Compute (CPU/GPU)

Storage

Networking

Databases

Software and platforms

Data is stored in provider-managed data centers rather than local devices

Accessed via internet using APIs, apps, or web consoles

Supports replication across locations for durability and availability

Applications run on remote servers (virtual machines, containers, serverless)

Users consume compute power without owning physical hardware

Resources can be scaled up/down automatically based on demand

Provision resources without human interaction from the provider

Accessible from diverse devices (laptops, phones, servers) over standard networks

Provider shares infrastructure among multiple customers with isolation controls

Scale out/in quickly; appears “unlimited” to the user

Usage is metered (e.g., GB stored, vCPU-hours, requests), enabling pay-per-use

What you get: virtual machines, storage, networking

You manage: OS, runtime, apps, data

Common uses: lift-and-shift migrations, custom stacks, dev/test environments

What you get: managed runtime/platform (app hosting, databases, middleware)

You manage: application code and data

Common uses: faster app development, reduced ops overhead

What you get: complete software delivered via web/app

You manage: configuration and usage

Common uses: email, collaboration, CRM, analytics

What you get: run functions on demand; no server management

You manage: function code and event triggers

Common uses: event-driven automation, APIs, background tasks

Infrastructure owned and operated by third-party providers

Shared environment with strong logical isolation

Cloud-like environment dedicated to a single organization

Can be on-premises or hosted

Combines public and private; workloads/data can move between them

Common for compliance, legacy integration, phased migration

Uses services from multiple public cloud providers

Motivations: resilience, best-of-breed services, vendor risk reduction

Providers operate regions and availability zones for redundancy

Virtual machines abstract physical hardware

Containers package apps consistently and efficiently

Tools manage scaling, health checks, deployment, and failover

Provisioning through consoles, CLI, SDKs, Infrastructure as Code

Reduced upfront capital expenditure (CapEx) and shift to operational expense (OpEx)

Pay only for what you use; economies of scale from providers

Handle traffic spikes and growth without buying hardware

Provision in minutes; faster experimentation and iteration

Built-in redundancy, backups, disaster recovery options

Deploy near users worldwide to reduce latency

Offload maintenance (patching, upgrades, monitoring) for many components

Cloud drives, archival storage, versioning

Auto-scaling frontends and backends

Warehousing, ETL, stream processing

GPU clusters, managed model deployment

Replication, failover environments, recovery testing

Build pipelines, test environments, artifact storage

Email, document sharing, conferencing

Physical security of data centers

Core infrastructure security (hardware, networking, hypervisor)

Availability of foundational services

Identity and access management (users, roles, permissions)

Data security (classification, encryption, key management)

Application security (code, dependencies, configurations)

Network configuration (firewalls, security groups, segmentation)

Encryption at rest and in transit

Logging, monitoring, threat detection

Least privilege access and MFA

Regular patching and vulnerability management

Compute time, storage capacity, network egress, requests/operations

On-demand (pay per use)

Reserved/committed use (discount for commitment)

Spot/preemptible (cheaper, can be interrupted)

Rightsizing, autoscaling, budgets/alerts, tagging, lifecycle policies

You buy and maintain hardware; fixed capacity; slower provisioning

More direct control but higher operational burden

Provider-managed infrastructure; elastic capacity; faster provisioning

Less hardware management; requires governance and cost discipline

You still own/are responsible for the data; provider hosts it

Can be, but poor architecture or uncontrolled usage can raise costs

Security depends on configuration, controls, and shared responsibility

Upload files to cloud object storage and access them from any device

Deploy a web app on cloud servers that auto-scale during peak traffic