All right. Oops.
Yeah, you're on.
Oh, yeah. We're on. Awesome.
We're live.
Anyway, good morning, everybody. Thank you for attending this session and obviously coming to the 44th Annual TD Cowen Health Care Conference. I'm my name's Matt Ramsay. I actually run the Semiconductor Practice and Research, so I'm a bit of a fish out of water here. So I guarantee you I know less about healthcare than anybody in this room, but hopefully we're gonna have a good discussion about about AI and healthcare. And been super pleased to be joined by Kimberly Powell, who runs the healthcare business for NVIDIA. You might have heard of NVIDIA. It's been a little bit topical in the last few months. I can't have a conversation without on any topic, literally, without AI infusing every end market in the economy and healthcare is certainly no different.
But, Steve and my friend JJ, we're gonna do some Q&A with Kimberly and hopefully explore a little bit about what NVIDIA's doing in the healthcare market, across a number of verticals. So, Kimberly, thank you for joining again this year. We really appreciate it.
Thank you. It's a pleasure to be here. I mean, this, this conference is just amazing. The sessions are very much painting the picture of what we're so excited about these, this next 10 years. So pleasure to be here, and thank you so much for the invitation.
Well, thank you. So I think I wanted to start, Kimberly, just, for this audience, I spend in my tech world, I spend all day talking about the hardware stack of NVIDIA, the software stack of NVIDIA, what you guys are doing at CUDA and libraries on top and driving the data center business towards what could almost be $100 billion in revenue this year, which is pretty astounding. No one's really seen that happen before. But I wanna see if you could tell us a bit about how the healthcare franchise fits in with the broader company at NVIDIA. So, what technologies you're deploying, what verticals within the healthcare industry are you trying to influence and infect, and then, like, how you go about building a healthcare franchise on the side of a huge technology company.
Yeah. No, I appreciate the, the question. I'll, I'll take you back a little bit. I'm working on my 16th year at NVIDIA. So we've been working in the healthcare space for a decade and a half. It was right at the time that we were pioneering a new computing approach called accelerated computing. We built these graphics, GPUs, graphics processing units for, for gaming. And if you think about what gaming is doing, it's simulating light. So we built a very, very powerful processor, and we realized that this processor could be used for a ton of different applications, but we didn't know them all. So Jensen, our CEO, made the decision, let's invite somebody to go figure that out. And that's kind of been my charter ever since. So we were pioneering accelerated computing back in 2008.
Some of the earliest applications of our accelerated computing technology were born right here at MGH down the street, working in the areas of medical imaging. If you think about what imaging is, it's really the cornerstone of all patient care, right? The first thing we need to do is early detection. We need to understand what's going on inside the body. The tools that we use to do that are largely in the area of imaging. Now imaging drives everything we do inside the walls of healthcare. On accelerated computing platforms, you're essentially creating sensor technologies, whether that's CT, ultrasound, MRI, the new photon-counting CT that is incredibly opening up new applications of being able to see the function of the anatomy, microscopy, pathology. All of these, think of them as sensors.
They need to have very, very powerful computers that allow you to take that sensor data and present it as information that can really drive decision-making, for clinicians. And so accelerated computing, imaging, some of the very first applications, and we're still pioneering new imaging applications, today. The second thing that NVIDIA has been pioneering with the world over the last decade is artificial intelligence. Back in 2012, a new modern approach to artificial neural networks leveraged was made possible by our accelerated computing platform but really spawned off this new computing approach called artificial intelligence. And right now, today, we're living in these two platform shifts of computing that are enabling whole new classes of applications. And so yes, we are built on top of the foundational pieces that NVIDIA does every single day. People know us for chips, absolutely, our GPUs. We also build CPUs.
We also have DPUs for data processing units or smart NICs. The complete system of computing is now something we've pioneered to allow for this next wave of artificial intelligence we're all living in, which is generative AI. And so these platform shifts are a complete new way of doing computing. And we recognize that, that, that applications are gonna be built in very, very different ways. And so our job in the healthcare business unit to talk about what is our mission statement, it's how do we take these pioneering computing approaches and apply it to the healthcare domain. In fact, because of generative AI now, we see this as the opening of an industry that is gonna be the world's largest technology industry. Healthcare will be the largest technology industry in the coming years.
At the end of the day, you don't know a lot about healthcare, but every single one of us is a patient, fact. We are all part of the healthcare ecosystem, whether we're patients or actually part of delivering care. So generative AI is giving these incredible new capabilities of opening new aspects of healthcare and to become technology. Let me give you some examples of where we're super, super excited. One is still in the area of medical devices and medtech. Every single day, new sensor technologies are being born or new applications are being able to be built with imaging. We're pioneering foundation models for imaging. You can do full segmentation of a 3D scan in seconds, full segmentation. It used to have to have humans who would sit there and contour the images so that they could prepare the plans for image-guided therapy.
