GCP VM Cost Calculator
Estimate the monthly and annual cost of a Google Cloud Compute Engine VM from its hourly rate and usage.
Formula last reviewed 4 August 2026 · How we verify our calculators
Monthly cost (USD)
$24.46
- Monthly cost (INR)
- ₹2,103
- Annual cost (USD)
- $293.46
Updates live as you type
Frequently asked questions
Compute Engine bills per second with a per-hour rate, so monthly cost is the hourly rate times hours run — roughly 730 hours for an always-on VM.
GCP automatically discounts VMs that run for a large share of the month, so steady workloads can cost less than the on-demand rate without any commitment.
Use committed-use discounts for predictable usage, Spot VMs for interruptible work, and right-size with e2 machine types for general-purpose workloads.
No — it covers VM compute only. Add persistent disk storage and network egress separately for a complete estimate.
The discount you get without asking
An e2-medium running continuously costs about $24.46 a month — 730 hours at roughly $0.0335 an hour — which is ₹2,103 at ₹86 to the dollar, or $293.46 a year.
Google Cloud's pricing has a feature the other major providers do not match: sustained use discounts are applied automatically, with no commitment, no reservation and no action required. Run an instance for a full month and the discount appears on the bill.
How the estimate works
Monthly cost = hourly rate × hours per month
730 hours is a full month (8,760 ÷ 12). GCP bills per second after a one-minute minimum, so partial usage is genuinely prorated.
Sustained use discounts
For general-purpose N1, N2, N2D and C2 machine types, Compute Engine automatically reduces the rate as monthly usage rises, reaching up to about 30% off for an instance running the whole month. The discount is computed on aggregate usage of a machine type in a zone, not on a specific VM, so two instances each running half the month earn the same discount as one running fully.
Note that E2 machine types, including the e2-medium in this example, are priced lower up front and do not receive sustained use discounts. That is the trade: a cheaper headline rate instead of a discount that accrues.
The other levers
Committed use discounts — one or three-year commitments to a quantity of vCPU and memory, saving up to 57% for general-purpose and more for memory-optimised. Crucially, commitments are to resources rather than instance types, so you can change machine shapes freely within the commitment. Newer flexible CUDs extend across regions and families.
Spot VMs — preemptible capacity at 60–91% off, reclaimed with 30 seconds' notice and a maximum 24-hour life on the legacy preemptible model. Well suited to batch, CI and fault-tolerant processing.
Custom machine types — a genuine differentiator. Rather than picking from fixed shapes, you specify vCPU and memory independently. A workload needing 4 vCPU and 8 GB does not have to buy a 16 GB standard shape, and paying only for what you configure often beats any discount programme.
Automatic rightsizing recommendations appear in the console based on observed utilisation, and are worth checking before any commitment.
Reading your estimate
The rate approximates a common US region. Prices differ by region, and asia-south1 (Mumbai) and asia-south2 (Delhi) carry their own rates. The estimate covers compute only.
Not included: persistent disks, billed by provisioned size and type and charged whether or not the instance is running; network egress, which is charged beyond the free allowance and is often the largest surprise on a GCP bill; static external IPs, charged when reserved but unused; and GST at 18% for Indian customers.
Common mistakes
Expecting sustained use discounts on E2. They do not apply to that family.
Leaving persistent disks attached to deleted instances. Disks survive instance deletion unless explicitly removed and keep billing.
Ignoring custom machine types. Standard shapes frequently force you to buy memory or CPU you do not need.
Forgetting egress between zones and regions. Traffic that crosses a zone boundary is charged, which catches multi-zone architectures designed for availability rather than cost.
Committing to resources before measuring. A three-year CUD on an over-provisioned footprint locks in the waste for three years.