Cloud computing history: A timeline of key milestones

Cloud computing did not begin with one inventor, one company, or one product launch. It emerged in stages as computers became shareable, virtualized, networked, programmable through APIs, and available as metered services. The clearest timeline starts with time-sharing in the early 1960s, passes through mainframe virtualization and packet-switched networking, and reaches the commercial cloud platforms, containers, and serverless services used today.

The history also explains what the cloud is not. Data and applications still run on physical machines in specific places. “Cloud” describes an operating and delivery model, not an invisible location. Our guide to where cloud data is stored covers that physical layer.

Key takeaways

• Modern cloud computing grew from several technologies: time-sharing, virtualization, networking, web delivery, APIs, automation, and usage-based metering.
• MIT demonstrated CTSS time-sharing in 1961, IBM developed foundational virtual-machine systems during the 1960s and early 1970s, and ARPANET connected remote computers in 1969.
• Emory University credits Ramnath Chellappa with the first scholarly definition of “cloud computing” in 1997. The expression and the cloud symbol evolved through several technical and business contexts, so a single-person origin story is too simple.
• Salesforce’s web-delivered CRM in 1999 and the 2006 launches of Amazon S3 and EC2 helped turn software, storage, and compute into commercial online services.
• Docker, Kubernetes, and function services changed how applications were packaged and operated, but they extended the cloud rather than replacing its physical infrastructure.

Cloud computing history at a glance

Year	Milestone	Why it matters
1961	MIT demonstrates CTSS	Multiple people can interact with one large computer through remote terminals.
1964–1972	IBM develops CP-40, CP-67, and VM/370	Virtual machines separate workloads from the underlying mainframe.
1969	ARPANET becomes operational	Geographically separated computers begin sharing resources over a packet-switched network.
1997	Ramnath Chellappa presents a scholarly cloud-computing definition	The phrase becomes associated with both technical and economic boundaries.
1999	Salesforce launches web-delivered CRM	Business software reaches customers through an early SaaS delivery model.
2006	Amazon launches S3 and opens EC2 beta	Developers can obtain storage and virtual compute through web-service interfaces.
2008	Google App Engine enters preview	A managed application platform reduces the infrastructure developers must operate.
2010	Windows Azure becomes generally available	Microsoft establishes a commercial cloud platform alongside its on-premises products.
2011	NIST publishes SP 800-145	A durable definition standardizes cloud characteristics, service models, and deployment models.
2013	Docker is demonstrated publicly	A developer-friendly container workflow improves application packaging and portability.
2014–2015	Kubernetes and AWS Lambda arrive	Container orchestration and event-driven functions expand cloud-native operating models.
2020–2021	Remote work accelerates adoption	Organizations rely more heavily on online infrastructure, collaboration, and application services.

1961: Time-sharing makes computing a shared service

Early mainframes commonly processed queued jobs in batches. Users submitted work and waited for results rather than interacting continuously with the machine. Time-sharing changed that relationship by rapidly allocating processor time among several users.

MIT’s Compatible Time-Sharing System was first demonstrated in 1961 and later operated as a service accessed from terminals. CTSS let people edit files, run programs, and use the same expensive computer interactively. Project MAC, founded in 1963, extended the research, and planning for Multics began in 1964 as a prototype computer utility.

These systems were not cloud platforms in the modern sense. They lacked internet-scale networking, self-service APIs, elastic regional capacity, and commercial metering. They established a crucial idea, though: many users could receive computing as a shared service instead of owning the whole machine.

1964–1972: Virtualization separates workloads from hardware

Time-sharing divided access to one operating environment. Virtualization went further by making one physical computer behave like several independent computers.

IBM’s historical timeline dates the experimental CP-40 project to 1964, CP-67 to 1967, and the announcement of VM/370 to 1972. Each virtual machine could receive an isolated view of processor, memory, and device resources. That abstraction later became one of the technical foundations for provisioning cloud instances.

