Google has a million machine types. Here are the ones that matter for small-to-medium workloads.
Cheapest. Shared vCPUs (you share the physical core with other tenants). Fine for bots, web servers, light workloads. Bursts when needed.
| Type | vCPUs | RAM | $/hour | $/month | Notes |
|---|---|---|---|---|---|
| e2-micro | 2 | 1 GB | $0.0084 | $6.11 | FREE TIER 1 instance/mo |
| e2-small | 2 | 2 GB | $0.0168 | $12.23 | FLEET STD What we run bots on |
| e2-medium | 2 | 4 GB | $0.0336 | $24.46 | WALTER Our beefiest shared-core |
Shared-core means 0.25 vCPU (micro), 0.5 vCPU (small), 1 vCPU (medium) baseline — can burst to 2 vCPUs temporarily. Google counts them as 2 vCPUs against your quota regardless.
Same E2 family but with dedicated cores. 2x the price of shared, dedicated performance.
| Type | vCPUs | RAM | $/hour | $/month | Notes |
|---|---|---|---|---|---|
| e2-standard-2 | 2 | 8 GB | $0.0671 | $48.92 | Dedicated 2 vCPU |
| e2-standard-4 | 4 | 16 GB | $0.1342 | $97.83 | Dedicated 4 vCPU |
| e2-standard-8 | 8 | 32 GB | $0.2684 | $195.66 | Getting serious |
| e2-standard-16 | 16 | 64 GB | $0.5369 | $391.32 | Very serious |
Newer Intel CPUs. Same price as E2-standard but better per-clock performance. Has sustained use discounts (SUD) — runs cheaper if you keep it on all month.
| Type | vCPUs | RAM | $/hour | $/month | Notes |
|---|---|---|---|---|---|
| n2-standard-2 | 2 | 8 GB | $0.0971 | $48.92* | JAMIE *with SUD |
| n2-standard-4 | 4 | 16 GB | $0.1942 | $97.83* | *with SUD |
| n2-standard-8 | 8 | 32 GB | $0.3885 | $195.66* | *with SUD |
SUD = Sustained Use Discount. Automatic ~30% discount when running 100% of the month. No commitment needed.
Same as N2 but AMD Epyc processors. ~10% cheaper. Good option if available in your region.
| Type | vCPUs | RAM | $/hour | $/month | Notes |
|---|---|---|---|---|---|
| n2d-standard-2 | 2 | 8 GB | $0.0585 | $42.66 | Cheapest dedicated 2-core |
| n2d-standard-4 | 4 | 16 GB | $0.1170 | $85.32 |
Latest generation. Best single-thread performance. Premium price. You don't need these unless you're doing actual computation.
| Type | vCPUs | RAM | $/hour | $/month | Notes |
|---|---|---|---|---|---|
| c4-standard-2 | 2 | 7 GB | $0.0969 | $70.69 | No SUD on C4 |
| c4-standard-4 | 4 | 15 GB | $0.1977 | $142.30 |
Running a bot, web server, or lightweight service?
→ e2-small ($12/mo, 2GB RAM). This is the fleet standard.
Need a bit more RAM for context windows or concurrent processes?
→ e2-medium ($24/mo, 4GB RAM). Walter runs on this.
Need guaranteed CPU performance (no sharing)?
→ e2-standard-2 ($49/mo, 8GB RAM). Dedicated cores. 4x the price of e2-small for 4x the guaranteed CPU.
Just need a heartbeat / DNS endpoint / webhook receiver?
→ e2-micro ($6/mo, 1GB RAM). One free per account. Vault runs on this.
Want to run AI inference locally?
→ Don't. Use API-based models. A GPU instance that can run anything useful costs $1,000+/mo.
| Type | $/GB/month | 10 GB | 20 GB | 50 GB | Notes |
|---|---|---|---|---|---|
| pd-standard (HDD) | $0.04 | $0.40 | $0.80 | $2.00 | Cheap, slow. Fine for archives. |
| pd-balanced | $0.10 | $1.00 | $2.00 | $5.00 | Middle ground. Good default. |
| pd-ssd | $0.17 | $1.70 | $3.40 | $8.50 | Fast. Our fleet default. |
| Snapshot | $0.026 | $0.26 | $0.52 | $1.30 | Incremental after first. Very cheap insurance. |
Static IP pricing:
Assigned to a running instance: free
Reserved but not in use: $0.01/hr (~$7.30/mo) — idle static IPs cost more than an e2-micro. Release them.
Every region has multiple zones (a, b, c). Not all machine types are available in all zones. Prices vary 10–30% by region — Iowa is cheapest, Europe is ~10% more, Middle East/Africa ~15% more.
Not exhaustive — Google has 40+ regions. These are the ones that matter for general-purpose workloads. See cloud.google.com/about/locations for the full map.
The one that bites you: CPUS_ALL_REGIONS
Default: 12 vCPUs globally. Every running VM counts against this, across all regions. Shared-core VMs (e2-micro, e2-small, e2-medium) count as 2 vCPUs each regardless of their actual CPU share.
To increase: GCP Console → IAM & Admin → Quotas → search "CPUS_ALL_REGIONS" → Request Increase. Google usually approves 24+ if you have billing history.
Other limits to know:
Per-region CPU: 32 (usually not the bottleneck)
Static addresses per region: 8
Instances per region: varies (usually not the bottleneck)
Snapshots: 1,000 (practically unlimited)
Persistent disk per region: 250 GB SSD (soft limit)
| Traffic | Cost | Notes |
|---|---|---|
| Ingress (data in) | Free | Always free. Upload all you want. |
| Egress to internet | $0.12/GB | First 1 GB/mo free. Then it adds up. |
| Same-zone egress | Free | VMs in same zone talk free. |
| Cross-zone (same region) | $0.01/GB | Basically free. |
| Cross-region | $0.01–0.08/GB | Depends on regions. |
| To Google APIs | Free | Cloud Storage, BigQuery, etc. |
Monthly cost reference points for running 24/7:
| What | $/month | Feels Like |
|---|---|---|
| e2-micro + 10GB pd-ssd | ~$8 | A fancy coffee |
| e2-small + 10GB pd-ssd | ~$14 | A lunch |
| e2-medium + 10GB pd-ssd | ~$26 | A dinner |
| n2-standard-2 + 20GB pd-ssd | ~$52 | A nice dinner |
| Forgotten GPU instance (A100) | ~$2,500 | Mom paying the AWS bill |
Google gives you one free e2-micro instance per month (us-central1 only), plus:
• 30 GB pd-standard (HDD) per month
• 5 GB snapshot storage per month
• 1 GB egress per month (most regions)
• Static external IP: free when attached to a running instance
This is how Vault runs for essentially free.