On October 27 in Berlin, at TechEd 2026, SAP did something not very typical of itself: it opened its most closed layer—ERP application storage—to Apache Iceberg. A November 5 in-depth report made the significance of this clear; Gartner Senior Director Analyst Aaron Rosenbaum said, in his exact words, that SAP is “moving from a very closed application storage model to an open lakehouse model.”
In the same week, there is another figure worth looking at alongside it. SAP's third-quarter report released on October 21 shows that current cloud contract backlog reached €24.1 billion (+26% year over year at constant currency), cloud revenue was €6.67 billion (+25%), total revenue exceeded €10 billion for the first time, and Cloud ERP Suite revenue rose 28%; however, cloud gross margin fell 0.7 percentage points year over year to 73.9% (SAP 2026 Q3 quarterly report, 2026-10-21).
On one hand, orders are booked years out; on the other, gross margins are squeezed by the infrastructure costs of cloud and AI. An open data layer is not a gesture—it is a calculated choice.
Core argument of this article: The most expensive wall for enterprise AI has never been that models are not smart enough, but that "data is locked in ERP applications, and to feed it to AI, you first have to do a major version migration." SAP uses open table formats to decouple "AI readiness" from the "ERP upgrade cycle"; for CIOs, the first criterion for selection in 2027 is shifting from a feature checklist to data portability and ownership of the context layer.
1. The Wall That Was Torn Down: Why AI Keeps Having to Wait for ERP Upgrades
Anyone who has worked on enterprise data projects is familiar with this chain: every time an analysis requirement is raised, it turns into an extraction project; every time an extraction is done, it turns into a round of governance debates over definitions, permissions, and compliance. Data is copied three times, definitions each speak for themselves, and before model training has even begun, half a year has already passed.
What is even harder is for enterprises still running ECC. To feed ten years of historical orders into a demand forecasting model, the implicit premise in the past was: first complete the S/4HANA Cloud conversion. Rosenbaum put this problem very directly—“tying the use of AI and the lakehouse to the ERP upgrade lifecycle has always been a problem.” And SAP this time clearly stated that the openness is not only for S/4HANA Cloud users, but also for ECC users (as reported by The Register, 2026-11-05).
The significance of this statement is that it moves ERP upgrades away from being the 'tollbooth for every data product.' Enterprises can now first prove whether an analytics or AI use case is worth doing, and then decide whether to schedule a major, disruptive version migration for it. The November 5 analysis piece gave a very precise summary: you no longer need a complete S/4HANA Cloud conversion to feed AI models; what you need is catalogs, access rules, lineage, and query contracts.
This is not a vendor slide about 'openness,' but a shift in where the work lands: in the past, the comparison was who could copy data faster; from now on, it will be who can 'certify' data more clearly. Zero-copy moves the burden from extraction to governance.
It is easier to understand in the domestic context. Gartner previously judged that by 2027, 62% of cloud ERP spending will flow to AI-related capabilities (in 2024 this proportion was only 14%, cited from the SAP Connect 2026 report, 2026-10-09). Money is moving toward AI, but if every AI budget must first cover an ERP migration, then the budget simply cannot be spent. This continues the same thread we discussed on November 3—'incremental system construction has peaked, and upgrading and retiring existing data have become the main battlefield': the bottleneck has shifted from 'whether to adopt new systems' to 'whether the data in old systems can be used securely and at low cost.'
II. What Exactly Did TechEd Launch: Four Layers from Tabular Format to Agent Gateway
Aligning the official announcement on October 27 with the report on November 5, SAP's “openness” this time is not a single-point feature, but four layers extending all the way from storage format to the agent invocation entry point.
| Hierarchy | Specific practices | Status / Source Date |
|---|---|---|
| Open table format | Business Data Cloud natively supports Apache Iceberg, with the Apache Polaris universal catalog; it also continues to support Delta Lake | 2026-10-27 announcement: GA dates were not provided for most capabilities |
| Zero-copy sharing | BDC Connect interoperates across Databricks, Snowflake, Google Cloud, AWS, and Microsoft, reducing duplicate copying | 2026-10-27 Announcement |
| Federated Query | Dremio (acquisition completed in 2026-05) SQL engine can query SQL Server, MongoDB, Oracle, and PostgreSQL without moving data; Reflections automatically precompute based on workload | 2026-10-31 / 11-05 Report |
| Semantics and Agents | Ontology Manager; SAP Reltio multi-domain master data management GA, using AI entity resolution to unify customer/supplier/product views; Joule and Databricks Genie planned to be integrated | 2026-10-27 / 10-31 |
| Development and Governance Entry | Joule Studio offers a choice of 70+ models and early access to n8n (open-source automation platform) workflows; the controlled gateway supports third-party agents calling Joule Agents via A2A / MCP; Enterprise AI Gateway unifies authentication, policy, routing, monitoring, and auditing; AI Governance Assistant aligns with the EU AI Act | 2026-10-27 Announcement |
Of these four layers, the one most easily overlooked — and the one CIOs should be watching most closely — is the last layer. SAP has opened up the channel for third-party agents to invoke its own agents, but at the same time has drawn authentication, policy, routing, monitoring, and auditing into a single gateway — which points in the same direction as the three hard constraints (controllability, full-process traceability, digital passport) of China's mandatory national standard on agent security issued in October: what is opened up is the right to invoke; what is tightened is the governance surface.
