Kumo AI
Category: Foundation Models / LLMs
Developer of KumoRFM, a Relational Foundation Model that delivers instant, zero-shot predictions on structured enterprise data without requiring feature engineering or model training Kumo AI was founded in 2021. The company is led by Vanja Josifovski. Based in Mountain View, California, United States. Team size: 101-500. Total funding raised: $37M. Latest round: Series B. Key investors include Sequoia Capital, SV Angel, A Capital, Frank Slootman, Sridhar Ramaswamy (Snowflake CEO), Ben Silbermann (Pinterest Founder), Matei Zaharia (Databricks CTO), Kevin Hartz (A*), Tristan Handy (dbt Labs CEO), Ron Conway (SV Angel), Michael Ovitz (Broad Beach Ventures).
- Founded
- 2021
- Headquarters
- Mountain View, California, United States
- Team size
- 101-500
- Total funding
- $37M
Value proposition
KumoRFM is a foundation model purpose-built for structured relational business data that makes accurate predictions in seconds — no feature engineering, no ML pipeline setup, no data science team required. It treats relational databases as temporal heterogeneous graphs, applying in-context learning to deliver zero-shot predictions rivaling months of traditional data science work.
Products and solutions
KumoRFM (Relational Foundation Model for zero-shot predictions), KumoRFM-2 (scaled foundation model, April 2026), Kumo Online Serving (real-time, low-latency predictions), Kumo Fine-Tune (task-specific customization), Kumo Research Agent (automated fine-tuning optimization), Predictive Query Language (PQL) — SQL-like syntax for predictions, Snowflake Native App, Databricks Lakehouse App
Unique value
The first Relational Foundation Model (RFM) that predicts anything on relational data instantly — like GPT for structured business data — achieving 89% accuracy vs 63% for LLM-based approaches and 75% for traditional AutoML on the SAP SALT benchmark.
Target customer
Enterprise data teams, data scientists, ML engineers, and business leaders at mid-to-large organizations across industries who need predictive insights (fraud detection, churn prediction, demand forecasting, recommendation systems) from their existing data warehouse
Industries served
Financial Services, Retail & CPG, Media & Entertainment, Insurance, Healthcare, Ad Tech, Telecom, B2B SaaS, Supply Chain, Manufacturing, Gaming, Travel & Hospitality
Technology advantage
Graph Neural Network (GNN) and Relational Deep Learning (RDL) architecture pioneered by co-founder Jure Leskovec (Stanford). Treats relational databases as temporal heterogeneous graphs. Hierarchical in-context learning extracts task-aware features at both individual and aggregate levels. Pre-trained on billions of relational patterns across diverse datasets. Table-agnostic coding scheme with Relational Graph Transformer. Achieves up to 95% time reduction vs traditional ML pipelines.
How they differentiate
Only foundation model purpose-built for relational/multi-table enterprise data using graph neural networks (GNN). Zero-shot predictions without training. 89% on SAP SALT benchmark vs 63% (LLM) and 75% (AutoML). PQL replaces months of code. Native integrations with all major data warehouses. Backed by GNN creator Jure Leskovec.
Main competitors
TabPFN / Prior Labs (single-table foundation model), Databricks ML / AutoML (integrated lakehouse ML), DataRobot (automated ML platform), H2O.ai (AutoML), LightGBM / XGBoost (traditional gradient boosting)
Key partnerships
Snowflake (Native App partner, jointly developed Snowflake Native App), Databricks (Lakehouse App partner), AWS, Google BigQuery, SAP, Samsara, Frank Slootman (Snowflake) and Sridhar Ramaswamy (Snowflake CEO) as advisors/backers
Notable customers
DoorDash (30% accuracy improvement), Reddit (5.5x conversion lift, 4-5 years of work in 2 months), Sainsbury's (UK #1 retailer), iFood (10% conversion lift), Chime, Databricks (5.4x conversion lift), Expedia Group
Major milestones
2021: Company founded in Mountain View, CA by Vanja Josifovski, Jure Leskovec, and Hema Raghavan, 2022-04: Emerged from stealth with $18.5M Series A led by Sequoia Capital, 2022-09: Raised $18M Series B led by Sequoia Capital, launched first product version, 2024-2025: Platform adopted by DoorDash, Reddit, Sainsbury's, iFood, Chime, 2025-08: Launched KumoRFM — world's first Relational Foundation Model, 2025: Named to Inc. Magazine's Best in Business list (Best Startups category) and Fast Company's Next Big Things in Tech, 2026-04: Released KumoRFM-2 (scaled foundation model), 2026-04: Launched Kumo Online Serving for real-time predictions, 2026-06: Acquired by Nvidia for over $400M, three co-founders transitioned to Nvidia
Growth metrics
Acquired by Nvidia for $400M+ (June 2026); ~69-73 employees (early 2026); 15 industries served; Hundreds of millions of users powered by Kumo models globally
Market positioning
Category-creator in Relational Foundation Models — uniquely positioned at intersection of graph neural networks, foundation models, and enterprise data warehousing. Differentiated from TabPFN (single-table limit) by handling multi-table relational databases at enterprise scale (hundreds of millions of rows). Competes against traditional AutoML/feature engineering approaches by eliminating months of pipeline work.
Geographic focus
United States (HQ), United Kingdom, Brazil, global enterprise market
Patents and IP
Multiple peer-reviewed publications including KumoRFM-2: Scaling Foundation Models for Relational Learning (Apr 2026); PLUREL: Synthetic Data unlocks Scaling Laws for Relational Foundation Models (Feb 2026); Learning Production Functions for Supply Chains with Graph Neural Networks (Mar 2026); graph Transformer innovations by co-founder Jure Leskovec (GNN co-creator)
About Vanja Josifovski
Ex-CTO of Airbnb; Ex-CTO of Pinterest; PhD in Computer Science. Previously held senior engineering leadership roles at Google.
Official website: https://kumo.ai