“The cloud” sounds weightless, but it is really one of the most physical things about modern computing: rows of humming server racks in warehouse-sized buildings, drawing enough electricity to power small cities, sliced up and rented out to millions of people who never see them.

There Is No Cloud, Just Someone Else’s Computer

Cloud computing means renting computing power, storage, or software that runs on servers owned and maintained by someone else, accessed over the internet instead of running on your own machine. Those servers sit in data centers, physical facilities that companies like Amazon, Microsoft, and Google operate around the world, and when you save a file to Google Drive or stream a show on Netflix, you are really reading and writing data on a server in one of these buildings, often hundreds or thousands of miles away.

The Trick That Makes It Affordable: Virtualization

A single physical server sitting idle most of the time would be a poor use of very expensive hardware, so cloud providers rely on virtualization: software that carves one physical machine into multiple virtual machines, each behaving like an independent computer with its own operating system, even though they all share the same underlying processors, memory, and storage. Every virtual machine is sandboxed from the others on the same hardware, so one customer’s data and applications stay invisible to the next. This is what allows a single data center to effectively act like thousands of separate computers serving thousands of separate customers at once, which is what makes cloud computing so much cheaper than everyone buying and running their own servers.

Three Layers You Can Rent

Cloud services generally break down into three tiers, distinguished by how much a customer manages versus how much the provider handles. Infrastructure as a Service (IaaS), like Amazon EC2, rents raw virtual servers and storage — the customer installs and manages their own operating system and software on top. Platform as a Service (PaaS), like Google App Engine, goes a layer further, handling the operating system, runtime, and databases so developers can just deploy their code. Software as a Service (SaaS), like Slack or Dropbox, hands over a finished, ready-to-use application where the provider manages literally everything and the customer just logs in.

Redundancy Is the Whole Point

Because hardware fails and buildings can lose power, cloud providers duplicate data and services across multiple physical servers and often multiple geographic regions. If one server, or even an entire data center, goes offline, traffic automatically shifts to a working copy elsewhere, which is why a cloud outage severe enough for you to notice usually means something has gone wrong at a scale far larger than a single broken machine. That built-in redundancy, more than any individual piece of hardware, is what lets services promise the kind of near-constant uptime that would be difficult and expensive for any single company to guarantee on its own.