A guided history of cloud computing — from 1960s mainframes shared by a handful of terminals to today's hyperscale, AI-saturated global infrastructure. Read it top to bottom as a course, or jump to any era below.
Long before anyone spoke of "the cloud," the core idea was already forming: computing was too expensive for any one person to own, so it had to be shared.
Computer scientist John McCarthy proposed sharing computing time among many users on a single machine — the seed of an idea he'd later describe as computing organized like a public utility, where people would simply "plug in" for processing power the way they do for electricity.[1][2]
IBM's CP-40 and later VM/370 systems let a single physical mainframe present itself as several independent "virtual machines," each running its own operating system — the direct technical ancestor of the virtualization that underpins every modern cloud platform.[5]
Between the mainframe era and the modern cloud lay three decades of quieter groundwork: theoretical foundations for virtualization, the rise of networking, and the first inklings of "cloud" as a word.
The U.S. Department of Defense's ARPANET connected research institutions' computers together, previewing the internet, while businesses experimented with early virtual private networks — showing that computing resources didn't have to sit in the same room as the people using them.[6][1]
Gerald Popek and Robert Goldberg published formal criteria for what makes a piece of software a true "virtual machine monitor" — turning what had been engineering intuition into rigorous computer science.[7]
The exact origin of the phrase is disputed, but Compaq executive George Favaloro is often credited with an early use of the term "cloud computing" around 1996, and technologist Ramnath Chellappa used it in a 1997 academic paper on computing economics. In 1998, VMware was founded at Stanford, and would go on to bring practical virtual machine technology to standard x86 hardware — the missing piece that made mainstream cloud infrastructure possible.[8][9][6]
Before anyone rented raw servers over the internet, a smaller idea proved the model could work: deliver an entire business application through a web browser instead of installed software.
Salesforce launched a customer relationship management tool delivered entirely through a website rather than installed on individual machines, marketing itself with the slogan "No Software." It proved that a company didn't need to install an application on every employee's computer — they just needed a browser — and became one of the first successful examples of what we now call SaaS.[10][6]
VMware released ESX 1.0, bringing genuine virtualization to standard x86 servers rather than specialized mainframes. This meant ordinary, off-the-shelf hardware could now be divided into many independent virtual machines — the technical foundation cloud providers would soon build entire businesses on.[6]
This is the moment most people mean when they say "the cloud" began: renting raw computing power and storage, by the hour, from someone else's data center.
Amazon Web Services launched Simple Storage Service (S3) in March and Elastic Compute Cloud (EC2) in August, letting anyone rent virtual servers and storage by the hour with no long-term hardware purchase. Amazon had noticed that its own data centers sat mostly idle outside of shopping peaks like the winter holidays, and realized that spare capacity could be sold as a service — establishing what we now call Infrastructure-as-a-Service (IaaS).[6][13][14]
Microsoft's Azure platform reached general availability, entering the market roughly four years after AWS. Because so many businesses already ran Microsoft software, Azure had a built-in advantage in reaching enterprise customers already comfortable with Microsoft's tools.[6][17]
Virtual machines had made servers divisible; the next innovation made applications themselves lightweight and portable, and then gave the field a way to manage thousands of them at once.
Docker's first release packaged an application with everything it needed to run into a lightweight, portable "container" that behaved identically across different machines. Containers had existed in Linux for years, but Docker made the technology dramatically easier to build, ship, and run, and it quickly became the industry standard.[6][18]
Google released Kubernetes, an open-source system for automatically scheduling, scaling, and restarting large numbers of containers across clusters of machines — work that grew out of Google's own internal container-management tools. Kubernetes reached its 1.0 release in mid-2015, the same day the vendor-neutral Cloud Native Computing Foundation was founded to steward it.[6][18]
Containers still needed to be provisioned and managed. The next step removed servers from the developer's mental model almost entirely — pay only for the moments your code actually runs.
AWS announced Lambda, a service that runs small pieces of code in response to events without the developer ever provisioning or managing a server. The word "serverless" didn't exist when engineers first sketched the idea in 2013, but Lambda's launch effectively created the category, and the company would later note it grew to power tens of trillions of invocations a month within its first decade.[19][20][21]
Cloudflare previewed Workers, running small pieces of application code on servers physically distributed close to users around the world rather than in a single distant data center — an early step toward what's now called edge computing.[6]
As Kubernetes made workloads more portable, tools like Google's Anthos and Microsoft's Azure Arc emerged to let organizations manage applications consistently across multiple public clouds and their own on-premises data centers at once, rather than being locked to a single provider.[22]
The most recent phase of cloud computing has been reshaped by a single force: the enormous, specialized compute demands of training and running artificial intelligence models.
Major cloud providers raced to weave generative AI directly into their core platforms — most visibly through Microsoft's deep partnership with OpenAI — turning GPU capacity and managed AI model access into one of the cloud industry's central products, alongside traditional storage and compute.[23]
Cloud providers expanded access to quantum hardware over the internet, with IBM offering cloud access to quantum processors exceeding 1,000 qubits — letting researchers experiment with quantum computing without owning specialized hardware, following the same "rent, don't own" logic that built the classical cloud decades earlier.[23]
By the mid-2020s, running applications as containers on Kubernetes, deploying serverless functions for event-driven workloads, and spreading systems across multiple providers had become standard enterprise practice rather than a cutting-edge choice. Global spending on cloud services was forecast to approach roughly $738 billion in 2026, with AI-driven workloads a major driver of that growth.[24][25][23]
Sources consulted while researching this page. Content above has been written in original language and independently organized — it is not copied or closely paraphrased from any single source. Consult these directly for authoritative detail.