SAP to run its tabular AI model SAP-RPT aboard the International Space Station

What SAP-RPT will do in orbit

SAP says onboard sensors and mission system telemetry on the ISS produce too much time-sensitive data to be routed back to Earth for analysis, and that latency has to be reduced to ensure the safety of the spacecraft. SAP-RPT is meant to address this by running directly on HPE Spaceborne Computer-2.

According to SAP, the model makes predictions directly from structured data, even with some data missing, and does so without the usual model training or fine-tuning. Onboard the ISS, it is intended to support several tasks:

• anomaly triage in real-time telemetry data
• predicting issues before failures escalate
• reconstructing missing sensor values based on historical data
• generating explainable mission-health reports for astronauts and ground teams

SAP lists the outcomes to be delivered onboard as real-time predictions for computer and mission health status, an overall mission risk score, an assessment of whether manual intervention is needed, human-readable explanations of AI-driven findings, and ad hoc reports for crew and ground teams.

Tabular foundation models versus LLMs

SAP-RPT belongs to SAP's portfolio of tabular foundation models, or TFMs, which also includes the TabPFN model by Prior Labs. SAP AI Core orchestrates, manages and governs the TFMs, predictions and workflows. SAP argues that large language models excel at language and reasoning, while TFMs are built for the structured tabular data that underpin real-world operations, such as the sensor-driven data in this project.

Energy use is a central claim. SAP states that SAP-RPT is about 50,000 times more energy-efficient than state-of-the-art LLMs, which it calls especially important in space, where every watt is constrained. The company also says the model is highly adaptable and requires no pre-training as conditions evolve.

Mayank Shrivastava, vice president of Generative AI and Foundation Models at SAP SE, said: "Space is the ultimate stress test for AI." Scott Wiest, Vice President and Global Sales Chief Technologist and HPE Fellow, said the deployment "shows what's possible with real-time detection and prediction using AI in the ultimate edge environment."

From orbit to the enterprise

SAP says the deployment on the ISS, planned by early 2027, will show how the model strengthens business functions on Earth. It argues that the constraints of space, including extreme energy limits, intermittent connectivity, radiation interference and zero tolerance for failure, are relevant to demanding use cases on the ground. Examples named by SAP are disaster relief operations, agriculture in remote areas, distributed manufacturing and spread-out supply chains.

The company draws a direct comparison with business data: a supplier whose deliveries are quietly slipping, a payment pattern that is drifting, or a compressor that is running hot. Shrivastava said the same model will predict supplier risk, cash flow anomalies and equipment failures across an enterprise.

SAP adds that insights from the ISS will accelerate its road map for enterprise edge AI and for use cases that rely on structured data in important analytical and operational processes. Looking ahead, it says the mission lays the groundwork to support space-based data centers, deep space exploration and other collaborations on space missions.

Customers can try SAP-RPT with their own data in the Tabular AI area of sap.com, according to SAP.