Compute Engine
GCP Compute Engine — commitment brief
Data sourced: March 2026. Verify current figures at the Google Cloud Pricing Calculator.
Coverage summary
Compute Engine supports two commitment instruments: resource-based CUDs (commit to a specific quantity of vCPUs, memory, GPUs, or Local SSD in a region) and compute flexible CUDs (spend-based commitments that apply across multiple VM families and regions). Resource-based CUDs deliver the highest per-resource discount and are best for stable, predictable workloads. Compute flexible CUDs offer a simpler, flat-rate discount that works even when the VM mix or region shifts. Both instruments are applied at the billing-account level; resource CUDs apply first, then flexible CUDs cover remaining eligible spend.
What is covered: vCPU and memory costs for eligible machine families, committed GPU quantities (A100, L4, and others on eligible families), and Local SSD capacity. OS license commitments are also available separately.
What is not covered: Shared-core machine types (f1-micro, g1-small, and similar), preemptible/Spot VMs, premium OS licenses (charged separately), networking, disk storage beyond Local SSD commitments, and machine families explicitly excluded from CUDs (see table below).
CUD types
Compute Engine supports both resource-based CUDs and compute flexible CUDs. Resource-based CUDs commit to a fixed quantity of vCPUs, memory, or GPUs in a specific region; 1-yr and 3-yr terms are available. Compute flexible CUDs commit to a minimum hourly spend across eligible VM families and regions within a billing account; discounts vary by family: 28% (1-yr) / 46% (3-yr) for most general-purpose and compute-optimized families; 17% (1-yr) / 38% (3-yr) for H3/H4D; no 1-yr discount / 63% (3-yr) for memory-optimized M-series. Note on Sustained Use Discounts (SUDs): SUDs do not apply to resources already covered by a CUD. On uncommitted resources, SUDs continue to accumulate automatically throughout the month.
Machine type / SKU coverage
N1 (General Purpose)
✅ Yes
✅ Yes
~37% (resource) / ~28% (flex)
~55% (resource) / ~46% (flex)
Eligible for SUDs on uncommitted usage; shared-core N1 excluded from CUDs
N2 / N2D (General Purpose)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
SUD max 20% on uncommitted usage
N4 / N4D / N4A (General Purpose)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
E2 (General Purpose)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
No SUD; cost-optimized family
T2D / T2A (General Purpose)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
Arm-based T2A; no SUD
C2 / C2D (Compute Optimized)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
SUD max 20% on uncommitted C2 usage
C3 / C3D / C4 / C4A / C4D (Compute Optimized)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
H3 / H4D (HPC Optimized)
✅ Yes
✅ Yes
~37% (resource) / ~17% (flex)
~55% (resource) / ~38% (flex)
H3/H4D flex CUD 1-yr 17%, 3-yr 38%; resource CUD up to ~55%
M1 / M2 (Memory Optimized)
✅ Yes
✅ Yes
~37% (resource) / no flex discount
~70% (resource) / ~63% (flex)
Flex CUD: no 1-yr discount; 3-yr only. SUD max 30% on uncommitted usage
M3 / M4 (Memory Optimized)
✅ Yes
✅ Yes
~37% (resource) / no flex discount
~70% (resource) / ~63% (flex)
Flex CUD: no 1-yr discount; 3-yr only
Z3 (Storage Optimized)
✅ Yes
✅ Yes
~37% / ~28%
~55% / ~46%
A2 (Accelerator Optimized — A100)
✅ Yes
❌ No
~37%
~55%
Resource CUD only; not covered by flex CUD
A3 (Accelerator Optimized — H100)
✅ Yes
❌ No
~37%
~55%
Resource CUD only; not covered by flex CUD
G2 (Accelerator Optimized — L4)
✅ Yes
❌ No
~37%
~55%
Resource CUD only; not covered by flex CUD
Shared-core (f1-micro, g1-small, e2-micro, etc.)
❌ No
❌ No
—
—
Excluded from all CUDs
Preemptible / Spot VMs
❌ No
❌ No
—
—
Spot pricing model; no commitment instruments
Resource CUD vs. Compute Flexible CUD
Resource-based CUD: Commits to a specific quantity of vCPUs, memory, GPUs, or Local SSD in a single region. Provides the highest discount ceiling — up to ~55% for most families and ~70% for memory-optimized (M1/M2/M3/M4) on 3-yr terms. Best when your resource footprint (family, size, region) is stable and predictable. Commitment is locked to a region; unused commitment is still billed.
Compute Flexible CUD: Commits to a minimum hourly spend (in dollars) across eligible VM families and regions within a billing account — similar in concept to AWS Compute Savings Plans. Flat-rate discounts for general-purpose and compute-optimized families: 28% on 1-yr terms, 46% on 3-yr terms. H3/H4D families: 17% (1-yr), 38% (3-yr). Memory-optimized M-series: no discount on 1-yr terms; 63% on 3-yr terms only. Covers Compute Engine, GKE, and Cloud Run within the same billing account. Best when your VM family mix or region footprint shifts, or when you need a simpler purchasing experience.
Sustained Use Discounts
Sustained Use Discounts (SUDs) are automatic discounts applied to N1, N2, N2D, C2, M1, and M2 instances that run for a significant portion of the billing month. SUDs accumulate incrementally — N1/M1/M2 machines earn up to 30% maximum SUD; N2/N2D/C2 earn up to 20%. Resources covered by a CUD do not also earn SUDs. The optimal strategy is to use CUDs for your committed baseline (earning the higher CUD discount), and let uncommitted burst usage earn SUDs automatically. Machine families without SUD eligibility (E2, T2D, T2A, C3, N4, Z3, A2, A3, G2) should be covered by CUDs or flex CUDs for maximum savings.
Regional availability
Resource-based CUDs and compute flexible CUDs are available in all commercial GCP regions for broadly-deployed machine families (N1, N2, N2D, E2, T2D, C2, C2D, C3). However, several machine series have restricted regional availability that directly limits where CUDs for those machines can be purchased.
Notable restrictions: M3 and M4 memory-optimized machines are available only in select regions and zones — verify availability in your target region before planning M3/M4 commitments. A3 (H100 GPU) machines are restricted to a very small number of zones globally (approximately 5 zones) and may require contacting your account team for quota. A2 (A100) machines are available in roughly 15 zones. G2 (L4 GPU) machines are the most broadly available GPU option, deployed in 40+ zones. H3 and H4D HPC-optimized machines are also limited to select zones.
⚠️ CUD availability varies by machine type and region. Always verify at GCP regions and zones before purchasing.
Archera
Google Cloud Compute Engine is a core service within Archera's commitment management scope. Archera can automate resource-based CUD and compute flexible CUD purchase, monitor utilization by region and machine family, right-size commitments as your fleet evolves, and wrap commitments in a GRI/GSP — eliminating downside risk on over-commitment while preserving the full CUD discount.
Sources
⚠️ Discount percentages are approximate and region/machine-type-dependent. Always verify with the Google Cloud Pricing Calculator.
Last updated
Was this helpful?