Legacy link
Extract → Copy → Governance debate → Wait for ERP upgrade → Only then can AI go live
New Link
Catalog → Access rules → Lineage → Query contract → Data available in place
Rosenbaum also pointed out an as-yet-undecided contest: Joule and Genie need to connect, which shows that where the “context layer” actually sits is becoming a point of tension among all vendors and enterprises. He himself does not believe there will be a single winner—“Customers are all grappling with: how exactly do we build these context layers, and how do we give consistent answers?” (2026-11-05). This statement matters more to enterprises than to vendors: whichever party holds the context layer is effectively given the right to interpret data for the next decade.
III. Openness does not mean no accountability: three constraints that are easily overlooked
First, most announcements did not include availability dates. An October 27 report explicitly noted that most of this batch of releases did not provide GA timelines: n8n workflows are only in early access; the proprietary agents run within SAP-managed environments and are free only to some customers until December 2026; the design tool is "free within fair-use limits." Before writing these into a 2027 launch plan, ask the vendor for written availability commitments.
Second, the old system will run in parallel for many years.At the same time, SAP updated the direction statement for BusinessObjects BI and Data Services: both remain supported and can stay on-premises or in a hosted cloud, with only Lumira Designer planned to retire with BI 2029. Most SAP customers will run both BusinessObjects and BDC simultaneously for a considerable time (IBA Group October review, 2026-10). "Open" does not mean "the old stack will automatically disappear."
Third, the security bill is rising in tandem. On October 13, SAP Security Patch Day released 16 new security notes at once, 4 of which were critical. Note 3805589 is an ABAP code injection in substitution and validation rules in the on-premise and private editions of S/4HANA (since 2020), sitting directly next to the finance core; Note 3797111 allows low-privilege portal users to reach connected remote systems. Agents use the same business roles as employees—permission reviews must cover "applications that interpret user input as logic," not just check segregation of duties (IBA Group, 2026-10). The more open the data layer, the tighter the role boundaries must be.
SAP 2026 Q3 cloud gross margin, down 0.7 percentage points year over year (SAP quarterly report, 2026-10-21); cloud contract backlog of €24.1 billion for the same period, +26% at constant currency
Only by looking at these three constraints together can one understand the nature of this opening up: it shifts competition from "who locks their data down more tightly" to "whose governance is more trustworthy and whose context layer is more worthy of being entrusted." For enterprises, this is of course a good thing, but the initiative will not automatically fall into their own hands just because vendors open up.
IV. Four Practical Recommendations for CIOs / Digital Leads
1｜Make "AI readiness" a separate initiative, detached from the upgrade roadmap. First, use the four-piece set of catalogs, access rules, lineage, and query contracts to take stock of the existing ECC data, assess which analytics scenarios can run in place, and then decide whether to schedule the S/4HANA conversion timeline. Don't let a single major version migration become a prerequisite for all AI scenarios.
2｜Write context layer ownership into the architecture decision record (ADR). Clearly answer: should the business semantic layer be placed within the ERP vendor's semantic layer, or on top of a replaceable open table format? The acceptance criteria must include at least three points—data can be exported, the engine can be replaced, and every access can be audited. This item determines whether you still have bargaining power five years from now.
3｜Open interfaces must always use open protocols + a controlled gateway. When third-party agents invoke ERP agents via A2A / MCP, authentication, policy, routing, monitoring, and auditing must be recorded on the unified gateway, and a Kill Switch and rollback path must be retained. This echoes the three hard constraints of the October mandatory national standard for agent security (plan number 20263116-Q-252); do not mistake "able to call" for "governed".
4 | Expand permission reviews to “business roles that can explain inputs,” and write the GA date into procurement acceptance clauses.In light of the October patch-day advisory on code injection in core financial code, include access permissions for configuration-type and rule-type applications in quarterly reviews; for newly announced AI capabilities, specify in the contract the availability date, metering method, and remedies for failure to meet standards, rather than relying on the roadmap from the launch event.
Finally, I want to say one thing: what this round of opening truly dismantles is not the database wall of some vendor, but the long-defaulted sequence of "upgrade first, then become intelligent." For digital leaders, once the sequence changes, budget structures, acceptance criteria, and negotiating leverage all have to be rewritten accordingly. Whoever first gets their data catalog, access rules, and lineage clear will be able to negotiate with vendors in the next round of selection using facts rather than PowerPoint.
This article is compiled and analyzed based on public information: SAP News 'SAP Empowers AI Developers to Build What Matters and Turn Intent into Autonomous Execution' and the TechEd 2026 Innovation News Guide (2026-10-27); an in-depth report on SAP TechEd 2026 open lakehouse and Iceberg/Polaris (The Register-affiliated / worldprogramming, glodaxia, 2026-11-05); an interpretation of SAP TechEd data and Joule Studio (alicloud, 2026-10-31; cloudnews.tech, 2026-10-27); a summary of SAP's 2026 Q3 quarterly report and October Security Patch Day (IBA Group, 2026-10); Gartner's forecast at SAP Connect 2026 that 62% of cloud ERP spending will flow to AI (2026-10-09). Amounts and growth rates in the text are quoted according to the original wording and basis of their respective sources, without conversion or consolidation. Data as of 2026-11-06.
Zhongdi Information · Specializing in Odoo and digital solutions for enterprise management
