Concept card

SAP AI Hub

As of 2026-07-24T14:00:00Z

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is SAP AI Hub?

The Hub is the model 'package registry' — a model can't reach Production until it clears the Hub's own evaluation suite against the deploying organisation's SAP-specific test dataset, not a generic benchmark.

The SAP AI Hub is the catalogue and lifecycle authority for every AI model available inside the SAP Business AI Platform — foundation models from third-party providers, SAP's own fine-tuned models, and models a customer or partner has fine-tuned and wants to use internally. It sits below the Model Gateway and above SAP AI Core: AI Core provides the raw compute and training infrastructure, the AI Hub is the registry that decides which model versions are trustworthy enough to be routed to, and the Model Gateway does the actual routing at inference time. A useful mental model is a software package registry: the Gateway is the application calling a dependency, the Hub is the registry that says which versions of that dependency are published, vetted, and safe to install.

What it actually does

Why it matters

Deprecated models stay queryable for a minimum 10-year EU AI Act documentation window rather than being deleted on retirement
Promotion to Production is gated by scorecards on the organisation's own datasets (Finance GL, Supply Chain anomaly, HR policy Q&A), not vendor leaderboards
The Model Gateway can only route to what the AI Hub has registered — governance sits one layer below routing, not bolted onto it

Key points

Model catalogue: browsable inventory of SAP and third-party models with metadata including EU AI Act risk classification and data residency region.
Lifecycle stages: Development → Staging → Production → Deprecated — gated transitions with evaluation-based approval workflow.
Evaluation framework: standardised SAP-specific test datasets (Finance GL, SCM anomaly detection, HR Q&A); scorecard required for Production promotion.
Deprecation registry: models retained for EU AI Act documentation period (10 years for high-risk); queryable by SAP Audit Journal for retrospective compliance.
Model card: SAP publishes EU AI Act Art. 53 model cards for all fine-tuned models — starting template for consultant risk classification analysis.
Drift detection: AI Hub monitors active Production model quality; alert when inference score drops >5% from promotion baseline.
Fine-tuning hosting: customer fine-tuned models registered and deployed via same Model Gateway routing as SAP-standard models.
SAP AI Hub is mastered only when it changes a named buyer decision.
Start with the semantic contract and control model before demonstrating the tool.
Use current SAP, analyst, study, KG, and news signals as evidence, not decoration.

Terms used on this page

Model lifecycle stage: A defined phase in the AI Hub model governance workflow: Development (experimental, not routeable), Staging (candidate for Production, undergoing evaluation), Production (live, routeable by Model Gateway), Deprecated (retired, retained for audit).
Model card: A structured documentation artefact for an AI model that describes its training data, intended use cases, limitations, evaluation benchmarks, and risk classification — required by EU AI Act Article 53 for GPAI model providers.
Data drift: The gradual divergence between the statistical properties of the production data the model processes and the training data the model was optimised on — causes model performance degradation over time without any change to the model weights.
Evaluation dataset: A curated test dataset used to measure model performance on a specific task type; the SAP AI Hub provides SAP-specific evaluation datasets (Finance GL classification, SCM anomaly detection, HR policy Q&A) for Production promotion gating.
Decision owner: The accountable person who accepts the trade-off and funds the next action.
Semantic contract: The shared definition of business terms, metrics, entities, and access rules used by tools and teams.
Control plane: The layer that applies policy, access, lineage, monitoring, and escalation across the operating model.
Evidence grade: A label that separates verified fact, directional signal, modeled assumption, and field observation.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →