Nvidia Pays $400 Million for AI Software Firm Kumo

NVIDIA

Nvidia acquired predictive artificial intelligence software startup Kumo AI for $400 million, The Information reported Wednesday (June 3).

    Get the Full Story

    Complete the form to unlock this article and enjoy unlimited free access to all PYMNTS content — no additional logins required.

    yesSubscribe to our daily newsletter, PYMNTS Today.

    By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions.

    The report, which cited an unnamed source, characterized the deal as something of a departure for Nvidia, which has invested heavily in open-weight models, while Kumo’s are proprietary. They’re designed to answer questions on structured business data, like customer information and payment data, which are typically tougher for AI models to decipher.

    It’s not clear how Nvidia would use Kumo’s models, the report said, positing that the company could employ them in its AI Foundry software, designed to help companies build custom AI models. In addition, Nvidia could use Kumo’s researchers to help create new, business-centric Nvidia foundation models.

    Kumo’s founders include Vanja Josifovski, former chief technology officer at Airbnb; Jure Leskovec, a Stanford University computer science professor; and Hema Raghavan, a former AI lead at LinkedIn. All three have been working at Nvidia since May.

    The acquisition is in keeping with Nvidia’s history of small-scale deals, with the $5 trillion company having spent around $3 billion on acquisitions in the last five years, according to the report. In late 2025, the company agreed to pay $20 billion to license technology from Groq, an inference chip designer.

    In other Nvidia news, the company partnered with Microsoft to create Windows PCs built for agentic computing. Shipping this fall, these laptops and compact desktops will be powered by Nvidia’s new superchip and designed to run AI agents directly on the device.

    “The pitch to enterprise buyers is straightforward,” PYMNTS reported Monday (June 1). Keep the data, the decisions and the agents inside the building. For enterprise buyers, the case centers on what stops leaving the building. Agents running locally can work across files, calendars and internal applications without sending data to a third-party server, cutting the latency, privacy exposure and usage costs that come with constant cloud inference.”

    The partnership is part of an agent push Nvidia began in March, when it introduced NemoClaw. This is an enterprise-hardened version of OpenClaw that adds security controls, policy enforcement and audit capabilities for businesses in regulated industries.

    For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.