What Kumo AI's Platform Does

Kumo's platform lets companies generate predictions directly from their existing relational databases—Snowflake, Databricks, SAP, and others—without manual feature engineering or fixed schemas . Its flagship product is KumoRFM, and the newer KumoRFM-2 launched in April 2026. These pre-trained relational foundation models can answer predictive questions almost instantly: customer lifetime value, churn likelihood, next-best-action, fraud detection, and credit risk scoring, among others .

A Kumo model points to a company's data, the user defines the prediction target, and the system produces results without additional training . It serves more than 15 industries including financial services, retail, media, and healthcare .

How Kumo's Graph Neural Network Technology Works

At the core of the platform is a proprietary Hybrid Graph Neural Network (Hybrid GNN) architecture . It treats entities such as customers, products, and transactions as nodes, and their connections—primary and foreign key relationships—as edges in a graph .

Key mechanics:

• Message passing: Each node aggregates information from its neighbors across the graph. Through multiple layers, nodes learn from the entire connected structure, not just their own attributes .
• No flattening required: Traditional machine learning flattens relational data into rows, destroying relationship context. GNNs operate directly on the native graph structure, preserving connections and temporal patterns .
• GPU acceleration: The models are designed to run best on Nvidia GPUs, which unlocks the performance needed for enterprise-scale predictions .

Kumo also combines GNNs with large language models (LLMs) for richer predictive modeling .

Deal Value

While Nvidia has not confirmed the price, The Information first reported the deal at “at least $400 million” . Other outlets including GuruFocus, Fortune, and regional financial news sources have repeated that figure . Because Nvidia has not disclosed terms, the number should be treated as a well-sourced estimate, not an official confirmation .

Notable Customers

Kumo's disclosed customers and ecosystem partners include DoorDash, Reddit, Databricks, Snowflake, and SAP . Its models serve more than a billion users globally through customer applications . Kumo has published performance results such as a 5.4x conversion lift with Databricks and a 5.5x lift in ad accuracy at Reddit .

The Founding Team

• Vanja Josifovski — Co-founder and CEO. Previously CTO at Pinterest during its IPO and CTO of Airbnb Homes .
• Jure Leskovec — Co-founder and Chief Scientist. A Stanford University professor who led the development of relational deep learning and GNN technology at Stanford and later served as Chief Scientist at Pinterest .
• Dr. Hema Raghavan — Co-founder and Head of Engineering. Former senior director of engineering and AI lead at LinkedIn, named to Inc.'s 2026 Female Founders 500 list .

All three co-founders hold PhDs and have joined Nvidia as part of the acquisition .

Venture Capital Raised

Kumo raised approximately $37 million across two rounds (some databases cite $36.5 million) :

• Series A (April 2022): $18.5 million led by Sequoia Capital, with participation from A Capital, SV Angel, and Ron Conway .
• Series B (September 2022): $18 million led by Sequoia, joined by A Capital, SV Angel, and several angel investors including Frank Slootman and Kevin Hartz .

According to PitchBook data, Kumo's post-money valuation after the Series B was approximately $250 million. The $400 million-plus acquisition price represents roughly 60% growth from that round .

Strategic Fit: Beyond Hardware Sales

The Kumo acquisition fits a pattern that Nvidia CEO Jensen Huang has been building for years: transforming Nvidia from a chip supplier into a full-stack AI platform company .

Several strategic threads converge in this deal:

• Selling software, not just silicon: Kumo gives Nvidia a ready-made, GPU-optimized enterprise SaaS product that can generate recurring revenue beyond hardware cycles .
• Locking in the hardware flywheel: Kumo's GNN models run best on Nvidia GPUs. When enterprises adopt Kumo for predictive analytics, they become more entrenched in the Nvidia ecosystem .
• Expanding into predictive AI: Nvidia is already dominant in training large language models. Predictive analytics aimed at business users—churn, fraud, recommendations—represents a massive adjacent market .
• Acquiring world-class talent and IP: Jure Leskovec is one of the pioneers of graph neural networks. Bringing his team and Kumo's Hybrid GNN architecture inside Nvidia strengthens the company's research pipeline and its ability to optimize future chips for relational AI workloads .
• Continuing an M&A pattern: Nvidia has steadily acquired AI software startups—Run:ai, Deci, Illumex, and the $20 billion Groq acqui-hire deal in late 2025—to assemble a platform that spans infrastructure, models, and applications .

The deal illustrates a familiar big-tech playbook: buy a smaller company with deep technical expertise and existing enterprise relationships, then scale its product on your own infrastructure. For Nvidia, Kumo isn't just about predictive AI—it's about ensuring that the next generation of enterprise AI workloads runs on Nvidia, end to end.

