Infrastructure as a service, usually abbreviated IaaS, is a cloud-computing service model in which a provider supplies fundamental computing resources such as processing capacity, memory, storage and networking on demand, rather than an organisation purchasing and operating all of the corresponding physical infrastructure itself. The customer normally obtains these resources through a management console, API, command-line interface or infrastructure-as-code system and can create, configure, resize and remove resources as required.
AWS describes IaaS as providing compute, storage and network resources on a pay-as-you-go basis while the provider maintains the underlying physical infrastructure and the customer remains responsible for deploying and maintaining applications. The fundamental idea is abstraction: the customer consumes virtualised or otherwise provisioned infrastructure without having to own the physical servers, storage arrays, switches, racks, power systems and data-centre facilities supporting it. IaaS is therefore the cloud service model that generally gives customers the greatest degree of infrastructure-level control while leaving the physical layer to the provider.
The principal resources supplied through IaaS are compute, storage and networking. Compute commonly appears as virtual machines or instances with specified numbers of virtual CPUs, memory allocations and other characteristics, although providers can also expose dedicated physical servers, GPUs and specialised accelerators. Storage may be supplied as block storage, which behaves similarly to attached disks; object storage, which stores discrete objects addressed through an API; or file storage, which exposes shared filesystem semantics.
Networking can include virtual networks, subnets, IP addresses, routing tables, load balancers, firewalls, gateways and other virtualised networking functions. AWS explicitly identifies compute, block/file/object storage and virtualised networking as broad categories of IaaS resources. These resources can be combined into complete virtual data-centre architectures, allowing an organisation to construct servers, databases, web applications, development environments, private networks and other systems without physically installing the underlying hardware.
IaaS is heavily dependent on virtualisation, although not every IaaS resource has to be implemented through conventional virtual machines. A physical server in a cloud data centre can be divided into isolated virtual machines by a hypervisor, allowing multiple customers to consume logically separate computing environments on the same physical host. Virtual networks similarly abstract physical switches and routers into programmable software-defined resources.
The customer sees logical infrastructure such as a virtual machine with a defined CPU and memory configuration, even though the provider determines which physical machine actually supplies those resources. This abstraction allows resources to be provisioned quickly and moved, replicated or resized without the customer physically manipulating hardware. Container infrastructure, bare-metal instances, specialised accelerators and other architectures can also be offered within the broad IaaS category. The essential characteristic is not the specific virtualisation technology but the delivery of configurable infrastructure resources as a service.
A defining characteristic of IaaS is the division of responsibility between provider and customer. The provider normally owns and operates the physical data centres, servers, storage systems, networking equipment, power infrastructure and physical security. The customer generally controls the operating system, applications, data, virtual networking configuration and many security settings within the resources they have provisioned.
AWS's comparison of cloud service models shows that IaaS leaves customers responsible for data, applications and runtime software while the provider manages the physical servers, storage devices, network appliances, virtualisation and operating-system layer supplied by the service. The exact boundary varies between services, particularly where managed components are introduced. This division is often described through the shared-responsibility model: moving infrastructure into the cloud transfers some operational responsibilities to the provider but does not transfer every responsibility for the customer's systems or data.
IaaS is particularly useful when an organisation needs substantial control over its computing environment but does not want to purchase physical infrastructure. A company can provision virtual servers for a web application, attach persistent storage, configure a virtual private network, install a chosen operating system and deploy its own software. A development team can create temporary environments for testing and delete them afterwards.
An organisation with highly variable demand can scale the number or size of instances rather than maintaining enough physical hardware for the highest possible workload. IaaS is also commonly used for application migration, disaster-recovery environments, development and testing, batch processing, enterprise workloads, specialised computing and workloads requiring operating-system-level control. The model is particularly valuable when the customer needs configuration freedom that a higher-level managed platform would not provide.
The economic model of IaaS changes the relationship between infrastructure capacity and expenditure. Traditional infrastructure frequently requires capital expenditure before demand is known: servers, storage and networking equipment have to be purchased and installed ahead of use. IaaS allows resources to be provisioned much more rapidly and generally charges according to resource consumption, reservation commitments or other commercial models. This can reduce the need for large upfront infrastructure purchases and can make temporary or highly variable capacity economically practical.
However, “pay only for what you use” does not mean cloud infrastructure is automatically cheaper. Persistent virtual machines, storage, network transfer, managed services, snapshots, reserved capacity, licensing and idle resources can all generate costs. Effective IaaS management therefore involves capacity planning, rightsizing, resource scheduling, monitoring, tagging, budgeting and cost governance.
IaaS provides considerable flexibility but also leaves customers with significant operational responsibilities. An organisation running virtual machines may still have to patch operating systems, configure firewalls, secure credentials, monitor processes, manage application dependencies, maintain backups and respond to vulnerabilities. Poorly configured cloud infrastructure can therefore create serious security or reliability problems even when the provider's physical infrastructure is highly protected. Infrastructure-as-code tools can reduce configuration drift by defining resources declaratively in version-controlled files.
Automation can also create repeatable environments, enforce policies and accelerate recovery. These practices are important because the scalability of cloud infrastructure means that configuration errors can potentially be reproduced across many resources very quickly. IaaS provides powerful control, but control also creates corresponding management obligations.
IaaS differs from PaaS and SaaS primarily in the level of abstraction and responsibility. With IaaS, the customer receives foundational infrastructure and generally manages much of the software stack above it. PaaS removes more infrastructure management and provides a managed application-development environment, while SaaS delivers a completed application managed largely by the provider. AWS describes IaaS as offering greater control than PaaS and SaaS because fewer components are abstracted away.
IaaS can therefore be understood as a programmable virtual data-centre foundation rather than simply “renting a server”. Its significance is that organisations can obtain scalable infrastructure resources through software interfaces, rapidly alter capacity, automate provisioning and avoid direct ownership of much of the physical infrastructure while retaining substantial control over the operating environment.
