Nvidia Buys Enterprise Model-Maker Kumo AI for at Least $400 Million
The Takeaway
- Nvidia acquires enterprise predictive AI startup Kumo AI for over $400 million.
- Kumo AI specializes in predictive AI for enterprise structured data.
- Acquisition expands Nvidia’s AI models optimized for its hardware.
Nvidia has bought Kumo AI, a five-year-old startup that sells predictive AI software to enterprises, for more than $400 million, said a person with knowledge of the deal.
The acquisition, first revealed by an Nvidia executive in a LinkedIn post on Tuesday, should expand Nvidia’s roster of AI models that can be optimized for Nvidia hardware and offered to enterprises for further customization.
Nvidia has invested heavily in its open-weight models, such as its Nemotron models, but Kumo’s models are proprietary. They’re tailored to answer questions on structured business data, such as customer information and payment data, which are generally harder for large language models to parse.
Kumo’s CEO and Nvidia declined to comment.
It’s unclear exactly how Nvidia will use Kumo’s models. It could use them in its AI Foundry software, which aims to help companies build custom AI models by combining their own data and domain-specific knowledge, such as genomic data, with Nvidia software, existing open-source models and synthetic, or AI-generated, data. Nvidia could also use Kumo’s researchers to help develop new, business-focused Nvidia foundation models.
Corporate data generally consists of structured layers of tables connected to other tables, and a company generally needs troves of proprietary data to train models on answering questions, such as on the likelihood that a customer will cancel. Kumo attempted to solve this problem by supplementing its training data with synthetic data from simulated enterprise environments and pairing that with graph machine learning techniques the founders developed at Stanford, the company has said.
Kumo released its latest model, KumoRFM-2, in April and says it counts DoorDash, Reddit, Databricks and Snowflake among its customers and partners. It raised $37 million from Sequoia Capital, Ron Conway’s SV Angel and others, including at a $250 million valuation in 2022, according to PitchBook.
Kumo founders Vanja Josifovski, former CTO of Pinterest; Jure Leskovec, a Stanford University professor, and Hema Raghavan, previously an AI lead at LinkedIn, have been working at Nvidia since May, according to their LinkedIn.
The Kumo deal is in line with Nvidia’s track record of smaller acquisitions. The company spent a total of around $3 billion on acquisitions over the last five years, according to its securities filings. Also, late last year it agreed to pay $20 billion to license technology from inference chip designer Groq,