NVIDIA GTC 2026 Live

This is NVIDIA GTC Live. Over the coming days, the world's leading developers, engineers, scientists, and founders will gather here at the center of the AI revolution.

For the next three hours, we'll hear from the pioneers, the architects, and the builders transforming every industry. From robotics and research to global AI platforms, every layer of the computing world is being reinvented. All leading up to the moment we've all been waiting for. NVIDIA founder and CEO Jensen Huang takes the stage to define what comes

waiting for. NVIDIA founder and CEO Jensen Huang takes the stage to define what comes next. Welcome to GTC! So settle in, you're not going to want to miss a

next. Welcome to GTC! So settle in, you're not going to want to miss a moment. NVIDIA GTC Live starts now.

moment. NVIDIA GTC Live starts now.

Welcome to GTC Live. We are kicking off NVIDIA's four-day GTC conference, where developers, researchers, startups, and industry leaders gather to explore the latest advances in artificial intelligence and accelerated computing. Everyone's starting to arrive for Jensen Huang's keynote address. But first,

we are here to bring you the pregame show. We have the key builders of the age of AI here with us today. Get ready for a lively and in-depth conversation about the technology shaping the AI era, creating intelligent agents, robotics, autonomous systems, and the infrastructure around the world to power it all. Sarah Gua,

founder and managing partner of Conviction, an AI native venture firm where we partner with founders building the next generation of intelligent software companies. And I'm Gavin Baker, founder and managing partner at Atreides Management, a crossover fund investing in high growth tech and consumer companies with a focus on AI, CIMIs, and other emerging technologies. AI isn't

just a feature of the global economy, it's the engine, reshaping how we build, explore, discover, and manufacture on a planetary scale. AI has moved from experimental to essential. It's rewriting the rules of industry, redefining security, accelerating science, and

essential. It's rewriting the rules of industry, redefining security, accelerating science, and retooling manufacturing all in real time. This show marks the one-year anniversary of the first GTC pregame show. Last year, the pregame show was a small production tucked into a corner of a balcony here at the SAP Center. But since then, GTC has traveled

the world to Paris, Taipei, Washington, DC, and back again today to San Jose, where it's literally bigger than ever. Visionary CEOs, founders, and scientists are here to build the future, connect with each other, and you. And there's one common thread. It's all accelerated by NVIDIA. The GTC pregame will bring you five panel discussions with tech's biggest names,

by NVIDIA. The GTC pregame will bring you five panel discussions with tech's biggest names, covering accelerated computing, how AI is becoming essential infrastructure, how open models power the ecosystem, how agentic AI is at an inflection point, and finally, how AI is entering the physical world. No question, this is the place to be if you

want to know where artificial intelligence is headed. Just how huge has the GTC pregame become? What better way to show you than from the sky? And how

become? What better way to show you than from the sky? And how

awesome is it that we are outside this year, Sarah? It's amazing. And you know, this stadium usually houses tens of thousands of Sharks fans screaming about hockey. Now they're

all equally like same level of energy about token throughput, compute access. I love it.

Same level of energy. I mean, this is, I feel like this is next level.

I'll do respect to the San Jose Sharks. I'm sorry. Better level of energy. Yeah.

We have an arena filled with token maxers. And what an amazing production. To help

us cover it all is our special GTC correspondent, Tiffany Jansen, who's out there in Arena Green. Tiffany? Thanks, Sarah. Hi, everyone. I am here in the pregame lounge, and

Arena Green. Tiffany? Thanks, Sarah. Hi, everyone. I am here in the pregame lounge, and can we just take a moment to appreciate this? I mean, first of all, look at all these awesome couches, chairs, coffee behind me. Not like it's necessarily needed, though, because the energy here already is incredible. Now, I got here... I

think it was around 7 a.m. this morning, and there was already a line of people waiting to get in, which is incredible. And now you can see the line behind me. It's even bigger, and it's going to continue to just get bigger and

behind me. It's even bigger, and it's going to continue to just get bigger and bigger before the keynote. Which, speaking of attendees, you know, they come from all over the world. I go to a lot of different conferences and events, but I can

the world. I go to a lot of different conferences and events, but I can tell you with certainty this is the biggest. People come from everywhere. When you go into a room, the energy is just electric. You hear people talking about different business deals, future of technology, starting companies. I mean, you can't help but feel so

inspired. Now, today I'm going to be taking you along for some really incredible conversations

inspired. Now, today I'm going to be taking you along for some really incredible conversations with some of the biggest names in tech. We are going to be talking about everything from agentic AI, physical AI, which is a topic I am very excited about right now. I mean, over the past year, the idea

of, you know, a humanoid in my house being able to do laundry for me, I need that, like, really badly. We're going to be talking from everything from energy, autonomous vehicles, and so much more. Now, these conversations are going to cover not only what these thought leaders, these tech leaders are focused on right now, but what they are thinking about when it comes to the future of technology. We see it moving

so quickly. Last year at GTC, I remember thinking to myself, there's no way

so quickly. Last year at GTC, I remember thinking to myself, there's no way NVIDIA can outdo themselves. But this year, they have. The energy here, one thing I really noticed is last year, the kind of energy was more so, okay, what's next?

What's coming up? Now this year, it's more so... We see the results, we see how fast it's moving, where are we going to take this, what are we going to build next? And that is so exciting. And I think a lot of the conversations today are really going to cover that so you know where to put your focus, what's coming up next. And as I mentioned earlier, the robotics, that is something

that I personally am very, very keen on. Now, I think with Jensen's keynote today, it's going to really focus and shape as to what's coming over the next decade, really, where these technologies are headed, how these enterprises are building with them, which is something extremely exciting. Now, I can hear behind me the crowd,

the line. I mean, I remember last year there's some bridges that are going like

the line. I mean, I remember last year there's some bridges that are going like 10 minutes, 20 minutes past the venue, and people are still lined up. It's really

incredible. So stay tuned for some of those conversations I'll be having. But in the meantime, I'm going to throw it back to the stage. Thanks, Tiff. Looking forward to it. Accelerated computing is driving breakthroughs in science, transforming manufacturing, and advancing health

it. Accelerated computing is driving breakthroughs in science, transforming manufacturing, and advancing health care. It's turbocharging how engineers imagine, simulate, and refine the technologies that shape our world

care. It's turbocharging how engineers imagine, simulate, and refine the technologies that shape our world before actual fabrication even starts. What began as a way to render pixels has become the foundation of a new era of computing. And its moment has truly arrived. For

30 years, we've been advancing accelerated computing. We invented CUDA and we observed. Its moment has now arrived, irrespective of AI. It does image

observed. Its moment has now arrived, irrespective of AI. It does image processing, computer graphics, it runs SQL, runs Spark. It does data processing at speed of light for our government, for national security, and for enterprises around the world to find insight from it. EDA to SDA, System

Design Automation to CAE, We're able to do simulations at a speed and scale unimaginable in the past. This is

going to expand the market of computing into the world of design and engineering for the very first time. You tell us what you need to have run, and I'm fairly certain we have an amazing library for you. NVIDIA's GPU is the only GPU that can do all of that, plus AI.

Here to unpack this pivotal moment for Accelerated Computing are Mark Edelstone, Managing Director at Morgan Stanley, Dinesh Nirmal, Senior Vice President, Software at IBM, Aki Jain, President and CTO at Palantir U.S. Government, and Anirudh Devgan, President and CEO at Cadence. Mark, I thought we'd start with you. You've been involved with

Cadence. Mark, I thought we'd start with you. You've been involved with NVIDIA as long as anybody outside the company is the analyst who took them public.

