The Shared Responsibility Model & Cloud Pricing
AZ-900 AZ-900 · Describe cloud concepts
Who is responsible for what?
The single most tested idea in the cloud-concepts domain is the shared responsibility model: in the cloud, security and management duties are divided between the cloud provider and the customer, and the dividing line moves depending on which service type you use. The exam tests this relentlessly, usually with a scenario: "A company runs a workload on IaaS. Who is responsible for patching the operating system?" If you can answer that family of questions, you have banked a significant share of the domain.
The model has a fixed anchor and a sliding line. The anchor: the customer is always responsible for its own data, accounts, and access — no service type hands those to the provider. The provider is always responsible for the physical infrastructure: the buildings, the power, the cooling, the physical hosts, and the physical network. Between those two anchors, responsibility slides with the service type. The more the provider manages, the less the customer manages — and the customer always keeps the top of the stack.
| Layer | On-premises | IaaS | PaaS | SaaS |
|---|---|---|---|---|
| Data and accounts | Customer | Customer | Customer | Customer |
| Applications | Customer | Customer | Customer (code) / Provider (platform) | Provider |
| Runtime / middleware | Customer | Customer | Provider | Provider |
| Operating system | Customer | Customer | Provider | Provider |
| Virtualisation | Customer | Provider | Provider | Provider |
| Servers, storage, network (physical) | Customer | Provider | Provider | Provider |
Read the table as a story of giving things up. On-premises, the customer owns everything. With IaaS (a virtual machine), the provider takes the physical layer and virtualisation, and the customer still owns the operating system and everything above it — including patching that OS. With PaaS (a web app service), the provider takes the OS and runtime too, and the customer owns its code and data. With SaaS (an email service), the provider takes nearly everything and the customer owns its data, its user accounts, and who can access what. The exam's favourite trap is a SaaS question that tempts you to hand the provider responsibility for the customer's data. Never do that: data is always the customer's job.
CapEx vs OpEx: the money model behind the cloud
The second big idea is how cloud changes spending. Traditional IT is dominated by capital expenditure (CapEx): buying servers, storage, and datacenter space upfront, then depreciating them over years. The money is spent before any value arrives, capacity must be guessed in advance, and unused capacity is wasted money. Cloud computing shifts spending to operational expenditure (OpEx): paying for services as they are consumed, month by month, with no upfront purchase. This is the consumption-based model — you pay for what you use, when you use it, and you stop paying when you stop using it.
The practical consequences matter more than the vocabulary. With OpEx, a company can start small and grow spending with usage, which lowers the cost of trying something new. It can delete a test environment on Friday and stop paying for it — there is no hardware to sell or write off. And it can match costs to revenue: a retailer pays more for computing in December because it sells more in December. The exam tests this with scenarios about a company that wants to avoid upfront hardware spending, or that runs a short-lived project, or whose costs should track its seasonal sales. In each case, the consumption-based model is the point.
That said, OpEx is not automatically cheaper. A workload that runs at a steady, predictable level 24 hours a day, every day, can cost less on hardware the company owns — or, in the cloud, less with a commitment. Which brings us to the pricing options.
Comparing cloud pricing models
Azure offers several ways to pay, and the exam expects you to match each to a situation:
- Pay-as-you-go: no commitment, billed per second or per hour of use. Best for unpredictable workloads, short projects, and anything you might delete soon. Highest per-unit price, maximum flexibility.
- Reserved instances: commit to one or three years of a virtual machine (or another resource) in a region and get a large discount — up to around 72% for VMs. Best for steady, predictable workloads you know will run for a long time.
- Azure savings plans: commit to a fixed hourly spend for one or three years, with the flexibility to apply the discount across different VM sizes and regions. Best when usage is steady in total but the exact resources may change.
- Azure Hybrid Benefit: reuse existing on-premises Windows Server or SQL Server licences in Azure instead of paying for new ones. Best for companies with licence investments they already own.
- Spot pricing: buy unused Azure capacity at deep discounts, with the catch that Azure can reclaim the capacity with short notice. Best for interruptible work such as batch processing, rendering, or test runs that can restart.
Notice the pattern: commitment trades flexibility for price. The more you commit, the less you pay per unit — and the more you lose if your needs change. Exam questions usually give you the workload's shape (steady, spiky, short-lived, interruptible) and ask which pricing model fits. Match the shape, not the brand name.
Serverless: the extreme end of the model
Serverless computing takes the shared-responsibility slide to its limit for code: the provider manages the servers entirely, and the customer supplies only code that runs in response to events. There are still servers — the name means the customer never sees or manages them. The billing matches the model: you pay per execution and for the compute consumed while the code runs, and you pay nothing when nothing runs, because the service scales to zero. Azure Functions is the exam's serverless example. Typical uses are event-driven work: processing a file the moment it lands in storage, responding to a message in a queue, running a short task on a schedule, or gluing services together. If a scenario describes code that should run only when something happens, and the team wants no servers to manage, serverless is the answer — not a virtual machine, which would sit idle (and billable) between events.
Worked example: putting the pieces together
A company runs a nightly report that crunches the day's sales data. It takes 20 minutes, runs once a day, and nobody should manage a server for it. The team proposes a virtual machine left running all day. The better design is serverless: an Azure Function triggered on a schedule, paying for 20 minutes of compute a day instead of 24 hours of a VM. The shared-responsibility lens explains why: with the VM (IaaS), the team owns the OS, patching, and the runtime; with the Function, the provider owns all of that and the team owns only the code and the data. The pricing lens agrees: pay-as-you-go serverless fits a short, scheduled burst far better than a continuously running VM, and a reservation would be pointless for 20 minutes a day. One scenario, three exam concepts, one answer.
Exam traps to watch for
- Handing data responsibility to the provider. In every service type — IaaS, PaaS, SaaS — the customer is always responsible for its own data, accounts, and access.
- Saying the customer patches the OS on PaaS. On PaaS the provider manages the operating system; the customer patches the OS only on IaaS (and on-premises).
- Thinking serverless means there are no servers. Servers exist; the provider manages them and the customer never sees them. Billing is per execution, and the service can scale to zero.
- Recommending reserved instances for unpredictable workloads. Reservations reward steady, predictable usage with a discount; unpredictable or short-lived workloads belong on pay-as-you-go (or spot, if interruptible).
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.