Sharing an expensive machine
In the 1950s a computer filled a room and cost more than most organizations could afford. Programs were usually run in batches: a user handed in a deck of cards and came back hours later for the printout. A single mistake could cost a day.
Time-sharing was developed to fix this. A fast computer could switch between many users so quickly that each seemed to have it to themselves, working from a terminal. At the Massachusetts Institute of Technology, the Compatible Time-Sharing System was demonstrated in 1961, and a larger project called Multics followed in the mid-1960s. Several companies began selling time on shared machines to customers who could not buy their own, which is the earliest commercial form of renting computing power remotely.
The utility idea
At about this time, people began comparing computing with electricity and water. John McCarthy, better known for his work in artificial intelligence, is often quoted as suggesting around 1961 that computing might someday be organized as a public utility. Other writers in the 1960s described similar ideas. The comparison is a recurring theme: customers pay for what they use and do not worry about the machinery behind it.
The idea did not take off immediately. Terminals connected over slow telephone lines were limited, and the arrival of the minicomputer and then the personal computer shifted attention toward owning your own machine. Time-sharing did leave a lasting technical legacy, though. IBM developed virtual machine software in the late 1960s that let one physical computer behave as several separate ones, a technique that later became central to cloud services.
Networks improve
Remote computing needed good connections. ARPANET and other early computer networks showed that machines could talk to each other across distances, and the growth of the history of the internet through the 1990s made high-capacity links common. The arrival of web browsers turned the network into something ordinary people used daily and made the idea of using software through a web page familiar.
In 1999 a company called Salesforce began offering business software entirely through a website, an approach later labeled software as a service. Others followed with email and office tools that ran in a browser instead of on the desktop.
Renting servers by the hour
Large internet companies had by this time built enormous data centers to run their own sites, and they had developed ways to share this hardware among many tasks. Amazon Web Services launched its Simple Storage Service and Elastic Compute Cloud in 2006, letting anyone rent storage and virtual servers over the internet and pay for the time or space used. Google began offering its App Engine platform in 2008, and Microsoft launched its Azure platform in 2010 after announcing it earlier. Many other providers appeared as well.
The word “cloud” itself comes from the habit of drawing the internet as a cloud shape in network diagrams, standing for parts someone else managed. It became a marketing term in the mid-2000s, and its exact first use is hard to pin down.
What changed for users
For businesses, the main shift was financial and practical. Instead of buying servers for peak demand and leaving them idle most of the year, a company could rent more capacity when needed and release it afterwards. Small start-ups gained access to computing that once required a large budget. Backups, photo libraries and shared documents moved from personal drives to remote storage, so files became reachable from phones and laptops alike.
Cloud services also made large-scale analysis easier, which links directly to big data work, and they supplied the computing power behind much modern machine learning. Many connected household devices rely on a cloud service to function, a topic covered in the story of the Internet of Things.
Trade-offs
The cloud is still someone else’s computers in someone else’s buildings. Users depend on a provider’s reliability, and outages at large providers have disrupted many websites at once. Moving to a different provider can be difficult. Storing sensitive data remotely raises questions about security and about which country’s laws apply, and cybersecurity depends on how carefully both provider and customer configure the service. Cost can also be hard to predict when pricing is based on usage.
Cloud computing is therefore best seen as a revival of an old idea, shared remote computing, made practical by cheap hardware, fast networks and software that can divide one machine into many. Whether the utility comparison fully holds is still debated, since computing services vary far more between providers than electricity does.