Abridge and Nvidia build foundation model for clinical intelligence from real doctor visits

Abridge wants to be the operating system for medicine -- and NVIDIA and Eli Lilly are helping build it | Fortune

On Thursday morning in New York City, Dr. Shiv Rao stood before a room of health system executives and made a case that ambient AI -- a technology that began largely as a transcription tool -- was ready to do something far more consequential than writing a doctor's notes. Abridge, the startup Rao cofounded in 2018, announced a strategic investment from drugmaker Eli Lilly and what it is calling the first AI-native clinician intelligence platform: a system that both documents the patient-clinician conversation and uses it as the foundation for billing, clinical decision support, payer adjudication, and pharmaceutical trial screening. More than 300 health systems -- including Northwestern Medicine, Emory Healthcare, and Johns Hopkins -- are already live on the platform, supporting upward of 100 million clinical conversations annually and serving more than 250 million patients. The company's platform captures conversation between patients and doctors in real time and automatically generates the clinical note, billing codes, and patient summary before the doctor has left the hallway. What's new is everything that flows from that moment. Before the visit, Abridge surfaces care gaps and prior clinical context for the clinician. During the encounter, the tech suggests discussion topics and surfaces relevant clinical guidelines without requiring the physician to switch applications. After the visit, it generates the documentation, flowsheets, and orders -- all grounded in the actual words spoken. "We've known all along we wanted to be able to connect the dots across the main stakeholders in healthcare, because the only thing that matters, I think, in terms of AI's impact on healthcare is business model innovation." Rao told Fortune. "If we can't actually improve how healthcare is delivered, how it's experienced, and how it's paid for, then we haven't really moved the needle on the problem." Behind the platform expansion is a war chest and a set of strategic bets. Abridge has raised approximately $1.1 billion to date, most recently closing a $316 million Series E extension in April 2026 at a $5.3 billion valuation. NVentures, NVIDIA's venture arm, is among its backers. On Thursday, NVIDIA announced it's also co-developing with Abridge the first foundation model for clinical conversations: an AI model trained on the specific dynamics of doctor-patient dialogue, not a general large language model adapted for medicine. Rao also announced a partnership with Artisight -- an NVIDIA-backed smart hospital company that uses computer vision to automate patient monitoring and nursing workflows inside hospital rooms. Together, the two platforms give care teams a continuous feed of room sensor data and ambient documentation across an entire hospital stay. Eli Lilly's bet tracks closely with its own AI ambitions. The pharma giant is simultaneously building what it has called the industry's most powerful AI supercomputer in partnership with NVIDIA, and Abridge's new life sciences module -- which can surface clinical trial eligibility from within the clinical conversation itself -- represents the kind of patient-facing pipeline Lilly needs to accelerate enrollment for next-generation therapies. "For us, the most important first priority that we would love to explore, that we are working to explore with them, is around clinical trials," Rao said. The ambient clinical intelligence mark Abridge is chasing was valued at $7.24 billion in 2025 and is projected to reach $56.61 billion by 2035. Microsoft, which acquired speech-recognition company Nuance for $19.7 billion in 2022, is the dominant enterprise incumbent. Ambience Healthcare, Suki, and Nabla are all also well-capitalized challengers. But the field is expected to consolidate within the next 12 to 18 months. The question is whether Abridge's expansion into payments and life sciences creates a defensible moat, or simply a larger target. "Now the priority is how much impact can we create, and speed is everything, so I think for the foreseeable future we're just going to focus on the algorithm," Rao told Fortune. The platform's ambition is also its risk surface. Recording protected health conversations requires updated security assessments, state-specific patient consent, and business associate agreements governing how audio is stored. And AI-generated notes that slip past physician review become part of the permanent medical record -- a liability that compounds as the platform moves from documentation into billing codes and clinical orders. Beneath it all is a deeper governance question. Abridge is now positioning itself as neutral infrastructure connecting providers, payers, and life sciences companies through some of the most sensitive data in medicine: the conversation between a sick person and their doctor. Whether that trust holds, at scale, is an untested hypothesis.

