Cloud Computing Models, Services and Data Centres

Cloud computing provides computing resources over the internet on demand, with users paying for what they consume. Its deployment and service models shape who manages the infrastructure, how resources are shared and how applications are delivered.

Cloud Deployment Models

Deployment models describe where cloud infrastructure is located and who can access it. They also determine how organisations balance control, flexibility, security and cost.

• Public cloud: A third-party provider offers services over the public internet using infrastructure shared by multiple customers. Google Cloud, Amazon Web Services (AWS) and Microsoft Azure are examples. Public clouds support rapid scaling and pay-as-you-go pricing.
• Private cloud: Infrastructure is dedicated to one organisation, providing greater control over resources and security arrangements. It may be located on the organisation’s premises or hosted by a third party.
• Hybrid cloud: Public and private clouds are connected so that data and applications can move between them. An organisation may retain sensitive workloads in its private cloud while using public-cloud capacity for other needs.
• Community cloud: Several organisations with shared requirements, such as security or compliance needs, use common infrastructure. Participating organisations share its costs and management.

Cloud Service Models

Service models distinguish how much of the technology stack the provider manages. As users move from IaaS to SaaS, they generally take responsibility for fewer underlying technology layers.

• Infrastructure as a Service (IaaS): Provides virtual machines, storage and networks. The provider manages the physical infrastructure, while users manage operating systems, applications and data. AWS EC2 and Google Compute Engine are examples.
• Platform as a Service (PaaS): Supplies an environment for developing and deploying applications, including infrastructure, operating systems, runtime environments, databases and web servers. Users manage their applications and data. Examples include Google App Engine and AWS Elastic Beanstalk.
• Software as a Service (SaaS): Delivers ready-to-use applications over the internet, commonly through a subscription. The provider manages the underlying infrastructure, platform and application. Gmail, Salesforce and Microsoft 365 are examples.

Core Cloud Concepts

Virtualisation creates virtual versions of resources such as servers, storage and network devices on physical hardware. It allows computing capacity to be divided and shared efficiently. Resource pooling describes the shared pool of configurable resources in a multi-tenant environment; capacity is assigned and reassigned among consumers as demand changes. Elasticity is the ability to scale resources up or down automatically in response to demand. Scalability is a system’s ability to handle increasing workloads by adding capacity. Vertical scaling increases the power of existing resources; horizontal scaling adds resources. Measured service means that cloud systems monitor and report resource use, supporting transparent billing based on consumption. These features can help organisations adjust capacity without maintaining all computing resources themselves, although effective management of usage remains important. Specialised processors can also support cloud workloads. Google’s Tensor Processing Unit (TPU) is an application-specific integrated circuit (ASIC) and neural-processing unit designed for neural-network machine learning. Compared with GPUs, TPUs favour high-volume, low-precision computation and more input/output operations per joule, without graphics-rendering hardware. Google selected a systolic-array design after three groups developed AI accelerators; its second-generation TPU added floating-point computation to integer processing and introduced bfloat16 for both training and inference.

Data Centres

A data centre is a physical facility housing computer systems and associated telecommunications and storage equipment. It processes, stores and distributes data, supporting the continuous operation of digital services and cloud infrastructure. Typical components include servers, storage arrays, network equipment, power distribution, cooling systems and security measures. Facilities may be enterprise data centres, owned and operated for one organisation’s use; colocation data centres, where organisations rent space for their own servers and hardware; or managed-services data centres, where a third party manages infrastructure for clients. Hyperscale data centres contain very large numbers of servers and are designed for high performance, efficiency and expansion. Major cloud providers, including Google, AWS and Microsoft, operate such facilities. Green data centres seek to reduce environmental impact through measures such as renewable energy and advanced cooling. For public administration and digital public infrastructure, data-centre planning also raises questions of data localisation, security and energy sustainability. These considerations can influence where workloads are hosted and how facilities are powered and cooled.

Data Centre Tier Classifications

The Uptime Institute’s Tier Standard classifies data-centre infrastructure by redundancy, availability and fault tolerance. TIA-942 is also a widely used data-centre infrastructure standard. The following availability and downtime figures are approximate.

• Tier I — Basic Capacity: A single path for power and cooling, with no redundant components; 99.671% availability and about 28.8 hours of annual downtime.
• Tier II — Redundant Capacity Components: A single power and cooling path with redundant components; 99.741% availability and about 22 hours of annual downtime.
• Tier III — Concurrently Maintainable: Multiple independent power and cooling paths with redundant components, allowing maintenance without interrupting operations; 99.982% availability and about 1.6 hours of annual downtime.
• Tier IV — Fault Tolerant: Multiple independent active power and cooling paths and dual-powered equipment, designed to withstand a single worst-case component failure without operational impact; 99.995% availability and about 26.3 minutes of annual downtime.

Key Prelims Takeaways

• Deployment models: Public, private, hybrid and community clouds describe how infrastructure is organised and accessed.
• Service models: IaaS provides infrastructure, PaaS provides an application development and deployment environment, and SaaS provides ready-to-use software.
• Resource management: Virtualisation and resource pooling enable shared use; elasticity is automatic adjustment, while scalability is the capacity to handle growth.
• Data centres: Hyperscale facilities support large-scale cloud services; green data centres focus on energy efficiency and reduced environmental impact.
• Resilience: Uptime Institute Tiers I–IV classify data-centre infrastructure from basic capacity to fault tolerance.
• TPUs: Google’s ASICs are designed for neural-network machine learning and use a systolic-array architecture.

Recent Context

On 9 November 2026, Google added TPU 8tx and TPU 8i Max, complementing TPU 8t for large-scale training and TPU 8i for inference. TPU 8tx targets confidential inference and small-scale training; TPU 8i Max doubles TPU 8i’s computing capacity and memory bandwidth. October’s TorchTPU enables native PyTorch execution.