SAP has agreed to acquire Freiburg-based Prior Labs and commit more than €1 billion over four years to turn the 18-month-old startup into a European frontier AI research lab, in a deal that combines corporate acquisition, long-term research funding, and an explicit bet that the next enterprise AI battleground will not be chatbots but tabular foundation models trained on structured business data, where SAP's distribution, workflow ownership, and customer lock-in give it an advantage that standalone AI startups in Europe and the US will struggle to match.
The specific identity of Prior Labs makes the deal more interesting than a generic corporate acqui-hire. The company was founded in 2024 by Frank Hutter, Noah Hollmann, and Sauraj Gambhir, all of whom bring serious technical credibility to the tabular AI problem. Hutter is one of Europe's most respected machine learning researchers, known for work in AutoML and model selection. Hollmann and Gambhir have worked on the practical challenge of making machine learning systems effective on structured data, which remains one of the least glamorous but commercially most important corners of enterprise AI. Prior Labs raised only a €9 million pre-seed before SAP stepped in with a transaction large enough to turn the startup into a lab with a four-year corporate research budget exceeding the total lifetime funding of many European AI companies. That asymmetry tells you everything about where the real commercial leverage is in this segment: not in raising a few more venture rounds to compete with frontier labs in chat, but in getting access to the proprietary business data, workflows, and sales channels that enterprise software incumbents already control.
Tabular data is the format in which enterprise life actually happens. ERP records, procurement histories, finance ledgers, supply chain events, invoicing systems, sales pipelines, inventory movements, and customer order data are all tabular in the broad sense that matters for machine learning. A company like SAP has built its business on systems that structure this information, and the value of any AI system that can reason over it effectively is not conversational fluency but the ability to predict, classify, impute, rank, and optimise business decisions from the data that already exists inside customer systems. Prior Labs' open-source TabPFN project has reportedly crossed 3 million downloads, which is a meaningful signal because it indicates real developer and researcher interest in tabular foundation models as a category rather than a one-off academic result. TabPFN-2.6 ranking highly on TabArena gives the project another credibility marker, because TabArena is one of the evaluation environments people in this space use to compare tabular models in ways that go beyond cherry-picked demonstration datasets. Taken together, those signals suggest that the research direction has already achieved a degree of technical validation before SAP decided to industrialise it.
SAP's strategic logic is unusually clear here. The company does not need a general-purpose frontier model to compete in enterprise AI because it already owns the software layer where enterprise data is generated and managed. Its competitive advantage is not a better chatbot. It is the fact that customers use SAP systems to run procurement, finance, inventory, manufacturing, and compliance workflows that are already structured around business objects and tables. A tabular foundation model that can improve forecasting, anomaly detection, risk scoring, and automated decision support inside those systems has a direct path to revenue because it can be embedded into the software customers already pay for. That is a more defensible business than a standalone AI product sold to enterprise teams through a generic API, because SAP can bundle the model into existing contracts, surface it inside workflows users already live in, and monetise it through the same enterprise sales motion that already reaches CFOs, operations leaders, and procurement executives.
The four-year, more-than-€1 billion commitment is the part of the deal that should change how people think about Europe's AI future. It is an acknowledgement that frontier capability in enterprise AI may emerge less from VC-backed startups trying to match OpenAI or Anthropic in a race for general intelligence and more from corporate-backed labs with access to real-world data, customer relationships, and a clear product path. Europe has long worried that it lacks the capital depth and consumer platform distribution that defines US AI winners. SAP's move suggests a different route to leadership: use incumbent enterprise distribution to fund frontier research in categories where the company already owns the workflow and the data moat. The startup ecosystem will interpret that differently depending on its position. Founders building standalone tabular AI startups may see a powerful incumbent absorbing the best research team in the category. Enterprise buyers may see a more credible path to deployed tabular AI than any venture-backed startup could offer. Investors may see proof that Europe can still produce frontier outcomes, but through corporate-backed labs rather than the familiar standalone startup model.
The consequences for SF readers are broad because the deal reframes what enterprise AI competition looks like. Chatbots have dominated the imagination around AI for two years, but the money in enterprise software is still in structured business data and the workflows around it. If tabular foundation models become the next serious battleground, the winners will be companies that can integrate model performance with the systems of record that enterprises already trust. SAP is one of the few companies in the world with that position. Its acquisition of Prior Labs is therefore not just a research story. It is a distribution story, a data story, and a reminder that the most commercially valuable AI breakthroughs in enterprise software may come from incumbents willing to fund long-term research on top of proprietary workflow platforms, not from startups trying to generalise across every use case from day one.
Also read: Altara Comes Out of Stealth With $7 Million From Greylock and Jeff Dean to Build AI Agents for the Scientific Data That General-Purpose AI Cannot Handle • Anthropic Has Reportedly Agreed to Spend $200 Billion With Google on Chips and Cloud and the Number Redefines What It Costs to Compete at the AI Frontier • AMD Beat Q1 Expectations and the 12% Share Jump Reflects Something More Important Than One Good Quarter for Nvidia's Most Credible Rival
