The TFM Gambit: SAP Buys Its Way Out of the LLM Trap
SAP is dropping a cool $1.16 billion on Prior Labs, an 18-month-old German startup that nobody outside of Freiburg had heard of until today. The enterprise dinosaur is betting that Tabular Foundation Models (TFMs) trained on spreadsheets and SQL databases will matter more than any chatbot ever could. This is the clearest signal yet that the AI industry’s obsession with language models is a dead end for actual business software. Prior Labs’ TabPFN models, already downloaded over three million times, do one thing well: they predict outcomes from structured data without all the hallucination and bloat that makes GPT a liability in a CFO’s hands. The founders get a massive cash exit. SAP gets a shortcut to pretending it isn’t behind.
The Walled Garden: NemoClaw Yes, OpenClaw No
Here’s where SAP’s strategy gets ugly. While it buys its way into the future with Prior Labs, it’s simultaneously locking down its entire ecosystem. SAP’s new API policy explicitly prohibits any AI agents it hasn’t personally endorsed. The only agent architecture getting a pass? Nvidia’s NemoClaw, which conveniently integrates with SAP’s own Joule Agents. This is classic incumbent behavior: embrace the open source community when it benefits you, slam the door shut when competition arrives. Salesforce is letting enterprises run OpenClaw through its Headless 360 architecture. SAP is building a moat. The message is clear: you can use our AI, but only the AI we approve. No CVE links are relevant here because the vulnerability isn’t technical. It’s strategic.
The SaaSpocalypse Survival Play
SAP’s stock has taken a beating in 2026 thanks to the SaaSpocalypse, and the company is clearly terrified of being disintermediated by agentic AI. CFO Dominik Asam admitted the game is about speed and economies of scale. But throwing cash at Prior Labs doesn’t solve the trust problem. SAP is simultaneously investing in Anthropic, Aleph Alpha, and Cohere while buying a startup that makes LLMs look irrelevant for enterprise use. That’s not a strategy. That’s panic disguised as diversification. The real question isn’t whether Prior Labs can build better models for spreadsheets. It’s whether any model developer can survive being acquired by a company that fundamentally doesn’t believe in open AI.
Source: Techcrunch