You can now have a computer completely do that automatically so that clinicians can actually think about the plan a lot more than the arduous work of segmenting, for example. So imaging, medical technology, exciting area. Imaging is also at the cornerstone of what we're seeing in minimally invasive and robotic surgery, right? You're sticking something, essentially a camera inside the body so you can give the clinicians much more visibility without having to have the effects of open surgery. And then we're moving into robotic surgery where you now have actually robots assisting the surgery and being able to again use imaging and artificial intelligence to augment what the clinicians are seeing in real time. We are pioneering complete full-stack computing platforms for being able to do that real-time AI. It's called Holoscan, NVIDIA Holoscan.
This is the capability to literally take and in milliseconds image information in, overlay AI on top so that a clinician can interact with that data in real time. A lot of this is leveraged, if you can imagine, from the self-driving car industry. It takes your mind a, a minute to think about this, but a self-driving car and a surgical robot are actually very, very similar in the job and the task that they do and the computing platforms that they need. So we're able to leverage a lot from these other core businesses of NVIDIA and apply it in here. So medical devices, imaging, medtech, surgical, robotics, incredibly exciting area. And you're gonna see this next phase of digital surgery coming. And the other area that we're super, super excited about is artificial intelligence and accelerated computing for drug discovery.
We see this as a sort of next generation of computer-aided drug discovery is in front of us for the very first time. Now that we have three things that has happened, the life sciences industry has created the digital biology moment. You have CRISPR technologies, cryo-electron microscopy, next-generation sequencing, spatial genomics, all creating these gigantic datasets of biology. Ingredient number one, you have GPT, generative pre-trained transformers, the ability to learn from large amounts of data and represent data inside of a computer, GPT. And the third ingredient, which is AI supercomputers, that allow for you to take that digital biology information, the method of generative pre-trained transformers, and representing that information in a computer and on supercomputers to train these new models.
And so this is the very first time we're able to take the language of drugs, DNA, sequence of characters just like our English language, but just in three billion characters long and four, four letters, amino acid sequence, again, 20 different characters in a sequence, chemistry, SMILES strings, also another sequence. So you can imagine we can now represent the language of drugs inside of a computer. And so we see this as the next generation of computer-aided drug discovery is in front of us. And there's a lot, a lot of evidence about that. So these are some of the areas that we're, very, very excited about.
There's obviously a lot going on. I just wanted to ask one more question for myself, and I'll turn it over to my healthcare colleagues. If you just think about the scale of your healthcare franchise today, right, can you talk a little bit about the size of the investments you're making within the company? Obviously, you're leveraging the technology from the data center group. You mentioned the automotive group. But what's the size? I mean, anything you can give us on the size of it from a revenue perspective, from a investment perspective, just what is the healthcare franchise inside of NVIDIA?
Yeah. We've talked about this recently. So, the healthcare business is well over now $1 billion, both direct and indirect, through our partners. So it's really ramping up now. We've been preparing for this moment for the last decade and a half. We have been hiring deep domain expertise. We have everything from applied research, dedicated applied research teams. You can see them on a lot of the publications that are happening in this area. We have dedicated engineering teams that go across the platforms that we're building. I'll talk a little bit more about NVIDIA Clara. After Clara Barton is our healthcare platform that has those different domains of computing platforms within it. She invented the American Red Cross.
We think of that as a platform to really facilitate these new computing approaches to the entire industry. So dedicated engineering. And now we have a lot of products. You're gonna see we GPU Technology Conference in about a week and a half. And you're just gonna see a ton of new products, largely software. And that's what is interesting. The healthcare business unit, as I say, it's the application of computing in this industry. So application synonymous with software. Most of what we build is software. We did decide to build a hardware computing platform for the real-time imaging. And that is NVIDIA Holoscan on IGX. And that is because we needed it. The industry needed it. They needed to have a general-purpose computer that can do that real-time processing. But otherwise, it's all about that.
And so we have thousands of people around the company doing everything from applied research, engineering, the product, all of the go-to-market. We have an Inception program. It's our startup program, AI startup program. It's kind of a virtual accelerator at NVIDIA. And I was just telling somebody about it. We have about 3,000 AI healthcare startups in that Inception program. And we are supporting them all over the world. They are everything from, you know, stealth. They were just a professor, and they have a great idea, all the way up to, pre-IPO. And we do everything from early access to our technology to, making them much more effective, and, and leveraging the technology, and even all of the go-to-market, that we're doing together. Am I not loud enough? Do I turn that one off?
