SAP completes its acquisition of Prior Labs
SAP just closed the deal that says enterprise AI’s next fight is tables, not chat. On July 17 the company said it completed its acquisition of Prior Labs, the Freiburg lab behind tabular foundation models, and restated a pledge to put more than €1 billion into scaling that lab over four years.
Purchase terms stay undisclosed. The capital commitment does not. SAP is buying a research stack for the structured data that already runs its customers’ businesses.
What Prior Labs brings to SAP
Prior Labs pioneered tabular foundation models (TFMs) — pretrained systems that predict business outcomes from rows and columns instead of generating prose. Its open TabPFN line targets payment delays, churn, supplier risk, demand forecasts, and similar table-native tasks without training a narrow model per dataset.
SAP’s own path into the category started with SAP-RPT-1, the relational foundation model it launched to prove predictions on enterprise tables beat bolting an LLM onto a spreadsheet. Closing Prior Labs, SAP says, accelerates that TFM push and pulls in one of the leading research teams in the niche. Cofounders Frank Hutter, Noah Hollmann, and Sauraj Gambhir stay central; the May agreement also put Yann LeCun and Bernhard Schölkopf on the scientific advisory board as the lab scales.
The €1B+ scale commitment
SAP’s completion note and the May agreement both frame the same number: more than €1 billion over the next four years to grow Prior Labs into a “globally leading frontier AI lab” for structured business data. That money is pitched at infrastructure, hiring, and multi-year research — not as a disclosed purchase price.
CTO Philipp Herzig’s line from May still defines the strategy: the untapped enterprise AI opportunity “wasn’t large language models; it was AI built for the structured data that runs the world’s businesses.”
Independence vs integration tension
SAP says Prior Labs will keep operating as an independent entity. The May release went further: own brand and research velocity, with a productization path through SAP AI Core, SAP Business Data Cloud, and the Joule agent layer. Hutter called the close the start of a chapter the 18-month-old lab could not fund alone.
That is the usual big-company paradox. Independence is the pitch to keep researchers; integration is how SAP monetizes TFMs inside ERP workflows. If product pressure wins too early, the “frontier lab” story thins. If independence wins forever, the €1B is a subsidy without distribution.
Why tabular FMs matter for ERP data
LLMs are weak at native table reasoning. TFMs are built for it: one pretrained model, in-context examples, instant predictions without a four-hour AutoML bake-off. Prior Labs’ TabPFN already had millions of downloads and topped public tabular benchmarks before the deal; SAP wants that accuracy on the data estates its software already touches.
Analysis: this is SAP refusing to let the LLM narrative own enterprise AI. Owning the TFM lab, pledging nine-figure multi-year spend, and keeping Prior Labs nominally independent is a bet that ERP value sits in predicting the next field in a row — churn, delay, risk — faster than a chatbot can summarize a memo. If the models land in Joule and Business Data Cloud with real lift, SAP gets a category lead peers cannot copy with another GPT wrapper. If they stay a research island, the €1B becomes a very expensive press release.
