Google Compute Engine

Google Compute Engine

by Google Cloud

Infrastructure-as-a-Service virtual machines on Google’s infrastructure: predefined and custom machine types, GPUs, disks, instance groups.

Machine families (per Google’s machine-families page, 2026-10-05)

• General purpose: N4, N4A, N2, E2, T2D, T2A (the page’s summary; newer C-series general-purpose types also exist, check the page).
• Compute-optimized: H4D, H3, C2, C2D (HPC and compute-heavy).
• Memory-optimized: X5, X4, M4, M3, M2 (up to 12 TB memory).
• Storage-optimized: Z3, Z4D. Network-optimized: C4N, M4N.
• Accelerator-optimized (GPU): A4X, A3, G4, G2.
• Custom machine types exist for select series (with a premium noted in the docs).

Features

• Persistent Disk and other block storage with snapshots, Local SSD (ephemeral), managed instance groups with autoscaling and rolling updates.
• Shielded VMs and Confidential VMs, sole-tenant nodes.
• Live migration moves most running VMs during host maintenance without reboot. It does not apply to GPU VMs (they stop with advance notice, typically 60 minutes, 10 minutes for A4/A3 Ultra), bare metal, most Confidential VMs, TPUs and some storage-heavy shapes.
• Spot VMs are the current discounted interruptible option; Preemptible VMs are still offered. Spot VMs have no 24-hour cap and preemption notice is 0 or 120 seconds depending on configuration.
• Integrates with VPC, Cloud Load Balancing, IAM, Cloud Monitoring, Cloud Storage, BigQuery, Vertex AI.

Billing model

Per-second billing for vCPU, memory, disks and accelerators; discount models include Spot pricing and committed-use discounts; egress extra. Pricing: https://cloud.google.com/compute/pricing (discount rules vary by machine family, so check the page).

When to choose (author’s view, opinion)

Lift-and-shift, licence-bound or stateful workloads, GPU/HPC jobs, anything needing OS-level control. For containers prefer Cloud Run or GKE; see also Google Cloud.

Caveats (author’s view, opinion)

Local SSD is described as ephemeral (see Features); machine and GPU availability can differ per zone, so check the docs for your zone.

Sources

• Machine families: https://docs.cloud.google.com/compute/docs/machine-resource (fetched 2026-10-05)
• Spot VMs: https://docs.cloud.google.com/compute/docs/instances/spot (fetched 2026-10-05)
• Live migration: https://docs.cloud.google.com/compute/docs/instances/live-migration-process (fetched 2026-10-05)

Open items

• Complete current machine-series list (e.g. C4 family) and 2026 AI-hypercomputer announcements not checked.