If you’ve ever tried to book a hotel in downtown Las Vegas during a major tech conference, you know the anxiety. You don’t just want a room; you want a guarantee that a room exists for you.
In the AWS cloud, we call this assurance Capacity Reservations (ODCRs). It’s the ability to pay for a “room” (EC2 capacity) just to know it’s empty and waiting for you, even if you aren’t sleeping in it yet.
But here’s the rub: Empty rooms cost money. A lot of it. And for years, Finance teams have looked at those empty, reserved slots and asked, “Why are we paying for air?”
Enter AWS’s latest game-changer from November 2025: Interruptible Capacity Reservations.
Let’s break down what this means, why it’s huge for your bill, and how to explain it to your boss without using the words “stochastic optimization.”
The Analogy: The Suite, The Friend, and The Couch
Imagine you (the Reservation Owner) have reserved a block of 10 hotel suites for a big event. You’re paying for them whether you show up or not because you cannot afford to be homeless if your flight arrives late.
Currently, 5 of those rooms are sitting empty.
Before this announcement: Those 5 rooms stayed locked and empty. You paid full price. Your friend from the Data Analytics team, who is broke and just needs a place to crash, had to go rent a separate, expensive room across the street.
- Result: Your company pays for 10 rooms + 1 friend’s room. Inefficient.
With Interruptible Capacity Reservations: You give the Data Analytics team a key to your empty rooms. You say, “Hey, you can sleep here. But if my team arrives, I’m kicking you out with 2 minutes’ notice.”
- Result: Your company pays for the 10 rooms. The friend pays for the time they used your room (offsetting your bill), and they didn’t have to rent that extra one across the street.
What Actually Happened? (The Tech Bit)
AWS now allows you to mark your On-Demand Capacity Reservations as “shareable” (technically, interruptible). This is a simple toggle you can flip in the AWS Management Console or via a quick CLI command update on your existing reservations.
- The Owner (You): You still hold the “rights” to the capacity. If you need it, you get it back instantly.
- The Scavenger (Consumer): Other teams in your AWS Organization can sniff out this unused capacity and launch their workloads into it.
When to Use This (and When to Run Away)
This isn’t a silver bullet. It’s a tool for specific scenarios.
✅ USE IT WHEN:
- You have “Wasteful” Guarantees: You run mission-critical apps (like a 911 dispatch system or a Black Friday checkout page) that must have capacity on standby. You have idle slots you’re already paying for.
- You have “Scavenger” Workloads: You have another team running batch processing, CI/CD pipelines, or training AI models, stuff that can stop and start without the world ending.
- Internal Cost Shifting: You want the Data Science team to “pay” for the capacity they use, effectively subsidizing the Production team’s reservation costs.
❌ DO NOT USE IT FOR:
- Databases: Do not put your primary SQL database on interruptible capacity. If the owner reclaims that instance, your database goes poof (or at least enters a painful recovery).
- Interactive Web Apps: If a user is filling out a form on your website and the server vanishes because Production needed the capacity back, that user is gone forever.
- The Owner’s Primary App: You, the owner, don’t use the “interruptible” flag. You use the standard reservation. The other team uses the interruptible part.
The Confusion Matrix: Spot vs. RIs vs. Capacity Reservations
This is where everyone gets tripped up. Let’s clear the fog.
1. The “Coupon” vs. The “Table Reservation”
- Savings Plans (SPs) & Reserved Instances (RIs): These are financial coupons. They say, “If you promise to spend $10/hour, we’ll give you a discount.” They do not guarantee that a server will be available when you click “Launch.”
- Capacity Reservations (ODCRs): This is a VIP velvet rope. It guarantees a server is physically waiting for you. It does not give you a billing discount (unless you combine it with an SP/RI).
Think of it like this: An RI is a Groupon for dinner. A Capacity Reservation is the Hostess holding your table. You can have a Groupon but no table (wait in line). You can have a table but no Groupon (pay full price).
2. Interruptible Capacity vs. Spot Instances
Both involve getting kicked out, but the source is different.
- Spot Instances: You are using AWS’s spare capacity. AWS can kick you out if anyone in the world needs that server. It is dirt cheap (up to 90% off), but risky.
- Interruptible Capacity: You are using your company’s spare capacity. You only get kicked out if your own team needs the server back. The reliability depends entirely on your own organization’s usage patterns, not the global market.
The FinOps Verdict
This new feature is a massive win for FinOps.
Previously, Capacity Reservations were a hard sell because they looked like “wasted money” on a spreadsheet. Now, they are a dual-purpose asset: Insurance for Prod, and cheap compute for Dev.
Go look at your bills. Find where you are paying for “On-Demand Capacity Reservations” that have <50% utilization. Turn on the “Interruptible” flag, point your CI/CD pipelines to it, and watch your efficiency score turn green.
