The short answer
SAP completed its acquisition of Prior Labs on 17 July 2026 and committed more than €1 billion over four years to scale it into a frontier AI lab for structured data. Prior Labs, an 18-month-old startup out of Freiburg, pioneered tabular foundation models — AI pre-trained to predict outcomes like payment delays, churn and demand directly from tables, a job large language models do badly.
The practical reading for engineering leaders: a big enterprise vendor just paid frontier-lab money to say the untapped opportunity in business AI is the structured data companies already sit on, not another wrapper around a chatbot. You do not have to wait for SAP to productize it — the underlying open models exist today — but the deal is a strong prompt to look for the prediction problems hiding in your own databases.
What did SAP actually buy?
Prior Labs is a research startup founded roughly 18 months ago in Freiburg, Germany, by Frank Hutter, Noah Hollmann and Sauraj Gambhir. Its work centers on tabular foundation models, and its TabPFN model series — published in the journal Nature — set the state of the art on tabular prediction benchmarks across hundreds of independent academic studies. SAP announced the acquisition in May 2026 and confirmed it had closed the deal on 17 July 2026, with the price reported at more than €1 billion.
The money does not stop at the purchase. SAP has committed to invest more than €1 billion over the next four years to scale Prior Labs into what it calls a globally leading frontier AI lab for the structured data that underpins the world's businesses. Crucially, Prior Labs keeps its own brand, leadership and research agenda, will continue to publish its research, and will keep making its models openly available — a structure aimed at preserving research velocity while wiring the lab into SAP's enterprise data reach and its GenAI integration pathway through products like SAP Business Data Cloud and the Joule agentic layer.
The strategic logic came straight from SAP's own framing: the company said it concluded the greatest untapped opportunity in enterprise AI was not large language models but AI built for the structured data that runs the world's businesses. That is a notable statement from a vendor whose customers store decades of exactly that data — the ledgers, orders, inventory and customer records sitting in ERP systems — and it explains why Prior Labs' models have already been used for things like predicting train failures at Hitachi and sharpening financial forecasting at TD.
What is a tabular foundation model?
A tabular foundation model is pre-trained to make predictions directly from structured data — the rows and columns of a table — rather than from text. Where a traditional approach trains a fresh model for each dataset, a TFM can be pointed at a new table and predict a target column with little or no task-specific training. Prior Labs describes a single foundation model capable of solving prediction tasks such as payment delays, churn, supplier risk and demand forecasting straight from structured enterprise data.
That matters because large language models, for all their fluency, have only a rudimentary understanding of tables, numbers and statistics. Serialize a spreadsheet into a prompt and an LLM will guess plausibly but predict unreliably; it was never built to be a calculator over your records. Tabular models treat the table as a first-class input and are optimized for exactly the quantitative jobs — forecasting, scoring, ranking risk — that dominate operational analytics. The two model families are complements, not rivals: LLMs read and write language; TFMs predict outcomes from data.
Why is SAP spending €1B on tables?
Because the value is where the data already is. Most enterprises have far more proprietary structured data — transactions, orders, claims, sensor logs — than proprietary text, and that data encodes the outcomes a business actually cares about: who will pay late, which customer will leave, how much stock to hold. A model that predicts those outcomes accurately and cheaply, straight from the tables, plugs into decisions that move money in ways a document chatbot rarely does.
There is also a sovereignty dimension. SAP is building this as a European frontier lab, and structured-data models run on data you can keep inside your own environment. For teams in FinTech and other regulated sectors, that is not a footnote: a prediction model that operates on your tables in your region keeps sensitive records in-jurisdiction end to end, which is far easier to defend under GDPR, DORA or HIPAA than shipping the same data to an offshore general-purpose API. The deal packages a technical bet and a compliance story in one.
What it means for US & EU software teams
The first lesson is that enterprise AI is not only about language. The industry spent two years wiring chatbots over documents; this deal is a large vendor pointing at the other half of the problem — prediction over structured records — and calling it the bigger opportunity. If your AI roadmap is entirely LLM features, you may be ignoring the models most likely to improve a forecast, a collections process or a fraud score.
