Fundamental's Large Tabular Model NEXUS is now available on Amazon SageMaker JumpStart

Today, we are announcing support for the NEXUS Fundamental model in Amazon SageMaker AI. With this introduction, you can use a basic model (FM) built for the purpose of predicting tabular data. This model helps your business generate accurate, decisive predictions from data organized in days instead of months.

In this post, we show you how to get started with NEXUS in Amazon SageMaker JumpStart, walk through the deployment process, and show how you can use forecasts against your business dataset.

What is NEXUS?

NEXUS is a basic model developed by Fundamental and designed to predict tabular data. Large-scale linguistic models (LLMs) are designed for text, and traditional machine learning (ML) methods require extensive engineering features and model training. NEXUS takes a different approach. It's pre-trained on billions of real-world predictive tasks on every programmed data set, so it already knows how to find a signal in your data.

As a Big Table model, NEXUS is designed for structured data analysis and offers these important innovations:

• Deterministic Architecture – Probabilistic LLMs may give different answers to the same questions. NEXUS produces consistent, repeatable results for each forecast.
• Native table understanding – Trained on billions of tables, NEXUS naturally processes numbers, categories, dates, and unstructured text without manual feature engineering.
• Non-sequential thinking – Many AI models predict sequential data (for example, the next word or the next pixel). NEXUS analyzes multidimensional relationships in business tables. For example, when predicting customer churn, NEXUS understands how multiple factors (transaction frequency, support tickets, and economic indicators) affect the likelihood of a crash.

Why existing methods fail

The most important business data resides in spreadsheets such as spreadsheets, enterprise resource planning (ERP) systems, customer relationship management (CRM) systems and relationship databases. Many critical business decisions depend on predictions made against this data. However, today's tools have important limitations:

• Traditional ML it takes teams of data scientists 3–6 months to build, train, and deploy a model in a single use case. You face a constant trade-off between the quality and quantity of predictions.
• LLMs they are inconclusive, producing different responses to the same dataset. They lose the context of the numbers during tokenization, which leads to incorrect results in structured data and requires complex monitoring to minimize these problems.

NEXUS is designed for tabular data and offers the following advantages:

• Variation of permissions – It recognizes that changing the order of a column does not change the meaning, which is different from the way transformers handle data.
• The ability of billions of lines – Processes large datasets without truncation or sampling.
• Cross-schema reasoning – Connects related data across different tables automatically.
• Automatic data cleaning – Resolves incomplete entries (for example, NEXUS can still guess even if entries are lost).

How NEXUS works in Amazon SageMaker AI

The following figure shows the end-to-end flow of deploying and running forecasts with NEXUS in SageMaker AI.

NEXUS uses a dedicated, single-tenant, network-separated GPU instance in a SageMaker AI managed environment. The workflow consists of the following steps:

1. Register and use – Subscribe to the NEXUS model package on AWS Marketplace, and use it as a SageMaker AI managed inference endpoint on an ml.p5en.48xlarge for example (8× NVIDIA H200 GPUs).
2. Install the SDK – Install the Fundamental Python SDK and connect it to your SageMaker repository. The SDK provides a compatible API for scikit-learn NEXUSClassifier again NEXUSRegressor they don't rate.
3. Upload data to Amazon S3 – The SDK organizes your table data and uploads it to an Amazon Simple Storage Service (Amazon S3) bucket in your account.
4. Train the model – Call clf.fit(X_train, y_train) to train. NEXUS handles data cleansing and engineering automatically, with no manual pipeline required.
5. Generate predictions – Call clf.predict(X_test) with deterministic predicates or clf.predict_proba(X_test) with probability estimates. The results are stored back in your Amazon S3 bucket.

Your data stays in your AWS environment throughout this process. The storage area is separate from the network and has a single tenant, making NEXUS suitable for business workloads with sensitive data.

Get started with NEXUS on Amazon SageMaker AI

To get started, navigate to Amazon SageMaker JumpStart, search Basic NEXUSand choose from the following:

• Basic model (pre-trained on more than 10B table rows).
• Industry-specific diversity (finance, healthcare, and manufacturing).

Industries for changing business use

Tabular data is central to business decision-making, from financial ledgers to patient records to chain logs. NEXUS is purpose-built for this data and helps you go from raw data to production-grade predictions without extensive feature engineering training or models. The following are a few proxy use cases where NEXUS can create value.

Financial services

• Fraud detection – Analyzes transaction patterns across millions of accounts.
• Modeling credit risk – Processes loan portfolios by removing the default feature.
• Compliance with the rules – Extracts structured data from random control fills.

Health care

• Clinical trial simulation – Identifies eligible patients across electronic health records (EHR).
• Drug availability – Analyzes biological assay data for integrated testing.
• Patient risk stratification – Predicts readmission risk using intensive care unit (ICU) time series data.

Manufacturing and supply chain

• Predictive maintenance – Predicts equipment failure from sensor data.
• Seek predictability – Anticipates inventory needs across global distribution networks.
• Supplier risk analysis – Checks the credibility of the seller using the purchase history.

Retail and ecommerce

• Churn is a forecast – Identifies vulnerable customers using purchase history and browsing behavior.
• Values ​​are strong – Optimizes prices based on competitor data and inventory levels.
• Analysis of cart abandonment – Helps you understand why customers leave items in online carts.

Why choose NEXUS over Amazon SageMaker AI

Modeling is only part of the equation. The infrastructure you use determines how quickly you can move from testing to production. SageMaker AI provides a controlled, secure, and scalable environment for running NEXUS at enterprise scale. Together, NEXUS and AWS reduce the unbundled heavy lifting so your data scientists can focus on business outcomes rather than infrastructure management.

• Accelerated time to value – Pre-built containers and scripts reduce deployment time.
• Cost effectiveness – SageMaker AI's managed infrastructure optimizes performance.
• Scalability – Automatically scales to petabyte scale datasets.
• Compliance is fine – Meets GDPR, HIPAA, and SOC 2 requirements automatically.
• Continuous learning – Native integration with Amazon SageMaker Pipelines for model retraining.
• Multiplex support – Supports multiple fit and predict operations in a single SageMaker AI endpoint, eliminating the need for dedicated resources for each use case.

AWS strategic partnerships

Core has entered into a strategic partnership with AWS to accelerate business adoption:

• Native integration – Install NEXUS directly from AWS Marketplace.
• Protect the infrastructure – Runs on a secure, compliant AWS cloud environment.
• Business support – Dedicated AWS Solutions Architects for implementation guidance.

Next steps

Are you ready to transform your data-driven decisions?

The conclusion

In this post, we've shown how the NEXUS model support in Amazon SageMaker AI helps you unlock new insights into your structured data assets. Whether you're predicting equipment failure, improving the supply chain, or detecting financial fraud, NEXUS provides decisive, uncontrollable capabilities for your business' predictive workloads.

To learn more, see the following resources:

About the writers