On-Demand and Reserved Instances
On-Demand Instances are the default. You pay a fixed hourly rate with no commitment, starting and stopping whenever you want. This is the most expensive per-hour pricing and the right choice when you genuinely cannot predict your capacity needs, when you are prototyping, or when workloads run for less than a month. On-Demand is also the baseline for understanding the discounts that other models provide.
Reserved Instances commit you to a specific instance type, region, and operating system for one or three years. In exchange, you receive up to 72 percent off the On-Demand price. Payment options let you choose between No Upfront (smallest discount), Partial Upfront, and All Upfront (largest discount). Standard Reserved Instances lock you into specific attributes. Convertible Reserved Instances let you change the instance type, OS, or tenancy during the term for a smaller discount of up to 54 percent.
Savings Plans work similarly to Reserved Instances but offer more flexibility. Instead of committing to a specific instance type, you commit to a certain dollar amount of compute spending per hour. EC2 Instance Savings Plans provide the deepest discount for a specific instance family in a region. Compute Savings Plans apply across instance families, regions, and even Fargate and Lambda, for slightly less discount.
Spot Instances, Dedicated Hosts, and Scheduled
Spot Instances use AWS's spare EC2 capacity at discounts of up to 90 percent off On-Demand. The catch is that AWS can reclaim the instance with a two-minute warning when capacity is needed elsewhere. Spot Instances are appropriate for fault-tolerant, stateless workloads that can be interrupted and restarted: batch processing, machine learning training jobs, rendering, data analysis. They are never appropriate for databases, session-based applications, or any workload that cannot gracefully handle interruption.
Dedicated Hosts give you a physical server fully dedicated to your use. This is required for certain software licenses that are tied to physical socket or core counts, and for compliance requirements that mandate physical isolation from other customers. Dedicated Hosts are the most expensive EC2 option. Dedicated Instances are a lighter version: your instances run on hardware not shared with other AWS customers but you do not control the specific host.
How to choose the correct answer
Workload is steady-state and runs continuously for 1 to 3 years: Reserved Instances or Savings Plans. Standard RI if the instance type will not change. Convertible RI if flexibility is needed.
Workload is unpredictable, short-term, or for testing: On-Demand.
Workload is fault-tolerant batch processing, ML training, or rendering with no time constraint: Spot Instances for maximum savings.
Workload requires per-socket or per-core software licensing: Dedicated Host.
Combination strategy: Reserved or Savings Plan for predictable baseline capacity, On-Demand for short-term spikes, Spot for flexible batch overflow.
Spot interruption: must design for it. Save work frequently, use Spot with a queue, checkpoint progress.