NVIDIA Kumo speeds predictions from structured data

What is changing?

NVIDIA expanded its Kumo open foundation-model family for structured enterprise data on October 21. The company aims to accelerate outcome forecasting, anomaly detection, fraud-risk scoring, and recommendations across relational tables and time series without building a separate machine-learning pipeline for every use case.

The portfolio includes Kumo Structured models for relational and tabular prediction, time-series forecasting, and anomaly detection. NVIDIA Kumo Engine provides an open library stack for developing and deploying these models. Teams can use pretrained models directly and fine-tune them with proprietary data when higher task performance is required.

NVIDIA lists Cognite, Coinbase, Databricks, DoorDash, Revolut, and Tubi among the early users and says Snowflake customers are also trying Kumo across industries. These are provider claims; model quality, data privacy, and total cost still need validation on each organization’s own data.