NVIDIA acquired Kumo AI on or around June 3, 2026, in a deal reported by The Information at more than $400 million. NVIDIA hasn’t confirmed financial terms. That caveat matters less than the move itself.
Kumo was built for one specific problem: prediction across relational data. Most enterprise databases don’t look like the clean flat tables that ML models prefer. They look like interconnected webs – customers linked to orders linked to products linked to suppliers. Graph neural networks navigate that structure natively. Kumo’s platform, according to the company, automates prediction workflows on relational data without requiring manual ML pipeline configuration. The use cases are mundane in the best way: customer churn, inventory forecasting, fraud detection. The kind of predictions that drive real enterprise value today.
The three co-founders are joining NVIDIA. Vanja Josifovski, former CTO of Airbnb and Pinterest, leads the team. Jure Leskovec is a Stanford professor and one of the academics most responsible for making graph neural networks a practical tool. Hema Raghavan was LinkedIn’s AI lead. That’s not a founding team assembled to build a product and flip it. That’s a team that understood enterprise data problems at scale before Kumo existed.
Kumo reportedly raised approximately $37 million in a 2022 round led by Sequoia Capital. The acquisition, if confirmed at $400 million-plus, would represent a meaningful return for Sequoia on a relatively modest early bet.
Why it matters
NVIDIA’s story for the last several years has been hardware dominance – GPUs, networking, inference infrastructure. That story is real. It’s also incomplete. Hardware margins compress as competition scales. Software margins don’t. Every acquisition NVIDIA makes in the application and tooling layer is a bet on a different revenue profile than selling chips. Kumo isn’t a research trophy. It’s a production-ready enterprise tool with documented customer use cases and a founder team that can sell into the Fortune 500.
This is the second confirmed software-layer acquisition in NVIDIA’s recent history following a pattern of moving up the stack through targeted M&A rather than internal build. Secondary reporting from PYMNTS confirmed the deal terms independently from The Information’s original exclusive.
What to watch
Three signals will indicate how seriously NVIDIA is treating Kumo as a platform play rather than a talent acquisition. First: whether Kumo’s product appears in NVIDIA’s AI Foundry announcements within the next two quarters. Second: whether Josifovski, Leskovec, or Raghavan take public-facing roles in NVIDIA’s enterprise go-to-market (conference appearances, product keynotes). Third: whether existing Kumo enterprise customers report continuity of their contracts or a migration push toward NVIDIA’s broader platform stack. Any of those three would confirm this is an integration, not an acqui-hire.
What to Watch
The real story is
this: NVIDIA is assembling an enterprise AI platform layer piece by piece, and it’s doing it with companies that already have paying customers. The Kumo deal doesn’t tell you NVIDIA will become an enterprise software company. It tells you NVIDIA doesn’t want the enterprise software layer owned by someone else.
Watch the next NVIDIA earnings call for any mention of enterprise software attach rates. That’s the first hard data point that will tell you whether this acquisition strategy is converting into revenue.