There is no actual cloud. Every photo, email and streamed show "in the cloud" is stored on a real computer in a real building, one that belongs to someone else. This guide explains how that works, step by step, with diagrams and worked examples.
When you press "back up" on a photo, where does it actually go? The honest answer is not "into the sky". It travels across the internet to a building full of computers owned by a company, and it is written onto a drive there. Cloud computing means using processing power, storage and software that live on those faraway machines instead of on the device in your hand. This article builds the idea from the ground up: the buildings, the machines, the clever trick that lets many customers share one machine, the three main kinds of service, and the trade-offs you should know about. No coding is needed.
Helpful background: the hardware parts mentioned here (processor, memory, storage) are explained in How Computers Work, and the way data travels in small pieces is explained in What Is the Internet, Really?. You can read this article without them, but they make it easier.
The quickest way to understand the cloud is to follow a single photo on its journey. Imagine you have taken a photo and your phone is set to back it up online.
A photo is 4 MB and your connection can send data at 8 Mbps.
Step 1: Match the units. File sizes use bytes (capital B). Internet speeds use bits (small b). There are 8 bits in 1 byte. So 4 MB = 4 × 8 = 32 megabits.
Step 2: Divide size by speed. Time = 32 megabits ÷ 8 megabits per second = 4 seconds.
Step 3: Scale up. For 100 such photos: 100 × 4 s = 400 s, which is 6 minutes 40 seconds.
Reality check: this is the best case. Real uploads usually take longer, because some of the data sent is labelling, many devices share the connection, and wireless speed rises and falls.
Network engineers have long drawn diagrams of connected computers. When part of a network was complicated or did not matter for the point being made, it was often drawn as a cloud shape. The shape meant "some network is here, and we do not need to show the details." The internet itself was commonly drawn this way, and the word "cloud" stuck to services that are reached through it.
A widely used formal definition comes from the United States National Institute of Standards and Technology (NIST). It describes cloud computing as a model for getting convenient, on-demand network access to a shared pool of computing resources, such as servers, storage and applications, which can be quickly provided and released with very little effort from the provider. Put simply: cloud computing is renting computing resources over a network, and using only what you need, when you need it.
Server: a computer whose job is to provide a service to other computers.
Client: the device or app that asks a server for something. Your phone is a client.
Data centre: a building designed to house many servers with power, cooling and network connections.
Latency: the delay between sending a request and starting to get a reply.
Provider: the company that owns and runs the cloud machines.
A server is less mysterious than it sounds. It usually contains the same kinds of parts as an ordinary computer: a processor, memory, storage and a network connection. The differences are that it is built to run all day and night, it normally has no screen or keyboard, and it is fitted into a metal frame called a rack alongside many others.
A data centre is not just a room of computers. Several systems work together so that the machines keep running:
Large providers run data centres in many places, and there are three main reasons:
Here is a puzzle. A small business may need only a fraction of a powerful server. If every customer received a whole physical machine, most of the hardware would sit unused. The solution is virtualisation.
Virtualisation uses a special program called a hypervisor. The hypervisor sits on top of the physical hardware and divides it into several separate virtual machines (VMs). Each virtual machine behaves like a complete computer with its own operating system and programs, but it is really a slice of one real server. The hypervisor keeps the slices apart, so one customer's virtual machine cannot see another's.
Virtualisation is one of the main reasons cloud computing is flexible. A virtual machine is just software, so it can be created in minutes, copied, moved to different hardware, or deleted when it is no longer needed. This is very different from buying, delivering and installing a physical computer.
Not every website is cloud computing. The NIST definition lists five essential characteristics. Here they are in plain language:
| Characteristic | What it means in everyday terms |
|---|---|
| On-demand self-service | You can get more computing resources yourself, when you want them, without phoning someone and waiting. |
| Broad network access | The service is reached over a network, from many kinds of device such as phones, tablets and laptops. |
| Resource pooling | The provider's machines are shared among many customers, and you do not choose or see which exact machine you get. |
| Rapid elasticity | Capacity can grow or shrink quickly as your needs change. |
| Measured service | Use is measured, so you pay for (or are limited to) what you actually use, like water or electricity. |
Elasticity solves a problem that is easy to miss. Demand for a website is rarely steady. An online ticket shop may be quiet for weeks and then be flooded with visitors for an hour when a popular event goes on sale. A school results website may be busy on one day of the year.
If a company owns its servers, it must buy enough for the busiest moment, and that equipment sits mostly idle the rest of the time. With the cloud, it can rent extra capacity for the busy period and release it afterwards. The animation below sweeps across a year of invented demand to show the difference.
An online shop needs 2 servers on a normal day, but 10 servers for 3 sale days each year. We count capacity in "server-days": one server running for one day.
Option A: own 10 servers all year. 10 × 365 = 3,650 server-days owned.
Option B: rent what is needed. Normal days: 2 × 362 = 724. Sale days: 10 × 3 = 30. Total = 754 server-days.
Compare: 3,650 − 754 = 2,896 server-days of capacity would sit idle under Option A. That is about 79% of what was owned (2,896 ÷ 3,650 ≈ 0.79).
Be fair: this ignores price. A rented server-day can cost more than an owned one, and owning has other advantages. The example only shows why elasticity is attractive when demand is uneven. The true answer depends on real costs and how steady the demand is.
