Skip to main content
Baseten Cloud is the default hosting option and the best fit for the vast majority of teams. Single-tenant and self-hosted are enterprise-grade hosting options for organizations that have exceptionally strict isolation or data-control requirements. Where your inference workloads run determines who owns and operates the infrastructure, how workloads connect to your applications, and which capacity and data-residency constraints apply. Baseten Cloud, single-tenant, and self-hosted use the same deployment and operations workflow, but place the workload plane in different environments. Baseten separates orchestration from inference:
  • Control plane: Manages your models, handling builds, scheduling, and lifecycle.
  • Workload plane: Routes inference requests and runs model replicas on GPUs.
Baseten operates the control plane for every option. The location, ownership, and tenancy of the workload plane change. For more information about the shared architecture, see How Baseten works.

​
Hosting option comparison

Choose Baseten Cloud unless an infrastructure, isolation, or networking requirement rules it out. Choose single-tenant when you need dedicated isolation or private connectivity without operating infrastructure in your own account. Choose self-hosted when inference must run in infrastructure you control.

​
Hybrid hosting

Hybrid combines self-hosted workload planes with Baseten Cloud. Use it when requirements differ across workloads and you want to place each workload in the appropriate environment. Hybrid extends self-hosted rather than replacing it as a separate infrastructure model.

​
Next steps