On-Demand vs Reserved vs Spot pricing models

WHAT THIS TESTS This checks whether you understand the three core compute purchasing models and can reason about cost versus availability tradeoffs rather than just naming them.

A GOOD ANSWER COVERS On-Demand pricing charges per second or hour with zero commitment and the highest unit price, making it right for spiky, short-lived, or unpredictable workloads. Reserved Instances on AWS, or Committed Use Discounts on Google Cloud, exchange a one or three year commitment for discounts often in the range of thirty to seventy percent, fitting predictable steady-state baselines like a core database or always-on API tier. Spot Instances and Preemptible VMs sell spare capacity at the steepest discount but the provider can reclaim them with only a short warning, so they fit fault-tolerant, stateless, or batch work. A strong answer also notes that these blend: cover the steady floor with reservations and burst with On-Demand or Spot.

COMMON WRONG ANSWERS Claiming Spot is simply cheaper with no caveats, ignoring interruption risk. Treating Reserved Instances as a different machine type rather than a billing construct. Forgetting that commitments lock capacity or spend and can become waste if usage drops.

LIKELY FOLLOW-UPS How do Savings Plans differ from Reserved Instances. How would you size a commitment. How do you make a Spot workload resilient to reclamation.

ONE CONCRETE EXAMPLE A video platform runs its always-on streaming API on three-year reservations for predictable savings, scales evening traffic spikes with On-Demand, and renders thumbnail batches on Spot, retrying any interrupted jobs. This blend cuts cost without risking the user-facing path.