SAP is paying unusually aggressively for a very young German AI company. The target is Prior Labs, a Freiburg startup founded only 18 months ago and previously backed by just $9.3 million in pre-seed funding.
The attraction is not another large language model team. Prior Labs is building foundation models for structured data, a field that sits closer to SAP’s core enterprise software business than the consumer chatbot race.
Deal terms
SAP has not publicly disclosed the acquisition price, but reports point to a commitment of EUR1 billion, or about $1.16 billion, over four years. The structure is described as almost all cash, with more than $500 million paid upfront to the company’s three founders.
Prior Labs was founded in Freiburg in 2024 by CEO Frank Hutter, Noah Hollmann and Sauraj Gambhir. Its February 2025 pre-seed round was led by Balderton Capital, with angel backers including Hugging Face co-founder Thomas Wolf and Black Forest Labs co-founder Robin Rombach. Yann LeCun, the former chief AI scientist at Meta, is listed as an adviser.
Why TabPFN matters
The company’s core product is TabPFN, short for Tabular Prior-data Fitted Networks. It is a family of foundation models trained for tabular data.
A text model is optimized to work with language. TabPFN is aimed at tables: sales records, missing fields, anomalies, distributions and relationships between columns. Instead of writing analysis code around a model, the model is meant to understand the structure of the table itself.
The open-source TabPFN models have passed 3 million cumulative downloads on Hugging Face. For a small German research lab, that level of adoption is a meaningful signal in the structured-data community.
SAP’s strategic line
SAP chief technology officer Philipp Herzig framed the thesis plainly: the biggest untapped opportunity in enterprise AI is not large language models, but AI built for structured data.
That view fits SAP’s position. ERP tables, finance documents, sales ledgers, inventory records, HR systems and production orders are not organized like natural language. Pushing them through a general-purpose LLM can be inaccurate or expensive. SAP, meanwhile, sits on some of the world’s most important enterprise structured data.
Prior Labs gives SAP a model layer designed for that data. If SAP can bind proprietary enterprise data, TabPFN-style models and its own application layer together, it gets a defensible AI wedge inside existing customers.
The NemoClaw clue
There was another detail in the announcement. SAP said its API policy blocks unauthorized AI agents, while Nvidia’s NemoClaw has been accepted as a SAP-recognized architecture through the Joule Agents platform.
That means SAP is not only adding AI features to ERP. It is also trying to control which agents get access and how that access is mediated. Prior Labs supplies the model layer, Joule Agents handles orchestration, and NemoClaw becomes an approved agent framework.
The pattern resembles moves by Salesforce with Agentforce and ServiceNow with AI Control Tower. Large enterprise software vendors are competing to define the standard interface and admission rules for workplace AI.
Can it work?
The risk is that TabPFN remains closer to research than production. Open-source downloads do not prove that large enterprises can deploy the technology at scale. Turning the product from demo to general availability for major customers could take one to two years.
The competitive window is also narrowing. Microsoft Fabric is reportedly exploring similar structured-data foundation models, while OpenAI has people looking at GPT-Tabular directions. SAP has moved early, but it still has to build a moat before larger model companies turn toward the same market.
Integration is the final test. A small German research team may seem culturally close to SAP, but the product pace of a large enterprise vendor is very different from the cadence of a lab.
If the bet works, though, SAP may have bought a candidate for the default model layer of enterprise structured data. At that point, EUR1 billion would look inexpensive.
Sources: SAP bets $1.16B on 18-month-old German AI lab and says yes to NemoClaw (TechCrunch); CocoLoop, SAP Acquires AI Startup Prior Labs With EUR1B Investment to Unlock Enterprise Structured Data (The AI Insider)