For SAP customers, the value of agentic AI hinges on access to trusted data that spans the entire business. In virtually all customer environments, that data estate extends well beyond SAP systems, but in most environments the data is hard to see, govern, access, and blend with enough confidence for an agent to act on it.

In less than six weeks, SAP has lined up three pending acquisitions that each target a distinct layer of the same data foundation problem. Reltio brings trusted entity resolution across SAP and non-SAP sources; Dremio brings an Apache Iceberg-native lakehouse with a universal catalog on Apache Polaris; and Prior Labs brings an AI architecture purpose-built for tabular business data. All three are intended to sit underneath SAP Business Data Cloud, the company’s unified data platform for AI, and Joule, its embedded AI agent and copilot layer.

It’s a big strategic bet, made with the recognition that even though enterprise systems of record have long been the center of gravity for business data, SAP cannot deliver agentic AI on a foundation built only from what runs inside its own walls.

Agentic AI’s Application Boundary Problem

Agentic AI has already reached production inside enterprise application suites, with embedded agents answering questions, drafting work, and quietly executing tasks across ERP, CRM, procurement, and HR modules at many of the largest enterprises in the world. What most vendors will not say out loud is how narrow that production footprint actually stays, because those agents tend to be scoped to whatever their host application can see and stop at the boundary of that one product family’s data. The problem is that decisions an enterprise actually wants an agent to take rarely fit inside that perimeter. They draw on customer records, supplier histories, financial state, and analytical context that sit across different systems and inside a lakehouse most of those systems were never wired into. The constraint on scaling agentic AI past the application boundary sits in the data foundation underneath, in whether an agent can see, govern, access, and blend the data living outside its native home.

Take seeing the data first. This requires knowing what exists across operational systems, partner feeds, and the analytical lakehouses where everything outside the application gets stored. Most enterprises have built that data ecosystem piecemeal, one application at a time, and the gaps show up the moment an agent can’t access those other applications or the underlying data stored in them. Governing it means lineage, entity resolution, and policy enforcement strong enough to hold up when the agent is committing transactions rather than offering suggestions. Access faces a latency and physics problem, because copies, conversions, and overnight batches collapse the responsiveness an agent needs to be most useful in real workflows. And blending depends on resolving the same customer, supplier, or product to one identity across systems that were built independently of each other.

SAP’s own study of 1,600 senior executives, The Value of AI, puts hard percentages under that diagnosis, with 75% of organizations naming incomplete or inconsistent data as a top readiness challenge, 68% naming siloed data, 55% lacking confidence in sharing data across business functions, and 60% lacking that confidence with external partners. Worse than that, planned data investment growth is running well behind planned AI investment growth, which means the foundation gap is widening rather than closing.

Closing that gap inside one platform has historically been hard for any single vendor to pull off. The data platform vendors with the strongest open lakehouse stories rarely own the application layer above them, and the application vendors with the strongest customer relationships have typically leaned on partner data platforms whose roadmaps did not move at AI speed. The vendor that closes it inside one integrated stack, owned end to end and tied together tightly enough that an agent can lean on every layer at once, will own the data foundation that will likely decide the next several years of enterprise application AI competition.

Buying the Open Data Foundation

On May 4, SAP announced its intent to acquire Dremio and Prior Labs on the same day, the second and third deals in a stretch of less than six weeks that opened March 27 with the announcement of the planned Reltio acquisition for cloud-native master data management. With Dremio, SAP Business Data Cloud becomes a lakehouse built natively on Apache Iceberg, the open table format that the broader data platform market has been consolidating around. In practical terms, this should mean that data sitting in SAP applications and data sitting in any other Iceberg-compatible store can be queried as one set of tables, without copies, conversions, or new pipelines to wire up. SAP plans to keep HANA Cloud handling real-time transactions while the Iceberg-native lakehouse handles analytical and AI workloads against that same shared layer.

Dremio also brings an open catalog built on Apache Polaris that is set to become the foundation for the SAP Knowledge Graph, a layer designed to give a Joule agent more than raw rows and columns to work with, attaching business meaning, organizational structure, regulatory context, and how data flows from one system to another so the agent can reason about what it is actually looking at. SAP has committed to continuing investment in Apache Iceberg, Apache Polaris, and Apache Arrow after the deal closes; these are open-source projects that Dremio has been one of the leading contributors to.

The Prior Labs deal brings a different category of AI architecture into SAP. The company pioneered tabular foundation models, a class of AI model trained natively on rows and columns rather than text. The bet behind that work is that the kinds of decisions enterprises actually run on every day — predicting payment delays, scoring supplier risk, identifying upsell opportunities, flagging customer churn — are structured-data problems that large language models were never optimized for. Prior Labs’ open-source TabPFN model, which has accumulated more than three million downloads, currently sits at the top of the TabArena benchmark for structured prediction, and reportedly delivers the accuracy of a multi-hour automated machine learning pipeline in a single model run. The deal includes the planned establishment of SAP’s frontier AI lab in Europe, with offices in Freiburg, Berlin, and New York — a commitment of more than €1 billion over four years. It also sets up Prior Labs to operate as an independent research entity whose models feed into SAP AI Core, Business Data Cloud, and Joule.