Virtualization is important to cloud history, but the terms are not interchangeable. A company can virtualize servers in its own data center without providing cloud characteristics such as on-demand self-service or measured usage. Cloud providers can also offer bare-metal machines, containers, and managed services alongside virtual machines.

1969–1990s: Networks make remote infrastructure practical

A shared or virtualized computer is useful only to the people who can reach it. Networking turned remote access into a broader infrastructure model.

DARPA traces the roots of ARPANET to research begun in the 1960s. The four-node network became operational in 1969, with the first computer-to-computer signal sent between UCLA and the Stanford Research Institute on October 29. ARPANET’s move to TCP/IP in 1983 helped establish the internetworking foundation from which the internet grew.

During the 1980s and 1990s, client-server systems, commercial internet access, and the web made remote applications practical for a much larger audience. Network diagrams often used a cloud symbol to hide internal network complexity. That visual convention is one reason attempts to assign the entire metaphor to one person are unreliable.

1997–1999: The term and the SaaS model take shape

Emory University credits information-systems professor Ramnath Chellappa with the first scholarly definition of cloud computing. He presented work on cloud-computing intermediaries at an INFORMS meeting in 1997, framing computing boundaries in economic as well as technical terms.

In 1999, Salesforce opened and delivered CRM through the web. Hosted business applications existed before Salesforce, but its model helped establish the expectation that customers could subscribe to software, access it through a browser, and let the provider operate the application stack. That pattern became known as Software as a Service.

2006–2011: Public cloud becomes a programmable utility

The decisive commercial shift came when developers could request infrastructure through web interfaces rather than negotiating for physical servers.

Amazon announced Simple Storage Service in March 2006. In August, it opened Amazon EC2 for limited beta testing, exposing virtual compute capacity through machine images and APIs. Together, storage and compute became programmable building blocks that could be provisioned without buying hardware.

Google launched an App Engine preview in April 2008. Instead of managing virtual machines directly, developers deployed applications onto a managed runtime with scaling, storage, authentication, and other platform services. Windows Azure then became generally available in February 2010, adding another major commercial platform.

NIST gives cloud computing a stable definition

In September 2011, the US National Institute of Standards and Technology published Special Publication 800-145. Its definition describes five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured service.

NIST also formalized three service models:

• Infrastructure as a Service (IaaS): the customer controls operating systems and applications while the provider operates the underlying infrastructure.
• Platform as a Service (PaaS): the customer deploys applications onto a provider-managed platform.
• Software as a Service (SaaS): the customer uses the provider’s application while the provider operates the stack.

It defined private, community, public, and hybrid deployment models. The market has since added labels such as multi-cloud, edge cloud, distributed cloud, and neocloud, but NIST’s characteristics remain a useful test of whether a service behaves like cloud computing.

2013–2015: Containers, orchestration, and serverless

Containers existed before Docker, but Docker made them easier for developers to build, distribute, and run. Solomon Hykes demonstrated Docker publicly in March 2013. A portable image could package an application with its dependencies while sharing the host operating-system kernel.

As organizations ran more containers, they needed a control plane for scheduling, networking, scaling, and recovery. Kubernetes began with a first commit in June 2014, was announced days later, and reached version 1.0 in July 2015.

Another abstraction removed most server management from the developer’s workflow. AWS introduced Lambda in preview in November 2014, running functions in response to events while the service managed the compute resources. “Serverless” never meant that servers disappeared. It meant that provisioning and maintaining them moved behind the service boundary.

2020–2021: Remote work accelerates adoption

The COVID-19 pandemic was an adoption milestone rather than a new technical foundation. Organizations needed remote access, online collaboration, elastic customer-facing systems, and faster digital delivery.

An OECD study of digital adoption during COVID-19 identified cloud and collaborative software among the applications that diffused most during the period. Eurostat separately reported that the share of EU enterprises using paid cloud services rose from 36% in 2020 to 42% in 2021. The data supports acceleration, although adoption varied widely by company size and country.

