A Virtual Private Cloud is a logically isolated network within a public cloud provider's infrastructure. Within a VPC, customers control the IP address ranges, routing tables, subnet divisions, and network access controls for their resources. VPCs are the default network primitive for cloud deployments — new cloud accounts come with default VPCs pre-provisioned, and most production deployments use custom VPC configurations.
VPCs enable several important capabilities: isolating sensitive workloads from internet-facing ones, connecting cloud resources to on-premise networks via VPN or direct connect, controlling which subnets can reach which services, and implementing defense-in-depth beyond what IAM alone provides.
For AI workloads, VPC design often has to accommodate hybrid patterns: inference endpoints that are public-facing, model training that must be isolated from the internet, and data ingestion pipelines that connect to internal data sources. Getting VPC architecture right early saves significant retrofit work when AI workloads expand from experimental to production.