On-Demand vs Reserved vs Spot pricing

WHAT THIS TESTS The interviewer wants to see that you optimize cost by aligning the pricing model with the workload's predictability and tolerance for interruption.

A GOOD ANSWER COVERS Describe the three models on two axes: commitment and reliability. On-Demand has no commitment and full availability, but the highest per-hour price, suiting unpredictable, spiky, or experimental workloads. Reserved Instances or the more flexible Savings Plans trade a one or three year commitment for a large discount, fitting a steady baseline you know you will always run. Spot draws on the provider's spare capacity at the deepest discount but can be reclaimed on short notice, making it ideal for stateless, interruptible, fault-tolerant work. A mature strategy blends all three: Reserved or Savings Plans for the baseline, On-Demand for normal variable load, and Spot for opportunistic burst.

COMMON WRONG ANSWERS Claiming Spot is just cheaper On-Demand with no downside, ignoring reclamation. Running stateful or latency-critical services on Spot. Paying On-Demand for a baseline that never changes, wasting money. Over-committing Reserved capacity you cannot actually use.

LIKELY FOLLOW-UPS How do you handle Spot interruptions gracefully. Savings Plans versus Reserved Instances. How do you size a Reserved commitment without over-buying. How do you mix Spot and On-Demand in one auto-scaling group.

ONE CONCRETE EXAMPLE A video platform reserves capacity for its always-on API tier via a one-year Savings Plan, serves variable user traffic with On-Demand instances in an auto-scaling group, and runs its nightly transcoding pipeline on Spot. The transcoder checkpoints progress to object storage, so when Spot capacity is reclaimed a replacement instance resumes from the last checkpoint, cutting compute cost dramatically without risking the live service.