Stop 04 · Compute ServicesTopics · 9/10

Spot & Reserved Instances

In short

Two companies can run the same servers and pay very different bills. The difference is often just the pricing model. On-demand is only one of three options, and usually the most expensive.

Spot instances use spare provider capacity at a steep discount but can be reclaimed with little warning; reserved instances commit to usage over 1-3 years for a discount — both are cost-optimization levers beyond on-demand pricing.

The blob character grins on a platform holding a monthly pass and a discount coupon.

Think of it like this

Monthly pass vs last-minute seat

A local train monthly pass costs less per ride because you commit upfront. A last-minute empty seat at a show is a steal, but you must move if the ticket holder arrives. Reserved is the pass. Spot is the bargain seat.

Watch it happen

Two ways to pay less than on-demand

Reserved / Savings Plan

Commit for 1 or 3 years
Steady discount, no interruptions
Wasted money if usage drops

Spot

Spare capacity, deepest discount
Can be taken back at short notice
No commitment at all

Cover your steady baseline with reserved capacity or savings plans, run fault-tolerant batch work on spot, and leave the unpredictable rest on-demand.

The big ideas

What you need to know

On-demand: no strings

Pay per second or hour, start and stop anytime. Best for new or unpredictable workloads.

Commit for a discount

Reserved instances and savings plans trade a 1 or 3 year promise for a lower rate on steady usage.

Spot is interruptible

On AWS you get a two-minute warning before a spot instance is reclaimed. Design for that.

Good spot workloads

Batch jobs, CI runners, image rendering, stateless workers in an auto scaling group with mixed instance types.

How it works

Building a sensible mix

1
Measure the floor
Look at a month of usage. The level you never drop below is your baseline.
2
Commit to the floor
Cover that baseline with a savings plan or reserved instances.
3
Spot for flexible work
Move batch and stateless workers that can restart to spot.
4
On-demand for the rest
Peaks and new projects stay on-demand until their pattern is clear.

Save yourself the pain

Common mistakes

Oops

Running a single-instance production database on spot to save money.

Do this instead

Spot can vanish with little warning. Use it only for work that can stop and resume, or that runs on many replaceable nodes.

Oops

Buying 3-year reservations for a project still being designed.

Do this instead

Commit only to usage you're confident about. Start with 1 year or a flexible savings plan.

Your mission

Price the same server three ways

~20 min

Quick check

Which workload is the best fit for spot instances?

Go deeper

1Compare on-demand, spot, and reserved pricing for the same instance type.
2Learn what kind of workloads fit spot instances well (fault-tolerant, interruptible batch jobs).
3Understand why reserved instances require confidence in long-term, predictable usage.