Enterprise AI
SAP Closes Its €1 Billion Bet on the AI Nobody Is Talking About
Newaiera Desk · 2026-07-22 · 5 min read
SAP finished acquiring Prior Labs on July 17 and committed more than €1 billion over four years to run it as an independent frontier lab. The target is not
While the industry argues about which chatbot writes better code, SAP just spent over €1 billion on a category most people have never heard of.
On July 17, SAP SE confirmed it had completed its acquisition of Prior Labs, the German research company behind Tabular Foundation Models. The deal was first announced in early May. Financial terms were not disclosed, but the commitment attached to it was: SAP is pledging more than €1 billion over the next four years to scale Prior Labs into what it calls a globally leading frontier AI lab — one built specifically for the structured data that underpins the world's businesses.
Crucially, Prior Labs will keep operating as an independent entity.
What is a Tabular Foundation Model?
Large language models are trained on text. Tabular Foundation Models (TFMs) are trained on tables — the spreadsheets, transaction logs, customer records and operational metrics that make up the overwhelming majority of enterprise data.
The conventional way to predict something from that data is to train a bespoke machine-learning model per dataset, per task, and retrain it as things drift. It works, but it needs data scientists and it does not scale across thousands of business questions.
A TFM flips that. One pre-trained model handles prediction tasks — payment delays, customer churn, supplier risk, demand forecasting — directly on structured data, without training a fresh model for each one.
SAP's blunt strategic insight is that the most commercially valuable predictions in an enterprise are not made on paragraphs. They are made on rows and columns.
Who Prior Labs are
Founded in 2024 in Freiburg by Frank Hutter, Noah Hollmann and Sauraj Gambhir, Prior Labs is best known for TabPFN, its open-source tabular model series. The work was published in Nature, and TabPFN has passed three million downloads with an active developer community around it. SAP has said it intends to keep supporting that open-source strategy.
The company is roughly 18 months old. Before this, it had raised a single €9 million pre-seed round.
That trajectory — academic paper to Nature to open-source adoption to billion-euro enterprise deployment partner, inside two years — is the reason this deal is getting attention in European tech policy circles well beyond SAP's customer base.
Why SAP, and why now
This is not a standalone bet. SAP had already built SAP-RPT-1, its own early proof that foundation models trained on structured data could complement LLMs rather than compete with them. Prior Labs brings in one of the leading research teams working on the same problem.
It also arrives eleven days after SAP completed a second acquisition: the data lakehouse company Dremio, closed on July 6. Buying the model layer and the data layer in the same quarter is not a coincidence — it is a platform strategy. SAP wants to own both the structured data and the models that reason over it.
For SAP customers, the practical endpoint is predictions surfaced through the products they already use, backed by models architected for business data rather than general-purpose language models retrofitted for the job.
The open questions
Two things are worth watching.
The first is independence. Large enterprise software companies have an uneven record of preserving the autonomy of research labs they acquire. SAP has been explicit in its public statements that Prior Labs keeps its independence, its brand and its open-source commitments. Similar promises have been made before elsewhere. Verification comes in the post-close years, not the press release.
The second is competition. Tabular foundation models are no longer a quiet corner of academic ML. If SAP has correctly identified where enterprise AI value actually sits, it will not be the only company that notices.
The bigger signal
Strip away the specifics and this deal says something about where enterprise AI is heading in 2026. The conversation is shifting from how capable is the model to how well does it do a real business task on real business data. Text was the first frontier because text was easy to scrape. Structured data is where the money is.
SAP is betting a billion euros that the second frontier is the one that pays.