Pricing & discounts
How Compute Engine charges for VMs, and the four ways to pay less. For the Spot/preemptible provisioning model see Special VM types; for the which lever do I pull decision see Cost and Utilization.
How you are billed
Every vCPU, GPU, and GB of memory is charged a minimum of 1 minute. Run a VM for 30 seconds and you still pay for 1 minute. After the first minute, billing is in 1-second increments.
Compute Engine uses resource-based pricing: each vCPU and each GB of memory is billed separately, not as one bundled machine-type price. You still create instances from predefined machine types, but the bill reports them as individual vCPUs and memory used. This is what lets sustained use discounts pool your usage across a region (see the example below).
Discount types
There are several discounts, but the discount types cannot be combined. A given resource gets one kind of discount, not several layered together.
| Discount | How you get it | Best for |
|---|---|---|
| Sustained use (SUD) | Automatic - just run the resource a large share of the month | A VM you leave running most of the month |
| Committed use (CUD) | Commit to 1-year or 3-year usage up front | Stable, predictable baseline load |
| Spot / preemptible | Choose the provisioning model at VM creation | Fault-tolerant, interruptible batch |
- Commitment term is 1 year or 3 years; you buy a fixed amount of vCPUs and memory.
- Up to 57% off for most machine types and custom machine types.
- Up to 70% off for memory-optimized machine types.
Sustained use discounts (SUD)
An automatic discount for running a resource (vCPU, memory, GPU) for a significant portion of the billing month. Once usage passes 25% of the month, each additional incremental minute is billed at a lower rate; the discount grows with use, up to a 30% net discount for a resource that runs the entire month.
The two rate schedules depend on the machine family:
| Usage level (% of month) | N1 / custom / memory-optimized / shared-core / sole-tenant (up to 30%) | N2 / N2D / compute-optimized (up to 20%) |
|---|---|---|
| 0% - 25% | 100% of base rate | 100% of base rate |
| 25% - 50% | 80% of base rate | 86.78% of base rate |
| 50% - 75% | 60% of base rate | 73.3% of base rate |
| 75% - 100% | 40% of base rate | 60% of base rate |
- Below 25% month use: no SUD.
- 50% of the month → ~10% effective discount.
- 75% of the month → ~20% effective discount.
- 100% of the month → ~30% effective discount.
SUD is calculated per calendar month and resets at the start of each month. To get the full 30%, create the VM on the first day of the month - a VM started mid-month can't reach 100% month-use.
Resource-based pooling (example)
Because billing is per-vCPU and per-GB, Compute Engine breaks every predefined machine type into raw vCPUs and memory and combines usage across all instances in a region to qualify for the largest SUD possible - it does not discount each machine type in isolation.
Two instances in us-central1 in one month: an n1-standard-4 (4 vCPU, 15 GB) for the first half, then an n1-standard-16 (16 vCPU, 60 GB) for the second half. Compute Engine re-groups this into 4 vCPUs + 15 GB running the full month (earns up to 30% SUD) and 12 vCPUs + 45 GB running half the month (earns ~10% SUD).
Custom machine types & sizing
Custom machine types let you tune exact vCPU and memory for finer pricing control - see Machine families & types. Compute Engine also surfaces VM sizing (rightsizing) recommendations; a recommendation for a new instance appears 24 hours after it is created. Acting on it is covered in Cost and Utilization.
Free usage limits
Compute Engine has Free Usage Limits - a small always-free allotment (e.g. one small e2-micro instance in eligible US regions, plus limited disk and egress) that isn't billed. Use the Google Cloud Pricing Calculator to estimate SUD and total cost for any workload.