Guide

SAP AI Launchpad — what it is, who uses it, and how it differs from SAP AI Core

As of 2026-09-25

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.

SAP AI Launchpad is a multitenant SaaS console on SAP BTP that manages AI scenarios running on one or more SAP AI Core instances and gives access to the generative AI hub. AI Core is the runtime that executes and serves models; Launchpad is the user interface over the same objects. Nothing runs in Launchpad itself.

The generative AI hub is not a third product. SAP documents it as a capability of AI Core and AI Launchpad, available only on AI Core's extended plan. If a client asks which of the three to buy, the answer is AI Core first; Launchpad is how people work with it.

Three names, one stack

SAP AI Core runs on Kubernetes, uses Argo Workflows for training and batch runs and KServe for serving, and exposes everything through the AI API, an open lifecycle specification. SAP AI Launchpad speaks the same AI API, so every object you see in it — scenario, executable, configuration, execution, deployment — is an AI Core object. The generative AI hub adds access to foundation models and the orchestration service on top of that runtime.

The practical split for delivery: why is inference failing is a runtime question you answer from AI Core; who deployed this, with which configuration, is a question you answer from the Launchpad view of the same objects. Since Sapphire 2026 all three sit under the SAP Business AI Platform umbrella, but provisioning is unchanged: you still entitle and subscribe them in the BTP cockpit.

What you actually do in Launchpad

The Generative AI Hub section groups the working surfaces. Model Library offers a model catalogue, a benchmark leaderboard and a model card per model with input types and cost information. Chat and prompt experimentation let you try prompts against a model, then save and version them. The orchestration builder composes a workflow — templating plus optional grounding, filtering, masking and translation — tests it and saves it as an orchestration config. Grounding Management creates document repositories, triggers syncs, previews chunks and runs test searches. Optimizations holds evaluations and prompt optimisation.

On the predictive side, Launchpad manages the same scenarios, configurations, executions and deployments that AI Core runs, per resource group. It is also where the SAP Certified Generative AI Developer exam (C_AIG_2604) is taken: the assessment is system-based and performed in this environment.

Who uses it, through which roles

Access is role-based and bundled in role collections. genai_experimenter can try prompts; genai_manager can also save and manage them; genai_administrator administers; orchestration_executor runs orchestration; grounding_manager creates and syncs repositories while grounding_viewer can only inspect.

That role set maps onto people on a project. A business analyst or prompt designer needs experimenter rights, not administrator rights. The platform engineer owns resource groups and deployments. The governance lead reads evaluations and model cards. Granting everyone genai_administrator in a pilot is the fastest way to lose the separation you will be asked to prove later.

How it fits the SAP analytics consultant's work

Launchpad is where most analytics consultants first touch SAP's custom AI stack, usually for three jobs: testing whether a model can summarise or classify business text before anyone writes code; setting up a grounding repository for an agent; and reading consumption and evaluation evidence for a governance review.

Two design decisions precede every click. First, resource groups: AI Core allows up to 50 per tenant by default, runtime objects cannot be moved between them afterwards, and the resource group is the finest cost grain the platform reports natively. Design one per use case and environment before the first deployment. Second, evidence: nothing is recorded by default. Inference observability captures model, tokens and latency only for requests that send its headers, with up to 16 labels such as scenario or cost centre.

Licensing — what is public and what is not

What our corpus can state: AI Core has two production plans. Standard covers custom predictive AI without generative AI; extended adds the generative AI hub. You can upgrade standard to extended but never downgrade. The free plan was discontinued on 22 May 2026 (SAP Note 3735945); a 30-day trial including the generative AI hub replaces it for exploration.

Billing runs in BTP capacity units. For custom AI, SAP meters node hours, storage and a baseline capped at 730 hours a month. For generative AI, compute and baseline are waived and model tokens are converted into GenAI tokens at model-specific rates in SAP Note 3437766, then into capacity units; grounding and observability carry their own metrics. SAP publishes a cost calculator.

What our corpus does not hold is a published list price for AI Launchpad's own subscription. We do not quote one. Use the calculator and the BTP cockpit usage export for a client estimate, never a number from a slide.

Decision table — Launchpad, AI Core API, or neither

Try models and prompts before building → Launchpad Chat and prompt experimentation (genai_experimenter). Compose and test an orchestration workflow → Launchpad orchestration builder; call it in production through the AI Core API at /v2/completion. Run inference in an application → AI Core API from code; Launchpad is not a runtime. Set up document grounding for an agent → Launchpad Grounding Management (grounding_manager). Enforce an approved-model list → allow list on the orchestration deployment, not a Launchpad convention. Only Joule and SAP-delivered AI in scope → none of these; those calls run under SAP's management and never appear in your AI Core tenant.

Dates that break integrations

Orchestration v1 (/completion) is decommissioned on 31 October 2026; anything built or tested against it in Launchpad must move to /v2/completion. Model versions carry deprecation dates listed in SAP Note 3437766: a pinned deployment stops working on its date, a deployment on latest changes behaviour without a ticket. Record each date when you pick a model in the Model Library, not when it fails.

What we cannot assert

We hold no published list price for the AI Launchpad subscription itself, so this page quotes none. Capacity-unit conversion rates live in SAP Note 3437766 and change per model; we do not restate them.

Frequently asked

Is SAP AI Launchpad the same as SAP AI Core?

No. AI Core is the BTP runtime that executes and serves models. AI Launchpad is a separate SaaS console that connects to one or more AI Core instances and manages the same objects through a user interface.

Do I need AI Launchpad to use the generative AI hub?

The hub is a capability of AI Core and AI Launchpad on AI Core's extended plan. Applications call it through the AI API; Launchpad adds the Model Library, prompt editor, orchestration builder and Grounding Management for people.

Is there a free version of SAP AI Launchpad?

AI Core's free plan ended on 22 May 2026. A 30-day trial including the generative AI hub is the documented route for exploration.

Can I see Joule's model usage in AI Launchpad?

No. Joule and SAP-delivered AI features run their model calls through AI Core under SAP's management; they are paid in AI Units or covered by Base AI and do not appear in your tenant.