Skip to main content

Deployment Targets for AI Products

An AIPDS must declare the deployment environments in which the referenced AI Product version can execute or be accessed.

Deployment target is an operational characteristic. It does not determine AI Product Type (APT), Intelligence Capability (IC), or AI qualification.


Deployment Target Categories​

1. Cloud / Managed Runtime​

Hosted in public, private, or sovereign cloud infrastructure and typically exposed through managed execution or service interfaces.

Typical declarations include provider, region, tenancy, resilience, service-level objectives, and regulatory constraints.

2. On-Premises / Private Data Center​

Executed inside enterprise-controlled infrastructure, often for data residency, security, latency, intellectual-property, or regulatory reasons.

3. Edge​

Executed close to the observation or action source, for example on mobile devices, cameras, gateways, industrial computers, or constrained edge nodes.

4. Embedded / Physical-Control Runtime​

Executed as part of an embodied or cyber-physical system such as a robot, vehicle, industrial controller, or autonomous machine.

Declarations SHOULD include, where applicable:

  • sensor and actuator dependencies;
  • control-loop frequency or timing envelope;
  • fail-safe / safe-state behavior;
  • hardware constraints;
  • offline/degraded-operation requirements;
  • safety interlocks and emergency-stop integration.

5. Hybrid / Multi-Environment​

Uses multiple execution environments as one governed realization, for example cloud planning plus edge perception or on-premise reasoning plus external model service.

6. Agent Runtime​

Executed inside an agent or orchestration environment where the AI Product may maintain context, invoke tools, pursue goals, or coordinate with other AI Products.

Agent runtime is a deployment target; Agentic behavior is separately described by AIPROD through IC/MO/AR semantics.


Required Declarations​

Every AIPDS must declare:

  • one or more deployment environments;
  • runtime and resource dependencies;
  • relevant locality/residency constraints;
  • environment-specific security and governance bindings;
  • supported execution modes;
  • portability or environment-locking constraints where material.

Physical or embedded deployments must additionally declare applicable safety and actuation constraints.


Example Contrast​

Fraud Detection AI Product

  • Cloud + on-premises deployment.
  • Event-driven execution.
  • No embodiment.

Autonomous Navigation AI Product

  • Embedded physical-control runtime.
  • Continuous execution.
  • Closed-loop physical actuation.
  • Safety binding required.

Both are AI Products, but their deployment realizations are fundamentally different.


Principle: Deployment targets describe where and within what operational envelope intelligence executes; they do not define what kind of intelligence Product it is.