It was quite a risk for you at the time, given 3D graphics was not maybe the best industry in the world. I'm just curious, what did you see in Jensen? What did you see in the company to make you take that risk? And

Jensen? What did you see in the company to make you take that risk? And

I'd love some anecdotes from the early years. Yeah, certainly it's been an amazing ride from the beginning. I first met the company back in 1996, and it was an interesting start. So they had just basically done their NV1 and NV2 chips. It didn't

interesting start. So they had just basically done their NV1 and NV2 chips. It didn't

go so well. And so I met some other people and began to think about them much like the other 60 or so 3D graphics and 2D graphics wannabe companies effectively that were coming to market at that period of time with their products. I

met Jensen a year later in 97, completely turned things around. It was pretty amazing what they did. They brought up their Reva product line that just completely transformed the industry. And so, you know, what I saw there was this company that had this

industry. And so, you know, what I saw there was this company that had this incredible view that you gotta build platforms. Basically, it's a combination of architecture, algorithms, software, compatibility, and do that over and over and over again. And so they really just changed the whole environment of what we had seen historically, where in this market

for graphics, everybody would leapfrog one another. And so companies could maybe hold together one or two generations of success, and then the next player would come in. And NVIDIA

really just changed that. And they changed that in part with a whole different approach to how they bring products to market. Jensen would basically call that three teams for two seasons. Back then, effectively, you had fall and spring refreshes for PCs

two seasons. Back then, effectively, you had fall and spring refreshes for PCs in the graphics marketplace. Ultimately, people would just miss a product window, and then therefore, they would lose their position in the market. That three teams for two seasons really changed it. You constantly could leapfrog yourself and do it faster and better than other

changed it. You constantly could leapfrog yourself and do it faster and better than other people. The vision at the beginning was basically that of computer vision that ultimately then

people. The vision at the beginning was basically that of computer vision that ultimately then transitioned into accelerated compute. And the story has just continued to get better and better ever since. So just tremendous combination of vision, technology,

ever since. So just tremendous combination of vision, technology, roadmap, execution that really has allowed this company to stand the test of time. So

at the end of the day, I didn't take that much of a bet. When

I met the people and understood what they were trying to do, it was pretty clear to me they'd be very successful. Wasn't that prescient to know that it would be the most valuable and most profitable company in the world, but you could see that they'd have long-term success. And now it's more like maybe nine teams for four seasons with all the products they have. It's such an amazing expression of a platform

company, and it was amazing foresight of yours, Marks. Dinesh, you have been part of a platform that's reinvented itself over multiple generations of technology. When did you start...

as a company paying attention to this generation of AI and say, we need to do it again? I think if you look at today, for the generative AI revolution as a platform to continue, I think one of the core elements is data. And the reason I say that is because if you look at the data in the past, there was always data as an endpoint, meaning you

have data stored in a database, you write a structured query language to extract the data, Today, data as tools, I call it, meaning you have MCP servers that is enabling access to data. There's multiple tools that's available for you to go access the data. Data is chunked and stored into a vector database, all

those things. And for the platform vision that you talked about to come alive, tomorrow

those things. And for the platform vision that you talked about to come alive, tomorrow I call it data as code or data as skills. The reason I say that is because, so let's say, you have a document that you want to extract something.

Today you just say, how much did I pay for my phone? It extracts from the document and you get an answer. Tomorrow you want to do the same thing.

When I talk about data as code or data as skills, what's going to happen is you can go to Cloud Code and say, create me a program or an application that enables to do that. That is the trick because that enables you to, one, flexibility, because now the next time you have the same set of questions

or different questions against the document, it can modify the code rather than regenerate the code. So the flexibility is there. Two, optimization and scale. Any platform, a data

code. So the flexibility is there. Two, optimization and scale. Any platform, a data platform for generative AI needs that. How do I optimize it? How do I scale it? And the third is the velocity and the veracity of the data that's coming.

it? And the third is the velocity and the veracity of the data that's coming.

I mean, if you look at enterprises, I would say 90% of the data is unstructured. How do you correlate with your transactional data is where the value is.

unstructured. How do you correlate with your transactional data is where the value is.

So that platform vision to come alive, I would say three things, right? I mean,

we have gone from endpoint today which is tools, and tomorrow how do I create data as tools, I mean data as skills, or data as code. So I think that transformation is what we are excited about, that's what I think the future is, and how do we unlock that value for enterprises where 90% of the data is sitting in unstructured, and how do we correlate to the transactional data to bring the

real value out. It makes total sense that you would have the early insight building, as IBM was, the data infrastructure for all of these enterprises and saying, aha, we finally see that, be it tools or skills, it's going to get a lot cheaper, a lot more accessible, a lot more powerful from here. Exactly. I think that's where the future is, because there's no shortage of data. I mean, the chats that you're

doing, the documents that you're creating, everything is data. But how do you really create the value for that data is where the enterprise is struggling. Because how do we make sure that there's so much transaction data happening, but the velocity of data or the volume of data is coming from unstructured side of things. So every

enterprise is looking to say, how do I really create value on the unstructured side, but also bring the transactional correlation to really make generative AI a reality, especially on real time? Awesome. Aki, first, thank you for keeping America and our allies safe. But second, would love to hear,

you know, Palantir is, I think, eponymous with AI at the application layer.

And would just love to hear when Palantir started, what AI could do, where we are today, and what do you think we'll be talking about next year at GTC from a capabilities perspective? Yeah, no, absolutely. Thanks for having me. I think kind of the points Nash made are really critical if you think go all the way back to the origin story of Palantir. We kind of as a bunch of software engineers

were doing in order to enable insights into that transactional and unstructured data. recognize

that we need to be able to treat data the same way we did software.

We needed things like revisioning. We needed the ability to actually put structure on it, similar to original object-oriented programming, and to make that really accessible to end users in order to enable humans. And at the time, AI was not mainstream.

There was a whole lot of AI exoticism back in the early 2000s. And unless

you were a really, really big internet company who had lots of data, you were Google. something like that, it was really hard to actually leverage machine learning in ways

Google. something like that, it was really hard to actually leverage machine learning in ways that drove ROI for your business or for your mission. So we really focused on the data problem and making data relevant to humans, curating it, and enabling large teams of humans to operate on that data together, similar to how software engineering teams worked.

And so if you fast forward to kind of like 2016 2017 more computer vision now on the back of Nvidia's investments and deep investments in CUDA and GPU All of a sudden became a lot more useful, right? This is when Elon said hey, you know We bring everything with computer vision autopilot and like we don't need these types of technologies We can do it in the low swap form factor on the

government side that became really interesting right and on top of that NVIDIA's GPU infrastructure, working with a number of companies, the ability to bring those computer vision models to keep soldiers and sailors safe when they're doing their missions kind of became a really core part of what we're doing. But the way you do that is with data, right? And both the unstructured, so unstructured isn't just text, it's video, it's audio, it's

right? And both the unstructured, so unstructured isn't just text, it's video, it's audio, it's pictures, as well as text. putting all that together with your transactional systems in order to enable the AI to actually make sense of your ecosystem. And I think where we are today with generative AI and a lot of where we've invested with AIP and closely with NVIDIA and the team has been a function of saying, we have

all these systems, we have the ontology, we have the ability to understand the organization through data right now. We know how to onboard humans to our organizations. How do

we onboard our AI teammates to work closely with humans hand in hand to actually accelerate a set of outcomes, but then to do it in a way that's governed, that's secure, and that ultimately scales, right? And we kind of think about it as we're moving into a generation of technology where the lymph act on scale for an organization is not just neurons. It's now electrons, actually, right? And the more electrons we

can pump through to actually accelerate and scale out our outcomes, the better. And that's

really kind of where we're focused right now with kind of the next couple of years. I am going to ruthlessly adopt Limfac. It took me a second to figure

years. I am going to ruthlessly adopt Limfac. It took me a second to figure it out, but I love it. Yes, limiting factor. But it would be great to hear a very specific example, maybe with your Ford deployed engineer model for the last three months where Palantir was able to really help kind of a mainstream Fortune 500

American company. Yeah, absolutely. So one of the things we've really started to invest in

American company. Yeah, absolutely. So one of the things we've really started to invest in is what we call AIFDE, right? So these are AI at Forward Employed Engineers. And

the whole concept of Forward Employed Engineers early on was to say a lot of organizations and companies don't understand how to model data, how to actually build an ontology in a way that's going to be useful to answer their long-term analytic questions. We've

all done this. We've built, you know, five-year schemas and roadmaps of organizing all our data and on day three you abandon it. It's very hard. Yeah, it's got to be iterative. And so really, you know, working a good example of this would be

be iterative. And so really, you know, working a good example of this would be some of our work with GE Aerospace as an example. In 2024, they're really focusing on actually modeling. So building the simulation modeling around a particular engine. They kind of presented this, I think, last week at AIPcon. They were basically explaining how they started with modeling and sim to actually have a digital twin of the capability. From there,

they actually took it and they built ontology around it to be able to understand all the parts components and different elements that went into it. And then they extended that all the way to their supply chain to have full end-to-end understanding of how should these engines perform when they don't and they're out of band. How do we actually start to get ahead of that from a supply chain perspective to build capability

to get ahead of it? If you can now do that not just with humans, but with FDEs, which we're really investing that are AI oriented, so it's a human plus AI, effectively an AI FDE, enabling them to now scale and do that for the next 10 components. So what used to take, it's all about compression, what used to take three months will now take three days because you have that ability to

scale out. That's a tangible example of something that we've seen in the last, just

scale out. That's a tangible example of something that we've seen in the last, just last three months leveraging generative AI. I have so many questions about reasoning in such a mission critical field and thinking about these different areas of engineering and operations that Palantir's working in. But maybe the other pinnacle of

humanity's engineering prowess is chips and chip design. Anurud, you run the platform that enables the platform, right? Can you talk a little bit about what Cadence has been doing to enable more innovation in chips and how AI has changed your workflow to date. I think everybody at this conference hears continually about coding agents

and the productivity and power that comes from them. But for this area of engineering, are you seeing the impact? Yes, absolutely. I mean, we have worked with AI and NVIDIA for some time. We have partnership with NVIDIA for the last couple of decades.

and it has really accelerated in the last five to seven years. And there are two parts to that partnership. One is what I call design for AI. Our software

is used to design NVIDIA's GPUs and most of the AI systems in the world.

And the second part is AI for design. We can apply AI to our own products, like you mentioned, and the effect can be quite transformative. I think one thing, of course, we've seen application of AI to coding. software, C++ coding.

But in chip design also, there is coding. There is system, Verilog, and RTL. It's

less known as registered transfer language. So once you have that, we have all kinds of tools over the last 20 years which optimize and generate the chip. But

in the past, we have not able to generate RTL code or RTL test benches.

So with the agentic AI flows now, we can generate that top layer. and then

put the automation that has always been there below it. So I always said for years that you have to look at it as a three-layer cake. And the top layer is AI, now agentic AI. The middle layer is ground truth, things like physics, chemistry, how molecules work. People forgot that's important too. And then the

bottom layer is compute and data, like it was mentioned. So any successful application, a vertical slice of this cake, By the way, the reason I call it a cake is unless you're two years old, you have to consume the slices together. So it

has to be AI plus ground truth, like the fundamental physics, chemistry, and then compute and data. And that definitely applies to chip design, just like it does to other slices of that cake. So Anurud, what will this mean in kind of the most specific way for semiconductor design? Are we going to have better

chips? Is the pace of progress going to accelerate? help us offset the slowdown or

chips? Is the pace of progress going to accelerate? help us offset the slowdown or into Moore's Law, depending on who you talk to? Are we going to be able to design more chips more cheaply? What is this going to mean? What does the future hold? Yeah, it has profound implications. So first of all, on Moore's Law, some

future hold? Yeah, it has profound implications. So first of all, on Moore's Law, some people say it's completely dead. Some people say it's alive. I think what you have to remember is what parts of Moore's Law is completely alive is area scaling. Maybe

performance is not improving. But the chips are getting smaller and smaller. You go from 5 to 3 to 2 to 1.4 to 1. So every generation, they're double the transistors on the chip. So that's likely to continue for the next 10, 15 years and further accelerated by 3DIC, just like Blackwell. You

can pack more chips in a package. So as a result, the size is going to explode. It's going to be 10, 20, 30 times bigger than now. And the

to explode. It's going to be 10, 20, 30 times bigger than now. And the

history of chip design is always the workload is exponential. See, sometimes people get worried, OK, if you are too efficient, does it change the way that things are done? Do you need less tools because tools are 10 times faster? But in

are done? Do you need less tools because tools are 10 times faster? But in

case of chip design, the workload itself is exponential. So we need 10x just to keep pace with that exponential. And this is the history of chip design. If you

look, Mark has witnessed this just like I have over the last 20 years. Our

software has made chips design 100 times better, but still there is more design and more activity. So I think to keep up with this exponential, we have to get

more activity. So I think to keep up with this exponential, we have to get 10 times better, at least 10 times better with AI, just to keep on that exponential complexity. Aki, I want to come back to you, just

exponential complexity. Aki, I want to come back to you, just because Palantir has been going for a while, serving some of these very risk-sensitive, or at least very serious customers. We talk about this period as the era of maybe the year of accelerated computing becoming mainstream.

How does more capacity or the exponential that Anirud and NVIDIA as a platform enables, how does it enable you to do more? Are you compute constrained in the use cases you have? Generally, we're not necessarily compute constrained, actually. I mean, really on the government side of things, we're generally capability constrained, right? Getting the same capabilities that we can pull out our phone and use into the hands of warfighters and analysts

and operators that need those tools to have some kind of compelling edge in their mission has traditionally been the area that we've seen be the limiting factor.

Sorry, when you say capability constrained, you mean in the product sense, like the military sense versus the model? Yeah, well, no, actually both traditionally. I think that's changed actually, which is why this is so exciting. I think if you, 15 years ago, when we were kind of getting off the ground, there was massive disparity in what was available to those users versus your traditional commercial or kind of public or private

sector users. Now what we're seeing is there's a very, very short lag. In fact,

sector users. Now what we're seeing is there's a very, very short lag. In fact,

it's incredible how quickly this technology is being leveraged. And so to that degree, then you do start to become compute constrained, right? And I think kind of on Arad's point, from a work perspective, there's only more work to do, there's only more data to analyze. These are not the things that we don't have. And so what's really

to analyze. These are not the things that we don't have. And so what's really cool we're seeing right now is that traditionally the problem has been there's been a lag in the software side, and there was tons of hardware, tons of comms. Now the software end is actually, the baseline is actually quite similar, and so now actually compute and scaled compute for the purposes of these missions are probably the limiting factor.

But the data, the mission, the space, the humans, those are all available in abundance.

Dinesh, I would love to hear from you. I've heard Jensen say that all enterprises in the future are either going to be token producers or token consumers. And I

think kind of the logical conclusion of this is if you're not a token producer, your fate as a business is going to be entirely determined by how efficiently and effectively you consume tokens as an organization. So I'd love to hear from you because I know how many companies IBM works with What are some common features

of your customers that are successfully consuming tokens?