Oh, wow.
I don't know. Oh, is this better? Okay.
Okay.
So pretty big, a decade and a half we've been building it. And we're ready to see this next decade and a half, with some real pioneering work.
Awesome. Josh, I don't know. Kimberly started going down the path of, like, the experience from imaging through surgical and post-surgical. I know that's an area that you've spent a lot of time on. So if you have any questions.
Thanks. Thanks for the mic. Appreciate the time here today. You know, I think medical device companies have been working on for years this incorporating AI and machine learning into digital robotics or our ecosystem. It consists of next-generation robotics, world-class instrumentation, advanced imaging visualization, and data analytic and analytics and digital solutions powered by AI, AI, and ML. But in order to accomplish the optimal approach to the interventions and so I mean, just from NVIDIA's standpoint, any stance on where we are in terms of the plan where robotics and AI-driven analytics will guide physicians through the best next steps intraoperatively or intra-procedure. I know NVIDIA has a big platform that's helping a number of partners get there. That's maybe just a foundational question.
Sure. Thank you. So, we think of this whole area as the transition it's going through right now is you think of a lot of these medical device companies as being hardware companies. And the reason we invented NVIDIA Holoscan is to create an industry of software-defined medical devices so that, as you know, being a clinician, this is the practice of medicine. It's not just the singular AI that can diagnose something. There are decision points that are happening continuously, especially in surgery. And so what do you need in order to do that? You need a real-time AI computing platform. And you need it to be continuously be able to take new applications that are coming to it, just like your Tesla car is getting over-the-air updates every morning, improving its performance.
We have now created the conditions for the medical device and surgical robotics industry to take advantage of that. Okay. So one, you needed the compute platform for these AIs to be able to land, live on, and deliver, the cognitive experience so that decision-making can be made better. Everything from, overlaying information, don't cut there, tool tracking so that you can understand the phase of surgery. You know, all of this incredible information now can be built into these systems. And, and it's not only that. There are new foundation models. Everything that you're seeing with, GPT, Midjourney, these other applications that you're seeing in the consumer internet space of, multimodal models, we're now seeing we're now right at the cusp of new models called, vision language models. So you're incorporating your, the ability to watch video and temporally reason over that video about what's happening.
In fact, there's some incredible new research where you can watch a surgery and predict what's gonna happen in the next 20 seconds. So imagine the decision-making that's going to transform within the walls of that operating room. That whole operating room is gonna become an AI, whether you're, you know, the anesthesiologist, whether you're the nurse going for the next tool, or whether you're actually the surgeon who's thinking about what his or her next move is for this particular patient.
You're gonna be able to talk to a computer and say, "Could you tell me about any pre-existing diseases that I should know about because the surgery's going a little bit long, and I need to know any, you know, comorbidities that might change or affect or complicate the surgery?" You could ask a computer to say, "Please bring me up her preoperative, you know, MRI so that I can, you know, see if I've extracted everything, you know, that I need to during the surgery." So the amount of information that's gonna now be able to enter the OR, the ability because now we have all these systems, automatic speech recognition, large language models to take that speech recognition, go retrieve information from all the hospital systems.
The OR and not even just the OR, but the OR is a very exciting space where we're gonna be able to start presenting that information in a much more friendly way for these clinical decision-makings. So this idea of intraoperative information is extremely exciting. And then you're gonna see it on both sides of it. Preoperatively, to prepare you, to remind you, "How does this robot work? What does that button do? How might I use that? What tool should I be using?" You're gonna have a conversation to refresh yourself, to retrain yourself.
Then postoperatively, you're gonna be able to use these artificial intelligence applications to summarize it, to reason over, "How could I deliver this surgery better?" to do all of the unfortunate paperwork that a lot of our clinicians have to do, you know, during their pajama time, if you will, right, because wouldn't it be great if 90% of what they have to summarize in reports and otherwise is done by a computer, and they're taking only the 10% to check it and make sure it's accurate and move on with their day? So it's going to span end to end, artificial intelligence. And that's why these systems need to be built.
Maybe one more follow-up just on NVIDIA Clara Holoscan. You know, already the computer-based platform is already bringing real-time decision-making to a number of different treatment settings. And I mean, I, I would imagine that medical device manufacturers are clamoring to partner up. And any, any help just understanding how, how medical device manufacturers can get access to, to the platform and, and partner with NVIDIA and anything you can share that's in the public domain about who those some of those partners are today. Thanks a lot.