The second lesson is that the tooling is already usable. You do not need to wait for SAP's productized version. Open tabular models such as TabPFN can be evaluated today against your existing gradient-boosting or hand-built baselines on a real problem. The right move is a scoped pilot on one high-value prediction, benchmarked on your own metrics — not a platform commitment made on the strength of a press release.
The third lesson is that model choice belongs behind an abstraction. The landscape is moving in both language and tabular AI, and the vendors, prices and hosting options will keep shifting. Keep prediction behind a clean interface so you can route a task to a tabular model, a gradient-boosted tree or an LLM depending on which clears your quality and cost bar — and swap the component underneath as the market changes, without rewriting the product.
What to do this quarter
Treat the SAP deal as a prompt to look at the data you already have the way Prior Labs' customers do.
- Inventory your prediction problems. List the outcomes hiding in your databases — churn, late payment, demand, capacity, fraud — and rank them by business impact.
- Pick one high-value target. Choose a single prediction where a few points of accuracy clearly move money, and where you already have clean historical rows.
- Benchmark a tabular model against your baseline. Test an open tabular foundation model against your current approach on the same data and the same metric — no vendor lock-in required to learn something.
- Decide model and hosting together. For regulated or sensitive records, run the model inside your own US or EU environment so data residency is a property you can prove.
- Keep prediction behind an interface. Route each task to the best-fitting model — tabular, tree-based or LLM — and keep the others available as fallbacks.
- Instrument accuracy and cost. Track prediction quality and per-call cost over time so the choice between model families is driven by evidence, not by which vendor announced last.
None of this is investment advice, and one acquisition does not settle how enterprise AI will be built. But the signal is hard to miss: a major vendor just paid frontier-lab money on the thesis that the next wave of business AI runs on structured data, not just language. Build so you can test that thesis on your own tables — and keep the freedom to route each problem to whichever model wins.
Frequently asked questions
What is a tabular foundation model?
A tabular foundation model (TFM) is an AI model pre-trained to predict directly from structured data — the rows and columns of tables that run most businesses. Instead of training a separate model per dataset, one TFM can be pointed at a new table and predict outcomes such as payment delays, churn, supplier risk or demand, often without task-specific training. Prior Labs' TabPFN series, published in Nature, set the state of the art on tabular benchmarks across hundreds of independent studies.
What did SAP acquire and how much is it investing?
SAP completed its acquisition of Prior Labs — an 18-month-old German startup founded by Frank Hutter, Noah Hollmann and Sauraj Gambhir — on 17 July 2026, in a deal reported at over €1 billion. SAP has committed to invest more than €1 billion over four years to scale Prior Labs into a globally leading frontier AI lab for structured data. Prior Labs keeps its own brand and leadership and will keep publishing research and releasing models openly.
Why do LLMs struggle with structured business data?
Large language models are trained on text and have only a rudimentary grasp of tables, numbers and statistics, so they are unreliable at quantitative prediction over structured records. Tabular foundation models are purpose-built for that data: they treat rows and columns as first-class inputs and are optimized to predict outcomes from them. For forecasting, scoring and risk, a TFM is designed to be both more accurate and cheaper than prompting an LLM over serialized tables.
What does the SAP deal mean for enterprise software teams?
It signals that a large share of practical enterprise AI value sits in structured data most companies already own, not only in chatbots over documents. Teams should inventory the prediction problems in their databases — churn, collections, demand, fraud, capacity — and treat tabular models as a distinct tool alongside LLMs. Because these models run on your own tables, they also keep sensitive data in-jurisdiction, which matters for GDPR, HIPAA and DORA.
Should teams wait for SAP's productized version or act now?
There is no need to wait. Open tabular models such as TabPFN are available today and can be evaluated against your own gradient-boosting or manual baselines on a real prediction task. The deal is a signal to prioritize structured-data prediction, not a gate. Pilot a tabular model on one high-value forecast or scoring problem now, keep model choice behind an abstraction, and adopt a productized platform later if it clears your quality and cost bar.
Sources
SAP News Center — SAP Completes Prior Labs Acquisition (company announcement)
Tech.eu — SAP acquires Prior Labs just 18 months after launch in €1B+ deal
TechCrunch — SAP bets $1.16B on 18-month-old German AI lab