Nvidia acquires predictive AI startup Kumo for over $400 million

According to Beating, NVIDIA has acquired Kumo AI, a five-year-old predictive AI startup, for over $400 million. The acquisition aims to leverage Kumo's large predictive models to provide enterprise clients with predictive analytics services based on structured data. Kumo's three co-founders—Vanja Josifovski, former CTO of Pinterest; Jure Leskovec, professor at Stanford University; and Hema Raghavan, former AI director at LinkedIn—joined NVIDIA in May 2026. Enterprise structured data consists of multiple interconnected tables, which are generally difficult to parse directly by large language models. Kumo combines graph machine learning techniques with synthetic data generated in simulated enterprise environments to train proprietary large models capable of predicting customer churn or payment defaults, thereby enabling intelligent prediction of enterprise data warehouses. NVIDIA may integrate Kumo's technology into its AI Foundry software services, helping enterprises combine private data with domain knowledge to customize proprietary models tailored to their business scenarios. In April 2026, Kumo released its latest predictive model, KumoRFM-2. Currently, Kumo's partners and customers include companies such as DoorDash, Reddit, Databricks, and Snowflake. Compared to the deal to acquire Groq's inference chip technology for $20 billion by the end of 2025, Nvidia prefers smaller-scale acquisitions. According to securities filings, Nvidia's cumulative acquisition spending over the past five years is approximately $3 billion.