So I go back to the data side of things. So if you look at enterprises, I would say there's three or four core things they look at. One is

security. Obviously, it's very critical. Scalability. How do I scale? And that's

where the tokens piece comes in because it's not just producing and consuming. How effectively

and efficiently are you doing that? Because there is a cost associated with it. The

third one is reliability, because once again, coming back to the token, if the answer is not correct, it causes even more damage for an enterprise. So if you are, you know, I go back to the billing question, which is, what was my bill last month and why did it go by $5? Now, when you ask that

question, obviously everything is a token, but when you ask that question, You're expecting the correct answer. Not somewhat correct answer, but accurate answer. If you don't get

correct answer. Not somewhat correct answer, but accurate answer. If you don't get that, you as a consumer is not happy. Immediately, what do you do? You cause

some damage by putting it on Facebook or wherever else. So for every enterprise, it's not just about the token, but the reliability of that answer becomes so critical. In

this world where we live in, where the data information flows so fast, It's so important that reliability. The last piece that I would say in that token producing piece is responsibility, meaning can I answer why that answer was given?

In the generative AI world, it becomes a bit more hard, and that's where the companies like IBM, Planter can come help because the explainability of that answer becomes so important. Why did I produce that answer, and how do I, if a regulator

so important. Why did I produce that answer, and how do I, if a regulator comes, especially in the regulatory industries, You've got to be able to explain why you made that decision or you gave that answer. So 100% right that token producing and token consumption is important, but the work that needs to get done, and

this is where Palandir, IBM, others are excelling, because you've got to take that data that's coming in, the ingestion phase, how do I clean that data, make it trustable, and make it available for these models to eventually for a consumer to use it as token consumer or producer. So one more example, and then I'll, so if

you go back to the past, right, it was all about events. Everything at an enterprise is an event, and events are causing a lot of data. But events also was a consumer and a producer. You have a producer of events, and you have a consumer of events, and that's where Kafka was born, Kafka took off. Same thing

is happening here, right? I mean, you have a token producer and a token consumer.

But in that process, the reliability of answers matter and the responsibility or explainability matters. So in the same way that effectively, you know, Claude code, to me the reason coding really exploded as this use case is people figured out how to have a really good harness for it that gave the model the context

and maybe not as much explainability, but there's a big opportunity to build harnesses for every use case imaginable. Correct. Even on that, to touch base, there's a really good point that you bring up, which is it's not about producing code. I mean, you can use cursor, you can use code code, you can use many tools out there as an IDE to produce code. But what is the quality of that code? Because

if I'm spending more time fixing that code, then I don't get the value. So

that's why many vendors don't talk about the quantity of the code, but they're more focused on the quality. Also, how do I get tasks done? Meaning I want to go from Java 7 to Java 17. What used to take me one month, can I do it two days. So I think it's moving from code production because everybody can claim I created 30,000 lines of code. But if that 30,000 lines of code

is not good quality, then you're spending even more time fixing it. So that's where the focus is going to be. How do I make sure? That's why I said the reliability matters. How reliable is the data? How reliable is the answers? How reliable

is the code? I think that is where the focus is and that's where we are really helping enterprises. As enterprises work through this hinge moment of struggling through the usefulness, productivity, interaction mode with these models, can we look ahead a little bit? Mark, you have been watching this company and this industry for 20 years.

little bit? Mark, you have been watching this company and this industry for 20 years.

There's never been more attention paid to semis and its impact on all of us.

If you were going to make a prediction on the next era of NVIDIA and what it means for all of the businesses that rely on it, what would it be? Well, I think we're still quite early in this revolution that's happening here on

be? Well, I think we're still quite early in this revolution that's happening here on AI. Obviously, it's been initially on training, moving to inference, but the whole world of

AI. Obviously, it's been initially on training, moving to inference, but the whole world of physical AI is yet to come. And when I look at what NVIDIA is so good at, which is driving parallel computing, taking advantage of their platform approach to things, where you really have to enable the entire ecosystem, that's what I think we'll see tremendous growth for them in the future. And, you know, Server's been amazing. You look

at this company and how they've evolved over time. It's been pretty spectacular. And I

think we're still early on in terms of this evolution. I can't wait to have my own multiple robots in my house. And hopefully that happens later this year or sometime in 27. Yeah, it is amazing. You were talking about the early days of having 60 plus competitors in the graphics space. It's kind of unthinkable to people looking at semis now that it would be so competitive. But the idea that you're just

chasing generation after generation, transitioning to a platform and an ecosystem.

What I'm hearing from you, Mark, is actually they're Jensen would tell us we're at the beginning of another S-curve around inference, and you're saying you definitely see one beyond that in physical AI. Absolutely. Yeah, I think inference is still very early days. We've

obviously taken a tremendous amount of infrastructure build that's been happening to drive training, but the reality is we all want to basically take advantage of those models. So that

will be infercing, but I think there's a whole other wave that comes past that.

And we see it now with early days of automated cars and things of that nature, but I think we're very, very early in terms of how we'll use all this intelligence at the edge. I'm a huge believer. Thank you so much, Mark and guests. It's been a great conversation. Our friend Tiffany will be with us all day

guests. It's been a great conversation. Our friend Tiffany will be with us all day with some amazing folks. Tiff, who are you with now? Thanks, Sarah. I'm here with Dan Ping, who is the VP of R&D at Siemens. Dan Ping, you know, AI isn't just transforming software. It's also transforming the physical world, and Siemens plays such

an important part in that. Can you expand on some examples of this at scale?

Oh yeah, sure. Tiffany, it's great to be on your show. You know, we live in a digital age. The cell phones we use, the cars we drive, the tools we use generate tons of data, which contains vital information to the success of a company. But to extract those information, we need a piece of smart

software, algorithms, so that we can make sense of the data. And that algorithm is called AI. You know, AI models like large language

called AI. You know, AI models like large language model has been very powerful and versatile, but at the same time, with its breadth, it comes with its weakness. It's like

Jack of every trade, master of none, right? It needs to be combined with the domain knowledge, and fortunately, we have this agentic AI. So today,

with agentic AI, We can actually build in the domain expertise of people working in the industry day in and day out. And then we can actually use the MCP, using the workflows, and we suddenly empower the people who do not necessarily to be tech-civic. This is a game changer, right? This is

very, very important. You know, at Siemens, we are in the manufacturing business. Our

software is run by thousands of companies manufacturers from the design simulation to the manufacture flow to the the Supply chain management and We know how hard it is to achieve that with AI with GPU We can do much more and do it faster right at Simons the at

CS the same as CEO Roland Bush and Jason was on the stage they announced that there is going to be a joint partnership between Siemens and Avigia in four areas. The AI in simulation, AI in supply chain management, AI in manufacture, and AI in EDA, right? So

I came from the EDA industry. I can say a little bit more about the EDA. So for years, the designers have been using the EDA software to design

EDA. So for years, the designers have been using the EDA software to design the chips that is faster and better. With GPU,

it's like a monster, right? Compared with the chips you find in the iPhone. And

today, we are using AI and GPU to design GPU. We do this by leveraging, you know, we run the physical simulation on the GPU with much improved speed. We

use AI to build the process model with unprecedented accuracy and speed. So

this is one area. The other one area is we do use generative AI, and genetic AI to control our workflow so that the software will think and take actions. That's very critical. It is. It is. And Danping, one thing you mentioned is just how quickly these workflows are speeding up now due to this. I

really want to cover quickly here about digital twins. Do you have an example of a company that is using digital twins and you've seen the workflows accelerate?

