According to informed sources, NVIDIA has completed its acquisition of the five-year-old startup Kumo AI, a deal for the enterprise-focused predictive AI software company valued at over $400 million.
An NVIDIA executive first disclosed the acquisition in a LinkedIn post on Tuesday. This move will expand NVIDIA's portfolio of AI models optimized for its own hardware, offering customized deployment services to businesses.
While NVIDIA's prior focus has been on open-source foundational models like Nemotron, Kumo's entire model suite is proprietary and developed in-house. Kumo's products are specifically optimized for structured enterprise business data, such as customer profiles and transaction records—categories that general-purpose large language models often struggle to process effectively.
Neither Kumo's CEO nor NVIDIA has officially commented on the acquisition at this time.
NVIDIA has not yet detailed the specific integration path for Kumo's models. One potential route is embedding them into the NVIDIA AI Foundry platform, a tool designed to help enterprises rapidly build custom large models by combining their proprietary business data (e.g., genomic data) and domain expertise with NVIDIA's proprietary software, open-source base models, and AI-generated synthetic data. Alternatively, NVIDIA could integrate the Kumo R&D team to aid in developing new, commercially-focused foundational models from scratch.
Enterprise data typically consists of multiple interrelated data tables. Training AI models to predict business outcomes, such as customer churn probability, often requires vast amounts of proprietary business data. Kumo's solution addresses the challenge of limited enterprise data sources by leveraging graph machine learning algorithms developed by its Stanford-founded team, supplemented with synthetic data generated to simulate enterprise environments for training.
Kumo launched its latest model, KumoRFM-2, in April of this year. Its clients and partners include DoorDash, Reddit, Databricks, and Snowflake. According to PitchBook data, the company had raised a total of $37 million prior to the acquisition, with investors including Sequoia Capital and SV Angel. Its post-money valuation was $250 million during a 2022 funding round.
Based on LinkedIn information, the founding team—former Pinterest CTO Vanya Yosifovski, Stanford University professor Jure Leskovec, and former LinkedIn AI head Hema Raghavan—have been employed at NVIDIA since May of this year.
This acquisition aligns with NVIDIA's established strategy of pursuing small to mid-sized mergers. Late last year, NVIDIA finalized a technology licensing agreement worth $20 billion to acquire related technologies from inference chip company Groq.
Comments