What if every wet lab had a digital twin? What if every surgeon had an AI assistant that could summarize patient data mid-procedure? And what if drug discovery no longer took years — but months?
All of these scenarios are part of the future for healthcare offered by Kimberly Powell, Nvidia's Vice President of Healthcare and Life Sciences, at the company’s GTC 2025 conference this week.
The event attracted more than 25,000 attendees, and featured thousands of speakers and demonstrations of self-driving cars, humanoid robots and clean energy systems. Powell’s presentation offered a sweeping view of how AI is reshaping scientific discovery and healthcare innovation, from genomic modeling to real-time clinical decision-making.
“This is just the beginning,” said Powell. “We’re starting to experience exponential levels of biological intelligence — being able to represent biology in a computer, not just to read it, but to write it, to generate it, to design it.”
That leap — moving from perception to generation to reasoning — is more than a technical milestone. It has the potential to reshape how hospitals treat patients, how scientists interrogate theories and how drugs are discovered and developed.
‘Digital Flywheel’ Upends Scientific Process
In Powell’s vision, the traditional order of scientific inquiry is being upended. Rather than designing an experiment and then collecting data, AI agents can now propose hypotheses, conduct virtual screenings and even generate structured research reports—all within a single workflow.
"It used to be that experiments would try to answer a singular question, move to the next decision point and do the next," Powell said. "You recognize that what we can do in the physical world can be hugely augmented by being able to do things in the digital world."
Powell described how this proximity creates what she calls a "digital flywheel" – where physical experiments feed data into AI systems that generate new hypotheses, which then guide the next round of lab work. This replaces the traditional sequential research process with a more dynamic, parallel approach to discovery.
Powell provided a hypothetical example of how this could work: a researcher interested in new treatments for cystic fibrosis might ask an AI agent to find recent discoveries in the field, identify promising therapeutic targets and screen billions of molecules against those targets — all without ever leaving their desk.
This kind of work, she said, is being made possible by tools like Nvidia’s BioNeMo platform and its suite of “biomedical research agents.”
The company’s latest foundation model, Evo 2, was trained on 9 trillion nucleotides and boasts a context length of one million tokens—an important advancement for working with the vast complexity of genomic data.
The company also announced the DGX Spark, a desktop-sized AI supercomputer delivering one petaflop of performance – that’s one quadrillion operations per second —enough power to simulate drug interactions, model DNA sequences or run advanced AI agents without needing a massive data center. The device will be available in July.
“Agents of all kinds are really going to transform not only the experience of healthcare, but also our ability to unlock the information that we've been digitizing for the last couple of decades, whether that be our electronic health records, whether that be digital wearable data, or real-world evidence,” she said.
The impact is already visible in the real world. Powell cited Insilico Medicine, which has reduced the process from target identification to drug candidate to just 13 months, while significantly decreasing the number of physical compounds that need to be synthesized.
Beyond the Lab: Using AI Applications in Clinical Care
While much of the AI conversation has centered around research and data analysis, Powell made it clear that patient-facing applications are also advancing rapidly.
Abridge, an AI-enabled communication platform, is already in use at more than 100 health systems across the United States. The tool uses AI reasoning to enrich the context behind clinician-patient conversations, potentially streamlining intake and reducing the cognitive load on physicians.
“Every time you visit a doctor and they ask your date of birth again — it’s a sign we’ve failed to integrate,” Powell said. “These agents can bring that context automatically into the room.”
Healthcare applications extend beyond the lab into clinical settings. Powell demonstrated how AI-enhanced surgical systems are creating operating rooms where voice-controlled assistants can recall patient information without surgeons removing their hands from instruments.
Powell highlighted several companies using AI for healthcare beyond the laboratory setting. ConcertAI is applying the technology to match patients with appropriate clinical trials, speeding up the process up to three times faster than traditional methods. Hippocratic AI has created what Powell described as "essentially a marketplace" of AI agents for healthcare applications. Since launching in January 2025, the company has developed agents spanning 25 different specialties with over 400 identified use cases.
Powell also mentioned Deloitte's work building lifelike digital avatars that can engage patients in natural conversations across multiple languages. Developed for hospital environments, the avatars can provide pre-operative information, answer questions about procedures and offer post-surgery guidance.
And she shared footage from Synchron showing how their brain-computer interface allowed a paralyzed man to control a computer through thought alone.
"This is what we can do when we're working together," Powell said, "taking advantage of technology and tools to solve incredibly important human problems."