Yes. You know, at Siemens, working with the NVIDIA, say, you know, Omniverse, we built this tool called Accelerator. Now, the

teams across the globe can use this software in a virtual environment with full tool realistic in the world. They can change the design, they can tweak it, make decisions. That is great. Damping, thank you so much for that example and for your

decisions. That is great. Damping, thank you so much for that example and for your time. I am so excited for what Siemens is doing and what comes next. But

time. I am so excited for what Siemens is doing and what comes next. But

for now, I'm going to pass it back to the stage. Thank you, Tiffany.

Thank you, Tiffany. Just as electricity powered the industrial age and the internet powered the digital age, AI infrastructure is the transformative driving force driving the next wave.

AI requires a new kind of computing, purpose-built systems designed to produce intelligence at massive scale. Here's a look at how AI is becoming the foundational infrastructure of the

massive scale. Here's a look at how AI is becoming the foundational infrastructure of the modern world.

NVIDIA is no longer only a technology company. We are an AI infrastructure company.

AI is essential infrastructure.

How can you plan your infrastructure, your land, your shell, your power, your electricity, all over the world? We described our company's roadmap in enough detail that everybody in the world can go off and start building

data centers. We're clearly in the beginning of the build out of this

data centers. We're clearly in the beginning of the build out of this infrastructure. Every industry will use it and every country will have it. I'm

infrastructure. Every industry will use it and every country will have it. I'm

certain of that.

We're joined now by leaders building that global infrastructure. Michael Dell, founder, chairman and CEO Dell Technologies, our second year in a row, Michael. Michael Entretter, co-founder, chairman and CEO of CoreWeave. Lin Chow, co-founder, CEO of Fireworks AI. And Joe

Creed, the CEO of Caterpillar. Michael, you are a legend at understanding the hardware business that comes with each technical revolution. You were there for the PCs, for the cloud. Now with AI, you are building these gigantic data centers, and yet there's also renewed interest in having a ton of compute at the consumer edge

with OpenClaw and these sites of projects. How do you think about where the distribution of compute is going forward? I think it's going to be everywhere.

The debates over the years about on what side of the network does...

computing power grow? And the answer is both. It's kind of always been both. But

now with autonomous agents, certainly we see tons of distributed inference growing. And two years ago, we introduced the Dell AI factory,

growing. And two years ago, we introduced the Dell AI factory, right? And we had customers that sort of knew what they were doing, but a

right? And we had customers that sort of knew what they were doing, but a lot of customers needed help. Now we have this AI data platform with an orchestration engine to organize all the data, that's incredibly important. Of course, the infrastructure continues to evolve and expand at an incredible pace. And

we have the GB300 in the workstation level and increasingly powerful machines that can run these autonomous agents all the way to the VARA projects that we're doing with Mike and many others. And

it's incredible to see the pace of the build out and the demand for compute and inference is highly distributed. Tons at the edge, tons in smaller data centers, and tons in the larger data centers too. I think one of the things that's really striking for everyone on this panel is the new level

of scale you are figuring out how to operate at and able to enable this ecosystem, right? It's not a single unit, it's a factory, the complexity and engineering in

ecosystem, right? It's not a single unit, it's a factory, the complexity and engineering in that is something that you're enabling tons of other people to take advantage of in partnership. Yeah, and it's a combination of, of course, all the

partnership. Yeah, and it's a combination of, of course, all the advances in compute, but also the network fabrics, the storage, we build a whole parallel file system, our Lightning file system, the fastest file system in the world. You

gotta feed these GPUs the data super fast. And as the networks grow, and of course, KV caching is incredibly important as you get more complex queries and multi-agent systems. And so having this sort of layered memory system with KV caching is, In some ways,

the growth of the memory and storage and data is more than the compute itself because of the way the algorithms work. And a big plot inside the enterprise is how do you unlock all this cold data that has never really been used? And this is what we see happening with these 4,000 plus customers now. They're

used? And this is what we see happening with these 4,000 plus customers now. They're

figuring out how to, you know, activate that data and really make this real inside their businesses. But it's a lot more than just hitting a button and all of a sudden you've got AI. You've

got to organize the data and curate it and also work on the processes in a company to make it effective. But we're now seeing thousands of companies do this. And I believe we're still at the beginning of it, actually.

this. And I believe we're still at the beginning of it, actually.

You make this really good point. I was thinking about just the overall scale of the clusters that we are being asked to build today, but the fact that it is a memory hierarchy problem, a system software problem, a model caching problem, a power and dirt and electricity management problem, I think it's

incredibly amazing what the ecosystem has put together. And I think we sometimes overly focus on this one really powerful thing, which is how much matrix multiplication can we do? And importantly, bringing all the data together because in the previous IT architectures, you

do? And importantly, bringing all the data together because in the previous IT architectures, you had silos of data, hundreds or thousands of them inside a large corporation, and there was never any reason to bring all the data together. Well, now there's a reason to do it. Absolutely, and an insightful observer who

together. Well, now there's a reason to do it. Absolutely, and an insightful observer who we're gonna hear a lot from maybe, I guess, in 90 minutes, described it as a five-layer cake. And Mike, you are CoreWeave Mike, You are an important part of that cake and would love to hear. I think a

big focus of today's keynote is probably going to be the disaggregation of inference into pre-fill and decode. What could this mean for your business?

Yeah, it's a great question. It's going to be one of the important themes that we're going to be focused on here and kind of as we move forward. It's

like such a fascinating panel because it's... It's giving you, and we were talking about this in the back, a real insight into how broadly this is impacting every aspect of the economy. So it really does stretch beyond compute, stretches beyond data centers, down into the energy structure, all the way up into inference. And so, you know, it's a fun panel because it really kind of touches on so many of

those thematic requirements to deliver the AI factories. I get really excited about the deconstruction of workloads because it allows a company like CoreWeave to really kind of laser focus on each aspect of that to be able to optimize in areas that perhaps in earlier waves of computing it didn't pay

because it wasn't at the order of magnitude that required that level of special specialization.

As far as the depreciation question you're asking, it's really interesting because what will happen is as you go about deconstructing these workloads, the ability to route the portion of the workloads into those part of the computing architecture progression will allow for extracting

more value from existing infrastructure. And so what you're probably going to see is folks begin to pair older generations of GPUs to pick up certain portions of the workload while the most bleeding edge tech forward components of the infrastructure stack are able to really deliver on those

computationally enormous workloads. And so I think that, you know, talk in my book a little bit here, of course, but I think what you're gonna get is once again, a realization that there's a lot of different workloads that are coming to the market and a lot of those different workloads will

enable us to extract value from the installed base as well as the base that we're installing as we move forward through time, which leads to the ability to extract value in excess of the depreciation cycle that we're working on, which we're really excited about. We've built our company

around being able to really target how we go about using this infrastructure.

Absolutely. I believe firmly the disaggregation of inference is going to lead to an extension of useful lives of GPUs. Love that. Because you can put a decode optimized solution in front of a three-year-old hopper, a five-year-old ampere, and get a lot more tokens out of those old GPUs. That's gonna be great for GPU residual values, useful lives.

And who would have thought that the desegregation of pre-full and inference was gonna save the private credit industry? Because those four to five-year useful lives people underwrote to might be eight or 10 years, and those financing rates are gonna drop, which will be great for everyone. Yeah, listen, the more value that can

be extracted from each dollar invested into the infrastructure, you know, certainly the better it is for the lenders, the better it is for the credit stack. But it's also really great for new use cases that are coming into existence

stack. But it's also really great for new use cases that are coming into existence because what it does is it allows ideas and new companies that may have been struggling to get their hands-on or access to infrastructure to build whatever their new vision was or whatever their new company was. And it's incredibly

important for all of us to be able to seed that next generation of companies. And if they can't get access to the compute that they need to build

companies. And if they can't get access to the compute that they need to build their companies, to drive their companies, to monetize their ideas, it really It's bad for the ecosystem at large. And so I think it's really great as you move through the stack to extract more value from what we have already

invested. I mean, those dollars are out the door. They're in the data centers. They're

invested. I mean, those dollars are out the door. They're in the data centers. They're

in the servers. They're in the supercomputers. They're in the energy infrastructure.

