Google Cloud Compute

Google Cloud compute is not one product. It is a set of execution models. You can operate virtual machines, schedule Kubernetes workloads, run managed containers, or submit finite work to a job queue.

Start with the workload, not the product name. Ask what must run, how it starts, how long it lives, how it scales, and which infrastructure controls you need.

Compute Engine gives you virtual machines and operating system control. Google Kubernetes Engine, or GKE, gives you managed Kubernetes. Cloud Run runs code or containers on a managed application platform. Batch queues tasks and provisions the compute needed to finish them.

Each step toward a more managed service removes some infrastructure work. It also narrows the execution contract. The best choice is the least operational surface that still meets the workload's requirements.

Choose the execution model

Use Compute Engine when the workload needs operating system or kernel control, a specific machine shape, attached accelerators, custom images, or software that expects a virtual machine.

Use GKE when Kubernetes is a requirement. GKE manages the Kubernetes control plane. Autopilot also manages worker nodes. Standard mode leaves more node and cluster configuration with you.

Use Cloud Run when an application can fit its managed container contract. A Cloud Run service handles requests or events through a managed endpoint. A Cloud Run job runs tasks to completion. A worker pool runs continuous non-HTTP background work.

Use Batch when work is finite, queueable, and described as one or more tasks. Batch provisions and removes Compute Engine resources for the job. It favors scheduled throughput over an interactive request path.

These choices can work together. A public service might run on Cloud Run while a data-processing stage uses Batch. A GKE application can call services on other Google Cloud products. A Compute Engine fleet can sit behind Cloud Load Balancing.