January 13, 2025 | Among the many partnerships and technical announcements made by NVIDIA during the company's Monday morning presentation at the 43rd Annual J.P. Morgan Healthcare Conference, NVIDIA focused several on innovation within the healthcare and life sciences industry.
In a private press briefing last week, Kimberly Powell, VP of Healthcare at NVIDIA, cast NVIDIA’s vision for the AI Industrial Revolution and outlined the computing advancements that are enabling it.
For the first half of the last century, computing was done by humans, she explained. Compute code was written as serial instructions because that’s how humans think. But that has changed as large language models became the operating system of the computer.
Gone are the innovations of perceptive AI and generative AI, Powell said. AI is “very rapidly accelerating into agentic AI,” she said, AI that acts as an agent, not only perceiving, but reasoning, planning, and even taking action and performing tasks.” AI agents can use tools (tool calling), choose from accessible information, and collaborate with humans. “Agents are becoming, you could imagine them, digital employees,” she said. “It’s happening, really, seemingly overnight. And right around the corner… is physical AI where you have physical robots and the physical environment all operating and embodying AI in a physical world.”
The AI Industrial Revolution, she says, marks a shift. Data is now simply raw material. Tokens are the new commodity—with NVIDIA serving as the token generator. Under the new paradigm: “You introduce data, the machine learns from that data, it builds models, and then it writes software that executes on a GPU,” Powell explained. “We call this the AI Factory.”
AI factories are a new form of computing infrastructure, she said. “They’re purpose built to process raw data, refine them into models, and then produce tokens with great scale and efficiency.” Every company will produce digital intelligence through tokens, she said. Tokens, she defines, are “the intelligent response and actions of digital nurses or tutors or customer service agents, chip designers, even autonomous cars and weather prediction agents.”
Mayo Clinic, NVIDIA Digital Twin Vision
Pathology has long been an early adopter of AI in computer vision, and NVIDIA plans to continue pushing that forward with a collaboration with Mayo Clinic. Mayo’s Clinic Digital Pathology platform, built from autonomous robotic labs and advanced imaging technology, already hosts a dataset of 20 million whole-slide images with 10 million associated patient records. Mayo Clinic and NVIDIA will work together to accelerate the development of the pathology foundation models using these data.
The work will “push the frontiers in personalized health experiences and really predictive and efficient treatment strategies,” Powell predicts.
Mayo Clinic will use NVIDIA’s DGX Blackwell 200 system featuring 1.4 terabytes of GPU memory, Powell said, which is ideal for handling large, whole-slide images, and it will integrate with Mayo existing digital platform and autonomous robot labs.
“Our ultimate goal is to create a human digital twin,” Powell said. “This is a dynamic digital representation including medical imaging, pathology, health records, and wearables. To achieve this, Mayo Clinic and NIVIDA will leverage some of the latest AI models and vision language models like Cosmos Nemotron and the NIVIDA NIM microservices. “This collaboration will be a cornerstone for new applications in drug discovery and diagnostic medicine,” Powell said.