Fundamental NEXUS on AWS SageMaker & Google TabFM: The Rise of Large Tabular Models for Structured Data

Large Tabular Models (LTMs) are rapidly emerging as a new AI frontier for structured data, with significant developments in 2026 from key players like Fundamental, AWS, and Google. Fundamental launched its NEXUS model in February 2026, followed by its integration into Amazon SageMaker in June, while Google introduced its TabFM model later that month, signaling a major shift in how enterprises handle complex, structured datasets. For broader context, explore our AI News.

The Emergence of Large Tabular Models

Large Tabular Models represent a distinct category of AI foundation models engineered to manage structured data, such as spreadsheets and databases. Unlike LLMs, which process data as token sequences, LTMs directly model tabular structures, enabling more deterministic predictions. This architectural difference allows LTMs to jointly learn numerical values, their semantic meaning, and the statistical relationships between entries, addressing a critical gap in AI's ability to handle the vast amounts of tabular data prevalent in industries like finance and healthcare.

Fundamental's NEXUS Model and AWS Integration

AI startup Fundamental officially launched from stealth in February 2026, backed by $275 million in funding. Its flagship product, the NEXUS model, is purpose-built for tabular data analysis. A significant development occurred in June 2026 when Amazon Web Services (AWS) integrated NEXUS directly into Amazon SageMaker. This integration makes NEXUS available as a native, first-party solution within SageMaker, providing enterprise users with direct access to its capabilities for structured data processing. This move underscores the growing recognition of LTMs as essential tools for enterprise AI.

Google's TabFM and Other Industry Players

Google also entered the LTM space with the release of its TabFM model in late June 2026. TabFM was trained on hundreds of millions of synthetic datasets, showcasing a different approach to developing these specialized models. Beyond Fundamental and Google, other major players are also contributing to the LTM landscape. In March 2026, Feedzai and Mastercard both launched their own finance-focused LTMs, highlighting the immediate practical applications of this technology within the financial sector. These developments collectively demonstrate a rapid expansion of LTM capabilities and adoption across various industries.

LTMs vs. LLMs: A Fundamental Difference

The core distinction between LTMs and LLMs lies in their approach to data. While LLMs like GPT and Claude excel at understanding and generating human language by processing sequential tokens, they are not inherently designed for the intricacies of structured, numerical, and categorical data found in tables. LTMs, conversely, are engineered to understand the inherent relationships and patterns within tabular data directly. This allows them to provide more accurate and reliable predictions for tasks such as fraud detection, risk assessment, and demand forecasting, where deterministic outcomes are crucial. Furthermore, LTMs often incorporate confidential computing, ensuring that providers cannot access customer data, which is a critical security feature for sensitive enterprise applications.

Key Takeaways for Enterprise AI

• Large Tabular Models (LTMs) are specifically designed for structured data analysis.
• Fundamental's NEXUS model is now natively integrated into Amazon SageMaker.
• Google introduced its TabFM model, trained on extensive synthetic datasets.
• Feedzai and Mastercard launched finance-specific LTMs in March 2026.
• LTMs offer deterministic predictions and often include confidential computing for data privacy.

Conclusion

The rapid advancements in Large Tabular Models, marked by the emergence of Fundamental's NEXUS, its integration with AWS SageMaker, and Google's TabFM, signify a pivotal moment for enterprise AI. As businesses increasingly rely on structured data for critical operations, LTMs offer a specialized and secure solution that complements the capabilities of existing LLMs. The continued development and adoption of these models are likely to reshape how organizations approach data analysis, forecasting, and decision-making in the coming years.

Sources

• https://aws.amazon.com/blogs/machine-learning/fundamentals-large-tabular-model-nexus-is-now-available-on-amazon-sagemaker-jumpstart/
• https://research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/
• Probing Memorization of Tabular In-Context Learning
• LaTable: Towards Large Tabular Models
• Why Tabular Foundation Models Should Be a Research Priority