Not a technology but a way of paying — and the bill is a design decision. Ends by separating all four subjects honestly.
That is genuinely all it is. The machines are ordinary servers in ordinary buildings. What changed was not the hardware — it was the billing model, and that change was enormous.
| Buying your own | Renting | |
|---|---|---|
| Money | Millions up front | Per hour, per gigabyte |
| Time to start | Weeks — order, ship, clear customs, install | Ninety seconds |
| Getting it wrong | You own the wrong machine for five years | Delete it, stop paying |
| Busy season | Buy for the peak, waste it the rest of the year | Scale up, scale back down |
Notice that none of those rows mention a technical capability. Cloud let people try things, and that is what changed the industry.
The romantic version is that Amazon had spare capacity after Christmas and sold it. The real reason is more instructive.
Amazon's own engineering teams were being held up for months waiting for servers to be provisioned. The company's actual problem was internal: their developers could not move. So they built a system where any team could get a machine in minutes.
Then they noticed every other company had the same problem. S3 launched in March 2006, EC2 in August. Within fifteen years it was more profitable than the shop.
You install everything, patch everything, secure everything. Most control, most work. Choose it when you need something unusual, or when the cost at scale justifies the labour.
No patching, no server administration. You give up some control and pay more per unit of computing. For most small teams this is the correct answer, because an engineer's time costs more than a server.
Gmail, Paystack, Supabase. You are not operating anything. Perfect until you need something the vendor will not do, and then you have no options at all.
Somewhere between PaaS and SaaS, and widely dismissed by people who have never had to pay for infrastructure themselves. For a Nigerian business it is often exactly right: a few thousand naira a month, no dollar exposure, no surprise bill, and support you can telephone. The six patterns are identical on it. Learn where it is cheap.
An engineer starts a large instance to try something on a Friday and forgets it. Nobody notices for two months. That is a completely ordinary, extremely common way to lose serious money. Set spending alerts before you set anything else up.
Uploading data is usually free. Downloading it out of the provider is charged per gigabyte. A media-heavy service can find that bandwidth, not computing, is its largest line. It is also, not coincidentally, what makes leaving a provider expensive.
From Lagos, a round trip to a European region is roughly 90–150 milliseconds. To the US east coast, closer to 200–300. On a page making ten sequential calls, that difference is felt as "this site is slow" — and no amount of code optimisation recovers physics. Choose the region nearest your users, and measure it rather than assuming.
A cloud bill fixed in dollars while you earn in naira means your infrastructure cost rises whenever the rate moves, with no warning and no relationship to your usage. This is a real business risk that engineering courses written abroad never mention, and it is a legitimate reason to prefer local hosting for steady workloads.
Almost nothing conceptually — and everything practically.
Training a large model needs hardware that costs a fortune to buy and sits idle between projects. Renting it by the hour is what put serious training within reach of anyone but a corporation.
A model on your laptop earns nothing. Cloud is how it becomes an endpoint that a stranger can call.
Fleet coordination and analytics belong in the cloud. The control loop does not — a robot that must wait for a network round trip before deciding to stop is a dangerous robot. Anything safety-critical stays local. Always.
Practical exercise. Take anything you have deployed and answer three questions in writing: what does it cost per month, what happens if it stops at 2am, and can you restore it from backup today. Most people cannot answer any of the three — and being able to is what makes someone trusted with production.
| Its subject | Its ancestor | Fails as | |
|---|---|---|---|
| AI | judgement under uncertainty | logic, philosophy | a wrong decision, confidently made |
| ML | rules learned from examples | statistics | a model that quietly stops matching the world |
| Robotics | sensing and acting physically | control engineering, 1788 | something breaks, possibly someone |
| Cloud | renting computing | economics and operations | a bill, or an outage at 2am |
These are four crafts. Different histories, different mathematics, different failures, different daily work. Someone excellent at one may be a beginner at another, and that is normal.
Module 1 taught you the shared skeleton so that you are never starting from nothing. These four modules taught you the differences so that you never mistake reading about a field for practising it.
The Practitioner track is built for adults. It assumes you are ready to deploy real systems that real strangers will use, and to be answerable for what they do.
We ask because in Nigeria you become an adult at 18, and this track is a paid commitment. Nothing here is stored for anyone under that age.