1.4 The Consumption-Based Model
Key Takeaways
- The consumption-based model means paying only for resources actually used, with no required upfront commitment — stop using a resource and you stop paying for it.
- Azure pricing options: Pay-As-You-Go (0% discount), Reserved Instances (1 or 3 years, up to ~72% off), Savings Plans (1 or 3 years, up to ~65% off), and Spot VMs (no commitment, up to ~90% off but evictable with 30 seconds' notice).
- Deallocating (stopping) a VM stops compute charges but disk storage charges continue; only deleting the VM and its disks stops all charges.
- Inbound data to Azure is generally free; outbound data leaving Azure is metered and billed after a free allowance.
- Cost drivers tested on AZ-900: resource type, resource size/SKU, region, bandwidth/egress, reservation commitments, and Azure Hybrid Benefit.
Quick Answer: The consumption-based (pay-as-you-go) model charges only for what you actually use — metered by the second, by the GB, or per transaction. No upfront cost, no wasted idle hardware; stop a resource and the meter stops. You can optionally commit to 1- or 3-year terms for large discounts.
Why It Matters
In traditional IT you buy hardware up front against a guess about future demand. Overestimate and you waste money on idle servers; underestimate and you cannot serve users. The consumption model removes the guess: capacity follows demand, and billing follows capacity. AZ-900 will not ask you to calculate a bill — it asks which pricing option fits a scenario and which billing rules apply.
Core principles
- No upfront cost — no hardware purchase or mandatory contract (unless you choose a discount commitment).
- No wasted resources — scale down or deallocate when demand falls.
- Pay for what you use — billing tracks real consumption.
- Stop paying when you stop using — deallocate compute and the compute meter halts.
Azure Pricing Options
Pay-As-You-Go (PAYG)
The default: pay per second/hour/transaction with no commitment.
- Best for: unpredictable or short-term workloads, dev/test, proofs of concept.
- Flexibility: maximum — change or stop anytime.
- Discount: 0% (full list price).
Reserved Instances (RIs)
Commit to a specific resource (e.g., a VM size in a region) for 1 or 3 years.
- Best for: steady-state, 24/7, predictable workloads.
- Flexibility: low — tied to the reserved capacity (exchange/refund rules apply).
- Discount: up to ~72% vs PAYG.
Azure Savings Plans (for compute)
Commit to a fixed hourly dollar amount of compute spend for 1 or 3 years; the discount flexes automatically across VM families and regions.
- Best for: consistent overall compute spend but changing VM mix.
- Flexibility: medium — locked to a spend rate, not a specific SKU.
- Discount: up to ~65% vs PAYG.
Spot VMs
Buy unused Azure capacity at deep discounts; Azure can reclaim it whenever capacity is needed.
- Best for: interruptible, fault-tolerant work — batch jobs, rendering, CI/CD, dev/test.
- Flexibility: none guaranteed — Azure evicts with a minimum 30 seconds' notice.
- Discount: up to ~90% vs PAYG.
Pricing Option Comparison
| Option | Commitment | Max Discount | Best For | Key Risk |
|---|---|---|---|---|
| Pay-As-You-Go | None | 0% | Variable/short-term | Highest unit cost |
| Reserved Instances | 1 or 3 yr | ~72% | Steady 24/7 workloads | Locked to specific capacity |
| Savings Plans | 1 or 3 yr | ~65% | Consistent compute spend | Committed to a spend rate |
| Spot VMs | None | ~90% | Interruptible workloads | Evicted with 30s notice |
How Azure Meters Usage
| Resource | Metered By | Example |
|---|---|---|
| Virtual Machines | Per second of compute | A D2s_v3 running ~730 hours/month |
| Storage | Per GB-month + transactions | 100 GB of Hot-tier Blob storage |
| Networking | Per GB of outbound transfer | 50 GB egress to the internet |
| Databases | Per vCore/DTU-hour + storage | Azure SQL with 2 vCores, 730 hours |
| Functions | Per execution + GB-seconds | 1M executions at 128 MB |
Billing rules you must memorize
- Inbound data (into Azure): generally free.
