Committed use discounts (CUDs) are the discounts Google Cloud grants for a one- or three-year commitment, either to a minimum quantity of resources in a region (resource-based) or to a minimum hourly spend on eligible services (spend-based). A committed use discount applies automatically, and the commitment is billed for the whole term whether usage occurs or not; it cannot be cancelled. For the bill, CUDs lower the price of predictable base load, and their value depends on usage filling the commitment every hour. Spend-based CUDs are also moving from credits to direct discounts, which changes the billing data.
How it works
Resource-based CUDs
Resource-based CUDs exist only for Compute Engine. You buy a quantity of vCPUs, memory, GPUs, Local SSD or sole-tenant nodes of one machine series in one region; operating system licences are separate commitments. The discount applies to the total of those resources in the region, whatever the VM size or machine type. The commitment is bought in a project; with CUD sharing, on by default for Cloud Billing accounts created on or after 2026-06-16, it covers every project of that account. It becomes active the day after purchase and is billed monthly. The plan, one or three years, sets the discount; the term may be longer, under three years on a one-year plan and under six on a three-year plan. Resource-based commitments can renew automatically, and only they can be bought with attached reservations that assure capacity.
Spend-based CUDs
Spend-based CUDs are bought on the Cloud Billing account, Google Cloud's billing account, and apply to eligible usage in every project it pays for; they cannot be shared across billing accounts. Compute flexible CUDs cover Compute Engine, GKE and Cloud Run with one hourly commitment, regardless of project, region and machine type. Service-specific commitments apply to one service, such as Cloud SQL, AlloyDB, Spanner, Bigtable or BigQuery; depending on the service only in the region bought (Cloud SQL) or in every region of the billing account (AlloyDB, Spanner, Bigtable); reserved capacity for databases and managed services deals with them. Flexible Savings Plans measure the commitment over a monthly rather than an hourly window.
The new consumption model
As read on 2026-10-16, Google states that some spend-based CUDs migrate automatically to a new consumption model with discounts instead of credits, announcing the date on the console's Billing Overview page; the spend-based CUD page also says most have already moved. Its table lists compute flexible CUDs, Cloud SQL, AlloyDB, BigQuery, Spanner, Bigtable, Memorystore and Dataflow among those affected; VMware Engine, NetApp Volumes, Backup and DR for VMware Engine and Backup for GKE stay on the credit model. Resource-based CUDs are not affected.
| Aspect | New model | Legacy model |
|---|---|---|
| Commitment stated as | discounted spend per hour | on-demand spend per hour |
| Covered usage | billed at the discounted price | billed at list price, offset by a credit |
| Fee SKU | new SKU priced at USD 1 per hour | old SKU priced lower |
| Savings in reports | net | gross credit amount |
Order of application
Compute Engine discount types are mutually exclusive. Every hour Google applies resource-based CUDs first, then compute flexible CUDs to the remainder; what is left pays on-demand and may still receive sustained use discounts. Among spend-based commitments, service-specific ones go before compute flexible ones and, among compute flexible commitments, the older before the newer; within one commitment the usage with the highest discount comes first, then the rest proportionally across Compute Engine, GKE and Cloud Run. Spot and preemptible VMs are excluded from compute flexible CUDs.
What it means for your cost
In the billing export to BigQuery, CUDs appear differently by type:
| CUD type | Covered usage | credits.type |
|---|---|---|
| Resource-based | on-demand price offset by a credit; the commitment fee alongside | COMMITTED_USAGE_DISCOUNT |
| Spend-based, legacy model | list price offset by a credit; the commitment fee alongside | COMMITTED_USAGE_DISCOUNT_DOLLAR_BASE |
| Spend-based, new model | discounted price with its own consumption_model; the credit offsets the fee | FEE_UTILIZATION_OFFSET |
In the new model the commitment appears every hour as Commitment Fee SKU, and FEE_UTILIZATION_OFFSET cancels it as far as usage is covered. At full utilisation the fee nets to zero; whatever remains is unused commitment. Usage above the commitment carries the Default consumption model and on-demand price. For compute flexible CUDs, consumption_model.id values D97B-0795-975B (one year) and 70D7-D1AB-12A4 (three years) are steadier than the description, written inconsistently as Compute Flexible CUD - 1 Year or Compute Flexible CUDs - 1 Year. In the standard export these fields are filled from 2025-07-15. Set up Google Cloud billing export for cost reporting shows how to create the export.
CUDs are billed as they run, so there is no upfront payment to spread as amortized cost. The question is attribution: Google distributes fees and credits to projects proportionally or by priority for resource-based CUDs, only proportionally for spend-based ones.
The CUD analysis report in the console, visible to Billing Account Administrators and Billing Account Viewers, has two tabs: CUD Coverage shows the covered share of eligible spend at on-demand rates, and CUD Utilization utilisation per commitment or scope, daily, or hourly for up to 30 days. The report uses cost data only and shows no utilisation for combined CUD types. Commitment coverage and utilisation explains how the two measures are read together.
| Programme | Binds | Applies to | Exit |
|---|---|---|---|
| Resource-based CUD | resource quantity of one series in one region | project or, with sharing, Cloud Billing account | none |
| Compute flexible CUD | spend per hour | Cloud Billing account, all regions | none |
| AWS Reserved Instances | instance attributes in one region | account, sharing in the organisation | sale or exchange, by class |
| AWS Savings Plans | spend per hour | account, sharing in the organisation | return only within narrow limits |
| Azure reservations, savings plans | attributes, or spend per hour | resource group up to shared scope | reservations within limits, plans none |
Where it goes wrong
- Wrong region. Symptom: a resource-based commitment sits unused while the same series pays on-demand in another region. Cause: resource-based CUDs and some service-specific ones, such as Cloud SQL, apply only in the region bought.
- Sharing off. Symptom: unused commitment in one project, on-demand usage of the same series in another. Cause: CUD sharing is not always on in older Cloud Billing accounts.
- Dashboards on the old schema. Symptom: after migration, credits seem to be missing and fees higher. Cause: the new model bills usage at the discounted price and offsets the USD 1 fee SKU with
FEE_UTILIZATION_OFFSET. - Commitment sized in on-demand terms. Symptom: a new spend-based commitment is larger than intended and stays partly unused. Cause: the new model states the commitment in discounted spend, the legacy model in on-demand spend.
- One-year commitment for memory-optimised VMs. Symptom: the commitment is consumed but usage still costs full price. Cause: compute flexible CUDs give M1 to M4 a discount only on three-year terms; their spend burns down the commitment without one.
Example
On 2026-10-16 the general-purpose VM pricing page showed, for an n2-standard-2 (2 vCPUs, 8 GiB) in Frankfurt (europe-west3):
| Pricing model | Price per hour |
|---|---|
| Default (on-demand) | USD 0.125124 |
| Compute flexible CUD, 1 year | USD 0.09008928 |
| Compute flexible CUD, 3 years | USD 0.06756696 |
| Compute resource CUD, 1 year | USD 0.078752 |
| Compute resource CUD, 3 years | USD 0.056246 |
The compute flexible prices are exactly the default price less the 28% and 46% documented for N2.
Sources
Read on 2026-10-16. Google Committed use discounts, Committed use discounts (CUDs) for Compute Engine, Resource-based committed use discounts, Spend-based committed use discounts, Spend-based committed use discounts: purchase and cancellation, Structure of Standard data export, Analyze the effectiveness of your CUDs, General Purpose VM pricing.