What Is AWS and How Does It Work? The Foundation for Everything Else
Amazon launched its web services platform in 2006 by doing something counterintuitive: a retail company started selling computing infrastructure to other companies. The first two services were S3 (Simple Storage Service) for object storage and EC2 (Elastic Compute Cloud) for virtual machines. Neither was immediately popular. Most businesses did not yet understand what they would do with rented virtual servers.
By 2024, AWS generates over $100 billion in annual revenue, holds roughly 31% of the global cloud infrastructure market, and operates infrastructure in over 30 geographic regions. The company that started with two services now offers over 200. Understanding how AWS is structured — not the individual services, but the underlying architecture — is the prerequisite for understanding everything else about cloud computing.
How AWS Organises the World: Regions and Availability Zones
AWS divides its global infrastructure into Regions, and each Region into Availability Zones. This two-level hierarchy is the physical foundation for how AWS provides reliability.
A Region is a geographic area containing AWS infrastructure. Examples: us-east-1 (Northern Virginia), eu-west-1 (Ireland), ap-southeast-1 (Singapore), ap-south-1 (Mumbai). AWS currently operates 33 Regions with more announced. Each Region is a completely independent cloud environment — it has its own power, cooling, networking, and is isolated from failures in other Regions. Data does not replicate across Regions unless you explicitly configure it to.
An Availability Zone (AZ) is one or more discrete data centres within a Region, each with independent power, cooling, and physical security. Each Region contains at minimum two AZs, typically three, and sometimes more (us-east-1 has six). AZs within a Region are connected by high-speed, low-latency private fiber links (typically sub-1ms between AZs).
AWS Global Infrastructure Hierarchy--------------------------------------World | +-- Region: us-east-1 (N. Virginia) | | | +-- AZ: us-east-1a (Data centre cluster 1) | +-- AZ: us-east-1b (Data centre cluster 2) | +-- AZ: us-east-1c (Data centre cluster 3) | +-- AZ: us-east-1d | +-- AZ: us-east-1e | +-- AZ: us-east-1f | +-- Region: eu-west-1 (Ireland) | +-- AZ: eu-west-1a | +-- AZ: eu-west-1b | +-- AZ: eu-west-1c | +-- Region: ap-southeast-1 (Singapore) +-- AZ: ap-southeast-1a +-- AZ: ap-southeast-1b +-- AZ: ap-southeast-1cThe AZ design is the core of AWS’s high availability model. If you deploy an application across two or three AZs in a Region, a power failure, network issue, or hardware problem in one AZ does not affect your application — it keeps running in the other AZs. Most AWS managed services (RDS Multi-AZ, ELB, ECS, ElastiCache cluster mode) handle multi-AZ deployment automatically.
Edge Locations: A Third Layer for Content Delivery
Below Regions and AZs, AWS operates over 450 Edge Locations worldwide. These are smaller facilities — sometimes a single server rack in a data centre run by a telco or internet exchange — positioned in population centres not served by full AWS Regions.
Edge Locations primarily serve two purposes:
- CloudFront: AWS’s CDN caches content (images, videos, static files) at Edge Locations so users receive data from a server close to them rather than from the Region that hosts the origin
- Route 53: DNS queries are resolved at Edge Locations for low-latency DNS response globally
Edge Locations are not full cloud environments — you cannot run EC2 instances in them. They are caching and DNS infrastructure.
How AWS Billing Works
AWS bills on a consumption model: you pay for what you use, typically measured per second or per hour, per GB, or per request. There is no minimum spend and no upfront commitment on standard on-demand pricing.
The main billing dimensions:
- Compute (EC2): Per-second billing (minimum 60 seconds). The price depends on instance type, operating system, and whether you use on-demand, reserved, or spot pricing.
- Storage (S3, EBS): Per GB-month stored plus per-request charges for S3.
- Data transfer: Inbound data to AWS is free. Outbound data (egress) is charged per GB. This is the billing line that surprises most new AWS users — data leaving AWS costs money.
- Managed services: Each managed service has its own pricing model. RDS charges per instance-hour. Lambda charges per million requests and per GB-second of compute. DynamoDB charges per read and write capacity unit.
The most important cost optimisation levers:
Reserved Instances: Commit to using a specific instance type in a specific region for 1 or 3 years. AWS gives 30–60% discount vs on-demand in exchange for the commitment.
Savings Plans: A more flexible version of reserved instances — commit to a dollar amount of compute spend per hour (across EC2, Lambda, Fargate) rather than a specific instance type.
Spot Instances: AWS sells excess capacity at 70–90% discount. The catch: AWS can reclaim the instance with a two-minute warning. Works for fault-tolerant batch workloads, not for production databases.
EC2 Pricing Comparison (m5.xlarge, us-east-1, Linux)------------------------------------------------------Pricing Type Monthly Cost Use CaseOn-Demand ~$140 Development, testing1-Year Reserved ~$84 Stable production workloads3-Year Reserved ~$56 Long-running stable workloadsSpot ~$28-42 Batch, stateless workersThe Core Services: What Most AWS Accounts Use
With 200+ services, knowing which ones matter most is itself valuable knowledge. The majority of AWS workloads are built on a small set of foundational services.
EC2 (Elastic Compute Cloud): Virtual machines. The first service, still the most fundamental. Thousands of instance type options. Used for anything that needs persistent compute.
S3 (Simple Storage Service): Object storage. The place to store files, backups, static assets, data lake files. Eleven nines of durability. Scales infinitely.
VPC (Virtual Private Cloud): The network container for everything else. Every EC2 instance, RDS database, and Lambda function runs inside a VPC.
RDS (Relational Database Service): Managed MySQL, PostgreSQL, MariaDB, Oracle, and SQL Server. AWS handles backups, patching, replication, and failover.
Lambda: Serverless functions. Write a function, deploy it, and pay only when it executes. No servers to manage.
IAM (Identity and Access Management): The control plane for permissions. Every API call in AWS is authenticated and authorised through IAM.
CloudWatch: Monitoring, logging, and alerting. Metrics from all AWS services flow here.
Where to Start
Creating an AWS account takes ten minutes and a credit card. The free tier covers most services for the first twelve months at usage levels sufficient for learning and experimentation — 750 hours of t2.micro or t3.micro EC2 per month, 5 GB of S3 storage, 1 million Lambda invocations.
The practical starting path: create an account, enable MFA on the root account immediately, create an IAM user for daily work (never use root credentials routinely), and explore the console. The AWS documentation is the best in the industry — thorough, accurate, and well-maintained. For structured learning, AWS’s own training at skill builder.aws covers every service with hands-on labs.
The single most useful thing to understand before deploying anything in AWS is the shared responsibility model and IAM. Getting identity and access management right from the start prevents the most common and costly cloud security failures.