Cloud providers offer different amounts of the "stack", which is the set of layers needed to run an application. The more the provider handles, the less you manage yourself.
| Model | What you get | Everyday picture |
|---|---|---|
| IaaS (Infrastructure as a Service) | Basic building blocks: virtual machines, storage and networking. You install and manage the operating system and everything above it. | A company renting virtual servers to run its own software. |
| PaaS (Platform as a Service) | A ready-made environment to build and run applications. The provider manages servers and operating systems; you manage your application and its data. | A developer uploading a web app and letting the platform run it. |
| SaaS (Software as a Service) | A finished application used through a browser or app. The provider manages nearly everything. | Web email, online documents, video calls. |
NIST also describes four ways a cloud can be set up, called deployment models:
Redundancy means keeping spare copies so that one failure does not destroy the only copy. Cloud storage systems commonly store data on more than one drive, and sometimes in more than one location. That is a big reason why losing a phone no longer has to mean losing every photo on it.
However, people often confuse two different features:
Sara keeps her school project in a synced folder on her laptop, her tablet and an online account.
Step 1: On Monday the file exists on all three. If the laptop breaks, she still has two copies. Sync helped.
Step 2: On Tuesday she accidentally deletes the project on her tablet.
Step 3: The sync system does exactly what it is designed to do. It copies the deletion to the laptop and the online account.
Result: all three places now lack the file. Many services keep a "recycle bin" or version history for a limited time, but this varies, so it should not be relied on as the only protection.
Lesson: sync is not a backup. A widely used rule of thumb is the "3-2-1" approach: keep 3 copies of important files, on 2 different kinds of storage, with 1 copy kept somewhere else (off-site).
Many services also use a content delivery network (CDN). A CDN is a group of servers in many places that keep copies of popular content, such as images, web pages or videos. When you request something, it is often sent from a copy that is physically nearer to you, which cuts the delay and reduces the load on the original server.
When many people watch the same popular video, the service does not need to send every viewer a copy from one distant machine. Large video services commonly keep copies on servers in many locations and direct each viewer to a suitable one. This is one reason a video can start quickly even when millions of people use the service.
Because a cloud service is used by so many people, how often it is available matters. Reliability is often described as a percentage of time the service works, such as "99.9%". People call these figures "nines". Redundancy makes failures less likely to cause trouble, but it does not make them impossible. Large providers have had outages from time to time, caused by faults, mistakes or events outside their control.
A year has 365 × 24 = 8,760 hours.
Step 1: 99.9% available means 0.1% unavailable.
Step 2: 0.1% of 8,760 = 8,760 × 0.001 = 8.76 hours of downtime per year.
Step 3: For 99.99%, the unavailable part is 0.01%: 8,760 × 0.0001 = 0.876 hours, which is about 52.6 minutes (0.876 × 60).
Meaning: each extra "nine" cuts the allowed downtime by a factor of ten. This is only arithmetic. It does not tell you what any particular provider promises or achieves, so always read the service's own terms.
Putting data in the cloud does not remove the need to protect it. Cloud security is usually described as a shared responsibility: the provider protects the buildings, hardware and core systems, and the customer protects what they put there and how they use it.
A very common cloud problem is not a clever attack but a simple mistake: a file or folder shared with "anyone with the link", and the link ends up somewhere public. Before sharing, check who can see it.
Privacy is a separate question from security. Providers have to follow data protection laws, which differ between countries, and each service has a privacy policy explaining what it does with your data. To go deeper, read Cybersecurity Basics and Digital Footprint & Online Privacy.
The cloud is useful, but it is not free of costs or risks. A balanced view includes:
Because heavy work can be done remotely, a modest phone or laptop can use services that would be too demanding to run locally. The device mainly shows results and handles taps and typing, while remote servers do the harder work. Many artificial intelligence services also run in data centres, because large AI models need far more computing power than a typical personal device has. This is one reason cloud computing and AI are so often mentioned together.
Try each question first, then tap it to reveal the answer. A scored multiple-choice quiz follows below.
It is computing resources (processing, storage, software) on servers in data centres owned by a provider and used over the internet, instead of on your own device.
Any two of: lower delay for nearby users; keeping the service running if one site fails; following laws about where data may be stored.
It divides one physical server into several separate virtual machines and keeps them isolated from each other.
6 MB × 8 = 48 megabits. 48 ÷ 12 = 4 seconds.
SaaS, because it is a finished application used through a browser or app, with the provider managing almost everything underneath.
Because deletions and mistakes are copied to every synced device, so all copies can lose the file together. A backup keeps separate copies and older versions.
Class activity (no accounts needed): let students act out Figure 1. Some are the "phone", some are routers, one is the "server". Write a short message on cards, split it into numbered packets, pass them along the line and ask the server to reassemble them in order. Then discuss what would change if one router were missing.
Discussion prompts: What files should have a backup? Who can see a file shared by link? What is stored offline on your device and what depends on the internet?
Safety note: many online services set minimum ages for accounts and may need a parent's permission, so check each service's terms before students sign up. Please do not have students upload personal photos or documents for classroom demonstrations.
This article is original EDUSAMBAM writing. The definition, five characteristics, three service models and four deployment models follow the standard reference below. The other ideas (data centres, virtualisation, sync versus backup, shared responsibility) are widely taught concepts of computing, explained here in our own words. All diagrams are original EDUSAMBAM schematics, and all numbers in the worked examples are invented or are simple arithmetic for practice.
Because technology services change, check any provider's current terms, limits and prices for yourself. This article does not recommend or compare particular companies.
10 questions. Select an answer for each, then submit to see your score instantly.