Architecture

Saturn Cloud runs inside your datacenter and integrates with your infrastructure provider APIs to orchestrate compute on behalf of your tenants.

Components

Saturn Cloud control plane

The Saturn Cloud control plane is the software that runs the platform. It handles:

• The tenant-facing portal (AI Studio and Token Factory)
• The operator management interface
• Authentication and multi-tenancy
• Usage metering and chargeback data

The control plane runs on a small set of servers in your datacenter, separate from the GPU servers you’re selling to customers.

AI Studio

AI Studio is the workspace environment your customers use. It provides:

• JupyterLab and VS Code in the browser
• SSH access to their resources
• Managed jobs (batch workloads with scheduling and failure handling)
• Managed deployments (long-running inference endpoints)

Each tenant’s AI Studio is isolated. Tenants cannot see each other’s resources, workloads, or usage.

Token Factory

Token Factory manages API endpoints for model inference. Tenants can deploy models and get API endpoints they can call directly or expose to their own customers.

Infrastructure integrations

Saturn Cloud calls out to your infrastructure provider APIs to do the actual work of provisioning and managing compute. There are two main integration points.

BMaaS integration

Saturn Cloud integrates with your bare metal as a service provider’s API. Through this integration Saturn Cloud can:

• Provision bare metal servers for a tenant
• Provision VMs on those servers
• Stand up managed Kubernetes clusters
• Stand up managed Slurm clusters

The BMaaS provider handles the low-level hardware orchestration (IPMI/BMC, OS imaging, Redfish calls). Saturn Cloud sits above that layer and calls the BMaaS API at the resource level.

If you do not already have a BMaaS provider, Saturn Cloud can recommend a partner. The BMaaS provider must be in place before Saturn Cloud can deploy.

Networking integration

Tenant isolation on the network layer requires a network provider API. Some BMaaS providers include this capability. If yours does not, a separate networking provider is needed.

Saturn Cloud uses the networking API to create isolated network segments per tenant, ensuring tenants cannot reach each other’s workloads at the network level.

Data flow

Customer traffic from tenant workloads does not pass through Saturn Cloud infrastructure and does not leave your network. The Saturn Cloud control plane handles management plane traffic (provisioning requests, portal interactions, usage metering). The actual GPU workload traffic runs directly on your hardware, on your network.

Deployment topology

Everything runs inside your datacenter. The table below shows what each piece is and who runs it.

The control plane calls the BMaaS and network provider APIs to orchestrate resources. Management plane traffic (provisioning, portal, metering) passes through the Saturn Cloud control plane. GPU workload traffic runs directly on your hardware and does not leave your network.