Nvidia and Abridge Will Build a Clinical AI Model From Scratch on Nemotron

Nvidia and the clinical-AI startup Abridge said on June 11 that they will jointly develop an artificial-intelligence model built specifically for healthcare, trained from the ground up on Nvidia's open Nemotron model family. Rather than adapt a general-purpose chatbot to medicine after the fact, the two companies are setting out to bake clinical knowledge into the model from its earliest training stages.

The new system is designed around clinical conversations — the back-and-forth of a patient visit — and will be folded into Abridge's platform to power documentation and clinical decision support. Abridge already builds ambient AI tools that listen to appointments and generate structured notes, visit summaries, and billing-code validation, the kind of paperwork that consumes a growing share of clinicians' time. A purpose-built model, the companies argue, should understand specialized medical terminology and the realities of clinical workflows far better than a model retrofitted for the job.

"Incorporating healthcare-specific expertise during the early stages of model development" beats bolting it on later, said Kimberly Powell, Nvidia's vice president of healthcare, who confirmed the model will be developed using the Nemotron family of open models. Abridge's director of applied science, David Liang, said the model is expected to be ready for use later this year.

The partnership is a tidy illustration of where Nvidia's ambitions now reach beyond silicon. By pairing Nemotron — its open, customizable model line — with a domain specialist like Abridge, Nvidia is positioning itself to seed vertical, industry-specific models, not just sell the chips they run on. For Abridge, co-owning a foundation model rather than sitting purely at the application layer deepens its moat as larger players circle the lucrative clinical-documentation market.