RI vs Spot Breakeven
Your CFO wants 3-year RIs at 40% off. Your CTO says spot is 65% cheaper and interruption risk is overblown. Both want a number, not an opinion. The breakeven depends on exactly one variable: how often your instances get reclaimed. Model it before the 36-month commitment locks you in.
📐 Pricing Scenario Configuration
Set instance pricing, RI commitment, spot discount, and interruption parameters. The engine computes annual costs across all three models and finds the breakeven interruption rate where RI and Spot equalize.
🎯 Common Scenarios
Click a preset to populate the calculator. Each models a real workload pattern: stateless web tier, batch processing, stateful analytics, and CI/CD runners.
RI vs Spot: The Commitment Trade-Off
Reserved Instances give you a guaranteed discount in exchange for a 1- or 3-year commitment. You pay for 8,760 hours per instance per year — whether your workloads run or not. Spot Instances give you a steeper discount with zero commitment — but AWS can reclaim them with a 2-minute warning. The economics are not a philosophical debate. They are a probability distribution over interruption events multiplied by the cost of recovering from each one.
How the Breakeven Math Works
This calculator computes annual costs across three models:
- On-Demand:
instances × monthly-hours × 12 × on-demand-price. Zero commitment. Maximum flexibility. Maximum cost. - Reserved Instance:
instances × 8,760 × on-demand-price × (1 − RI discount). You pay for 24/7/365. The discount is guaranteed regardless of utilization. If your workload runs less than ~60% of the year, RI overpayment eats the discount. - Spot:
instances × effective-hours × on-demand-price × (1 − spot discount). Effective hours = actual runtime + time lost to interruption recovery. Interruption rate is per-1,000-instance-hours, drawn from AWS's historical frequency data for your instance type and AZ.
The Interruption Rate Is the Only Variable That Matters
AWS publishes spot interruption frequency data but not per-customer prediction. In practice, interruption rates range from 5–20% for GPU instances (p4d, p5, g5 in us-east-1) to <1% for less-contested instance families (m6i, c6i). The cost of an interruption — minutes of lost work — varies by workload: stateless web servers lose ~10 seconds of in-flight requests; ML training runs with checkpointing lose ~5 minutes since last save; stateful analytics without fault tolerance can lose hours. This calculator lets you model both dimensions independently.
When Spot Wins (and When It Loses)
Spot wins when interruptions are rare and recovery is cheap. Stateless web tiers behind a load balancer, batch jobs with idempotent processing, and CI/CD runners with retry logic are textbook spot candidates — savings of 60–70% with near-zero operational impact. Spot loses when interruptions are frequent or recovery is expensive. GPU training runs with hourly checkpointing, stateful databases, and latency-bound streaming pipelines should strongly consider RIs or On-Demand. The breakeven line is where spot savings = RI savings + interruption cost.
The Dirty Secret: You Can Mix Both
Teams that win the cloud pricing game do not pick one model. They run a base fleet on RIs (covering the minimum steady-state load) and burst capacity on Spot (covering variable or elastic demand). A 10-instance workload might put 6 on 3-year RIs, 2 on spot, and leave 2 on on-demand as a safety buffer. The math for mixed fleets is more complex but the principle is the same: commit to what you know you will use; float the rest on spot. Use this calculator to find where your breakeven line sits, then allocate accordingly.