Three CUD variants: resource based (deepest, rigid), flexible (best balance, up to 46 percent three year), and spend based (managed services). The p20 baseline sizing rule. BigQuery slot commitments at 40 percent below on demand. Eleven buyer moves.
Google Cloud Committed Use Discounts reward buyers who commit only to the consumption floor they can prove, and punish those who commit to forecasts that never arrive.
Key takeaways
A CUD trades a one or three year spending commitment for a discounted rate. The deeper the rigidity, the deeper the discount.
The three variants sit on a flexibility curve. Resource based is cheapest and most rigid, spend based is most flexible and shallowest.
Resource based three year commitments reach the deepest rates, with current bands published on the Compute Engine pricing page. Flexible sits below that, and spend based is shallower again.
The published CUD rate is only half the picture. The platform wide discount and the commitment architecture move more money.
Three levers carry most of the value, and they stack.
Google Cloud CUD variant comparison, illustrative
| Variant | Discount depth | Flexibility | Best fit |
|---|---|---|---|
| Resource based | Deepest, up to 70% three year | Low, fixed family and region | Stable known workloads |
| Flexible | Moderate | High, any family or region | Mixed Compute Engine estates |
| Spend based | Shallowest | Service specific | Managed services like BigQuery |
The standard account team pitch is to maximize the resource based commitment because it carries the deepest discount, so buyers commit to the full forecast. We disagree. In roughly 25 of the 35 Google Cloud estates we benchmarked in 2024 and 2025, the deeper rate was wiped out by 15 to 25 percent of unused commit that billed anyway when the forecast missed. The buyer side move is to commit resource based only to the p20 consumption floor you can prove from twelve months of telemetry, cover the variable band with flexible CUDs, and burst the rest on demand, because an unused deep discount costs more than a used shallow one.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
An unused deep discount costs more than a used shallow one. Commit to the floor you can prove.
Take twelve months of hourly consumption, find the level you exceed 80 percent of the time, and commit resource based to that floor. That is the p20 baseline.
Everything above the floor goes to flexible CUDs or on demand. The floor is almost never wrong, so the deep discount is almost never wasted.
Resource based CUDs only apply to the exact family and region you commit. Classify consumption first, or the discount strands when workloads move.
Bring telemetry, not a forecast. The commitment that survives an audit is the one built from history.
Sustained use discounts apply automatically to steady running instances with no commitment, while CUDs require a one or three year commit for a deeper rate. The two coexist on the same estate.
Bring twelve months of hourly usage classified by machine family, region, and service. That history, not a sales forecast, sets a commitment that survives the term.
Get the buyer side framework our advisors use on live Google Cloud engagements. The CUD math, the p20 baseline rule, and the commitment sequence.
Download the GCP FrameworkResource based CUDs commit to specific vCPU and memory in a region for the deepest discount, flexible CUDs commit to an hourly spend across families and regions for a moderate discount, and spend based CUDs commit to a managed service spend for a shallower service specific discount.
Resource based three year commitments reach the deepest rates, up to roughly 70 percent on covered compute, while a layered architecture across resource based, flexible, and on demand typically moves 15 to 30 percent of total compute spend in practice.
Use resource based for the stable consumption floor where the machine family and region are known with high confidence, and flexible for the variable band because it applies automatically across families and regions even when workloads move.
The p20 baseline is the consumption level you exceed about 80 percent of the time over twelve months. Committing resource based only to that floor means the deep discount is almost never wasted on capacity you do not use.
The shortfall trap is committing to a forecast that never arrives. A resource based CUD bills the committed amount whether or not you consume it, so an over sized commitment quietly erases the discount it was meant to deliver.
Yes. The platform wide discount applies on top of the CUD architecture, so the two are complementary rather than alternatives, and the strongest position stacks both.
Negotiate CUDs alongside the broader platform discount conversation, after you have twelve months of consumption telemetry classified by machine family, region, and service, so the commitment is built from history rather than a sales forecast.
BigQuery slot commitments are sized to steady state query demand and typically land 30 to 40 percent below on demand pricing when committed at the right level, which is why steady state sizing matters more than peak.
