SAP AI Core Resource Group & Orchestration 404 Fix

If the Generative AI Hub in AI Launchpad shows no models, or your orchestration call comes back 404, the cause is almost always one of five things: the wrong service plan, no quota, no resource group with an orchestration deployment, a mismatched resource group header, or a service key that points at a different AI Core instance. Work through them in that order before you open a ticket. This guide walks through that path for functional leads, architects and test teams who have to prove the hub works before a custom AI feature goes near a business process.

What the Generative AI Hub is, and where it sits

The hub is not a separate product you switch on in S/4HANA. It is a capability inside SAP AI Core, and you reach it through SAP AI Launchpad. The Generative AI Hub incorporates generative AI into your AI activities in SAP AI Core. You can access Generative AI Hub via SAP AI Launchpad. AI Core is the runtime. AI Launchpad is the screen where you manage connections, resource groups and deployments.

That distinction matters for scope. Joule features inside S/4HANA are licensed and metered as Business AI. Your own AI apps on SAP BTP, such as a side-by-side extension that summarizes exception messages or drafts a purchase-order justification, run on the hub. Those apps sit outside the core, which fits the clean core approach we cover in SAP Clean Core: What It Means and How to Achieve It. ECC had no equivalent. Any AI in an ECC estate was custom code or a third-party add-on, so this is new ground for most teams moving to S/4HANA on RISE.

For the wider picture of what Business AI, Joule and the platform mean for your job, start with SAP AI for Practitioners. This guide stays on the hub plumbing.

Step 1: Confirm the plan, the entitlements and the quota

The first failure is usually commercial, not technical. The generative AI hub is available only with the extended service plan in SAP AI Core. The standard plan gives you production inference, but not the hub. Check the plan on the service instance before anything else. SAP's service plan comparison sets out the difference between standard and extended.

Then check the entitlements in the subaccount. One practitioner post notes that the default quota allocation for AI Core is zero — you must explicitly set it. That post is community guidance, not an SAP Note, so confirm the quota in your own subaccount. Set it, save it, and check that the SAP AI Launchpad subscription exists too. SAP's own tutorial treats AI Launchpad as the recommended interface, and the hub UI is only reachable through it.

Step 2: Get the resource group and orchestration deployment right

The resource group is where most people lose an afternoon. A resource group is the isolation boundary for models, deployments and executions. Without a configured resource group, model deployments will not be visible even after AI Core and AI Launchpad are correctly set up. SAP's own guidance on onboarding is worth reading against that: as part of the SAP AI Core onboarding process, an orchestration deployment is automatically created in the default resource group.

Here is the catch. SAP's orchestration tutorial adds a second rule for any group you create yourself: for any new or additional resource groups, you must deploy a separate orchestration setup. So the default group may work on day one, while a project group you created last month shows nothing. Every deployment is built on a configuration, so a group needs at least one configuration before it can deploy anything. Treat "default works, custom group is empty" as the signature of this problem.

1. EntitleExtended plan for AI Core, standard plan for AI Launchpad, quota set above zero.
2. ConnectAdd the AI Core service key as a connection in AI Launchpad under Workspaces.
3. GroupCreate or select the resource group. Note its ID exactly.
4. ConfigureCreate an orchestration configuration in that group.
5. DeployDeploy it and wait for RUNNING status.
6. ProveSend one call with the same group name in the header.

1. EntitleExtended plan, AI Launchpad, quota above zero
2. ConnectService key added as a connection
3. GroupResource group ID recorded
4. ConfigureOrchestration configuration in the group
5. DeployWait for RUNNING
6. ProveOne call with the same group header

Two more rules to write into the runbook. First, you must use the same resource group for all of your generative AI activities. Second, if your team uses grounding, a document-grounding resource group carries a label. SAP's sample sets the key "ext.ai.sap.com/document-grounding" with the value "true" when the group is created for that purpose. Grounding is covered in the free course RAG for SAP.

Step 3: Separate the 403 from the 404

Teams often treat every failed call the same way. They are not the same. A 403 means access was refused. A 404 means the path was not found, which in AI Core usually means the request points at the wrong place.

For 403, check the roles first. To manage a deployment you need either the mloperations_editor or scenario_deployment_editor role, or you are assigned a role collection that contains one of these roles. A user who can open the hub may still be unable to create a deployment. In a 403 on an inference call, one community thread shows a team sending the right bearer token and the default group header and still getting 403 RBAC: access denied. That points back to role assignment or the instance the token was issued for.

For 404, the community troubleshooting series lists the common causes. Three common causes: (1) the deployment belongs to a different AI Core instance than your service key (see Error 8); (2) the token expired between the deployments list call and the inference call; (3) the AI-Resource-Group header is missing or pointing to the wrong group. The first cause is the one people miss. The deployment list can show the ID while the inference call still fails, because the deployment lives under a different instance. In AI Launchpad, check the instance label on the resource group to see which instance owns it.

Step 4: Run one orchestration call and record the result

Do not judge the hub from the UI alone. Send one request to the orchestration deployment with the deployment ID, the group header and a valid token. Then record the response.

Use the orchestration format for orchestration deployments. A common mistake is calling an orchestration deployment with the foundation-models format. The community guide puts it plainly: empty response or format error on inference · Orchestration called with foundation-models format · Check scenarioId; use /completion for orchestration. Confirm the scenario ID on the deployment list before you blame the model.