We need to milk all the value from that we can. to be able to invest in the next wave. Absolutely. Getting more compute out of what we have is lower financing rates. It'll help us get the 1,000x more compute we need, the 100,000x more compute, or even more. We need to finance this. I think one of the things that is sometimes forgotten is that there is several magnitudes more incredible

engineers working on the efficiency of models and the ability to use one generation behind or smaller clusters in terms of running models that are incredibly powerful. There's two magnitudes more people working on this today. I think you have these two curves of people chasing capability and people making it accessible. And I

think the ability to benefit the entire industry from making it accessible is a really important thing. So, Lynn, if inference is gonna increase a

important thing. So, Lynn, if inference is gonna increase a thousand X, or as Jensen says, maybe a hundred thousand X from here, today it's still dominated by a couple large labs. Your business is about enabling other people to run their own inference. When does that transition happen? When do they begin to

care about owning it, or what other attributes matter? Right, we start to see increasing, significant, explosive increasing. adoption of customized inference.

Fireworks is on the critical path of delivering that value. And we provide customized inference with customized model for purpose built for applications. And before we talk about, like, 2026 is the year of efficiency, I would like to talk about 2026 is the year of explosive innovation, AI innovation that's happening. And

we can see that from demo to production scale, it drops from multiple quarters to multiple days. And OpenClaw Bot is just the poster child of the beginning of this journey. And Fireworks, we massively agentize our day-to-day workflow. This can generalize the whole entire industry,

where we build our recruiting agent to source top-tier candidates.

And our finance team build agents use Fireworks to automate finance forecasting and budget planning. Our legal team are using Fireworks to build their legal tech. We obviously develop inference software and we use coding copilot massively. And what we see is across the whole entire

industry, the tools are getting better. And therefore, agents can build more efficiently on top of new tools. And new tools get further better, and the new agents get emerged to replace old agents. So there will be a huge amount of energy in the application agentic space to continuously innovate. And

the whole ecosystem will take some time to settle down. And we're really happy to power all this innovation at light speed. Now, talking about 2026 as the year of efficiency, we have seen the explosive usage consumption of token from our own company and from all of our customers.

And it's kind of interesting phenomenon that now today, having product market fit doesn't mean having a durable business. It's very strange. Because in the typical SaaS era, having a product market fit, you just scale as fast as you can. And today, there are so many cases, whether they're startups or incumbents, they will literally scale to bankruptcy.

So this is because fundamentally, inference is not as cost efficient yet as before because the whole entire supply chain, we are going through the evolution of being operational efficiency, all the way from the software stack to hardware to data center to energy, the whole entire system.

going through this revolution. I think we are all working together. Everyone is all working together to bring down the cost to the cost of ownership. We are

approaching this in a very unique way. What we think about this is we don't think application or agent builder should be API wrappers. They should

activate their private data produced by their application or locked inside enterprise to bring their model to the next level of intelligence.

And use that to also make the model faster and a lot more cost efficient.

So this is something we call three-dimensional optimization that was produced as a unique way of approach a problem. And we can bring down your TCO by five to ten times. At the same time, we're also working on cutting edge technology. We're thinking,

ten times. At the same time, we're also working on cutting edge technology. We're thinking,

why we separate out model training from inference? We should all mix it together.

While your model works with your application to do inference, your model should continue to learn the specific pattern from your application, therefore to make it faster, better, and more cost efficient. So we are working on a lot of those cutting edge technologies to

cost efficient. So we are working on a lot of those cutting edge technologies to bring customized inference into fruition. And obviously, our innovation will not happen without significant innovation from all of the technology at basic infrastructure level. And particularly, we are very excited about the new

and Grog integration because this is a massive disaggregated deployment at hardware level.

But we also think, hey, when we blend training and inference together, we can also do a lot more aggressive disaggregation, as in every single bottleneck, they shouldn't be blend together. They can scale completely independently to drive the maximum amount of efficiency.

together. They can scale completely independently to drive the maximum amount of efficiency.

So this is an extremely exciting time for us and to see all the impact building on top of us and work with all of you to drive the innovation here together. Awesome, so Joe, thanks for being with us here. The world is

here together. Awesome, so Joe, thanks for being with us here. The world is structurally short wafers and watts. And we just heard a lot from your fellow panelists about how we're gonna get a lot more tokens out of the wafers we're making because of the new technologies that we're gonna hear about in a little bit. More tokens per watt means Jevons Paradox.

going to consume way more tokens and way more watts so if you thought the world was was um lacking in power before um you know wait you know look out look out 12 to 18 months but um i know you guys have been increasing capacity um so joe can you just talk about how caterpillar you know in in that five layer cake nothing can happen without that uh bottom layer

of Watts. Sure, I mean, I think it's amazing, again, this panel to be up

of Watts. Sure, I mean, I think it's amazing, again, this panel to be up here. You think about founders of tech companies and everybody that's here, and that as

here. You think about founders of tech companies and everybody that's here, and that as a hundred year old industrial company, it shows really how widespread AI and what we're seeing with technology in the world. I mean, the infrastructure build out is gonna be like nothing I've seen in my almost 30 years of career at Caterpillar. And I

love how Jensen lays out the five layers and you talk about energy and chips and infrastructure, those first three layers. I mean, that's, that's what we enable at Caterpillar, right? We are construction, we are mining, and we're power and energy. So when you

right? We are construction, we are mining, and we're power and energy. So when you think about what our customers do, we need critical minerals for chips, that's through mining. We need iron ore and copper to build data centers. Our construction equipment

through mining. We need iron ore and copper to build data centers. Our construction equipment builds the data centers. Our power and energy equipment, we're massively short on enough energy and watts to keep up with this build out. So we kind of refer to ourselves as the invisible layer and kind of supports all that infrastructure build out. And

I think it's just an incredible time for a hundred year old company to be looking at what we have in front of us and how we play into this, into the bottom three layers of that. What's also kind of fascinating is for Caterpillar, we're also using the top two layers. We're big users of the technology as well.

So when you think about, we just launched an AI assistant for our customers, we're gonna launch it in the cab of the machines to be an operator coach later this year. And so our mission is to solve our customers' toughest challenges. And the

this year. And so our mission is to solve our customers' toughest challenges. And the

things that I hear from our customers are, hey, first off, job site safety. Same

in our factories, we want our employees and our customers to be safe. Shortage of

operators, we need to be able to quickly train operators and have coaches in the cab for them. And then also job site efficiency. There's a lot of inefficiency in our factories and on job sites where we can use digital twins and AI to really unlock a lot of value here. And so, as users of the technology, I mean, we are just at the cusp of really getting out there, putting people at

the forefront and moving them from the dangerous jobs to where they're overseeing equipment working instead of necessarily in the cab. And just having coaches. I mean,

one of the things Mike and I talked about this last night, there's a lot of talk about the power and having enough power infrastructure to keep up with the growth that we have, but we need more skilled labor. Our customers are struggling to get operators, we're trying to help them through autonomy solve that challenge, but we need more electricians, we need more technicians to work on our equipment. And so we're making

big investments in the workforce at the same time because right behind power, I believe, is having enough skilled labor to really keep up with this huge demand that we have for infrastructure across the globe. That's awesome. So AI is not only creating jobs, it's creating better, safer jobs. Absolutely. And we need that AI to meet the demand we have for watts and wafers. Absolutely. It makes sense that you would be

investing in these supply gaps on labor that we can't solve.

