SAP and Microsoft have detailed bi-directional, zero-copy access between SAP Business Data Cloud (BDC) and Microsoft Fabric’s OneLake — so Fabric can read SAP’s semantically rich data products without duplication, and BDC can also consume datasets from OneLake. Microsoft positions this as a way to preserve SAP meaning while cutting duplication and latency.
But timing matters. SAP’s announcement sets BDC Connect for Microsoft Fabric on a Q3 2026 GA path, while Fabric platform upgrades already make OneLake a credible home for SAP-adjacent analytics today. Pilot now; scale as features harden.
On the AI front, SAP’s TechEd updates underscore an expanding Business AI footprint (including Joule Agents) and a developer-centred, agentic approach across the suite.
“AI isn’t starving for models; it’s starving for context. Zero-copy is how we feed it — securely and fast.”
— Jon Pepper
What Changed This Week (and What’s Timelined)
- Zero-copy, both ways. Fabric can read SAP BDC data products without duplication, and data can flow back into SAP apps — simpler access, semantics intact.
- Availability & runway. GA target is Q3 2026 for BDC Connect; meanwhile the November 2025 Fabric wave (including SQL Database GA) strengthens the landing zone now for SAP-adjacent workloads and Copilot.
- Integration pulse. Data Factory and platform posts highlight OneLake as the backbone for zero-copy / zero-ETL patterns across partners — a practical way to shrink pipelines and latency.
Field Architecture: A Simple Pattern That Works
The key is to not overcook it. This is the lightest path from SAP meaning to AI outcomes, with governance built in. Here’s a simple 3-step field architecture you can follow:
- Publish SAP-meaningful data products in BDC. Start with curated CDS views and expose them as governed data products. Open Resource Discovery (ORD) is the standard for describing and discovering these products across hubs.
- Expose to OneLake via zero-copy. Fabric reads the SAP data product directly in OneLake for Power BI, Copilot and downstream AI — no forklifting, less drift.
- Layer agents on one foundation. Joule drives SAP-native actions; Copilot narrates and collaborates — both referencing the same governed truth.
Risks & Governance (Don’t Skip This)
Speed without discipline is how programs stall. Lock down scope, security and cost controls.
- Timelines & scope. Capabilities are still maturing, so design for pilot → scale and watch what’s actually GA in Fabric’s rolling updates.
- Security posture. Treat OneLake as an extension of your SAP trust zone: identity, lineage, audit, masking. November’s SAP Patch Day carried multiple critical notes; analyst write-ups flagged HotNews issues in SQL Anywhere Monitor and Solution Manager.
- FinOps and guardrails. Track refresh costs, query latency and permissions from day one; align RLS/ABAC and PII masking early so governance accelerates — not blocks — delivery. Platform posts outline new OneLake sources, security and capacity controls worth adopting.
“Move like a recon team: prove the path with one dataset, harden security, then scale deliberately.”
— Jon Pepper
For Executives (Decision-Makers)
If you own a P&L, this is what moves the needle:
- Business case. Shorter data supply lines; self-service analytics on SAP truth; faster AI outcomes tied to margin, cash and stock — with less duplication and lower compliance risk. Fabric’s unified stack (OneLake + Copilot + SQL DB GA) is built to cut latency and copy sprawl.
- 90-day outcome. One funded pilot, one automated KPI (e.g., margin-erosion alert), and a Joule ↔ Copilot demo leaders can use in Teams — value the board can actually see. TechEd briefings show Joule expanding into more operational flows, making the hand-off practical.
For Consultants & Architects
Make it real in days, not months.
- Ship 2–3 products. Map 2–3 CDS views → BDC data products, validate lineage and RLS, and prove zero-copy into OneLake. ORD metadata and explicit ownership improve discoverability and trust.
- Guardrails early. Align RLS/ABAC, enable PII masking in Fabric, and set prompt-grounding standards for agents. Keep an eye on Fabric’s “What’s new” roll-ups so behaviour changes don’t surprise users.
Leadership in Motion
My advice is to treat this like a field exercise: tight scope, clear owners, visible metrics, and a business-driven demo.
- Pick one high-value flow (stock-at-risk, ageing WIP, cash leakage).
- Publish once in BDC as a governed data product with clear semantics and row-level security.
- Prove zero-copy into OneLake with end-to-end lineage and access tests.
- Demo the AI hand-off: Joule proposes an SAP action; Copilot explains the “why” in Power BI/Teams. Ignite guidance points to this cross-assistant pattern as a core design point.
The Bottom Line
Zero-copy isn’t a slogan — it’s a logistics upgrade for AI on SAP. Shorter supply lines, fewer breakpoints, and business meaning kept intact from source to decision.
Build clean, governed data products once and reuse them everywhere — analytics, Copilot, Joule and downstream apps — to lift time-to-insight, cut duplicate spend, and reduce operational risk. Aim for one pilot, one measurable KPI lift, and one Joule ↔ Copilot hand-off your team can run without IT. Prove it, secure it, then scale.
