Large Tabular Models Bring AI Analysis to Spreadsheets

https://spectrum.ieee.org/large-tabular-models-nexus

Publish Date: 2026-07-09 08:00:01

Advancements in Data Analysis with Large Tabular Models

Generative AI models like ChatGPT have demonstrated impressive capabilities in generating human-like text, but they struggle significantly when handling structured data found in spreadsheets. This limitation is primarily due to their reliance on sequential data processing, which is incompatible with the non-sequential nature of tabular data. To address this, AI startup Fundamental has developed a new type of generative AI foundation model called large tabular models (LTM), exemplified by its model NEXUS. Unlike existing machine-learning algorithms, LTMs model the structure of tabular data directly, enabling more accurate reasoning and predictive capabilities. NEXUS has secured partnerships, including with Amazon Web Services, leveraging confidential computing to handle sensitive customer data securely. These advancements indicate a trend towards automated data processing systems that can outperform human data analysts, with LTMs filling a niche that large language models cannot occupy.

Key Points:

Large language models (LLMs) excel with text but are not suited for structured data in spreadsheets.
Fundamental in collaboration with AWS developed NEXUS, an LTM capable of analyzing structured data by modeling its non-sequential structure.
NEXUS uses confidential computing to ensure that customer data remains secure, unlike typical LLM approaches.
Similar technologies are being developed by major tech players and research institutions to enhance the automation of data analysis tasks.
The convergence of LLMs and LTMs is expected to lead to more powerful systems for data analysis, complementing each other as distinct but complementary components like the human brain’s analytical processes.