NVIDIA Acquires Predictive AI Startup Kumo AI for Over $400 Million
NVIDIA has acquired Kumo AI, a predictive artificial intelligence company founded five years ago, for reportedly over $400 million.
The three co-founders of Kumo—former Pinterest CTO Vanja Josifovski, Stanford professor Jure Leskovec, and former LinkedIn AI head Hema Raghavan—joined NVIDIA in May 2026. Kumo uses graph machine learning and synthetic data to train proprietary large models that can predict customer churn and payment defaults.
NVIDIA plans to integrate Kumo's technology into its AI Foundry services to help businesses customize proprietary predictive models based on structured data warehouses. Kumo's clients include DoorDash, Reddit, Databricks, and Snowflake. This acquisition aligns with NVIDIA's preference for mid-sized acquisitions, with total spending on acquisitions reaching approximately $3 billion over the past five years.
Through this acquisition, NVIDIA enhances enterprise AI predictive capabilities, with large tech companies becoming major buyers. The Kumo founding team and early investors achieve an exit, while funding in the AI infrastructure sector continues to concentrate on companies with structured data processing capabilities.
Source: Public Information
ABAB AI Insight
NVIDIA has previously strengthened its full-stack AI capabilities through acquisitions of companies like Mellanox and Run:ai. The acquisition of Kumo AI continues its path of expanding from chips to enterprise-level applications, particularly addressing the traditional weaknesses of large language models in structured data. In late 2025, NVIDIA acquired Groq's inference chip license for $20 billion, indicating a focus on complementary technologies in its acquisition strategy.
In terms of capital strategy, NVIDIA is leveraging its cash reserves and stock valuation advantages to acquire mid-sized AI teams, rapidly internalizing Kumo's graph machine learning and synthetic data technology. The goal is to enhance the enterprise customer stickiness and customized revenue of the AI Foundry platform, creating a closed-loop service chain from GPUs to predictive models.
This acquisition is similar to NVIDIA's previous acquisitions of DeepMap and Omniverse-related assets to strengthen industrial AI. The AI industry is currently undergoing a transformation phase from large models to vertical enterprise predictive applications.
Essentially, this represents a restructuring of the industrial chain: the predictive capabilities of enterprise structured data are shifting from general large models to specialized graph learning models, as Kumo's technology addresses the parsing bottleneck of multi-table relational data, shifting pricing power from pure computing power suppliers to integrated platforms that master both data modeling and inference.
ABAB News · Cognitive Law
Chips sell hardware, acquisitions sell the future; true pricing power lies in the ability to turn data into predictions in a closed loop.
Mid-sized acquisitions are a powerful tool, while large licenses are a supplement; capital builds full-stack barriers through precise acquisitions.
Structured data is a gold mine; whoever cracks it first will dominate the next battlefield of enterprise AI.