Global – October 29, 2026 -- NVIDIA has released Kumo, a family of open models, libraries and tools for predictive AI on structured data, and Snowflake customers can run it on GPU compute without moving data out of their accounts. The release is part of NVIDIA's new Agent Knowledge Foundation, a set of open models and libraries meant to help enterprise agents retrieve evidence, apply consistent business meaning, predict outcomes and optimize actions.
Kumo trains graph neural networks on full relational structure
The NVIDIA Kumo Relational model and NVIDIA Kumo Engine, an open-source package, train graph neural networks across an enterprise's relational graph. Kumo learns from tables, the keys connecting them and the timing of events. Teams do not need to flatten these links into a single wide table.
NVIDIA lists four target use cases: customer churn, fraud detection, demand forecasting and product recommendation. In its example, a team predicts which active customers will buy again within 90 days using four Snowflake tables and a few lines of Python.
Vanja Josifovski, VP of Graph Engineering at NVIDIA, said "Enterprise data is inherently relational," and that the most accurate predictions come from learning on that structure directly.
Training, scoring and data stay inside Snowflake's governance perimeter
Customers install the Kumo package into Snowflake Notebooks on Container Runtime and run it on Snowflake ML. Kumo reads from Snowflake tables, trains on Snowflake GPU compute pools and writes predictions back to Snowflake tables. Teams can orchestrate the workloads with Snowflake ML capabilities such as ML Jobs, under the platform's existing security controls.
NVIDIA's verified agent skills for Snowflake CoCo let teams describe a prediction task in natural language. They receive a runnable notebook covering data connection, GPU training and writing predictions back to Snowflake.
Affirm, NOV, VideoAmp and Yieldmo already use Kumo on Snowflake ML
Lynn Steffen, VP of Finance Business Operations at National Oilwell Varco (NOV), said Kumo has accelerated how quickly the company turns data into action across several business areas. She added that CoCo removes friction by letting teams build and iterate where the data already lives.
Craig Simmons, VP of Product at Yieldmo, said the ad-tech firm applies Kumo on Snowflake ML to its Companion Audiences product, which builds act-alike audiences. Buyers set the similarity threshold and decide whether Companion leads or backs up a campaign. Managed service buyers can train directly from conversion data with no seed spend.