EC2 instance families are lettered to advertise their shape: C is compute-heavy, M is balanced, R is memory-rich, and T is burstable and cheap for variable workloads. Choosing among them is part of architectural fitness, and so is choosing how to pay. Spot Instances offer up to 90% discount but can be reclaimed with a 2-minute interruption notice, which your application can learn about through the instance metadata at the well-known spot/instance-action path or via an EventBridge event. Reserved Instances commit you to a specific instance type for one or three years for roughly a one-third discount, while Savings Plans trade instance-type specificity for family flexibility at a similar discount.
AWS Lambda runs code in response to events without provisioning servers. Each invocation is capped at 15 minutes, and the default regional concurrency limit is 1,000 (raiseable on request). To reach resources inside a VPC, you attach a configuration that includes subnets and security groups; AWS then creates ENIs in those subnets for the function. EventBridge cron or rate rules are the standard way to schedule Lambdas, and Lambda can also be invoked by SQS, SNS, API Gateway, and dozens of other sources.
AWS Fargate applies the same serverless model to containers, removing the need to manage EC2 hosts underneath ECS or EKS. ECS is the AWS-native container orchestrator and EKS is managed Kubernetes, both of which can run on EC2 or on Fargate. For batch workloads such as genomics or rendering, AWS Batch schedules jobs across dynamically provisioned compute and is the right answer when you need queue-driven parallelism. To move existing servers, Application Migration Service (MGN) performs continuous block-level replication and orchestrates cutover. AMIs tie together an EBS snapshot and the launch metadata needed to boot a new instance, and they can be shared across accounts with explicit launch permissions, enabling cross-account image distribution.