SAP Business Data Cloud Adds Iceberg, Polaris and Dremio: What Changes for Enterprise AI
Published: 29 October 2026 | SAP TechEd 2026 | Status: SAP announcement; verify tenant availability and entitlements
Most enterprise AI teams have a data problem before they have a model problem. An agent can reason brilliantly and still produce the wrong answer when customer, supplier, and transaction records live in different systems with inconsistent definitions. SAP's latest Business Data Cloud announcement tackles that less glamorous but essential problem: how to query distributed enterprise data, preserve business meaning, and avoid building another copy of everything.
On 29 October 2026, SAP described a broader open lakehouse foundation for SAP Business Data Cloud (SAP BDC), bringing together native Apache Iceberg support, an Apache Polaris-powered universal catalog, and Dremio's federated SQL engine. This is more consequential than a new dashboard. It changes the architecture discussion for teams building analytics and AI agents across SAP and non-SAP systems.
What SAP actually announced
According to SAP's 29 October product announcement, SAP BDC now supports Apache Iceberg and a universal catalog powered by Apache Polaris. Dremio's SQL query engine is positioned to federate queries across lakehouses and databases, including SQL Server, MongoDB, Oracle, and PostgreSQL. SAP also describes Dremio Reflections, which can precompute frequently needed results to improve performance as usage patterns evolve.
There is an important chronology here. SAP announced its intention to acquire Dremio on 4 May 2026 and confirmed the acquisition was completed on 6 July 2026. The October TechEd material explains how SAP intends to put that technology to work in its data platform. The acquisition itself is not new news today.
Why Iceberg, Polaris and Dremio matter together
Apache Iceberg is an open table format for analytical datasets. Apache Polaris supplies catalog capabilities so engines can discover and govern those datasets. Dremio adds a query layer that can reach data across systems without first requiring a conventional extract-and-load pipeline for every use case. These are complementary layers, not three interchangeable products.
Consider a procurement analyst asking an AI assistant why a supplier's delivery performance deteriorated. The relevant facts might be scattered across SAP purchase orders, a non-SAP quality system, logistics events, and supplier reference data. A federated architecture can reduce the need to copy all of those records into a separate AI-specific warehouse. But federation does not automatically reconcile conflicting supplier identifiers or fix authorization gaps. Those remain design responsibilities.
Likewise, zero-copy is an architectural capability, not a promise that every query has zero latency, zero cost, or zero data movement. Caching, materialization, source-system load, network boundaries, and data residency still matter. SAP's product claims should be tested against actual workloads rather than treated as performance guarantees.
The second announcement: business context for agents
SAP also introduced SAP Business Data Cloud Ontology Manager, which helps teams create domain-specific views of SAP Knowledge Graph. A collections agent, for example, needs a coherent understanding of customers, invoices, payment terms, and credit risk, rather than an unrestricted view of every enterprise object. SAP's TechEd keynote coverage identifies Ontology Manager as available in beta. Do not present it to customers as universally generally available.
Separately, SAP states that SAP Reltio is generally available in SAP BDC, providing multidomain entity resolution across SAP and non-SAP records. That is a different layer from federated querying: it addresses whether two records describe the same customer, supplier, product, or asset. SAP's Reltio product description explicitly positions it alongside SAP Master Data Governance, rather than as a blanket replacement for MDG.
What architects and SAP customers should verify before committing
- Confirm availability, region and commercial scope. SAP's October announcement describes platform capabilities, but does not establish that every feature is active in every customer's tenant. Check your BDC edition, region, entitlements, product documentation, and roadmap with SAP.
- Map the real source landscape. List SAP and non-SAP systems, ownership, data sensitivity, refresh needs, and the queries that would benefit from federation. Separate workloads that require transactional consistency from those suited to analytical access.
- Test a representative zero-copy use case. Measure query latency, concurrency, source-system impact, cost, and behavior during network or source outages. Compare against an existing replicated pipeline rather than assuming federation is always cheaper.
- Design identity and policy propagation. Confirm how row-level access, masking, consent, audit events, and service identities behave across the catalog, query engine, and source systems. An agent should never acquire broader access simply because it uses a new data layer.
- Separate data access from data truth. Decide which master data needs entity resolution and stewardship. Federated access alone cannot establish a reliable golden record.
- Pilot the agent with measurable acceptance criteria. Use a narrow question such as supplier-risk investigation or collections prioritization. Evaluate answer accuracy, traceability to source records, permission enforcement, response time, and cost per request.
What is available, and what remains to be checked?
Verified announcement (29 October): SAP describes native Iceberg support, Polaris catalog integration, Dremio federation, expanded SAP Databricks capabilities, and SAP Reltio general availability. Verified beta status: SAP BDC Ontology Manager is identified as beta in SAP's TechEd keynote report. Not verified for an individual customer: regional rollout, licensing, feature activation, connector-specific limits, and production performance. Treat those as implementation checks, not assumed entitlements.
My practitioner takeaway
The strategic shift is from moving data into yet another AI repository toward making governed data usable where it already lives. That is promising, but the architecture only succeeds when the business meaning of a supplier, employee, invoice, or customer survives the journey. I would start with one cross-system business question, prove the access controls and economics, and only then scale the lakehouse and agent architecture.
For broader platform context, see SAP Business Data Cloud and SAP's full 29 October technical announcement. SAP's TechEd keynote recap provides the availability distinction for Ontology Manager.
No comments:
Post a Comment