Beyond that, how else are you making Caterpillar go faster? Because you

can't train new people to match a 1000x or 100x demand. What else are you doing to compress the bottleneck in the physical construction? Yeah, so we have a number of things going on. One, we're just trying to add enough capacity to keep up with the product and demand that we have. But internally, changing how we work and how do we use technology and now we're moving to inference. How can we embed

it inside our company to help our people develop product, faster, how do we take the inefficiencies out and get people off of sort of mundane tasks? You'd be surprised as a hundred year old company, the processes we have and how many opportunities we, I think, that we have in front of us that we've barely even scratched the surface on of really making our people more efficient and kind of elevating them out

of what I would say is kind of mundane work and putting them into where they use their mind and creativity to really move the needle. So I think for us, it's going to really be super impactful for our culture and just how our people work every single day, both in the office and the factory and in the field. You think about our technicians that have to support and our dealers technicians supporting

field. You think about our technicians that have to support and our dealers technicians supporting customers in the field, having quick access to an AI assistant to diagnose problems, remote monitoring and diagnostics for us. So they show up ready to get customers back to work. And that's what we're focused on. I think it's a really important point because

work. And that's what we're focused on. I think it's a really important point because organizational structures and systems are a function of the tools that were available at the time they were created. And now we have this incredible breakthrough technology that is evolving rapidly, continuing to advance. And so how you get to the outcomes in 2029

is very different than you did in 2019. And so you really have to reimagine the entire process processes from a tops down standpoint to get to the outcomes and then you get incredible speed inside your company and you better do that or else you know you're going to get destroyed by somebody who does 100 the third

quarter was actually the first time that we saw kind of real um S&P 500 companies in a variety of industries talk about AI-driven productivity, have it show up in their financial statements and have the stocks go up significantly. You know, C.H. Robinson, a

trucking company, was a big AI winner in the third quarter. But it'd be really helpful, kind of maybe a little bit of a lightning round. You know, maybe Michael, since you posed it, you could go first. But what's the most specific, mind-blowing productivity gain you've seen at your own company in the last six months through AI, where you were just like, wow, I didn't think we'd get that until 2030

and it's here now. So many examples, I could take it the rest of the panel, I won't do that. But the one that I'm most excited about is the speed of innovation from idea to product delivery in our product groups, right? So 20,000 plus engineers enabled with the latest tools

and they're able to do things in two weeks that would take nine months, right?

And so that is making us a far better company and we're just innovating at scale much faster and that is super valuable. And when

I think about all companies sort of on that path, I mean, that's just an enormous speed up for drug discovery, solving all sorts of unsolved challenges out there. And ultimately, as we get these better tools, we're going to do more. The economy is going to expand. Absolutely. Michael, are you seeing that kind

do more. The economy is going to expand. Absolutely. Michael, are you seeing that kind of speed up in such a real world, operationally intensive business? Absolutely.

So one of the great things about this panel here is that, like, for me, is that I'm getting affirmation that every single part of the stack is seeing similar types of challenges, similar types of hurdles that they have to address in order to maintain the pace of the build out of infrastructure. And whether or not we

ultimately hit Jensen's lofty numbers of how much more compute is required or not, directionally, that is where we are going. And so the challenges that all of us are going to face are going to be sustained. One of the things that I do believe is that when you introduce new technologies like this, they tend to come from

the fringe inward towards the core of what a business does, right? And that there's numerous different- Who's the fringe right now? So you start out with, you know, hey, I'm going to make my email a little bit better. I'm going to do this.

I'm going to do that. And then all of a sudden you start to see AI presenting itself through engineering, right? In the coding and you know, when you ask me about like, you know, what's astounding or what's magical around artificial intelligence, when I sit with my CTO and he says that like, basically, unless I'm on

a phone call, I'm 10x more effective. These are pretty talented engineers to be able to magnify the impact of those engineers who are building this by 10x unless they're working with a client, going through the kind of normal cadence of how you build a business, that's incredible, right? And so my expectation of how our company will feel

the impact of this was that you start on the margin and kind of work your way down towards the core of what a business does. And so you're going to see just enormous strides in terms of the velocity with which we're able to drive our technology stack, the software layers that sit on

top of it, how we go about configuring our data centers to optimize and future-proof them so that we're able to support not just the current iteration of technology but into the future. You see it working its way through HR so that we are not hiring for tasks but really hiring for people that are going to be able to carry the company forward. All of these things are driving more productivity

and more scaling within our organization. And I think that that's probably a progression that you see across industries. And I'm, you know, like we're still years away from where it's going to really get to the core of some of our heavy industries. Just gonna take some time and it's just really exciting. And absolutely, I

heavy industries. Just gonna take some time and it's just really exciting. And absolutely, I think we're tracking well ahead of Jensen's lofty predictions from previous GTCs. So let's hope we stay on that path. It'll keep us busy. Thank you so much to this group. You guys are the builders behind the builders, so we are counting on you.

group. You guys are the builders behind the builders, so we are counting on you.

And thank you to my awesome co-host, Gavin. Thank you, Sarah. Thank you, team. Thank

you. Thanks, everybody. Thanks, guys.

Let's check in with Tiff and see who she's talking with backstage. Thanks, Sarah. I'm

here with Claudia Blanco, who is the Chief Innovation and AI Officer at GE Vernova.

Claudia, when we think AI, we're really starting to realize just how much AI and energy are converging. What are some opportunities there for GE Vernova? Thank you for the question, Tiffany. And first of all, I'm really glad to be here today with all of you. So we are in a unique moment in time, right? So we

see large... AI data centers are demanding more power, more energy, and is where we came into the intersection of things, right? So we need AI for AI, actually, is what I like to say. So how those large data centers will require energy when the power system requires the combination

of things to happen, right? So from how to generate more energy to how to transfer this energy into with more capacity, but also how the distribution can be done more efficiently, right? And also combining with sustainable solutions. So

it's a large mix of things that has to happen at the same time to make it work. And AI at the same time play a key role. So at

the time that we have to provide this power system and all that is required to make it work, it's also very important to know that AI can help a lot to manage the power systems, right? Absolutely. I love that you said AI for AI. I think I'm going to steal that from you. I'm going to use

for AI. I think I'm going to steal that from you. I'm going to use that one. I'm curious to get your perspective on this. What are some parallels when

that one. I'm curious to get your perspective on this. What are some parallels when it comes to the energy grid and AI infrastructure that you're seeing?

Very interesting question, because actually we see that parallelism. If we think about how the power system infrastructures were built 100 years ago, it was a unique moment as well. Many things had to happen, right? From the power architecture to be built to

well. Many things had to happen, right? From the power architecture to be built to also the transmission systems and also the subsystems to the distribution, as well as the automation and control at the edge. So all of this had to happen 100 years ago. And we see something very similar today with the AI

infrastructure, right? How the large data centers will require a combination of multiple things to

infrastructure, right? How the large data centers will require a combination of multiple things to happen as well. That will require a big transformation. So how to merge the power supply, the power systems with computing, with also, you know, all that is going to be the cooling systems and the infrastructures management of an AI data center, it will

require also a big transformation. So I see very important parallelism there.

