How Committed-Use Discounts (CUDs) work on Google Cloud, the difference between resource-based and flexible CUDs, and when buying a commitment makes sense.
What Are Committed-Use Discounts?
A Committed-Use Discount (CUD) is a pricing agreement where you commit to paying for a specific amount of compute usage for 1 or 3 years. In return, you get a significant discount compared to on-demand pricing. For steady, non-interruptible workloads, CUDs are usually the deepest practical discount.
You pay for the commitment whether you use the resources or not. The trade-off is lower per-unit cost in exchange for predictability.
Resource-Based CUDs
Resource-based CUDs commit you to specific Compute Engine resources (vCPUs, memory, GPUs, Local SSD) in a particular region, project, and machine series. You define exactly what you are committing to.
Eligible resources:
- vCPUs and memory (most machine series)
- GPUs (A100, H100, L4, T4, etc.)
- Local SSD disks
- Sole-tenant nodes
- OS licenses (SUSE, RHEL)
Discount rates for hardware (resource-based, as of May 2026):
| Category | 1-Year Commitment | 3-Year Commitment |
|---|---|---|
| Most eligible machine series and hardware resources | 37% | 55% |
| Memory-optimized machine series | 37% | 70% |
| OS license commitments | Varies by license | Varies by license |
Key constraints:
- Tied to a specific region, project, and machine series by default
- Can be shared across projects in the same Cloud Billing account if resource-based CUD sharing is enabled
- Cannot be canceled after purchase
- Can be combined with a reservation for capacity guarantee
Compute Flexible (Spend-Based) CUDs
Flexible CUDs commit you to a minimum hourly spend on Compute Engine, GKE, and Cloud Run. Instead of specifying exact hardware, you commit to a dollar amount of spend.
Discount rates (flexible CUDs, as of May 2026):
| Service / Machine Series | 1-Year | 3-Year |
|---|---|---|
| E2, N1, N2, N2D, N4, N4D, N4A, C2, C2D, C3, C3D, C4, C4A, C4D, Z3 | 28% | 46% |
| GKE (Standard and Autopilot) | 28% | 46% |
| Cloud Run (instance-based billing, jobs, worker pools) | 28% | 46% |
| Cloud Run (request-based services and Cloud Run functions) | 17% | 17% |
| Local SSD disks | 28% | 46% |
| Sole-tenancy premium | 28% | 46% |
| H3, H4D (newer compute-optimized) | 17% | 38% |
| M1, M2, M3, M4 (newer model) | No discount | 63% |
Key advantages:
- Not tied to a specific project, region, or machine series
- Covers eligible spend across your entire billing account
- Covers broader eligible GKE and Cloud Run spend; resource-based CUDs apply only to matching Compute Engine resources
- For Compute Engine, eligible spend includes vCPUs, memory, Local SSD, and sole-tenancy premium; GPUs are not eligible
Resource-Based vs Flexible CUDs
| Aspect | Resource-Based | Flexible (Spend-Based) |
|---|---|---|
| What you commit to | Specific resources (vCPUs, memory, GPUs) | Minimum hourly spend |
| Scope | Specific region, project, and machine series by default; project sharing can be enabled | Entire billing account |
| Flexibility | Low (must match committed specs) | High (covers eligible spend across services) |
| GPUs included? | Yes | No |
| GKE / Cloud Run? | GKE VM resources can benefit when they match the committed Compute Engine resources; Cloud Run is not covered | Yes, for eligible GKE and Cloud Run spend |
| Deepest discount | Yes (especially memory-optimized, 3-year) | Moderate |
| Best for | Known, stable hardware requirements | Mixed workloads, GKE, Cloud Run, evolving architectures |
CUDs vs Reservations
These are separate concepts that are often confused:
| Concept | What It Does | Provides Discount? | Provides Capacity? |
|---|---|---|---|
| CUD | Pricing discount for committed usage | Yes | No |
| Reservation | Holds capacity in a specific zone | No (by itself) | Yes |
| CUD + Reservation | Discount + capacity guarantee | Yes | Yes |
A CUD gives you a lower price but does not guarantee that capacity will be available when you need it. A reservation holds capacity for you but does not change the price. For production workloads where both cost certainty and availability matter, attach a reservation to a resource-based CUD.
How Discounts Apply
Discounts are mutually exclusive per resource. Google applies them in this order:
- Resource-based CUDs first (if the VM matches a committed spec)
- Compute flexible CUDs next (if the spend is eligible)
- Sustained-Use Discounts on any remaining on-demand usage
- Spot VM pricing is separate and does not interact with CUDs or SUDs
A single VM hour can only receive one type of discount.
When to Buy CUDs
Good candidates for CUDs:
- Production VMs that run 24/7 with stable resource requirements
- Databases, application servers, API backends with predictable load
- Workloads you are confident will exist for 1-3 years
- Organizations with FinOps practices that can forecast usage
Risks of over-committing:
- You pay for the commitment whether you use the resources or not
- If you migrate to a different machine series, resource-based CUDs do not follow
- If you move workloads off Google Cloud, you still owe for the commitment term
- Flexible CUDs mitigate some of this risk but still require minimum spend
When not to buy:
- Dev/test environments with sporadic usage
- Startups or projects with uncertain timelines
- Workloads already covered by Spot VMs (CUDs do not apply to Spot)
- Situations where SUDs alone provide sufficient savings
Tip: Start with SUDs (automatic, free) and Spot VMs (for interruptible workloads). Buy CUDs only when you have 2-3 months of usage data showing stable, predictable demand. The Recommender can analyze your usage and suggest CUD purchases.
TL;DR
- CUDs give deep predictable-workload discounts (up to 55% for most hardware resources and up to 70% for memory-optimized machine series) for 1 or 3-year commitments.
- Resource-based CUDs are tied to specific hardware in a region and project by default. Best for known, stable requirements.
- Flexible CUDs commit to hourly spend across your billing account. Best for mixed workloads, GKE, and Cloud Run.
- CUDs and reservations are separate: CUDs = discount, reservations = capacity. Combine for both.
- Discounts are mutually exclusive. CUDs apply first, then SUDs on remaining on-demand.
- Do not over-commit. You pay whether you use it or not. Use 2-3 months of data before buying.
Resources
Committed-Use Discounts Overview Official documentation for resource-based and flexible CUDs.
Combine Reservations with CUDs How to attach reservations to commitments for both discount and capacity.
Share Resource-Based CUDs Across Projects How resource-based commitment discounts can be shared across projects in the same Cloud Billing account.
A FinOps Guide to Updated Spend-Based CUDs Google Cloud Blog post on the July 2025 spend-based CUD model changes.
Sustained-Use Discounts Automatic discounts for long-running VMs that require no commitment.
Spot VMs Up to 91% off for interruptible workloads.
Cost Optimization Overview of all cost levers on Google Cloud.