NVIDIA Buys Kumo AI, Bringing Structured-Data “Prediction Foundation Models” Into Its Enterprise Stack

NVIDIA has acquired Kumo AI, a small startup known for “foundation models” designed to make business predictions from structured data such as customer or transaction records, according to a report published June 3 by Fortune. Kumo, founded about four years ago, had been building model-based forecasting tools that can answer prediction questions without the back-and-forth typically required in traditional analytics workflows, the report said.

The deal appears to have been completed quietly. Fortune reported that Kumo’s three co-founders, Vanja Josifovski, Hema Raghavan and Jure Leskovec, transitioned to NVIDIA last month, with their LinkedIn profiles updated to reflect NVIDIA employment. However, Kumo’s website still made no mention of the change at the time of the report, and NVIDIA declined to comment on the transaction.

Financial terms were not disclosed. Fortune reported that it was unable to learn the deal value, and said NVIDIA has completed numerous AI acquisitions in recent years. Kumo’s early backers included Sequoia Capital, and the startup raised $37 million across two venture rounds in 2022, Fortune reported. Public reporting also linked Kumo’s models to use by companies including DoorDash, Reddit and U.K. grocery chain Sainsbury’s.

Kumo’s technology, as described in earlier coverage quoted by Fortune, was built around faster predictive modeling for business use cases. In that explanation, Kumo’s approach was framed as a way to define a prediction target using a company’s data, such as what constitutes “churn,” and then generate the forecast shortly afterward, without additional training. The model was also described as capable of handling a range of enterprise outcomes, from churn to credit default risk.

NVIDIA’s interest fits into its broader “full-stack” enterprise strategy, where it markets not just chips but also software, model tooling and deployment pathways. The company’s NVIDIA AI Foundry is presented as a platform and service to help organizations build custom generative AI models using enterprise data and domain knowledge, including access to NVIDIA-created models, NVIDIA NeMo customization software, and dedicated compute on NVIDIA DGX Cloud. It also describes an output called NVIDIA NIM, a containerized inference microservice with a standard API for deploying custom models.

How Kumo’s specific models will land inside that stack remains unclear. A separate report via PYMNTS, citing The Information, characterized Kumo’s technology as a proprietary system aimed at structured business data, and suggested NVIDIA could incorporate the models into its AI Foundry software or use Kumo researchers to develop business-centric NVIDIA foundation models. That matters because NVIDIA has also invested heavily in open-weight model ecosystems, so adopting proprietary components could represent a more targeted expansion into enterprise prediction workloads.

What’s still missing is the level of disclosure investors typically look for after a major acquisition. NVIDIA did not provide integration timelines, whether Kumo’s offering will remain a standalone product, or how customers using Kumo models might be supported going forward. It is also not clear whether NVIDIA will emphasize Kumo’s foundation-style prediction methods as part of AI Foundry, or fold the capability into internal tooling and model development.

For now, the clearest announcement is strategic. NVIDIA is moving to strengthen its enterprise software narrative by adding talent and technology geared toward accurate prediction on structured organizational data, an area that often resists “prompt-based” approaches used in general chatbots. The next checkpoints to watch are any NVIDIA announcements about AI Foundry updates, deployment support for structured-data prediction, and further changes to Kumo’s public footprint as the founders settle into their new roles at NVIDIA.