What are tabular foundation models? Tabular foundation models are a new class of artificial intelligence built specifically for structured data, the rows and columns found in spreadsheets and databases, rather than the free-flowing text that large language models handle. The category jumped from research to strategy on May 4, 2026, when SAP announced a definitive agreement to acquire Prior Labs, the pioneer behind the TabPFN model series, and committed more than 1 billion euros over four years to build a frontier AI lab for structured data. For commercial real estate investors, this matters because almost every number that drives a deal lives in a table. To see how AI fits the broader workflow, start with our complete guide to AI multifamily underwriting.
Key Takeaways
- Tabular foundation models are AI built for structured data such as rent rolls, trailing twelve month statements, and comparable sales, the core inputs of CRE underwriting.
- SAP agreed to acquire Prior Labs on May 4, 2026, investing more than 1 billion euros to scale tabular AI across SAP AI Core and SAP Business Data Cloud.
- Large language models like ChatGPT, Claude, and Gemini reason well over text but make weaker predictions on raw numeric tables without extra tooling.
- For CRE, tabular models point toward faster, more consistent forecasts of occupancy, expense growth, and default risk straight from operating data.
- Tabular AI complements rather than replaces language models, and CRE teams will likely run both inside the same underwriting pipeline.
Tabular Foundation Models Real Estate Investors Should Understand
The reason tabular foundation models matter to real estate is simple. A rent roll is a table. A trailing twelve month operating statement is a table. A rent comparable set, a debt schedule, and a sources and uses summary are all tables. SAP CTO Philipp Herzig framed the Prior Labs deal around exactly this point, arguing that the greatest untapped opportunity in enterprise AI was never large language models but AI built for the structured data that runs the world's businesses. Prior Labs, founded in 2024 in Freiburg, Germany by Frank Hutter, Noah Hollmann, and Sauraj Gambhir, published its TabPFN approach in the journal Nature and set state of the art results on tabular benchmarks across hundreds of academic studies. Former Meta chief AI scientist Yann LeCun has served as an advisor.
Traditional large language models struggle here. As powerful as ChatGPT, Claude, and Gemini are with documents and prose, they have only a rudimentary native understanding of tables, numbers, and statistics, which is why most CRE workflows today route spreadsheet math through code interpreters or plug-ins rather than the raw model. A tabular foundation model is trained to read a table as a table, learning the statistical relationships between columns so it can predict a missing value or a future outcome directly. That is a different and arguably better fit for the numeric heart of underwriting. For a practical comparison of where spreadsheets still win, see our breakdown of AI vs Excel for CRE underwriting.
How Tabular AI Changes the Underwriting Stack
Underwriting a commercial asset is a sequence of predictions dressed up as a spreadsheet. What will occupancy be in year two? How fast will operating expenses grow? Which tenants are most likely to default or vacate? Tabular foundation models are designed to answer questions like these from historical structured data, the same way Prior Labs technology is positioned to predict payment delays, supplier risk, and customer churn for SAP's enterprise clients. Translate those use cases to real estate and you get default probability on a rent roll, expense ratio drift on a T12, and lease renewal likelihood by unit.
Consider net operating income, which is gross revenue minus operating expenses and excludes debt service, capital expenditures, and income taxes. A tabular model trained on hundreds of comparable operating statements can flag when a projected NOI looks optimistic relative to peers, before that number flows into your cap rate, where cap rate equals NOI divided by purchase price, or into your debt service coverage ratio, where DSCR equals NOI divided by annual debt service and is expressed as a ratio such as 1.25x. The promise is fewer manual errors and more consistent assumptions across a portfolio. To put structure around those outputs, pair the model with a scoring approach like the one in our guide to AI deal scoring frameworks for CRE investors.
Key Benefits of Tabular Foundation Models for CRE
- Native numeric reasoning: The model reads rent rolls and T12s as structured tables, reducing the brittle workarounds language models need to do spreadsheet math.
- Faster forecasting: Occupancy, expense growth, and renewal predictions can be generated directly from operating history instead of hand-built assumptions.
- Portfolio consistency: One model applied across assets standardizes how risk is scored, which helps when comparing deals or reporting to a fund.
- Risk surfacing: Tabular AI can flag default likelihood, expense anomalies, and outlier line items that a human reviewer skimming a data room might miss.
- Data you already own: The fuel is your existing accounting and property management exports, not a new data source you have to buy.
Real-World CRE Applications
The near-term path runs through the platforms CRE already uses. SAP plans to integrate Prior Labs technology across SAP AI Core and SAP Business Data Cloud, and a large share of institutional real estate runs accounting and ERP data through SAP and similar systems. As tabular models reach those layers, expect property management and asset management software vendors to follow, much as the wave of AI-native underwriting startups already has. The recently funded platform we covered, where Fifth Dimension raised 26 million dollars for CRE AI underwriting, is a sign of how quickly capital is moving toward decision intelligence on structured deal data.
For a hands-on team, the workflow does not have to wait for a vendor. You can already clean and normalize messy operating data before any model touches it, a step we walk through in our tutorial on how to automate rent roll cleanup with Claude Projects. Clean tables in means trustworthy predictions out, regardless of which model you run. If you are weighing which engine to standardize on, our overview of how to compare the leading AI models for CRE lays out the tradeoffs. The broader market context is favorable: research firms peg the AI in real estate market at roughly 1.3 trillion dollars by 2030 at a 33.9 percent compound annual growth rate, and CBRE and other advisors have documented underwriting teams using AI completing preliminary analysis materially faster than peers. If you want help turning these tools into a working underwriting pipeline, The AI Consulting Network specializes in exactly this kind of implementation for CRE firms.
A note of realism belongs here. SAP's own framing is that tabular models excel at prediction over structured data, not at reading leases or narrating an investment memo. Those text-heavy tasks remain the domain of language models. The practical 2026 architecture is therefore hybrid: a tabular model for the numbers, a language model for the documents, and a human underwriter making the final call. Only about 5 percent of firms report achieving most of their AI program goals, so the edge will go to teams that operationalize carefully rather than chase headlines. For a structured rollout, CRE investors looking for hands-on AI implementation support can reach out to Avi Hacker, J.D. at The AI Consulting Network. You can read SAP's official deal announcement on the SAP News site.
Frequently Asked Questions
Q: What is a tabular foundation model in plain terms?
A: It is an AI model trained to understand structured data, meaning the rows and columns of spreadsheets and databases, so it can predict missing or future values directly. Prior Labs' TabPFN is the best known example, and SAP is investing more than 1 billion euros to scale the technology after its May 2026 acquisition agreement.
Q: How is this different from using ChatGPT or Claude for underwriting?
A: Large language models are optimized for text and reasoning, and they handle spreadsheet math through code tools rather than native numeric understanding. Tabular foundation models read a table as a table, which can make them more accurate and consistent at predicting outcomes like occupancy or default risk from operating data.
Q: Can CRE investors use tabular foundation models today?
A: Direct access is mostly through research tools and the open-source TabPFN ecosystem for now, but the capability is heading into enterprise platforms like SAP AI Core and SAP Business Data Cloud. In the meantime, you can prepare by cleaning and standardizing your rent rolls and T12s so any model produces reliable results.
Q: Will tabular AI replace human underwriters?
A: No. The realistic 2026 model is a hybrid in which tabular AI handles numeric prediction, language models handle documents and narrative, and an experienced underwriter validates assumptions and makes the final decision. With only about 5 percent of firms achieving most of their AI goals, disciplined human oversight remains the differentiator.