Cloud computing is the on-demand delivery of computing resources over a network, such as servers, storage, databases, networking and software, from a shared pool that a provider operates and the customer provisions by self-service and pays for according to use. Among cloud fundamentals, it is the umbrella concept: service models such as IaaS, PaaS and SaaS and deployment models such as public, private and hybrid cloud all describe particular forms of it.
The most widely used definition, published by the US National Institute of Standards and Technology in NIST SP 800-145, names five essential characteristics, three service models and four deployment models. What distinguishes cloud computing from traditional hosting is not where the servers stand, but how resources are obtained: through an API or console in minutes, without a purchase order or a technician installing hardware.
The essential characteristics of cloud computing
A service counts as cloud computing when it shows all five characteristics from the NIST definition. The table below lists them and what each means for a team that uses the cloud.
| Characteristic | Meaning | Practical consequence |
|---|---|---|
| On-demand self-service | Customers provision resources without human interaction at the provider | A server is available in minutes through an API |
| Broad network access | Resources are reachable over standard network protocols | Management from any location and device |
| Resource pooling | Many customers share physical infrastructure through multi-tenancy | Lower cost, but isolation matters |
| Rapid elasticity | Capacity scales out and in quickly, often automatically | Systems follow demand instead of peak sizing |
| Measured service | Usage is metered and billed by consumption | Cost visibility per resource and team |
These characteristics explain why a rented physical server with a monthly contract is hosting, not cloud: it lacks self-service elasticity and metered billing. Conversely, a company can run a private cloud in its own data center if internal teams obtain resources by self-service from a pooled, metered platform.
Service models and the division of responsibility
Cloud services are grouped by how much of the stack the provider manages. The higher the model, the less the customer operates and the less it can customize. The comparison below shows the three classic service models.
| Model | Provider manages | Customer manages | Typical examples |
|---|---|---|---|
| IaaS | Hardware, virtualization, network | Operating system, runtime, application, data | Virtual machines, block storage, virtual networks |
| PaaS | Everything up to the runtime | Application code, data, configuration | Managed databases, application platforms |
| SaaS | The complete application | Users, data and settings | Email, office suites, CRM systems |
Between and beyond these models sit newer offerings such as containers as a service and serverless functions, where customers deploy code or container images and pay per request or per second of execution. Regardless of the model, the shared responsibility model applies: the provider secures the infrastructure, while the customer remains responsible for its data, identities and configuration.
Deployment models
The deployment model describes who uses the infrastructure and who owns it. Many organizations combine several models, which is why hybrid and multi-cloud designs are common.
| Model | Who uses the infrastructure | Main advantage | Main drawback |
|---|---|---|---|
| Public cloud | Many unrelated customers | Scale, breadth of services, no hardware | Less control, risk of cost overruns |
| Private cloud | One organization | Control, data residency, predictable cost | Capital investment and own operations team |
| Hybrid cloud | One organization across private and public | Workloads placed where they fit best | Integration and networking complexity |
| Community cloud | A group with shared requirements | Shared compliance effort | Rarely available, limited scale |
Major public providers, including Amazon Web Services, Microsoft Azure and Google Cloud, organize their infrastructure into regions and isolated availability zones within each region. Distributing a workload across zones protects it from the failure of a single data center, while choosing the region determines latency to users and the legal jurisdiction over the data.
Cloud computing compared with traditional infrastructure
The move from owned servers to cloud computing changes finance, operations and architecture at the same time. The table below summarizes the most important differences.
| Aspect | Traditional on-premises | Cloud computing |
|---|---|---|
| Cost model | Capital expense, depreciation over 3 to 5 years | Operating expense, billed per second to month |
| Provisioning time | Weeks to months for hardware | Seconds to minutes through an API |
| Capacity planning | Sized for expected peak | Scaled to actual demand |
| Operations | Hardware, facilities and software in-house | Provider operates the lower layers |
| Cost predictability | High once purchased | Variable, requires active cost management |
Cloud computing is not automatically cheaper. Steady, predictable workloads running around the clock can cost more in the cloud than on owned hardware, especially with large data egress charges. The economic advantage is strongest for variable or uncertain demand, fast experimentation and services that would be expensive to build in-house.
Limitations and common pitfalls
Cloud adoption brings new risks that organizations often underestimate during cloud migration. Experienced architects plan for the following issues from the start.
- Misconfiguration is the leading security risk, for example storage buckets left publicly readable or overly broad access permissions.
- Uncontrolled spending from idle resources, oversized instances and forgotten test environments; tagging and budgets are essential.
- Vendor lock-in through proprietary managed services, which makes later moves between providers costly.
- Lift-and-shift without redesign, which moves old inefficiencies into a more expensive environment and misses elasticity.
- Data residency and compliance requirements that restrict which regions and services may be used.
- Dependence on network connectivity, since every interaction with the platform travels over the network.
In plain words
Cloud computing means renting computing power instead of owning it. A company does not buy and install its own servers. It asks a provider for what it needs, gets it within minutes and pays only for what it uses.
It is similar to electricity from the grid. Few businesses run their own power plant. They plug in, use as much as they need and pay according to the meter. The provider takes care of the machines, the buildings and the spare capacity.
A real-world example
A ticket-selling company sees quiet traffic for most of the year, but when a major concert goes on sale, demand rises about fiftyfold within minutes. With its own servers, the company would have to buy hardware for that rare peak and leave most of it idle the rest of the year.
After moving to a public cloud, its web tier runs on virtual machines in three availability zones with automatic scaling. Before a large sale, capacity grows from 6 to about 300 instances, and it shrinks back within hours afterward. A managed database replaces the self-run cluster. Infrastructure costs now follow ticket sales, and the site stays online during peaks that once caused outages.