> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/ai-enterprise/ai-factory/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/ai-enterprise/ai-factory/_mcp/server.

# Inference Hub

> How the published Inference Hub Reference Design complements the NVIDIA Enterprise AI Factory core architecture.

Inference Hub is an optional **Capability Reference Design** for delivering governed inference to enterprise applications and agents. This page explains how to compose it with the Enterprise AI Factory core, Secure Agent Workspace, and Industry Profiles.

#### Read the Published Inference Hub Reference Design

Use the authoritative NVIDIA publication for the detailed architecture, request flows, security controls, adoption path, and implementation guidance.

## When to Apply Inference Hub

Apply this capability when multiple applications or teams need a shared inference service rather than independently managed model endpoints. Begin with workload outcomes, service objectives, ownership, and data constraints; use the published design to plan the service.

Inference Hub composes existing core responsibilities. It does not replace the core stack or require every AI Factory workload to use one serving engine, gateway product, or deployment model.

## Relationship to the Core

Select [Inference Hub in the interactive architecture](/ai-enterprise/ai-factory/start-here/interactive-architecture?view=core\&capability=inference-hub) to see the responsibilities it adds, extends, or inherits:

| Core Responsibility                                                                                                            | Integration Decision                                                                                                                                                                  |
| ------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [Workload Demand](/ai-enterprise/ai-factory/core-reference-design/workloads#representative-workload-families)                                           | Identify consuming applications and agents, accountable owners, and service objectives. Application demand remains separate from serving implementation.                              |
| [AI Platform](/ai-enterprise/ai-factory/core-reference-design/ai-platform#optimize-and-serve)                                                           | Connect the service to model lifecycle, serving, and resource orchestration. Request routing does not take ownership of GPU scheduling or model execution.                            |
| [Platform Services](/ai-enterprise/ai-factory/core-reference-design/platform-services#service-responsibilities)                                         | Add the Inference Portal and extend gateway routing and usage attribution. Reuse enterprise identity, policy, asset lineage, delivery, and telemetry services.                        |
| [Cloud-Native Platform](/ai-enterprise/ai-factory/core-reference-design/cloud-native-platform)                                                          | Establish platform and runtime ownership for enterprise-operated backends. For managed destinations, define the provider handoff instead of assuming control over its infrastructure. |
| [Hardware Foundation](/ai-enterprise/ai-factory/core-reference-design/enterprise-hardware-ra-foundation)                                                | Translate approved demand and deployment choices into infrastructure requirements; use authoritative hardware and OEM guidance for sizing and configurations.                         |
| [Governance and Assurance](/ai-enterprise/ai-factory/core-reference-design/security-and-governance) and [Operations](/ai-enterprise/ai-factory/core-reference-design/operations) | Carry the factory's data, risk, access, incident, and recovery requirements into the service and its provider boundaries.                                                             |

## Compose with Other Capabilities and Profiles

* **Secure Agent Workspace:** [SAW](/ai-enterprise/ai-factory/capability-reference-designs/secure-agent-workspace#required-interfaces-and-dependencies) governs agent execution, workspace isolation, tool authority, and egress. Inference Hub governs access to inference services. Using both does not remove workspace isolation or network-perimeter requirements.
* **Government:** Apply the [Government profile](/ai-enterprise/ai-factory/industry-profiles/government#sovereign-and-restricted-environments) to backend eligibility, residency, external dependencies, and operating boundaries. Do not assume that a published implementation supports disconnected operation or satisfies sector authorization requirements.
* **Implementation selection:** Choose components using the publication's implementation guidance and current publisher support documentation. [Catalog candidates](/ai-enterprise/ai-factory/ecosystem-integrations/solution-catalog?layer=ai-platform) support discovery, not interoperability or support claims.

## Detailed Design and Adoption Guidance

Continue to the published design for implementation-level guidance.

#### Architecture

Portal, Gateway, and inference-backend responsibilities and boundaries.

#### Request Flow

Administrative, developer-access, and runtime handoffs.

#### Security and Operations

Trust boundaries, hybrid operations, attribution, and telemetry privacy.

#### Adoption Path

The published sequence for introducing and operating the service.

## Published Example Implementations

Use these published examples to evaluate component choices and their operating boundaries:

#### Inference Hub Component Examples

Explore the published mapping of portal, gateway, routing, serving, resource orchestration, platform lifecycle, and observability components. These are implementation examples, not required products.

#### Red Hat AI Factory with NVIDIA

Consult the published example for its component mapping, feature availability, dependencies, and support boundaries. It is an implementation choice, not the required Enterprise AI Factory stack.

Confirm component availability, deployment assumptions, and support in the published example and current vendor documentation.