What Kumo AI's Platform Does

Kumo's platform lets companies generate predictions directly from their existing relational databases—Snowflake, Databricks, SAP, and others—without manual feature engineering or fixed schemas . Its flagship product is KumoRFM, and the newer KumoRFM-2 launched in April 2026. These pre-trained relational foundation models can answer predictive questions almost instantly: customer lifetime value, churn likelihood, next-best-action, fraud detection, and credit risk scoring, among others .

A Kumo model points to a company's data, the user defines the prediction target, and the system produces results without additional training . It serves more than 15 industries including financial services, retail, media, and healthcare .

How Kumo's Graph Neural Network Technology Works

At the core of the platform is a proprietary Hybrid Graph Neural Network (Hybrid GNN) architecture . It treats entities such as customers, products, and transactions as nodes, and their connections—primary and foreign key relationships—as edges in a graph .

Key mechanics:

• Message passing: Each node aggregates information from its neighbors across the graph. Through multiple layers, nodes learn from the entire connected structure, not just their own attributes .
• No flattening required: Traditional machine learning flattens relational data into rows, destroying relationship context. GNNs operate directly on the native graph structure, preserving connections and temporal patterns .
• GPU acceleration: The models are designed to run best on Nvidia GPUs, which unlocks the performance needed for enterprise-scale predictions .

Kumo also combines GNNs with large language models (LLMs) for richer predictive modeling .

Deal Value

While Nvidia has not confirmed the price, The Information first reported the deal at “at least $400 million” . Other outlets including GuruFocus, Fortune, and regional financial news sources have repeated that figure . Because Nvidia has not disclosed terms, the number should be treated as a well-sourced estimate, not an official confirmation .

Notable Customers

Kumo's disclosed customers and ecosystem partners include DoorDash, Reddit, Databricks, Snowflake, and SAP . Its models serve more than a billion users globally through customer applications . Kumo has published performance results such as a 5.4x conversion lift with Databricks and a 5.5x lift in ad accuracy at Reddit .

The Founding Team

• Vanja Josifovski — Co-founder and CEO. Previously CTO at Pinterest during its IPO and CTO of Airbnb Homes .
• Jure Leskovec — Co-founder and Chief Scientist. A Stanford University professor who led the development of relational deep learning and GNN technology at Stanford and later served as Chief Scientist at Pinterest .
• Dr. Hema Raghavan — Co-founder and Head of Engineering. Former senior director of engineering and AI lead at LinkedIn, named to Inc.'s 2026 Female Founders 500 list .

All three co-founders hold PhDs and have joined Nvidia as part of the acquisition .

Venture Capital Raised

Kumo raised approximately $37 million across two rounds (some databases cite $36.5 million) :

• Series A (April 2022): $18.5 million led by Sequoia Capital, with participation from A Capital, SV Angel, and Ron Conway .
• Series B (September 2022): $18 million led by Sequoia, joined by A Capital, SV Angel, and several angel investors including Frank Slootman and Kevin Hartz .

According to PitchBook data, Kumo's post-money valuation after the Series B was approximately $250 million. The $400 million-plus acquisition price represents roughly 60% growth from that round .

Strategic Fit: Beyond Hardware Sales

The Kumo acquisition fits a pattern that Nvidia CEO Jensen Huang has been building for years: transforming Nvidia from a chip supplier into a full-stack AI platform company .

Several strategic threads converge in this deal:

• Selling software, not just silicon: Kumo gives Nvidia a ready-made, GPU-optimized enterprise SaaS product that can generate recurring revenue beyond hardware cycles .
• Locking in the hardware flywheel: Kumo's GNN models run best on Nvidia GPUs. When enterprises adopt Kumo for predictive analytics, they become more entrenched in the Nvidia ecosystem .
• Expanding into predictive AI: Nvidia is already dominant in training large language models. Predictive analytics aimed at business users—churn, fraud, recommendations—represents a massive adjacent market .
• Acquiring world-class talent and IP: Jure Leskovec is one of the pioneers of graph neural networks. Bringing his team and Kumo's Hybrid GNN architecture inside Nvidia strengthens the company's research pipeline and its ability to optimize future chips for relational AI workloads .
• Continuing an M&A pattern: Nvidia has steadily acquired AI software startups—Run:ai, Deci, Illumex, and the $20 billion Groq acqui-hire deal in late 2025—to assemble a platform that spans infrastructure, models, and applications .

The deal illustrates a familiar big-tech playbook: buy a smaller company with deep technical expertise and existing enterprise relationships, then scale its product on your own infrastructure. For Nvidia, Kumo isn't just about predictive AI—it's about ensuring that the next generation of enterprise AI workloads runs on Nvidia, end to end.