SAP has just sent a strong strategic signal to the enterprise software market, almost simultaneously announcing the acquisition of Dremio, a specialist in lakehouse architectures based on Apache Iceberg, and Prior Labs, a young German company specializing in AI models applied to tabular data. Above all, the German publisher is trying to regain control of the future value chain of corporate data, at a time when artificial intelligence is redefining the very architecture of business software.
The amounts of the acquisitions have not been officially communicated. However, according to several sources cited by Pathfinders and TechCrunch, the operation around Prior Labs would represent an exit of more than 425 million euros, mainly in cash, with more than 255 million euros paid immediately to founders Frank Hutter, Noah Hollmann and Sauraj Gambhir. SAP also announced an additional investment of one billion euros over four years to transform Prior Labs into a global laboratory dedicated to AI applied to structured data.
For several years, a growing portion of the value generated around enterprise data has gradually moved outside of traditional ERPs. Large cloud and analytical platforms, such as Databricks, Snowflake or Google Cloud, have captured an essential share of uses linked to analysis, AI and advanced data exploitation. In this model, ERPs risk gradually being relegated to the rank of simple transactional systems powering external platforms that have become much more strategic, and it is precisely this scenario that SAP is today trying to avoid.
Founded in 2017, the American company Dreamio has established itself as one of the major players in open data architectures, capable of unifying data dispersed between different clouds, data lakes and analytical bases. Before its acquisition, the startup had raised around 306 million euros from investors such as Sapphire Ventures, Lightspeed Venture Partners, Cisco Investments and Adams Street Partners. Its last estimated valuation exceeded 850 million euros.
Its technology makes it possible to query, structure and govern large volumes of data via a unified SQL layer, while relying heavily on open source standards, in particular Apache Iceberg. This point is central because Iceberg is becoming for data architectures what Kubernetes has become for cloud infrastructure, namely a standardized layer allowing interoperability between different environments. The format makes it possible to manage massive volumes of analytical data while maintaining a detailed history of modifications, fine governance and better traceability of flows.
The other operation is that of Prior Labs, founded less than two years ago in Fribourg, the startup had raised around 7.9 million euros during a pre-seed announced in February 2025. The round was led by Balderton Capital, with the participation of XTX Ventures, Atlantic Labs and Hector Foundation. Among the business angels were Peter Sarlin, founder of Silo AI, Christopher Lynch of AtScale, Guy Podjarny, founder of Snyk and Tessl, Edward Grefenstette of Google DeepMind, Robin Rombach of Black Forest Labs, Ashutosh Kulkarni of Elastic, Thomas Wolf of Hugging Face, as well as Steve Anavi, co-founder of Qonto.
The company develops AI models specialized in tabular data, i.e. data structured in rows and columns that constitute the heart of ERP systems. Their TabPFN-2.5 model is designed to produce predictions, detect anomalies or automate certain business analyzes directly from massive databases or spreadsheets.
This positioning differs greatly from major language models. Since the explosion of generative AI, the market has primarily focused on LLMs capable of generating text or interacting in a conversational manner. But in companies, the majority of critical data remains deeply structured: accounting, supply chain, purchasing, human resources, treasury, supplier management or inventory control.
SAP seems to believe that the next wave of value in enterprise AI will come less from conversational interfaces and more from the ability to intelligently exploit these gigantic transactional bases. The group had already started this trajectory with RPT-1, its Relational Pre-Trained Transformer launched last year. The acquisition of Prior Labs now allows it to suddenly accelerate this strategy, while recovering a research team particularly recognized in the field of tabular models.
But behind these acquisitions there is also a much broader battle at play: that of AI agents.
Large software publishers now know that the traditional interface of business applications could gradually disappear in favor of agents capable of interacting directly with information systems. In this model, whoever controls the data, permissions, and workflow orchestration layer potentially controls the future interface of the enterprise.
SAP therefore seeks to prevent a third party, whether OpenAI, Anthropic or a new generation of agentic platforms, from capturing this strategic relationship.
The recent tightening of its API policy goes in this direction. The publisher now prohibits certain unapproved AI agents on its systems and favors architectures validated by SAP, particularly around its own Joule agentic layer.
Basically, these acquisitions reflect the same concern among the large historical publishers: to prevent AI from transforming their software into simple invisible infrastructures powering much more powerful external platforms. With Dremio and Prior Labs, SAP seeks to rebuild a central position in the company’s future technology stack, no longer just as an ERP publisher, but as a unified infrastructure for data, AI and tomorrow’s autonomous agents.
