoft Learn and often appear in scenario questions.
| Option | You commit to | Best for |
|---|---|---|
| Pay-as-you-go | Nothing | New, variable or short-lived workloads |
| Reservations | A specific resource type, and for some products a region, for one or three years | Stable workloads that run around the clock |
| Savings plan for compute | A fixed hourly spend on eligible compute services for one or three years | Steady spend that moves between VM sizes, regions or compute services |
| Spot pricing | Nothing, but Azure can evict the workload when it needs the capacity back | Interruptible work such as batch jobs and test runs |
For OrdersPortal, the production database that runs 24/7 is a reservation candidate, the web tier that may move from virtual machines to containers fits a savings plan, and the nightly report-rendering job that can restart is a Spot candidate.
Pitfall: a reservation is a billing discount, not capacity you “own”. Unused reserved hours are lost; they don’t roll over.
Serverless
Section titled “Serverless”Serverless computing means you deploy code or configuration and the provider runs it on demand, scales it automatically and bills only while it runs; you never see or manage the servers. The servers still exist, but provisioning, patching and scaling them is the provider’s job.
On Azure, Azure Functions is the main serverless compute service: an event (an HTTP request, a timer, a message on a queue or a new blob in storage) wakes the function, and resources are released when it finishes. Functions are stateless by default; Durable Functions add state across calls. OrdersPortal could use a function to generate a PDF invoice each time an order message arrives, instead of keeping a virtual machine running all day to wait for orders.
Pitfalls
- Serverless suits short, event-driven work. Long-running or always-busy processes may cost less on reserved compute.
- “Serverless” is not a cloud model like public or private; it is a way of consuming compute within a cloud.
Benefits of cloud services
Section titled “Benefits of cloud services”The skills outline groups the benefits into four pairs.
High availability and scalability. High availability means keeping a service running despite failures; Azure states its uptime guarantees per service in service-level agreements (SLAs). Scalability means adjusting resources to demand. Vertical scaling (scale up or down) changes the size of a resource, such as more CPUs or RAM for a virtual machine. Horizontal scaling (scale out or in) changes the number of resources, such as adding virtual machines or container instances. Elasticity is scaling that happens automatically as demand rises and falls.
Reliability and predictability. Reliability is the ability of a system to recover from failures and keep working; the cloud’s global, decentralized design lets you run in several regions. Predictability covers performance (autoscaling and load balancing keep the user experience consistent) and cost (real-time usage tracking and tools such as the pricing calculator help forecast spend). Both are pillars of the Azure Well-Architected Framework.
Security and governance. Cloud platforms offer templates, policies and built-in tools that keep deployments compliant with corporate standards and regulations, and the provider handles physical security and much of the patching, depending on the service type.
Manageability. Management of the cloud means you can autoscale, deploy from templates, monitor health and replace failing resources, and get alerts automatically. Management in the cloud means you can do all of that through a web portal, a command-line interface, APIs or PowerShell.
Pitfall: availability, durability and disaster recovery are different measures. A highly available single-region app still needs a recovery plan if the region is lost.
Cloud service types
Section titled “Cloud service types”The three service types describe how much of the stack the provider manages.
- Infrastructure as a service (IaaS) gives you virtual machines, storage and networking; you manage the operating system and everything above it. It offers the most control and the most responsibility. Use cases: lift-and-shift migration of existing servers, and test and development environments that you create and delete quickly. Azure examples: Azure Virtual Machines, Virtual Machine Scale Sets, SQL Server on Azure Virtual Machines.
- Platform as a service (PaaS) gives you a managed environment for building and running applications; the provider manages the operating system, middleware and runtime. Use cases: a development framework for web apps and APIs, analytics and business intelligence. Azure examples: Azure App Service, Azure Functions, Azure SQL Database, Azure Container Apps.
- Software as a service (SaaS) gives you a finished application, usually rented per user. Use cases: email and messaging, business productivity, finance and expense tracking. Examples: Microsoft 365 and Dynamics 365.
| You want to… | Type |
|---|---|
| Move a Windows server unchanged to Azure | IaaS |
| Deploy the OrdersPortal web app without patching servers | PaaS |
| Give staff email and calendars | SaaS |
Pitfalls
- Azure Kubernetes Service is a managed orchestration service, but you still manage the node pools and workloads, so it sits between IaaS and PaaS. Exam questions usually test the clear cases.
- Functions-as-a-service (FaaS) is a subset of PaaS, not a fourth type in the skills outline.
Practice questions
Section titled “Practice questions”-
Which costs are capital expenditure: buying servers for a new datacenter, or paying a monthly bill for Azure virtual machines? Buying servers is CapEx: an up-front purchase of physical assets. The Azure bill is OpEx, paid as the service is consumed.
-
Under the shared responsibility model, who is responsible for the data in a SaaS application? The customer. Data, devices and accounts and identities always stay with the customer, whatever the service type.
-
A company must keep customer records on its own hardware for regulatory reasons but wants to run its public website in Azure. Which cloud model is this? Hybrid cloud: connected private and public environments, with the company choosing where each workload runs.
-
What is the difference between scaling up and scaling out? Scaling up (vertical) gives one resource more CPU or memory. Scaling out (horizontal) adds more instances of the resource. Scaling out is what makes elastic, highly available web tiers possible.
-
Scenario: OrdersPortal runs a reporting job every night that can restart if interrupted, a database that runs 24/7, and a web tier that the team expects to move from virtual machines to containers within a year. Which pricing option fits each? The report job fits Spot pricing, because it tolerates eviction. The always-on database fits a reservation, because its size and region are stable. The web tier fits a savings plan for compute, because the commitment follows hourly spend across eligible compute services rather than one VM size, so the move to containers does not waste the discount.
Related pages
Section titled “Related pages”- AZ-900 study guide overview
- Next: Azure architecture, compute and networking
- Azure basics
- Azure vs AWS vs GCP