# Nvidia acquires Kumo AI predictive model maker > Source: > Published: 2026-06-04 05:21:37.914230+00:00 # Nvidia acquires Kumo AI predictive model maker Nvidia has acquired **Kumo AI**, a four-year-old startup that builds predictive foundation models for business data, reporting by **Fortune** and **SiliconANGLE** shows. **Fortune** reports the three Kumo co-founders - **Vanja Josifovski**, **Hema Raghavan**, and **Jure Leskovec** - moved to Nvidia last month; **Fortune** also reports Nvidia declined to comment. **The Information** reported the deal was for at least **$400 million**. **SiliconANGLE** describes Kumo's platform as using **graph neural networks**, a proprietary Predictive Query Language, and integrations with data warehouses such as **Snowflake** and **Databricks**; SiliconANGLE also reports the platform automates data preparation and can cut manual setup effort by as much as **95%**. **Fortune** reports Kumo raised **$37 million** in 2022 from investors including **Sequoia Capital**, and that customers have included **Reddit** and **Sainsbury's**. ### What happened Nvidia has acquired **Kumo AI**, a four-year-old enterprise startup that builds predictive foundation models for business data, reporting by **Fortune** and **SiliconANGLE** shows. **Fortune** reports the three Kumo co-founders - **Vanja Josifovski**, **Hema Raghavan**, and **Jure Leskovec** - transitioned to Nvidia last month and now list Nvidia on their profiles, per Fortune. **Fortune** reports Nvidia declined to comment on the transaction. **The Information** reported the deal was for at least **$400 million**. **Fortune** also reports Kumo had previously raised **$37 million** in 2022 from investors including **Sequoia Capital** and that Kumo's customers have included **Reddit** and **Sainsbury's**. ### Technical details SiliconANGLE reports Kumo's models rely on **graph neural networks** to represent rows and entities in business data as nodes and edges rather than treating rows in isolation. SiliconANGLE describes a proprietary Predictive Query Language that translates natural-language-style questions into model queries, and reports Kumo integrates with data warehouses such as **Snowflake** and **Databricks**. SiliconANGLE further reports the platform automates data cleaning, joins, and feature preparation and claims it can reduce manual setup effort by as much as **95%**. **Fortune** cites an earlier on-record comment from cofounder **Jure Leskovec**: "With the foundation model, you point it to your data, you define what you mean by churn, and a second later, you get the prediction," from an interview with Jeremy Kahn. ### Editorial analysis - technical context Companies building predictive capabilities for enterprise data increasingly combine representation-learning techniques with automated data plumbing to reduce time-to-value. Industry-pattern observations: graph-based representations are a common choice when relationships across records matter, because they let models encode cross-entity signals without manual feature joins. Industry-pattern observations: product teams frequently pair tight integrations with prominent data warehouses (Snowflake, Databricks) and a declarative query or DSL layer to simplify adoption for analytics and BI teams. ### Context and significance Editorial analysis: Nvidia's acquisition fits coverage that the company has been assembling a broad AI stack via deals reported by **Fortune** and others; Fortune notes Nvidia has completed many acquisitions in recent years, and recent deals cited by Fortune include purchases of Illumex and Run.ai and a high-profile Groq transaction. Editorial analysis: For practitioners, the acquisition is noteworthy because it signals continued vendor interest in packaging model capabilities together with data connectors and automation. Kumo's emphasis on predictive foundation models and graph learning illustrates a marketplace for specialized, data-connector-first model products rather than general-purpose LLM-only offerings. ### What to watch Editorial analysis: Observers should watch whether Nvidia makes Kumo's technology available as a standalone service, embeds it into GPU-accelerated inference products, or exposes it through partnerships with Snowflake and Databricks. Editorial analysis: Metrics to monitor include how Nvidia surfaces Kumo models in developer tooling, whether existing Kumo customers retain access under the new ownership, and whether the Kumo team publishes benchmarks or technical notes on their graph-based predictive approach. ### Reported limitations Fortune and SiliconANGLE report that Nvidia did not disclose full terms (Fortune reports Nvidia declined to comment; The Information reports a figure of at least **$400 million**). SiliconANGLE's efficiency claims (up to **95%** reduction in manual setup) are reported as vendor statements about the platform's automation capabilities. Overall, the acquisition is a concrete example of enterprise-model consolidation: an attractively packaged predictive model plus connectors and a DSL being folded into a major AI vendor's ecosystem, according to the published coverage. ## Scoring Rationale This acquisition is notable because it brings a specialized enterprise predictive-model product into Nvidia's expanding AI stack. It matters to practitioners focused on productionizing models with tight data-warehouse integrations and graph-based representation learning. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)