AZ-900
Deep Dive
Formal definition, NIST characteristics, real benefits, consumption model, and economies of scale. Everything the AZ-900 tests about cloud fundamentals.
Contents
Cloud computing is the delivery of technology services — including servers, storage, databases, networking, software, analytics, and intelligence — over the internet, with pay-as-you-go pricing.
Microsoft's official definition for the AZ-900. Memorize the three pillars: delivery over the internet, IT resources, pay-as-you-go.
Before (On-premises)
Problem
Solution: Cloud
The National Institute of Standards and Technology (NIST) defines cloud with 5 characteristics that distinguish it from traditional hosting. The AZ-900 tests them directly.
On-demand self-service
On-demand self-serviceA consumer can unilaterally provision computing capabilities — such as server time and network storage — as needed automatically, without requiring human interaction with each service provider.
In Azure
Azure Portal → create a VM in 3 clicks. Azure CLI → vm create in 30 seconds. No calling anyone.
Exam trap
Don't confuse this with "self-managed" — you still need Microsoft for the hardware.
Broad network access
Broad network accessCapabilities are available over the network and accessed through standard mechanisms that promote use by heterogeneous thin or thick client platforms (e.g., mobile phones, laptops, tablets).
In Azure
Azure from any country, any device, standard HTTPS. Azure Portal on Safari on iPhone or Firefox on Linux.
Exam trap
"Broad" means standard/universal, NOT that you have unrestricted access to all resources.
Resource pooling
Resource poolingThe provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically assigned and reassigned according to consumer demand.
In Azure
Your VM shares the same physical server with VMs from other companies. The Hyper-V hypervisor isolates them. You have no idea where your VM is physically located.
Exam trap
You can't choose which physical rack. But you can choose the region (and sometimes the availability zone).
Rapid elasticity
Rapid elasticityCapabilities can be elastically provisioned and released — in some cases automatically — to scale rapidly outward and inward according to demand. To the consumer, the capabilities available often appear unlimited.
In Azure
Azure VM Scale Sets: Monday 9am → high load → 50 VMs. Sunday 3am → low load → 2 VMs. Auto-scaling based on CPU/RAM/custom metrics.
Exam trap
"Unlimited" in the definition = unlimited from the consumer's perspective. Azure does have subscription limits (soft limits Microsoft can raise).
Measured service
Measured serviceCloud systems automatically control and optimize resource use by leveraging a metering capability at some level of abstraction appropriate to the type of service. Resource usage can be monitored, controlled, and reported, providing transparency for both the provider and the consumer of the service used.
In Azure
Azure Cost Management: see exactly how many GB of storage you used, how many VM hours, how many API requests. Granularity from hours down to seconds.
Exam trap
"Measured" does not mean "limited." You can use as much as you want; it just gets measured (and billed) in full.
The AZ-900 tests these benefits directly. Learn to distinguish them and choose the right one for each scenario.
High availability
Uptime guarantee through an SLA. Azure offers 99.9% to 99.99% depending on the service. Measured in "nines": 99.9% = 8.7h downtime/year. 99.99% = 52min/year.
Azure example
Azure VM SLA with Premium SSD + Availability Zones = 99.99%
Scalability
Ability to increase resources to handle greater demand. Vertical: more CPU/RAM to one server. Horizontal: more servers.
Azure example
Azure VM Scale Sets: scales from 2 to 200 instances in minutes.
Elasticity
Automatically scaling both up AND down based on actual demand. The distinction vs scalability: elasticity is automatic and bidirectional.
Azure example
Black Friday: 500 instances. Tuesday 3am: 2 instances. No intervention.
Agility
Speed to deploy and change technology resources. From weeks/months (on-premises) to minutes (cloud).
Azure example
MVP in production in hours. Test environment in 5 minutes. Rollback in seconds.
Geo-distribution
Deploy across multiple global regions to be close to users and meet data residency requirements.
Azure example
Global app: instance in East US for LATAM, West Europe for Europe, Southeast Asia for APAC.
Disaster recovery
Built-in BCDR (Business Continuity and Disaster Recovery). Azure Site Recovery replicates workloads to another region for automatic failover.
Azure example
RPO of minutes, RTO of hours — without maintaining a second physical datacenter of your own.
Security
Microsoft invests +$4B/year in security. SOC 1/2/3, ISO 27001, PCI DSS, HIPAA certifications. Automatic security updates.
Azure example
Microsoft Defender for Cloud detects threats 24/7. You don't need a dedicated SOC team.
Manageability
Reduced operational work. Automatic patches, platform updates without downtime, built-in monitoring.
Azure example
Azure PaaS: zero OS patches. Azure SaaS: zero app updates. You just write code.
Predictability
Two types: cost predictability (Azure Pricing Calculator, Cost Management) and performance predictability (SLAs, Azure Monitor).
Azure example
Monthly budget with alerts at 80%/90%/100%. No hardware surprises.