- Outbound data (leaving Azure): metered and billed after a free monthly allowance.
- Deallocated (stopped) VM: compute charges stop, but disk storage charges continue.
- Deleted resources: all charges stop. To zero out a VM's cost, delete the VM and its disks.
On the Exam: The classic trap question is "a stopped VM stops all charges." False — deallocation halts compute only; the OS/data disks still cost money until deleted.
Factors That Affect Cost
| Factor | Effect on Cost |
|---|---|
| Resource type | Different services price differently (compute > storage) |
| Resource size/SKU | Larger VMs and more storage cost more |
| Region | Prices vary by region (East US vs West Europe) |
| Bandwidth/egress | Outbound transfer is metered |
| Reservations/Savings Plans | Commitments cut the unit price sharply |
| Azure Hybrid Benefit | Reuse existing Windows Server / SQL Server licenses to save |
Cost-management tooling (also tested): use the Azure Pricing Calculator to estimate before deploying, the Total Cost of Ownership (TCO) Calculator to compare on-prem vs Azure, Azure Cost Management + Billing to track and analyze actual spend, and budgets/alerts plus resource tags to govern and attribute costs.
On the Exam: You do not compute prices. You match a workload to the right option (steady 24/7 ⇒ Reserved Instance; interruptible batch ⇒ Spot; unpredictable/short ⇒ PAYG) and recall the billing rules above.
Worked Scenario: Picking the Right Commitment
An analytics company runs three distinct workloads. Its production database servers run 24 hours a day, every day, with steady load — these are textbook Reserved Instance candidates, locking in up to ~72% savings for a 1- or 3-year term because the demand is fully predictable. Its data-science team runs large, restartable model-training jobs overnight that tolerate interruption — perfect for Spot VMs at up to ~90% off, since an eviction merely reschedules the job.
Its developers spin up short-lived test environments a few hours at a time on an unpredictable schedule — best left on Pay-As-You-Go, because committing would waste the discount on capacity they do not consistently use. If the company has consistent total compute spend but keeps changing VM families, a Savings Plan captures up to ~65% without locking a specific SKU.
Azure Hybrid Benefit and Other Savings Levers
Beyond the four pricing options, AZ-900 expects you to recognize Azure Hybrid Benefit: organizations that already own Windows Server or SQL Server licenses with Software Assurance can apply them to Azure VMs and pay only the base (Linux) compute rate, avoiding a second license charge. Combined with reservations, Hybrid Benefit can stack for substantial savings. Other levers that affect the bill include choosing a cheaper region, right-sizing the VM SKU, and reducing egress by keeping traffic within a region.
Cost-Management Tooling Summary
| Tool | Purpose | Before or After Deploy |
|---|---|---|
| Pricing Calculator | Estimate the cost of a planned configuration | Before |
| TCO Calculator | Compare on-premises cost vs Azure | Before |
| Cost Management + Billing | Track, analyze, and report actual spend | After |
| Budgets & Alerts | Notify or trigger actions when spend nears a threshold | After |
| Azure Advisor (Cost) | Recommend right-sizing and reservation purchases | After |
Common Trap: The Pricing Calculator estimates a proposed deployment, while Cost Management analyzes existing spend, and the TCO Calculator compares on-prem to Azure. Match the tool to whether the scenario is planning, comparing, or monitoring — exam stems hinge on that distinction.
A company runs a fault-tolerant nightly batch job that can be safely interrupted and restarted. Which Azure pricing option minimizes cost?
An administrator deallocates (stops) a Virtual Machine through the Azure portal. Which charges CONTINUE to accrue?
Which statement about Azure data transfer billing is correct?
An organization wants to estimate the cost of an Azure solution BEFORE deploying any resources. Which tool should it use?