The Bets, the Risks, and What to Watch

Of the three pending acquisitions, Dremio carries the most strategic weight. The most persistent criticism of SAP’s data platform story has been that SAP data lives behind a wall, even as the rest of the market has rallied around Apache Iceberg as the open table format that lets enterprise applications and external lakehouses share one set of tables. Buying the leading Iceberg-native lakehouse, plus the open Polaris catalog underneath it, removes that objection in one move. Once the deal closes, SAP can stand credibly inside the open data conversation in a way it could not have a quarter ago.

Prior Labs is the more interesting research bet of the two May 4 deals. For two years, the enterprise AI conversation has been built around large language models, focused on unstructured content like documents, transcripts, and chat, and on non-deterministic outputs like summaries, drafts, and conversational responses. The market is now starting to move toward marrying that unstructured side with the structured data that actually runs the business. That is SAP’s wheelhouse, and Prior Labs gives SAP a credible position to extend that lead into the AI architecture that reasons over it. The questions enterprises actually pay to answer (who is going to pay late, which supplier might fail, which customer is about to churn) are tabular questions that LLMs handle poorly by design. A foundation model class purpose-built for that data shape is a real category, and SAP putting €1 billion behind it over four years demonstrates that the company is treating tabular AI as its own architectural priority with its own lab and its own roadmap. It also marks one of the more ambitious sovereign AI moves a European software vendor has made in years, planting a frontier-grade tabular AI lab on European soil at a moment when the region is openly searching for AI capability it doesn’t have to import.

Naturally, Reltio can anchor both as the cloud-native master data play, because an open lakehouse and a tabular model would both fall apart without a clean entity and context layer to ground them. It’s when they’re considered together that these three deals add up to a vertically integrated data and AI stack that almost no enterprise application vendor would have the cash, appetite, or conviction to assemble this fast.

That said, there are clear risks. Integrating one company is hard. Three? That’s a different beast and the hardest question mark for SAP. Can the company stitch all three acquisitions together into something coherent fast enough for the integration to matter? The integrated solutions make complete sense on paper, but individually, each deal carries a specific risk that goes beyond standard post-merger friction. Dremio’s open-standards posture has to stay intact post-close, because the Iceberg-native pitch only works if the open community treats SAP as a credible partner; if contributions to Apache Iceberg, Polaris, and Arrow were to slow down, it would collapse that story. With Prior Labs, the value of a frontier lab depends on its ability to keep publishing, recruiting, and attracting the academic partnerships that make it a frontier lab, all of which can fade quickly once a research team gets absorbed into a corporate roadmap. With Reltio, the question is whether its golden records become the entity ground truth that Business Data Cloud and Joule actually rely on, or whether Reltio ends up running parallel to SAP Master Data Governance, the company’s existing master data product, and creating new fragmentation inside the SAP estate.

The proof points worth watching through year-end include Iceberg interop demonstrations against external lakehouses with no data movement; named Joule customers running tabular foundation model predictions against production data; and Reltio golden records showing up as the entity resolution source feeding live agent decisions. The next several quarters will tell whether those land as pieces of one integrated foundation or three product lines under shared corporate cover.

Beyond the Application Boundary

Every prior wave of enterprise software has taught that capability without a trusted data foundation underneath it stalls before it ships, and enterprise AI buyers have spent eighteen months relearning the same lesson. The agentic wave is repeating the pattern at a faster cadence and with higher stakes, because an agent that acts on bad data can quietly commit a wrong payment, release the wrong shipment, or expose the wrong record before anyone notices. The market is already sorting vendors into the ones that treat the data foundation as a precondition and the ones that treat it as someone else’s problem. The smart bet is that this sorting is going to define enterprise application AI competition through the rest of the decade.

SAP is one of the application-centric vendors moving most aggressively to own the trusted, open data foundation underneath agentic AI, and the only one to do it through three pending acquisitions in less than six weeks. As the enterprise application vendor with the most money and inventory passing through its systems every day, SAP has the most to lose from agents acting on bad data. The Reltio, Dremio, and Prior Labs deals are SAP making that argument with its checkbook. And that bet paying off depends on how cleanly the company integrates three independent companies into one platform story over the next year.

Whether this puts SAP ahead of the application-vendor pack or catches it up to where the rest of the market has been heading is the open question, but the bet matters because of where SAP starts from, managing the enterprise applications and business data that runs the world’s largest enterprises. The next phase in the evolution of the category will be defined by how the answers come back, in customer commitments, in open-source contribution velocity, and in whether tabular AI emerges as a real architectural lane alongside the LLM stack.