Cloud architecture today

The cloud is now a family of operating models rather than one architecture. Public cloud, private cloud, hybrid environments, multiple providers, edge locations, and distributed systems can coexist in one application estate.

• Public cloud provides shared services operated by a third party.
• Private cloud applies cloud characteristics to infrastructure dedicated to one organization.
• Hybrid cloud connects distinct private and public environments with enough integration or portability to support a workload across them.
• Multi-cloud uses services from more than one cloud provider, whether or not the environments are tightly integrated.
• Edge computing places selected processing or storage closer to users, devices, or data sources.
• Distributed cloud spreads managed resources across locations while coordinating them as a service.

Cloud and distributed computing overlap, but they answer different questions. Cloud describes how resources are delivered and operated; distributed computing describes how components coordinate across machines. Our cloud-versus-distributed-computing comparison separates the concepts, while the distributed cloud architecture guide examines nodes, coordination, data consistency, and failure handling.

Architecture decisions now follow the workload. A hybrid design may preserve an existing system while using cloud capacity; the hybrid cloud file-transfer guide shows how control and data paths cross that boundary. AI workloads have also produced specialist infrastructure categories, explained in our guide to neocloud AI infrastructure.

Sustainability requires measurement

Resource pooling and higher utilization can reduce waste, but no cloud deployment model is automatically sustainable. Electricity sources, hardware utilization, cooling, water, equipment life, data movement, and reporting boundaries all affect the result.

The International Energy Agency estimates that data centers used about 1.5% of global electricity in 2024, with much larger effects in some local markets. The European Commission now requires reporting from larger EU data centers because energy and water performance need comparable evidence. Provider sustainability claims should therefore be evaluated with disclosed methods and current data, not inferred from the word “cloud.”

What this history does not prove

• Cloud is not always cheaper. It changes the purchasing and operating model. Cost depends on utilization, architecture, data movement, commitments, support, and staff time.
• Public cloud is not automatically secure or compliant. Providers and customers divide responsibilities, and the result depends on configuration, identity, data handling, contracts, and operations.
• Virtualization is not the whole cloud. It is one enabling technique among networking, automation, APIs, orchestration, and metering.
• Distributed and decentralized are not synonyms. A system can run across many machines while ownership and control remain centralized.
• New abstractions do not remove hardware. Containers and functions still consume processors, memory, storage, networks, electricity, and physical space.

Frequently asked questions

Who invented cloud computing?

No single person invented it. Time-sharing researchers, mainframe engineers, networking teams, software providers, and infrastructure platforms contributed separate layers over several decades.

When was the term “cloud computing” first used?

Cloud symbols appeared in network diagrams before the phrase had a stable modern meaning. Emory University credits Ramnath Chellappa with the first scholarly definition in 1997. Claims that one person coined the entire metaphor should be treated cautiously.

What was the first cloud service?

The answer depends on the service model. Time-sharing offered remote shared computing decades earlier. Salesforce helped establish web-delivered SaaS in 1999. Amazon S3 and EC2 made programmable storage and virtual compute broadly available in 2006. It is more accurate to identify milestones than one universal “first.”

What are the three main cloud service models?

NIST defines Infrastructure as a Service, Platform as a Service, and Software as a Service. They differ mainly in which layers the provider operates and which layers the customer controls.

How is cloud computing different from distributed computing?

Cloud computing is a service and operating model built around network access, pooled resources, elasticity, self-service, and measurement. Distributed computing is an architectural approach in which components coordinate across multiple machines. A cloud platform is usually distributed internally, but a distributed system does not have to be sold or operated as a cloud service.

The cloud timeline is still expanding

Cloud computing grew by hiding one layer of operational complexity at a time: first shared processor access, then virtual hardware, remote networks, programmable infrastructure, managed platforms, portable application packages, orchestration, and event-driven functions. Current hybrid, edge, distributed, and specialist compute models continue that progression. The useful question is no longer whether a workload belongs “in the cloud,” but which layers should be managed, where the resources should run, and who should control them.