Skip to main content
Pricing models

Spot vs Reserved vs On-Demand

Match each pricing model to your workload's interruptibility and stability.

Executive Summary & Quick Answer

Every major provider sells the same compute through three pricing lenses: on-demand flexibility, spot/preemptible auctions, and reserved commitments. Picking the wrong lens is one of the most common cloud bill mistakes.

Reviewed

2026-08-22

On-demand: the default you pay a premium for

On-demand is billed per second or hour with zero commitment. It is the right price for spiky, stateful, or hard-to-replicate workloads, and the wrong price for anything that runs 24/7 for months.

Spot: 60-90% off with an eviction notice

Spot instances are excess capacity sold at deep discounts with a short termination warning (30 seconds to 2 minutes depending on provider). They shine for batch jobs, CI runners, rendering, and fault-tolerant worker pools that checkpoint state externally.

Reserved and savings plans: commit to save

One-year and three-year commitments discount steady-state compute substantially on hyperscalers. Commit only after usage has been stable for several months, and prefer convertible terms if your instance mix may change.

A simple decision rule

Runs occasionally or must never be interrupted: on-demand. Interruptible and parallelizable: spot. Runs 24/7 with predictable size: reserved. Re-run this analysis every quarter as workloads change.

Key takeaways

  • Spot discounts range from 60% to 90% across AWS, Azure, GCP, and Alibaba.
  • Reserved commitments reward predictable steady-state workloads.
  • Match the pricing model to workload interruptibility, not to habit.

Explore Related Cloud Pricing Guides

Compare alternative regions, providers, spec tiers, and workload architectures.