Yeah. Appreciate it. One of the early adopters of NVIDIA Holoscan was actually Medtronic. We announced our partnership with Medtronic. They see this computing platform as their future ability to innovate in a software-defined way. The very first application is their FDA-approved AI colonoscopy. They published last year that that AI allows them to see 50% more polyps, which obviously they wanna take care of during the procedure so that they don't develop into any cancer. And that is the first application. And now they have the ability to continuously add applications to it. And so this is exactly what we're building for the industry.
Instead of a Medtronic or a GE HealthCare or otherwise having to reinvent, re-architect, redesign the computing system itself, write all the system software, write the real-time application framework, NVIDIA has embodied. We've codified the last 16 years of our experience into this computing platform so that rapid innovation can happen on top of it. And so we have well over 30 partners that are working with the Holoscan platform, everything from the sensor technology to get the data in, so 4K, 240-hertz cameras that just bring amazing fidelity when you're doing any kind of this robotic surgery, for example, ultrasound, lots of sensor technology partners. We have hardware partners as well. So we create the system architecture. They can actually build the compute system so that it fits in an endoscopy rack or it goes inside your robot.
And then we have, of course, a whole bunch of solution development partners, so people that can facilitate the actual application development in a lot of the medical certifications. Now, we've built the system so that it's medical-grade in mind. So both the hardware is 60601 architected. The software stack is 62304, if you guys are familiar with these. And again, this is an effort to take a lot of that heavy lifting away or reinvention of the wheel inside of these otherwise hardware companies that are looking to transform their business into software-as-a-service companies. And this is a tremendous uplift for them. I mean, you see the economics that have gone on with the autonomous vehicle market. The absolute same economics, if not much, much more, will happen in the area of healthcare.
We simply don't have enough healthcare professionals to serve the community of patients. We need to make healthcare more equitable. We need this technology to reach more of the world. I mean, still about one-third of the world's population has access to imaging or surgery, right? And so the only way we're gonna get there is through artificial intelligence platforms. So we've done a ton of the heavy lifting on that side. We have over 40 reference applications out on GitHub in the open source. So you can literally get applications up and running in a matter of minutes. So it's a tremendous platform. Medtronic is an early adopter. We also partner with a really great startup out of France, Moon Surgical. They've gone from, you know, they're a very young company. 2019 is when they started.
They have approved essentially robotic-assisted devices on the market built on NVIDIA Holoscan in just a few, you know, 18 months to two years. And that's the kind of acceleration factor that you're gonna see in robotic surgery. No longer should it take 5 - 8 years to get a new type of device to market, because you have the compute platform already sort of architected. We should really, really see a massive innovation cycle there. So super excited about these next couple of years. And just like the AV market, it took off. We're kind of right at that moment where they're realizing that they can, you know, these medtech companies are realizing AI is right there for them to take and grab and be able to innovate and drive new business models for their companies.
Now, that stuff is super interesting. I wanted to shift the conversation just a little bit 'cause we got the little blinky lights to say we only have eight minutes. But our team in the semiconductor franchise and the tech franchise has done a couple of 400 or 500-page reports over the last couple of years with our biotech team on AI and drug discovery. And it's, I mean, I've spoken with Jensen, NVIDIA's founder, about this a little bit and talking about even things that go down to the individual patient's genome as the absolute first variable in drug discovery in certain cases, right? So it's, Steve, I know that's an area that you've spent a ton of time on.
Yeah.
If you have any questions for Kimberly, that'd be great.
Yeah. No, that'd be great. Yeah. No, I appreciate you doing this. So I was told, years ago by Sean McClain at Absci and kind of an early adopter of your chips that, you know, the chips were gonna be kind of the coin of the realm in terms of AI and drug discovery. But maybe just give us a sense of, you know, how broad and extensive your partnerships in the AI drug discovery is, and mostly in terms of, like, numbers. I mean, I appreciate you gave the billion-dollar reference. But, and then maybe talk about how those customers are using your chips, maybe beyond, you know, the kind of usual better shots on goal, you know, reduce time, reduce timelines to get, you know, clinical candidates. You know, any interesting ways that they're using your chips?
Yeah. Exactly. So, part of the NVIDIA Clara platform, we invented the ability. It's called NVIDIA BioNeMo. And this is an application framework that is speaking the language of biology. So just like I described, we needed to take what was the tools today that you use for natural language processing, and we needed to help them speak the language of biology. 3 billion-long sequence has a much different architecture going into a computer than our spoken word. So we had to understand the data formats. There's n different model architectures. They're GPT-like, but they're not GPT. When you go and you train a foundation model on, let's say, DNA, RNA, or otherwise, they're different model architectures. And so we're taking the state-of-the-art model architectures, and we're making them available in this application framework.
