On-premise AI
AI systems deployed inside an organisation’s own infrastructure — its private cloud tenancy or physical data centre — rather than calling out to a vendor-hosted AI service.
On-premise AI (also called “private deployment” or “VPC deployment”) refers to AI systems whose models, vector databases, application layer, and data storage all run inside the customer’s own infrastructure. This contrasts with “cloud AI” where the customer sends data to a vendor-managed service and receives outputs back.
For regulated financial services, on-premise AI is structurally different on three axes: data egress (zero, versus the vendor managing the data), regulator boundary (the customer’s, versus the vendor’s), and continuity (the customer controls the lifecycle, versus dependent on the vendor’s SLA).
On-premise AI is what makes AI deployable for institutional fund managers, AFSL holders with CPS 234 obligations, and any firm where the data classification cannot leave the regulated perimeter. The trade-off is operational lift to provision and maintain the deployment — which is why on-premise vendors typically ship infrastructure-as-code templates (Terraform, CloudFormation, ARM) rather than expect customers to engineer the deployment themselves.