AI Glossary / Data engineering and architecture / Beginner

Infrastructure as a service

Definition

Infrastructure as a service (IaaS) is a cloud service model that provides fundamental computing resources such as processing, storage and networking. Customers can deploy operating systems and applications on those resources without managing the underlying physical cloud infrastructure. The provider and customer retain different responsibilities, and the exact division depends on the chosen service and configuration.

What it means

Instead of buying and operating every physical server, a team provisions computing resources through a service interface. It still makes important decisions about its software, data and network access. IaaS therefore offers flexibility while leaving substantial operational work with the customer. It differs from software as a service, where the customer primarily uses a finished application, and from platform services that take over more of the application runtime or operating environment.

How it works

A provider operates physical facilities and resource-management systems. Customers request resources using a console, API or automation tool and configure the resulting environment. They install or select software, manage access and monitor workloads. Usage is measured according to the service's charging model. Capacity can often be adjusted, but availability, quotas and provisioning time still apply. Designing a reliable application requires explicit decisions about failure recovery, data protection and dependencies rather than assuming those arise automatically from renting infrastructure.

Three practical examples

• A development team provisions virtual machines to test an application and removes them when the test is complete.
• An AI team rents suitable compute resources for a training workload while managing its own environment, datasets and job execution.
• A business hosts an internal application on cloud infrastructure, retaining responsibility for its guest software and access configuration.

Think of it this way

Think of renting a workshop with electricity and space already provided. You can choose your tools and working methods without constructing the building. You still need to operate the tools safely and look after your materials. The analogy has limits: cloud resources are programmable, can be created in many locations and may incur charges through several usage dimensions rather than a single monthly rent.

Worked example

Suppose a fictional compute service charges 0.50 currency units per running instance-hour. Four instances used for ten hours consume 40 instance-hours and incur a compute charge of 20 units under that simplified rate. If they remain running for another 20 hours, the additional compute charge is 40 units. Storage, data transfer, licenses and other services are excluded from this example and may add costs. The arithmetic illustrates why lifecycle management matters; it is not a current provider quote, purchasing recommendation or full project budget.

Illustrative case study

In a fictional project, Sequoia Design needs temporary computing capacity for a model evaluation. Engineers define the environment, restrict access and track which datasets and software versions are used. They schedule a review of unused resources and verify that results are saved before shutting down the compute instances. The pilot clarifies both operational effort and cost categories. It does not demonstrate that cloud infrastructure is always cheaper or that a single machine is sufficient for production availability.

Common misconceptions and mistakes

IaaS is not a finished application and does not remove all system-administration work. The provider's protection of physical infrastructure does not automatically secure customer software or permissions. Elastic capacity is not unlimited capacity on demand in every region. Stopping compute may not remove every associated billable resource. Replication, backup and disaster recovery are also separate design concerns. A service's availability commitment should not be interpreted as a guarantee that every application built on it will meet the same outcome.

Business use

Organizations choose IaaS when they need control over the operating environment or flexible access to computing resources. Evaluate total operating effort alongside direct service charges. Assign ownership for patching, access, monitoring, backup and incident response, using the provider's service-specific responsibility model. For AI work, check accelerator availability, memory, data movement and software compatibility. Preserve reproducible environment definitions so a later experiment can be understood even if the underlying rented resources have been removed.

People also ask

Does IaaS always mean public cloud?

No. Service models and deployment models are different classifications; infrastructure services can be offered through different cloud arrangements.

Does the provider update my application?

Not automatically. In a typical IaaS arrangement, the customer manages deployed application software and much of the guest environment.

Can I avoid buying hardware entirely?

You may avoid buying the rented infrastructure, but local connectivity, devices and other operational needs still remain.

Check your understanding

A provider secures the physical host, but a customer leaves its application account publicly accessible. Is physical host security enough to protect the application?

Answer: No. The customer must configure and operate its part of the service appropriately. Responsibilities span layers rather than transferring entirely to the provider.

Related terms

• Cloud computing
• Infrastructure as code
• virtual machine
• platform as a service
• software as a service
• shared responsibility
• elasticity

Learn next

Draw the responsibility boundary for one specific IaaS service, marking who manages each layer. Then estimate a small workload using all relevant cost categories and define how resources will be retired. Compare that operational burden with a managed platform option for the same application.

More in Data engineering and architecture

• Information integration - Information integration brings information from different sources into a coherent form or view so that people and applications can use it together.
• Information model - An information model describes the concepts in a chosen domain, their properties and relationships, and the rules that give them meaning.
• Information repository - An information repository is an organized environment for storing, describing, managing and retrieving information.
• Input/output automaton - An input/output automaton is a mathematical state-machine model for describing components of asynchronous concurrent systems.
• Insertion Anomaly - An insertion anomaly is a database design problem in which a legitimate fact cannot be recorded without also supplying another, logically separate fact.
• Instruction-level parallelism - Instruction-level parallelism (ILP) is the opportunity to overlap the execution of multiple machine instructions from a program's instruction stream.

Sources

Sources: NIST, “The NIST Definition of Cloud Computing,” SP 800-145, https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-145.pdf; NIST CSRC, “Infrastructure as a Service,” https://csrc.nist.gov/glossary/term/Infrastructure_as_a_Service; AWS, “Shared Responsibility Model,” https://aws.amazon.com/compliance/shared-responsibility-model/; AWS, “Security in Amazon EC2,” https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-security.html. Primary-source search excerpts consulted; full-document and service-specific editorial review remain pending. Pricing numbers are fictional teaching assumptions.

Previous: InfoVAE Next: Infrastructure as code