GCP Sustained Use Discounts (SUDs) aren’t just a passive discount; they’re an active, automatic mechanism that rewrites your bill in real-time based on your actual usage, often making your first thought about them, "Wait, it just does that?"
Let’s watch it in action. Imagine you have a Compute Engine instance running n1-standard-1 in us-central1. Without any discounts, the on-demand price for that instance is $0.030 per hour.
Here’s the raw pricing for a n1-standard-1 in us-central1 (as of this writing, always check the official GCP pricing page for the absolute latest):
us-central1-n1-standard-1: $0.030/hour
Now, let’s say this instance runs 24/7 for the entire month, which has 730 hours.
- On-Demand Cost: 730 hours * $0.030/hour = $21.90
But GCP automatically applies Sustained Use Discounts. The first 25% of usage in a month gets a 7% discount. Usage beyond that, up to 50%, gets a 20% discount. And if you run something for more than 50% of the month, the discount jumps to 55%!
Here’s how the discounts stack for that n1-standard-1 instance:
-
First 25% of hours (approx. 182.5 hours):
- Discounted price: $0.030/hour * (1 - 0.07) = $0.0279/hour
- Cost for this tier: 182.5 hours * $0.0279/hour = $5.09
-
Next 25% of hours (approx. 182.5 hours), up to 50% total:
- Discounted price: $0.030/hour * (1 - 0.20) = $0.024/hour
- Cost for this tier: 182.5 hours * $0.024/hour = $4.38
-
Remaining 50% of hours (approx. 365 hours), over 50% total:
- Discounted price: $0.030/hour * (1 - 0.55) = $0.0135/hour
- Cost for this tier: 365 hours * $0.0135/hour = $4.93
-
Total Monthly Cost with SUDs: $5.09 + $4.38 + $4.93 = $14.40
That’s a savings of $7.50, or about 34%, compared to the on-demand price, all without you lifting a finger.
The system fundamentally works by tracking your commitment-free usage of specific resource types (like vCPUs, memory, GPUs, SSDs, etc.) within specific regions. As your usage of a particular resource type in a region crosses certain hourly thresholds within a billing month, the discount rate for that resource type automatically increases. It’s not a single discount applied at the end of the month; it’s a tiered, per-resource, per-region calculation that adjusts the effective hourly rate for that resource as the month progresses.
The key levers you control are not applying the discount (that’s automatic), but rather how much you use and where you use it.
- Resource Type: You get SUDs on vCPUs, memory, GPUs, and persistent disk storage. The discount rates are specific to each.
- Region: Discounts are regional. Running 100 vCPUs in
us-east1doesn’t impact the SUDs for vCPUs ineurope-west2. - Commitment Duration: The discount tiers are based on the percentage of the month you’ve used a resource. For example, running a resource for 75% of the month will get you the 55% discount for that entire 75% of usage.
This automatic nature is GCP’s way of rewarding consistent, predictable workloads without forcing you into complex upfront commitments for every single resource. It’s designed to capture the value of your baseline, always-on infrastructure. You don’t need to provision a specific machine type to get the discount; the discount applies to the usage of that machine type.
The most counterintuitive aspect is how the discounts are calculated retroactively within the billing cycle. If you have a resource running for the first 20 days of the month, and then you double its size for the last 10 days, the discount applied to the first 20 days’ usage will update. The system doesn’t just apply a discount based on your usage at the end of the month; it re-evaluates the discount rate for each hour based on the cumulative usage up to that hour. This means your bill for the first 20 days might be adjusted on day 21 if your usage pattern changes significantly, even though the discount for those first 20 days was technically determined before the end of the month.
The next thing you’ll want to understand is how Committed Use Discounts (CUDs) interact with SUDs, as they represent a more aggressive, but also more structured, way to save money.