Nvidia has acquired Kumo AI, a startup that builds predictive AI models for enterprise customers, in a transaction valued at more than $400 million. The purchase was confirmed on Tuesday, and it folds yet another AI company into Nvidia's expanding portfolio of software assets as the chipmaker pushes to grow beyond selling hardware alone.

What Kumo AI Actually Builds

Kumo's core idea is pretty simple once you cut through the technical language. It creates foundation models trained directly on a company's own business data, then uses them to answer practical questions. Things like "Will this customer churn?" or "What demand should we expect next quarter?" And it does this without the months of traditional data science work those forecasts usually demand.

The platform connects straight into data warehouses such as Snowflake and Databricks. From there, it handles data preparation automatically, cutting out as much as 95% of the manual effort that normally goes into building a predictive model. That's the real selling point here: speed and a lot less grunt work.

The Graph Neural Network Advantage

What sets Kumo apart technically is its reliance on Graph Neural Networks. Instead of analyzing rows of data one at a time in isolation, this approach maps relationships across an entire dataset. So the system looks at how customers, products, and transactions all connect to one another. The result is more context-aware predictions, because the model understands the web of relationships rather than just the individual data points sitting alone.

KumoRFM-2 and a Roster of Real Customers

In April, Kumo launched KumoRFM-2, a foundation model the company described as the first to outperform supervised machine learning on enterprise prediction tasks. It's built to scale, handling more than 500 billion rows of data. And the customer list gives the technology some credibility: Reddit, DoorDash, and the U.K. grocery chain J Sainsbury are all using it.

Inside the Deal and What It Signals

Kumo, based in Mountain View, California, had raised $37 million in venture capital across two funding rounds back in 2022, with Sequoia Capital as its most prominent backer. As part of the acquisition, co-founder and CEO Vanja Josifovski has moved over to Nvidia, along with co-founders Hema Raghavan and Jure Leskovec, according to their LinkedIn profiles.

Nvidia didn't publicly disclose the financial terms. But The Information reported the price tag came in above $400 million, citing a person with knowledge of the deal. This isn't a one-off move, either. It follows Nvidia's purchase of data semantics startup Illumex for roughly $75 million in February, as well as its reported $20 billion acqui-hire of the technology assets belonging to AI inference company Groq.

A Strategy Built on Bundling

Look at the bigger picture and the logic becomes clear. The Kumo deal reflects Nvidia's broader strategy of bundling customized AI models with its hardware, giving enterprises a more complete AI stack rather than just chips. It's a way to lean less heavily on chip sales alone, and it comes as the company reported record first-quarter revenue of $81.6 billion.