Common traps about benefits
The consumption model is the economic foundation of cloud. You pay only for the resources you use, when you use them. It's the reason startups can compete with enterprises.
Traditional model (fixed)
You pay for what you MIGHT need
Consumption model (cloud)
You pay for what you ACTUALLY use
Real example: e-commerce during Black Friday
January–October
5 servers
$500/month
November (BF)
50 servers
$5,000/month
December
8 servers
$800/month
With on-premises: you would have bought 50 servers for BF. Fixed cost = $5,000/month all year even though you only use 5 in January.
With Azure: you pay exactly what you use each month. Savings of ~$50k over a year.
Economies of scale are one of the fundamental reasons cloud can be cheaper than on-premises for most businesses.
Why can Azure offer cheaper resources than if you bought them yourself?
Massive hardware purchases
Microsoft buys millions of servers per year. They negotiate prices no individual company can get. A server you would buy for $5,000 costs Microsoft ~$800.
Operational efficiency
A Microsoft datacenter has thousands of servers managed by a few highly specialized technicians. A typical enterprise datacenter has dozens of servers managed by a relatively costly team.
High average utilization
On-premises: companies use 20-40% of capacity. Microsoft: with millions of different customers, workloads "smooth out" → utilization of 60-80%. More efficient = cheaper per unit.
Energy and cooling at scale
Microsoft builds datacenters in places with cheap energy (Sweden hydro, Iowa wind). They negotiate electricity prices no company can get. PUE (Power Usage Effectiveness) of 1.12 vs a typical enterprise 1.5+.
R&D amortization
The cost of developing Azure is amortized across millions of customers. You get access to cutting-edge technology (Cosmos DB, Fabric, OpenAI) without paying the full development cost.
For the exam
"Economies of scale" → answer = the cloud provider can offer lower prices because it buys and operates at massive scale, spreading costs across millions of customers. The benefit is passed on to the customer as lower prices.
| Dimension | On-Premises | Azure Cloud |
|---|---|---|
| Initial investment | $500k–$2M for a medium datacenter | $0. Pay as you use. |
| Spending model | CapEx: depreciable assets | OpEx: monthly operating expense |
| Provisioning time | 2–6 months (purchase + installation) | Minutes or seconds |
| Scaling | Buy hardware → 6–12 weeks | Auto-scaling in seconds |
| Hardware maintenance | Your team: patches, replacements, cooling | Microsoft (invisible to you) |
| OS patching | Your team: plans, tests, applies | Automatic (in PaaS/SaaS) |
| Physical security | You: CCTV, biometrics, guards | Microsoft: ISO 27001, SOC2 certifications |
| Availability | SLA you define (and pay for) | 99.9%–99.99% on Microsoft SLA |
| Unused capacity | Keeps costing the same | Don't provision it = $0 |
| Global access | Requires costly dedicated WAN | 60+ regions. Deployment in hours |
| Compliance | You obtain and maintain certifications | Azure has 90+ inheritable certifications |
| Innovation | Limited to your R&D budget | Immediate access to AI, ML, IoT, Quantum |
| Control | Total. Every byte, every config. | Shared depending on model (IaaS/PaaS/SaaS) |
| Vendor lock-in | Specific hardware, but you own the data | Provider dependency (mitigable with containers) |
| Internet dependency | Works without internet (LAN) | Requires connectivity (ExpressRoute for high availability) |
The AZ-900 does NOT say cloud is always better
Cloud is better for MOST cases. On-premises is still valid when: (1) ultra-low latency is critical and you can't tolerate the internet, (2) regulatory requirements demand data PHYSICALLY on your premises, (3) stable and predictable long-term workloads where amortized CapEx is cheaper, (4) data so sensitive you don't trust it to any external provider.
Common myths that appear as distractors on the exam.
Cloud is always cheaper than on-premises
FALSE. Without optimization, cloud can be more expensive. A depreciated on-premises server can be cheaper than an Azure VM for stable 24/7 workloads. The key is active cost management.
Cloud means "no maintenance"
FALSE. It eliminates hardware maintenance. But you remain responsible for your code, your data, your configurations, your IAM. Maintenance shifts, it doesn't disappear.
Cloud is less secure because data is "on the internet"
FALSE in most cases. Azure has more security resources than most companies. The problem is misconfigured permissions on the customer's side, not the platform.
Cloud = no servers (serverless)
FALSE. "Serverless" means you don't manage servers, but servers still exist. IaaS has servers. Cloud includes IaaS, PaaS, SaaS, and serverless.
Moving to cloud means total loss of control
FALSE. In IaaS you have almost the same control as on-premises. In PaaS you give up the OS. In SaaS you give up more. It's a spectrum, not binary.
Cloud is only for large companies
FALSE. The consumption model makes cloud accessible to one-person startups. In fact, economies of scale benefit small companies more (they can't invest in their own datacenters).
Did you get this topic?
Test what you just learned
A company needs to automatically adjust its server capacity based on daily traffic. Which cloud benefit describes this capability?
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.