Model versions need the same discipline. In the Launchpad prompt tooling, SAP's user guide says that instead of pinning a version you can specify "latest" to use the latest version of the model available in SAP AI Core. For a regulated test cycle, pinning is safer. A "latest" setting can change the output between your test run and go-live, so record the exact version either way.

The free course Generative AI Hub in SAP AI Core covers models, orchestration and content filtering in more depth. For the test side of non-deterministic output, see Testing Joule Agents and AI Features in SAP.

The 404 and 403 quick reference

Symptom	Likely cause	First fix
Hub tiles missing or greyed out	Standard plan, or AI Launchpad not subscribed	Move to the extended plan; add AI Launchpad
Custom group shows nothing; default works	No orchestration configuration or deployment in the new group	Create configuration, then deployment, in that group
404 on a RUNNING deployment	Wrong group header, expired token, or deployment in another instance	Refresh token; match header; check instance label
403 RBAC on a call or action	Missing editor or deployment role collection	Assign mloperations_editor or scenario_deployment_editor
Empty response or format error	Foundation-models format sent to orchestration	Check scenario ID; use the orchestration path
Group stuck in Provisioning	Creation still in progress or failed	Wait for PROVISIONED or ERROR; raise a support ticket if it persists

The last row matches the community guide's advice to raise a ticket with component CA-ML-AIP when a group stays in provisioning beyond about 15 minutes. Treat that as a starting point, and confirm the component in the current support guidance before you file.

Where the cost lands

Cost is part of the setup, because a hub that works in a sandbox can stall when the budget runs out. The published picture is split. One partner post says two separate budgets: AI Units pay for the Joule products; BTP credits pay for AI Core, AI Launchpad and the Generative AI Hub. Another reports the hub is priced in capacity units per year. SAP's own material on AI Units and capacity units changes often, so get the current order form and metering terms from your SAP account team and check them against the BTP cockpit usage view before go-live.

What changes for an S/4HANA program

The hub does not change your ERP test scope by itself. It changes what you must prove. If a custom app calls the hub from a Fiori extension, your regression suite now includes model output, grounding results and the authorization path. The Joule side has its own rules. For agent permissions, read Joule Agent Authorization Testing. For the S/4HANA Private Edition Joule view by line of business, see Joule in S/4HANA Private Edition. For prompt practice with Joule for consultants, the Joule prompts guide is the companion piece.

Sign-off checklist for the hub

• Service plan, AI Launchpad subscription and quota recorded for each subaccount
• Resource group IDs listed per environment, with the owning instance
• Orchestration configuration and deployment ID recorded for each group, with RUNNING status captured
• Model name and version recorded, with the decision on pinning versus "latest"
• Editor and deployment roles assigned to named people, not shared accounts
• One orchestration call per environment, with the request, group header and response saved as evidence
• Grounding repository and label documented where grounding is used

Resource groups and deployments are configuration in each AI Core landscape, so they do not move with a transport. Plan to rebuild them, or script them, in each environment. Use the same group names and document any difference between dev, test and production.

If your team is new to the hub, the free SAP Business AI Explained course gives you the vocabulary first. The full set of courses is on the SAP Academy page.

Frequently asked questions

Why does the Generative AI Hub show no models in AI Launchpad?

The most common causes are a standard plan instead of extended, a missing AI Launchpad subscription, or a resource group with no orchestration deployment. SAP's tutorials say the hub is available only with the extended plan. Check those three first, then confirm the group you are viewing is the one that holds your deployment.

What is the difference between a 403 and a 404 on an orchestration call?

A 403 means the caller lacks permission, so check the editor or deployment roles and the token's instance. A 404 usually means the request points at the wrong place: a stale token, a different resource group header, or a deployment in another AI Core instance. Fix the header and instance first, then look at roles.

Do I need a separate orchestration deployment for each resource group?

Yes, for any group you create yourself. SAP's onboarding creates an orchestration deployment in the default group, but SAP's orchestration guidance says additional groups need their own setup. Each deployment is based on a configuration, so create the configuration first.

Should I pin the model version or use "latest"?

For a test cycle, pin the version and record it. SAP's tooling lets you specify "latest," which picks up the newest model version available, so output can change between test and production. Document whichever you choose and retest when the version changes.

Sources

1. Set up Generative AI Hub in SAP AI Core, SAP Developers tutorial
2. Setup Generative AI Hub in SAP AI Launchpad, SAP Developers tutorial
3. SAP AI Launchpad user guide: Generative AI Hub, SAP Help Portal
4. Service Plans - SAP AI Core, SAP Help Portal
5. Consumption of GenAI models Using Orchestration (V2) service, SAP Developers tutorial
6. Getting Started with Generative AI Hub, SAP Learning
7. SAP Generative AI Hub: A Practitioner's Setup Guide, SAP Community
8. SAP AI Core: Deployment and Configuration Errors and How to Fix Them (Part 2 of 2), SAP Community
9. Mastering SAP AI Core: Deployment Best Practices, Configuration Pitfalls, and Troubleshooting, SAP Community
10. RBAC: access denied on inference call to SAP AI Core deployment, SAP Community
11. SAP Business AI Enablement as an SAP Partner, SAP Community
12. Managing SAP AI Core using PowerShell, SAP Community