Nvidia snaps up Kumo AI, a predictive AI startup known for its extreme accuracy

Nvidia Corp. has bagged itself another artificial intelligence startup, acquiring four-year-old model maker Kumo AI Inc.

The company designs AI models focused on making extremely accurate business predictions. It’s the latest in a string of startups acquired by the chipmaker in the last couple of years. Nvidia didn’t disclose the terms of the acquisition, but the deal has already been finalized, according to reports by Fortune and The Information.

Kumo AI is known for its extremely accurate machine learning models that are trained and then fine-tuned on customers’ own business data. One of its main advantages is convenience. Rather than having companies spend months building complex data pipelines and feature engineering, they can simply plug Kumo’s models into their data warehouse, such as Snowflake or Databricks, and start predicting outcomes using simple, Structured Query Language-like queries.

Kumo’s platform automatically handles all of the work involved in preparing, cleaning and joining historical business data, eliminating the manual effort required to set up predictive models by as much as 95%. The company also translates user’s natural language queries into its proprietary Predictive Query Language. This means they can ask simple questions such as “Will this customer churn?” or “What demand can we expect to see next quarter?” and instantly generate answers grounded in their own corporate data.

The secret sauce in Kumo’s predictive models is its Graph Neural Networks. Whereas traditional AI models look at each data row in a database in isolation, Kumo’s graph learning will consider them altogether, creating a massive interconnected network of nodes and edges that map the relationships between them. By mapping business data like this, the model can see how each customer is connected to the various products and services a company sells, as well as other users. That’s what enables it to generate more accurate, context-aware business insights.

Kumo co-founder and Chief Executive Vanja Josifovski appeared on SiliconANGLE Media’s mobile livestreaming studio theCUBE during its coverage of the 2024 Data Cloud Summit. He was joined by Brian Rikuda, the executive vice president of strategy, operations and programming scheduling at one of its biggest customers, Black Entertainment Television LLC, who talked about how beneficial the platform is.

“With services like Kumo … [it allows us to] look at customer behavior, it allows us to look at the inventory of content that we have and start making more dynamic recommendations,” Rikuda said. “If you could solve the problem of retention in the weekly or monthly leakage of your subscribers, you actually solve a huge problem for the growth of the overall business.”

It’s not immediately clear how Nvidia intends to put Kumo’s predictive models to work, or what motivated it to buy the startup. However, Josifovski and his co-founders Hema Raghavan and Jure Leskovec have already transitioned to the chipmaker, based on their LinkedIn profiles, which describe them as Nvidia employees. However, Kumo’s website does not mention the acquisition.

Mountain View, California-based Kumo had raised $37 million in venture capital funding via two rounds, which both took place in 2022, with the most prominent investor being Sequoia Capital. Its customers include Reddit Inc., DoorDash Inc. and the U.K. grocery chain J Sainsbury plc.

Nvidia has acquired dozens of startups in the last few years, with many of its deals going under the radar. In one of its biggest recent acquisitions, it bought the technology assets and plucked key personnel from the AI inference company Groq Inc., in what was reported to be a $20 billion “acqui-hire.” More recently, in February, it acquired a data semantics startup called Illumex Ltd. for about $75 million.

Image: Kumo AI