Nvidia acquires Kumo AI to bolster predictive foundation model capabilities

Nvidia acquires Kumo AI to bolster predictive foundation model capabilities

Nvidia has acquired Kumo AI, a four-year-old startup specializing in foundation models for business prediction, according to sources familiar with the deal. Kumo's three co-founders—Vanja Josifovski, Hema Raghavan, and Stanford computer scientist Jure Leskovec—have transitioned to Nvidia. The Mountain View-based startup had raised $37 million from Sequoia Capital across two 2022 rounds and counts DoorDash, Reddit, and Sainsbury's among its customers. Financial terms were not disclosed.

This acquisition continues Nvidia's strategy of absorbing AI talent and technology through small-scale deals, accelerating its expansion beyond hardware into full-stack AI services. Kumo's key asset is a foundation model for relational forecasting that can handle predictions like customer churn or credit default without custom training—a capability that directly extends Nvidia's enterprise AI platform. The deal aligns with the acqui-licensing pattern Nvidia has refined across 100+ startup purchases, most notably the $20 billion Groq acqui-hire in December 2025 and the $700 million Run.ai deal in 2024.

By internalizing Kumo's prediction engine, Nvidia gains a differentiated offering for its enterprise and cloud customers, reducing reliance on independent predictive analytics vendors. The acquisition also pulls Stanford researcher Leskovec into Nvidia's orbit, strengthening its research bench. While the terms are undisclosed, the pattern fits Nvidia's capital-efficient approach: buying small teams with proven technology at single-digit-million costs, then layering their capabilities into its broader infrastructure stack. This strategy deepens the context-engineering moat around Nvidia's ecosystem, making its platform stickier for business users who need instant predictions from relational data.

#Nvidia #KumoAI #PredictiveAnalytics #AcquiHiring #EnterpriseAI #FoundationModels