Nvidia Taps Abridge to Train AI on Real Doctor Visits | PYMNTS.com

Abridge's platform listens to doctor-patient exchanges and automatically generates clinical notes. The model will run exclusively within Abridge's platform. The deal gives Nvidia something it lacks: a large set of real physician-patient conversations recorded inside hospital settings. What makes that data valuable is precisely what makes it contested. Abridge's already operates across 100 health systems, including Kaiser Permanente, Mayo Clinic, Johns Hopkins and Yale New Haven Health. Private market research platform Sacra found that Kaiser Permanente alone has deployed the platform to 24,600 physicians across 40 hospitals and 600 clinics. Every one of those encounters produces usable conversational data. Nvidia Vice President of Healthcare Kimberly Powell told the Journal the new model will bring clinical intelligence into an earlier stage of development than the industry has previously attempted. Ambient AI Tools Became Healthcare's Most Valuable Data Source Ambient AI tools work by listening to conversations in the room and turning them into structured records without anyone typing. Hospitals bought them to cut the hours doctors spend on paperwork after patient visits. What became clear faster than most expected is that the platforms capturing those conversations were sitting on something more durable than a documentation business. Abridge had already built its own speech recognition model because off-the-shelf tools weren't accurate enough for medical conversations, Davis Liang, the startup's director of applied science, told the Journal. The Nvidia partnership extends that same logic further. Abridge CEO Shiv Rao told the Journal that even powerful general models need clinical shaping to work in practice: "Generic models are powerful, but clinical intelligence -- it still has to be trained, it has to be shaped, and it has to be evaluated against real-world conditions." AI models trained on published medical literature miss what happens in the room. Clinical notes are compressed and shaped by billing requirements. Conversations carry what's underneath: treatment options weighed aloud, patient history surfaced before it reaches the chart. That's the signal the new model is built to learn from. Patient Consent Laws Are Reshaping How Clinical AI Gets Built Using that conversational data carries obligations that don't apply to other AI training sets. Censinet reported in February that healthcare data breaches affected over half the U.S. population in 2024. Abridge trains on data with patient identifiers removed, but researchers found that stripped records "remain statistically tethered to identity through the very correlations that confirm their clinical utility." Removing names doesn't fully sever the link. Patient consent adds another layer. The Texas Medical Liability Trust noted that several states now require explicit consent before AI systems can record and process clinical encounters. The American Bar Association reported that legal challenges around ambient scribes are reshaping what vendor contracts must cover when patient conversations become training data. Forvis Mazars noted that more than 250 AI-related bills were introduced across 46 states in the past year. That regulatory pressure shapes how Abridge built the model's infrastructure. Running it on Abridge's own hardware rather than outside cloud services limits how far patient data travels, Liang told the Journal, and cuts costs. Nvidia Builds Inside Abridge's Walls as Microsoft Moves on Mayo Clinic The model is expected to be ready later this year, the Journal reported. It will sit alongside other models Abridge already uses rather than replace them. Powell told the Journal the Abridge work is one example of how Nvidia's models can be adapted for specific healthcare uses, from drug discovery to medical devices. The Journal also reported that Microsoft last week announced a similar collaboration with Mayo Clinic, drawing on Mayo's clinical data. Joon Lee, CEO of Emory Healthcare, which has deployed Abridge to more than 3,000 physicians, told the Journal he expects the technology to move toward real-time decision support as it matures. Last June, Abridge announced a $300 million Series E led by Andreessen Horowitz, valuing the company at $5.3 billion. Nvidia's venture arm has participated in multiple Abridge funding rounds.

Nvidia Accelerates Healthcare AI Race With Transcription Model | PYMNTS.com

The model will use Nvidia's Nemotron open models and will be deployed exclusively in Abridge's app that transcribes conversations between doctors and patients. The model is expected to launch later this year, according to the report, which cited interviews with executives from the two companies. The collaboration demonstrates how Nvidia's open models can be used for healthcare and life sciences applications, such as drug discovery, medical devices and digital health, the report said. PYMNTS reported in January 2025 that Nvidia launched a series of partnerships designed to boost the healthcare sector via AI. The company said at the time that it was working with partners to develop solutions that included AI agents that reduce administrative burden during clinical trials, AI models that advance drug discovery and digital pathology, and physical AI robots for surgery, patient monitoring and operations. In March, Nvidia released what it described as the first open platform built specifically for healthcare robotics, a stack of datasets, simulation tools and vision-language-action models designed to train AI systems on surgical environments and deploy them in real clinical workflows. The company said the platform's use cases fall into two categories: AI that watches surgery and surfaces information to clinicians in real time, and AI that handles the coordination work that fills a hospital's hours between procedures. The PYMNTS Intelligence report "Healthcare Firms Going Long on GenAI Investment" said in October 2024 that among healthcare firms generating at least $1 billion in annual revenue, about 90% expected a positive return on investment from the generative AI they were deploying in product innovation, fraud prevention and customer service. Anthropic announced in May that it teamed with the Gates Foundation on a $200 million health-and-education-focused AI project. The partnership will combine grant funding with technical support and usage credits for Anthropic's Claude model to "extend the benefits of AI in areas where markets alone will not," Anthropic said. OpenAI said in April that it introduced a version of ChatGPT called ChatGPT for Clinicians that is designed to help clinicians by supporting tasks such as documentation and medical research. For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.