The deep learning revolution has largely overlooked a critical component of the modern economy: the spreadsheet. While Large Language Models (LLMs) excel at processing text and image generators create stunning visuals, the structured, relational data underpinning global commerce – the rows and columns of ERP systems, CRMs, and financial ledgers – has traditionally been treated as a simple file format. This has forced businesses to rely on manual data science processes and older machine learning algorithms for forecasting.
Today, February 9, 2026, San Francisco-based AI firm Fundamental is launching with $255 million in total funding to address this gap. The company is emerging from stealth with NEXUS, a Large Tabular Model (LTM) designed to analyze business data not as a sequence of words, but as a complex network of interconnected relationships.
“The most valuable data in the world lives in tables and until now there has been no good foundation model built specifically to understand it,” said Jeremy Fraenkel, CEO and Co-founder of Fundamental. He emphasized that while the AI world focuses on text, audio, and video, tables represent the largest data modality for most enterprises. “LLMs really cannot handle this type of data very well,” he explained, “and enterprises currently rely on very old-school machine learning algorithms in order to make predictions.”
Most current AI models operate on sequential logic – predicting the next word or pixel. However, enterprise data is inherently non-sequential. A customer’s likelihood to churn isn’t a simple timeline, but a complex interplay of transaction history, support interactions, and regional economic factors. Existing LLMs struggle with this due to the size and dimensionality of enterprise-scale tables.
NEXUS was trained on billions of real-world tabular datasets using Amazon SageMaker HyperPod. Unlike traditional models like XGBoost or Random Forest, which require data scientists to manually define relevant variables, NEXUS is designed to ingest raw tables directly, identifying patterns and relationships that human analysts might miss.
A key challenge in applying LLMs to tabular data lies in how numbers are processed. Fraenkel explains that LLMs tokenize numbers like words, breaking them into individual components. “The problem is they apply the same thing to numbers. Tables are, by and large, all numerical,” he noted. “If you have a number like 2.3, the ‘2’, the ‘.’, and the ‘3’ are seen as three different tokens. That essentially means you lose the understanding of the distribution of numbers. It’s not like a calculator. you don’t always get the right answer because the model doesn’t understand the concept of numbers natively.”
tabular data is order-invariant, unlike language. The position of columns shouldn’t affect the prediction. For example, predicting diabetes risk from a patient table shouldn’t change if height and weight columns are swapped. LLMs, sensitive to word order, are not naturally suited to this characteristic, while NEXUS is designed to overcome it.
Fundamental differentiates its approach from recent integrations like Anthropic’s Claude in Microsoft Excel, which operate at the “formula layer” – essentially treating formulas as text-based code. “We aren’t trying to allow you to build a financial model in Excel. We are helping you make a forecast,” Fraenkel stated. NEXUS is designed for rapid, automated decisions, such as fraud detection during a credit card transaction, rather than manual analysis.
The core value proposition of Fundamental is a significant reduction in time-to-insight. Traditionally, building a predictive model could take months of manual data preparation. “You have to hire an army of data scientists to build all of those data pipelines to process and clean the data,” Fraenkel explained. “If We find missing values or inconsistent data, your model won’t work. You have to build those pipelines for every single use case.” Fundamental claims NEXUS eliminates this manual process with a single line of code. Its pre-training on a billion tables reduces the need for task-specific training and feature engineering.
As Fundamental transitions from stealth, it’s leveraging a commercial structure designed to simplify enterprise adoption. The company has secured several seven-figure contracts with Fortune 100 organizations through a strategic partnership with Amazon Web Services (AWS). NEXUS is offered on the AWS Marketplace, allowing enterprises to use existing AWS credits for predictive intelligence, treating it as a standard utility.
For developers, implementation is designed to be high-impact but low-friction. NEXUS operates via a Python interface at the predictive layer, requiring developers to connect raw tables and specify target columns – such as credit default probability – to trigger forecasts. The model then delivers predictions directly into the enterprise data stack, functioning as a silent, high-speed engine for automated decision-making.
Beyond commercial benefits, Fundamental emphasizes the societal impact of predictive intelligence. The company highlights applications in preventing catastrophic outcomes by identifying hidden signals in structured data. Analyzing sensor data can predict equipment failures like pipe corrosion, potentially preventing incidents like the Flint water crisis, which cost over $1 billion in repairs. During the COVID-19 pandemic, PPE shortages cost hospitals $323 billion. Fundamental argues NEXUS could predict such shortages 4-6 weeks in advance, enabling timely manufacturing. The model is also being used to predict flood and drought risks, and to identify patients at risk of hospital readmission.
Fraenkel acknowledges that the optimal balance between performance and latency varies by industry. For some applications, accuracy is paramount, while others prioritize speed. “Increasing the prediction accuracy by half a percent is worth billions of dollars for a bank,” he said.
The $225 million Series A, led by Oak HC/FT with participation from Salesforce Ventures, Valor Equity Partners, and Battery Ventures, reflects strong confidence in the potential of tabular data. Annie Lamont, Co-Founder and Managing Partner at Oak HC/FT, stated: “The significance of Fundamental’s model is hard to overstate—structured, relational data has yet to see the benefits of the deep learning revolution.”
Fundamental’s partnership with AWS includes a unique engineering achievement: the ability to deploy fully encrypted models – both architecture and weights – directly within the customer’s environment, addressing data privacy concerns. “Customers can be confident the data sits with them,” Fraenkel said. “We are the first, and currently only, company to have built such a solution.”
Fundamental aims to redefine the operating system for business decisions. If NEXUS delivers on its promise – handling financial fraud, energy prices, and supply chain disruptions with a single model – it could mark a turning point where AI finally unlocks the power of the spreadsheets that drive the global economy. The ability to predict is no longer about analyzing the past; it’s about deciphering the hidden language of tables to anticipate the future.