And then finally, what we're doing is there are literally many papers every single day now, every single day, many of them popping out of MIT but also out of, you know, Meta and otherwise on foundation models in biology. And so what we're doing is creating the ability for this industry, the biotech industry, the biopharma industry, who are not AI scientists, they're not computer scientists, but be able to leverage the AI, AI in computer science that's going on, taking the models that are being built, state-of-the-art models, and we're turning them into cloud services. So literally, a biologist can become an AI biologist just by logging into a website. So this is going to be a, a huge exponential in terms of their ability to access and leverage this technology. So we built BioNeMo, the application framework, and we built it for two purposes.
So one way that we're partnering with the industry is, most of the data in biotech and biopharma is proprietary data as it should be. And so we want to create the capabilities for them to create their own foundation models because essentially, what you're doing with foundation models is you're creating your IP. You're representing your company's intelligence in a model. You want that to be yours. You know, some of these companies go back 2,300 years. All of that data is actually part of the company's DNA or part of their intelligence. And so now you can codify. You can represent that into these foundation models so that they can continue to do their work. So the building of these models, large-scale training. So Amgen, for example, is one of the early adopters of BioNeMo.
They had all of their, you know, proprietary and, you know, very useful antibody data. You know, they're at the top of top of the tier for antibody design. And they wanted to create foundation models so that they could think differently about how they do the drug discovery process. They thought they were gonna take, you know, six months to a year to develop these foundation models. They developed five models in about four weeks on our platform. They took these models. They integrated them into their antibody design process. And I love the readouts that Amgen is doing. And actually, a lot of the pharma companies are starting to do it. They took their Design-Make-Test process that, and then they measured it. It's about two years long. And the output is about a 50% clinical candidate likelihood. Okay? That's their traditional approach.
They inserted generative AI upstream of a lot of the work that they do. They use generative AI to generate new proteins, new ideas outside of the scientist's thinking box. They use generative AI to predict their properties so that they're manufacturable, that they're non-toxic, all of the things that you need to do. And then they put it into their lab. And you call it there's this. This is coming into the industry really into view of this idea of lab-in-the-loop, generative AI, lab-in-the-loop, where you do use generative AI to really help you generate new ideas but fine-tune those ideas, then put them in the lab, every data point that comes out of the lab, inject it back into the models. And you kinda go in this iterative active learning cycle.
And what Amgen readout now is they're getting to reduce that time of Design-Make-Test from two years to nine months. And they're also increasing the clinical likelihood from 50% to 90%. And this is the idea behind our excitement in generative AI is this, this ability to what I say is play the game, not the score at this moment. Playing the game means, how can we change our strategy around drug discovery so that, yes, we can increase our chances of our, the ability to enhance what's going into the clinic to have a better success rate? I don't even think it's about the time. It's not even about the time, right? Because at the end of the day, this has to be safe and effective in clinical trials.
And so what we're really trying to do is open the search space, right, because it's essentially an infinite search space, and fine-tune our ideas by using these generative AI methods. So, Amgen is an early adopter. Late last fall, we announced our Genentech partnership. We work with all of the large pharma in Japan. We helped Mitsui build a supercomputer called Tokyo-1 that Astellas and Ono and Daiichi use. And, we're helping them leverage the supercomputer to build foundation models. So we have, you know, over 50 partners throughout the biopharma ecosystem are actually using BioNeMo. And then we have some of the deeper partnerships that we've talked about, Amgen and deCODE for foundation models in biology. We are just at the early innings. And that's the exciting part. The enthusiasm's fantastic.
And the readouts that are coming from the pharma companies are showing this has real value in terms of their drug discovery process. And that's when we, we say the words, "The next generation of computer-aided drug discovery." I think the whole process of drug discovery is transforming because AI is being injected in different places. And we're rethinking the serial process and looking at active learning in the loop process, to really facilitate drug discovery.
Awesome. Super fascinating stuff. I just wanna say it's a pleasure to be in front of this audience. Thank you, Kimberly, and your team for coming and spending some time with TD Cowen. It's not too often you see a $2 trillion company where they're telling you they're just entering their largest market. So, it's gonna be an exciting time. And, Kimberly's brought a few team members from her healthcare team here to the conference. I'm sure they'll be about so you guys can all meet them. But we really do appreciate the partnership, Simona, for you as well. And thank you so much, Kimberly, for your time.
Thank you, Matt, Steve, Josh. Thank you so much. Thank you. Appreciate it.