Google Cloud Committed Use Discounts are the workhorse savings instrument across GCP. Compute, memory, GPU, BigQuery, and Spanner each carry distinct commit math. The buyer side view reads the math, the leverage, and the exit posture.
GCP CUDs sit at the center of the Google Cloud commercial estate. A buyer side read of the CUD landscape changes what an enterprise pays across the three year horizon.
Read this pillar alongside the GCP CUD 2026 guide, the CUD negotiation tactics, the FinOps CUD playbook, and the Google Cloud practice page.
The CUD framework matters because the commit posture decides the bill across three years. A buyer that signs the standard public CUD without negotiating the custom layer leaves discount on the table.
Google sells two CUD families. Resource based and spend based. Each family carries its own math, its own term structure, and its own override posture.
Resource based CUDs apply to compute and memory. The commit buys a specific quantity of vCPU and memory. The discount applies to that quantity across the term.
Spend based CUDs apply to specific services. The commit buys a dollar spend per month. The discount applies to that service spend across the term.
Flex CUDs are spend based commits that apply across multiple compute families and across regions. The flex layer trades depth for flexibility.
Compute CUDs are the workhorse. Most GCP estates anchor the commit posture on Compute Engine vCPU and memory. The discount math is well documented but the negotiation layer is often missed.
Public CUDs run 37 percent off list at one year and 55 percent off list at three years for general purpose compute. Memory commits run at similar tiers.
Custom CUDs unlock additional discount through bespoke negotiation. The custom layer typically lands 5 to 12 points deeper than the public ladder.
GCP CUD discount ladder by family
| CUD family | 1 year discount | 3 year discount | Custom layer above |
|---|---|---|---|
| General purpose compute | 37 percent | 55 percent | +5 to +12 points |
| Memory optimized | 37 percent | 55 percent | +5 to +10 points |
| GPU A100 | 30 percent | 55 percent | +8 to +15 points |
| GPU H100 | 25 percent | 52 percent | +10 to +18 points |
| TPU v5 | 30 percent | 50 percent | +5 to +12 points |
| BigQuery annual slots | 20 percent | 40 percent | +4 to +10 points |
Memory commits run separately from vCPU. The two commits work as a pair. Buying one without the other leaves on demand exposure on the unbalanced side.
The vCPU commit and the memory commit must align with the actual deployed ratio. A 1 to 4 vCPU to memory ratio on the deployed estate maps to the same ratio on the commit.
GPU commits are a distinct category. AI and ML workloads drive demand. The 2024 and 2025 cycles saw GPU CUDs become a major line item across the customer base.
GPU CUDs cover specific GPU families. A100, H100, L4, and TPU each carry separate commit math. Mixing GPU types in one commit is not supported.
GPU commits include capacity reservation. The commit guarantees the GPU is available when needed. The capacity layer matters as much as the discount layer for AI workloads.
The standard Google Cloud account team pitch is that a five year custom CUD captures the deepest discount tier and signals strategic commitment. We disagree. In roughly five out of seven GCP estates we have advised, the five year CUD locked in compute generations that aged out before the term, leaving stranded commit on retired SKUs while AWS and Azure competed for the same workload class. The buyer side move is to default to three years on resource based CUDs, refresh the CUD mix every 12 to 18 months as the SKU map shifts, and reserve five year terms for highly visible workloads where the technology roadmap is settled.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
“The public CUD page is the floor. The custom layer is where the negotiated reality sits. A buyer that only reads the public page is reading the marketing brochure.”
BigQuery commit posture sits outside the CUD framework. Flex slots are reserved capacity for BigQuery. The reservation math is slot based not vCPU based.
Flex slots reserve query capacity. The reservation is measured in slots. One slot equals one unit of query processing capacity per second.
The break even between on demand BigQuery and flex slot reservation depends on the query volume and the slot utilization. Most enterprise BigQuery estates cross break even at 500 to 2000 slot equivalent demand.
The negotiation posture decides whether the buyer captures the custom layer above the public CUD ladder. Posture matters more than tactical moves in the room.
Leverage sits in three places. The total spend commitment. The competitive alternative. The term length. A buyer that brings all three to the table captures the deepest custom rates.
Step down clauses, custom rate freezes, and the no worse than language sit at the contract layer. A buyer that negotiates only the price misses the contract protection.
Exit options exist but are contractual. Step down rights, partial termination, and change of control protection vary by deal. Default Google paper does not include strong buyer side exit.
Step down rights let the buyer reduce the commit at defined milestones. The right is negotiated, not standard. Most custom deals carry some step down.
A Google Cloud Committed Use Discount is a contractual commitment to use a specific quantity of a Google Cloud resource for one or three years in exchange for a discount on the public price. CUDs apply to Compute Engine, memory, GPU, TPU, and select database services.
A resource based CUD commits to a specific quantity of vCPU, memory, or GPU. A spend based CUD commits to a dollar spend per month on a specific service such as Cloud SQL or Spanner. The two families carry different math and different override rules.
The custom layer typically sits 5 to 12 points deeper than the public CUD ladder for general purpose compute. GPU custom layers can sit 8 to 18 points deeper depending on family and term. The custom layer requires direct negotiation with Google.
No. Step down rights are negotiated. Default Google paper does not include step down. A buyer side negotiation typically secures 20 to 30 percent step down at year 2 and a larger step down at year 3 with notice.
BigQuery on demand prices per terabyte scanned. Flex slots reserve query capacity per slot per second. Break even sits between 500 and 2000 slot equivalent demand. Above 2000 slots, the annual slot commit dominates.
Standard CUDs cannot be exited early without forfeiting the discount. Negotiated step down clauses allow planned reductions. Change of control language can protect the buyer in acquisition scenarios.
Open with an inventory and entitlement baseline before any vendor conversation. Pull trailing twelve months of usage data, score it against contracted scope, and document the gap. The single most common reason buyers leave money on the table is opening the negotiation without a defensible baseline. The buyer side calendar starts at 270 days out, not at 60.
Redress is 100 percent buyer side. We hold no publisher partnerships, take no publisher commissions, and operate no referral revenue. The advisory practice is funded entirely by enterprise buyer subscriptions and engagement fees. That means the recommendations on this page reflect what we have measured in client engagements, not what a publisher or partner wants the market to believe.
Google Cloud commitment posture, custom discount mechanics, marketplace strategy, and the buyer side moves across the GCP estate.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
“CUDs are not a finance instrument. They are a contract instrument. The buyer that treats commit as math alone misses the negotiated lever above the public CUD rate.”
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
CUD coverage moves, custom commit posture, marketplace strategy, and the buyer side playbook. One email per month.
