A risk adjusted GCP commitment strategy starts with the right instrument and ends with the right coverage. Resource based committed use discounts bind you to a specific machine family and region for one or three years and give the deepest rates, often in the region of 37 percent for one year and up to 55 percent or more for three years on general purpose (indicative, confirm current GCP pricing). Spend based committed use discounts commit to an hourly dollar amount and float across machine types or services, trading some discount depth for flexibility. Both are use it or lose it and cannot be cancelled. Sustained use discounts sit underneath for free on eligible families. Cover the baseline you can defend with billing history, choose resource based for settled workloads and spend based where shape is uncertain, and leave the volatile top on demand.
This is where most GCP savings live and most stranded commitments are created. The mechanics below decide which.
Resource based versus spend based: which CUD should you buy?
The two committed use discount types trade depth against flexibility.
Resource based CUDs commit you to a quantity of vCPU and memory in a specific machine family and region. They carry the deepest discount, but the commitment is specific: it only covers that family in that region, and if you migrate families or regions the commitment does not follow. They suit a settled, predictable workload that will not move.
Spend based CUDs commit you to an hourly dollar spend, and any eligible usage draws it down regardless of machine type. The flexible variant covers a broad set of compute families, and spend based commitments also exist for other services. The discount is usually a little shallower than the equivalent resource based commitment, and that gap buys you the freedom to rebalance your fleet across the term.
The buyer rule mirrors the other clouds: resource based for the hardened, settled baseline that will not move; spend based for the steady but evolving band where you expect to change machine families. Many estates layer both.
Why does use it or lose it set the real risk?
Both committed use discount types are commitments in the strict sense. You cannot cancel a CUD. Once purchased it bills for the full one or three year term whether or not you use it. There is no exchange and no refund equivalent to some other clouds' reservation escape hatches. That makes oversizing a permanent cost, not a recoverable one.
Resource based commitments add a second dimension of risk: because they are scoped to a family and region, a migration or a region change can strand a commitment that still bills while covering nothing you run. Spend based commitments are more forgiving on that axis because they float across families, which is why they are the safer choice when your architecture is still moving.
The implication is the same discipline as elsewhere: size commitments to the floor of your usage, the spend you are confident exists for the whole term, not the average and never the peak.
How do sustained use discounts change the math?
GCP is unusual in giving you an automatic discount before you commit anything. Sustained use discounts apply for free to eligible families that run most of the month, up to about 20 or 30 percent depending on family, as covered in sustained use discounts explained. That changes the commitment calculation in two ways.
First, the real comparison for a steady workload is the committed use rate against the sustained use rate it already earns, not against full on demand. The incremental saving from committing is the gap between the two, and it is still usually worth it because committed use discounts go meaningfully deeper. Second, usage under a CUD does not also earn the sustained use discount, so you are not losing the free discount by committing, you are replacing it with a deeper one. For workloads too variable to commit, the sustained use discount remains the free floor.
A worked example: layering CUDs over a free floor
A European SaaS company ran a steady GCP production fleet plus a variable analytics tier. A vendor model proposed three year resource based CUDs sized to the total run rate for the deepest headline discount. Billing history showed only about 60 percent of that run rate had held for a full year on settled families, while the analytics tier swung and was mid migration to a different machine series. We covered the settled 60 percent with three year resource based CUDs, used a spend based commitment for the evolving band so it would follow a family change, and left the variable analytics on demand where sustained use discounts applied for free. Coverage stayed near fully utilized, and nothing was stranded when the analytics tier later changed families. Figures are verified against billing data and anonymised.
How does this fit a wider GCP cost program?
Commitments reward a clean estate. Rightsize first, schedule non production off as covered in scheduling non production environments off on GCP, then commit the baseline that remains. Choose resource based for settled families and spend based where shape is moving, let sustained use discounts cover the free floor, and review coverage and utilization monthly. The full sequence sits in the GCP cost optimization guide, and the same forecast discipline carries across providers in the cross cloud cost optimization guide.
Frequently asked questions
Get the buyer side commitment playbook
We size GCP committed use discounts to a forecast your finance team can defend and choose resource based versus spend based deliberately, with zero provider commissions. It is the same discipline behind the 31 percent median reduction we see in the first 90 days. Our guarantee: we reduce your cloud spend or we reimburse our service fee.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.