AZ-900
Deep Dive
The CapEx/OpEx decision isn't technical — it's financial and strategic. Understanding the consumption model means understanding why cloud changed the economics of technology.
Contents
CapEx — Capital Expenditure
Capital expenditure: acquiring physical or intangible assets that will generate value for more than a year. Recorded as an asset on the balance sheet and amortized/depreciated over its useful life.
OpEx — Operating Expenditure
Operating expense: day-to-day costs to keep the business running. Recorded directly as an expense in the period it occurs (immediate P&L). No amortization.
An analogy so you never forget it
CapEx = Buying a car
You pay $30,000 today. The car is yours. You depreciate it over 5 years = $6,000/year on the books. If you don't use it, you still paid $30k.
OpEx = Uber/car rental
You pay $50 for the ride. You only pay when you use it. If you don't need to go anywhere, you don't pay. Scales instantly.
How each model impacts a company's financial statements.
| Concept | CapEx (On-premises) | OpEx (Azure) |
|---|---|---|
| Balance sheet | The asset appears as "Property, plant and equipment." Increases total assets. | No asset recorded. Doesn't appear on the balance sheet. |
| Income statement (P&L) | Only annual depreciation (e.g., $100k/year if you bought $500k over 5 years) | The full expense for the period (e.g., $8k/month = $96k/year) |
| Cash flow | Large outflow in Year 1. Years 2–5: maintenance only. Predictable long-term. | Small, regular outflows. Variable based on usage. Predictable short-term. |
| Tax advantage | Deduction spread across years (depreciation). Lower tax benefit in year 1. | Full deduction in the year of the expense. Higher immediate tax benefit. |
| Impact on debt / leverage | If financed with debt, increases liabilities. Can affect financial ratios. | No additional debt. Better leverage ratios. |
| EBITDA | Depreciation adds back to EBITDA (excluded from earnings before D&A) | Operating expense reduces EBITDA directly |
| Flexibility for changes | Low. Purchased hardware is hard to reduce or change. | High. Reduce or stop services at any time. |
The CFO's angle
For many CFOs, moving IT to OpEx is attractive because: (1) it eliminates large capital outflows that affect the balance sheet, (2) it improves return-on-assets (ROA) metrics by not having IT assets depreciating, (3) it increases financial agility — you can grow or reduce spending without long-term commitments.
Azure implements OpEx through the consumption model: you pay exactly for what you use, with second-level granularity in many services.
How consumption is measured in Azure
VM B2s = $0.0416/hour. Powered off = $0/hour (only pays for disk)
LRS Blob: $0.018/GB/month + $0.004 per 10k operations
General Purpose 2vCore: $0.3716/hour. Paused = minimum (serverless)
First 1M executions/month: FREE. Then $0.20 per M executions
First 5 GB/month free. Then $0.087/GB (zone 1)
Tools to predict and control costs
Azure Pricing Calculator
azure.microsoft.com/pricing/calculator
Estimate before creating resources. Combines services, regions, and configurations. Exports budget as PDF/Excel.
Before migrating or starting a new project
Azure Cost Management + Billing
Portal → Cost Management
Analysis of actual spend. Breakdowns by subscription, resource group, tag, service. Historical trends.
Continuous monthly monitoring of actual spend
Azure Budgets and Alerts
Portal → Cost Management → Budgets
Set a monthly/quarterly budget and get alerts at 70%, 80%, 90%, 100%. Doesn't stop resources automatically by default.
To avoid bill surprises
Azure Advisor — Costs
Portal → Advisor
Automatic recommendations: underused VMs, Reserved Instances where worthwhile, orphaned disks, unused IPs.
Periodic (monthly) spend optimization
Mid-size company: 50 servers, 500 users, standard workload. Not a huge enterprise, not a startup.
On-Premises (CapEx)
Year 5: hardware depreciated, refresh cycle = another $750k
Azure (OpEx / Consumption)
With 3-year Reserved Instances on compute: additional ~35% savings
Conclusion of the analysis
Year 1
Azure is ~62% cheaper. The large upfront investment in on-premises hardware is the dominant factor.
Years 1–5 (total)
Azure is ~22% cheaper in this scenario. The gap narrows over time as hardware amortizes.
Year 6+
On-premises can be cheaper IF the workload is stable and predictable. But hardware needs a refresh (new CapEx cycle).
The AZ-900 does NOT say cloud is always cheaper. These are the situations where on-premises can win.
Stable 24/7 workloads with no variation
If you have a database server running at 80% CPU constantly, with no spikes, every day of the year — the consumption model gives you no advantage. You pay the same amount always. On-premises can be cheaper because you amortize the hardware.
Real example
A SAP ERP server that runs exactly the same on Monday at 9am as on Saturday at 3am.
Solution/mitigation
For this case: Reserved Instances in Azure (up to 72% discount with a 1-3 year commitment).
No active cost optimization
Azure without active management can be extremely expensive. VMs nobody shuts down at night, orphaned disks from deleted resources, accumulating snapshots, unused public IPs — it all adds up.
Real example
A company migrates 50 VMs to Azure via "lift and shift" without rightsizing. Pays for D8s_v3 when it needed B2s. Spends 4x more than necessary.
Solution/mitigation
FinOps: Azure Cost Management + Azure Advisor + a culture of shutting down non-productive resources outside business hours.
Unexpected egress (data transfer out)
Transferring data out of Azure (egress) costs money. The first 5 GB/month are free. Then ~$0.087/GB. For applications that move terabytes of data out of Azure (e.g., large-scale content distribution), this cost can be significant.
Real example
A streaming app serving 50 TB/month of video to users → egress cost ~$4,000/month in bandwidth alone.
Solution/mitigation
Azure CDN reduces egress costs. Inbound (data coming into Azure) is free.
Azure offers a hybrid model: you can commit to 1 or 3 years of usage and get massive discounts. Technically this comes close to the CapEx model within cloud.
Pay-as-you-go (PAYG)
No commitment. Pay by the hour. More expensive per unit.
Savings: 0%
Development, testing, unpredictable workloads
Reserved — 1 year
1-year commitment. Upfront or monthly payment.
Savings: ~40% vs PAYG
Stable workloads that have run unchanged for 6+ months
Reserved — 3 years
3-year commitment. Bigger discount, less flexibility.
Savings: ~60–72% vs PAYG
Stable core infrastructure (DBs, AD, critical servers)
Are Reservations CapEx or OpEx?
An interesting accounting dilemma: if you pay for a 3-year Reserved Instance upfront ($10,000), many accountants treat it as CapEx (upfront capital commitment). If you pay monthly even though it's "reserved," some treat it as OpEx. The AZ-900 mentions Reservations as a way to reduce costs without going deep into the accounting classification — but knowing this sets apart those who truly understand the topic.
Did you get this topic?
Test what you just learned
A company migrates its infrastructure to Azure to eliminate purchasing physical servers. How does its spending model change?
Put what you learned to the test with original AZ-900 exam questions.
Start AZ-900 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.