Databricks Mosaic AI inside SAP Business Data Cloud

What "SAP Databricks" is, concretely

SAP's February 2025 announcement of the partnership describes SAP Databricks as a version of the Databricks platform "included natively as a service within SAP Business Data Cloud" — a first-party data service, fully managed by SAP, not a bring-your-own-Databricks integration a customer has to wire up separately. The same announcement names native Delta Sharing as the mechanism that lets SAP data products harmonize "bidirectionally" with existing lakehouses, and gives a concrete example of the kind of use case this unlocks: predicting the payment date on open receivables — a finance ML problem that needs both SAP transactional context and a model trained on historical patterns.

The zero-copy path that feeds it

SAP Databricks, the SAP-managed edition inside BDC, reads SAP data products natively. For a customer's own Databricks workspace, SAP Business Data Cloud Connect for Databricks reached general availability on 6 October 2025, built on Delta Sharing and secured with mTLS combined with OAuth. Activated SAP data products mount directly into the Databricks Unity Catalog and can be queried alongside any other data already governed there, with no replication, custom API export or manual upload — replacing exactly the fragile extract-and-load pattern that previously created governance gaps, data silos and high maintenance cost. This is the plumbing; Mosaic AI is what sits on top of it once the data is there.

Why it matters

• SAP Databricks is a first-party, SAP-managed service — not a customer-wired integration — which changes both the contracting conversation and who is accountable for the underlying platform.
• Because activated data products mount into Unity Catalog rather than being copied, Mosaic AI jobs inherit BDC's existing access controls automatically — there is no second permission model to design or audit.
• Mosaic AI needs genuine Databricks/Python/Spark skills; scoping it as a low-code extension of SAC or Datasphere modelling under-resources the engagement from day one.

Key points

• SAP Databricks: a first-party, fully managed Databricks service natively included in Business Data Cloud (SAP News, February 2025).
• Two paths: SAP Databricks reads SAP data products natively inside BDC; a customer's own Databricks workspace uses BDC Connect for Databricks (GA 6 October 2025: Delta Sharing, mTLS + OAuth, mounted into Unity Catalog).
• Mosaic AI is Databricks' own machine-learning/generative-AI toolset — a Databricks product, not an SAP one — running inside the SAP-provisioned workspace.
• Governance carries over: Unity Catalog access controls apply to Mosaic AI jobs and endpoints without a second permission system.
• SAP's own named use case: predicting payment dates on open receivables, a finance ML problem needing both transactional context and trained models.
• Reach for Mosaic AI when the job needs custom model training, feature engineering or model serving; reach for Just Ask or a BDC intelligent application when it does not.
• The zero-copy connect is the precondition for a Mosaic AI engagement, not the deliverable itself.

Terms used on this page

SAP Databricks

A first-party, fully managed Databricks service included natively as part of SAP Business Data Cloud.

Mosaic AI

Databricks' own machine-learning and generative-AI product line — model training, serving, feature store, vector search.

Unity Catalog

Databricks' unified governance layer for data and AI assets; SAP data products mount into it rather than being copied.

Delta Sharing

The open protocol underlying zero-copy, bidirectional data sharing between BDC and Databricks.

Feature Store / Feature Engineering client

Databricks' central registry for ML features, registered as Unity Catalog objects, giving lineage, point-in-time joins and consistent feature values between training and inference.

Model Serving endpoint

A Databricks-managed REST API deploying a Unity-Catalog-registered MLflow model for real-time or batch inference, scaling serverlessly.

DBU (Databricks Unit)

Databricks' unit of processing capability, used to meter Mosaic AI compute consumption — the cost dimension that has no SAP AI Units equivalent.

Mosaic AI Agent Framework

Databricks' environment for building, evaluating and deploying multi-step, tool-calling agents on top of governed Unity Catalog data — the Databricks-side analogue to a Joule Agent.

Sources

1. SAP News Center — SAP and Databricks: A Bold New Era of Data and AI (February 2025 — SAP Databricks as first-party service, Delta Sharing, receivables use case)
2. Databricks Blog — Announcing General Availability of SAP Business Data Cloud Connect to Databricks (6 October 2025 — Delta Sharing, mTLS + OAuth, Unity Catalog mounting)
3. Databricks — Unity Catalog product overview (fetched 2026-09-27)
4. Databricks documentation — Feature Store / Feature Engineering client (fetched 2026-09-27)
5. Databricks documentation — Model Serving overview (fetched 2026-09-27)
6. SAP AI Core — generative AI hub / orchestration service (help.sap.com)
7. SAP — AI Units pricing model
8. CBS Consulting — the modern data stack: SAP Business Data Cloud meets Databricks and Snowflake (press, corroborating)
9. SAP Databricks in Business Data Cloud — SAP
10. SAP Databricks: Building an Intelligent Enterprise with AI Unleashed (Part 2) — SAP Community
11. SAP Databricks is now GA: skilling yourself with Mosaic AI — SAP Community
12. AI and machine learning on SAP Databricks — Databricks documentation
13. Mosaic AI: Build and Deploy Production-quality Compound AI Systems — Databricks Blog
14. Mosaic AI Agent Framework: build GenAI apps — Databricks documentation
15. The POWER of SAP Business Data Cloud with Existing Third-Party Investments — SAP Community
16. Databricks and SAP Join Forces to Ready Data for AI with SAP Databricks — SAPinsider