Absolutely, I think it's so important to be able to look back, you mentioned a hundred years of energy grids and taking learnings from that into our future and what we are building next. I'm curious to get your take, how is software really reshaping energy management? Yeah, as well as

well this is something that is going to be critical for You know that with AI actually we can give for the first time the capabilities to our operators to manage the grid, to manage the power system in a more intelligent way. So which

means that if we provide this intelligence, power system is a complex system. So it's

a large amount of data to be managed at the same time that system, hardware, physical system at the edge, and then how can you operate that in a larger scale, right? So AI will provide ways to handle this massive data

scale, right? So AI will provide ways to handle this massive data in a more comprehensive way, more intelligent, and it's where it will play a key role. And connecting to the previous question, I think the AI factory concept play a

role. And connecting to the previous question, I think the AI factory concept play a key role here. because we will start designing from the beginning, the infrastructure require end to end, merging all those systems, power systems with what the AI data centers will require. Absolutely. Claudia, thank you so much for shedding some insight on that. I am

require. Absolutely. Claudia, thank you so much for shedding some insight on that. I am

very excited to what the future holds, but for now, I'm going to throw it back to the stage.

Thanks Tiff, we'll check back with you in a bit. I'm now joined by Sequoia Capital partner, Alfred Lin, who will help lead us for the remainder of the show.

Welcome Alfred. Thank you Sarah, thanks for having me here. Thanks Jensen, thanks the NVIDIA team. You guys did a great job leading up to this. This is awesome. It's

team. You guys did a great job leading up to this. This is awesome. It's

a bright day, it's sunny. It's awesome to spend time with you too, because usually I spend more time with your husband. Well, you know, one of us has to work today. It's great to be here with a friend, a OG NVIDIA investor,

work today. It's great to be here with a friend, a OG NVIDIA investor, and a co-traveler as we try to figure out what's happening in the next wave of the ecosystem, which we're going to talk about now. Yeah, well, Sequoia has been very, very fortunate to be an early investor, seed investor in NVIDIA. And Mark Stevens made that investment, and he's still, even though he's retired from Sequoia, still on the

board. And so we're very, very fortunate to be associated with NVIDIA. I wasn't there

board. And so we're very, very fortunate to be associated with NVIDIA. I wasn't there when the NVIDIA investment was made. That was just genius. But when I joined Sequoia, I did check NVIDIA stock price, and it was a quarter. I think Jensen's added $180 to that, which is a 721 times

return in the last 10 years, 15 years, 16 years. So it's amazing. It's an

amazing company. It's pretty cool to see, also, hearing from Mark at the beginning of the day, what the journey is from a competitive market of 60 vendors fighting about graphics to platform company with three S-curves in front of it, and we're maybe in the beginning of the second one. I cannot wait to talk to some of the companies we are speaking to in the rest of the pregame about

the platform journeys they're going on. Yeah. And you guys set it up so well.

We had the accelerated compute, the infrastructure, and now we're going to talk about open weight models and systems about AI agents, and then we're going to end with physical AI. Great to have you here. Before we get to the next chapter, Tiffany's got

AI. Great to have you here. Before we get to the next chapter, Tiffany's got another great guest on deck. Tiff? Thanks, Sarah. I'm here with Alex Atala, who is the co-founder and CEO of Open Router. Alex, why is Open is so important when it comes to AI innovation? All, a vast majority of the economy is gonna

move towards inference and compute. And a major principle we wanna make sure is true in that future is that we have competition in the economy around inference.

Open models allow multiple companies to compete to build incredible inference for the models that come out of Model Labs. And what that results for the end consumer and for businesses is faster inference at higher quality and more interesting integrations. And you just need an open economy to enable

a really competitive future for that. And open models are the way to do that.

Absolutely. I couldn't agree more. Open models. On that note, when it comes to developers, how are you seeing developers navigate so many different models today? Developers

tend to follow the vibes, which most people are familiar with. I like that.

They do look at benchmarks, but we find that happens when a new model launches.

We've observed something that we call the Cinderella effect, which we wrote a paper about last year, where the first cohort, the first week of developers, there's a section of it that figures out what the model is good for. And they stick to it. They have the highest retention among all cohorts that follow them in the

to it. They have the highest retention among all cohorts that follow them in the future. So they discover these novel use cases or what these open models are good

future. So they discover these novel use cases or what these open models are good at, and then they stick to them. And what we do is try to expose those insights to the rest of the world so people can see what the power users and what the LLM enthusiasts are doing with these new models. We also see that closed models tend to boost the total addressable market for a particular use

case. And then open models come in and help people reduce costs, which becomes a

case. And then open models come in and help people reduce costs, which becomes a huge thing when you realize like, wow, our costs are ballooning. All of our developers are doing crazy things we didn't expect. Absolutely. And was this Cinderella effect? Is that

what you call it? That's what we called it. I like it. I need to read that paper. That sounds very fascinating. I'm curious, what startups are you starting to see emerge? We're seeing a lot of startups try to

see emerge? We're seeing a lot of startups try to re-envision how to organize a company now with AI agents.

Like what will it look like when you actually try to give them responsibility and when you try to give them full context? Because these two things are the two big gaping holes in AI today. Like AI agents don't have responsibility for anything and they don't have the full context of the company. So we're seeing more startups try to work on those two problems. We're seeing a lot of startups just integrate, try

to get more integrations to AI agents so that they can act on behalf of humans in different places. We're seeing a lot of interesting agents emerge to just help developers improve their productivity. help developers manage multiple workflows that are happening simultaneously. A lot of startups are realizing, whoa, I have

the problems that 100 person companies usually have. And much more. When it

comes to AI agents, what is an example, just a quick specific example with AI agents that you're seeing a startup use them for? Very specific example, well, I could just think of all the cases our own company is using them for. Yeah. I mean, we're building agents to help our sales team prep

for. Yeah. I mean, we're building agents to help our sales team prep for calls, very bespoke agents that run automatically before the calls happen so that they know everything they need to know about who they're talking to. We have agents around auditing for security. We have like security audits running constantly all the time on

all PRs for code that was merged in a week ago. And we have, Now we have interesting roll-up agents that are just running ambiently in the background and looking for interesting insights in our data. What apps are emerging? We have data about public apps,

apps that opt into public visibility, and we expose all this data to the world in our new apps model. rankings and in our models rankings. Thank you for those examples. I love hearing specific examples because it really gets me so excited as to

examples. I love hearing specific examples because it really gets me so excited as to what other companies are working on, what people are thinking about. And we're going to continue a lot of these great conversations. But for now, I'm going to give it back to the stage.

We'll check back with you soon. AI's impact won't come from chatbots alone. The real

breakthroughs happen when developers take open models and build on top of them, adapting them to new industries, new problems, and entirely new kinds of applications. That's why open models are so important. They give developers, startups, researchers, and enterprises the freedom to experiment, specialize, and push AI into new frontiers. So what happens when open models meet a global community of builders? Let's dive into our next chapter.

Open source models have become incredibly useful for developers. They are now the lifeblood of startups.

Researchers need open source. Developers need open source. Companies around the world, we need open source. And so NVIDIA is dedicating ourselves to go do that.

source. And so NVIDIA is dedicating ourselves to go do that.

Each one of the industries have its own use case, its own use cases, its own data, its own use data, its own flywheels. All of that capability, that domain expertise needs to have the ability to embed into a model. Open source makes that possible. Here to discuss how

possible. Here to discuss how open models drive innovation, expand the AI ecosystem are Aravind Srinivas, co-founder and CEO of Perplexity AI. Arthur Munch, co-founder and CEO of Mistral AI, Robin Rombach, co-founder and CEO of Black Forest Labs, and Aidan Gomez, co-founder and CEO of Cohere.

Welcome. Welcome, everyone. First, let me start with Arvind. You had a very, very successful launch with Perplexity Computer. and you've described it as the personal computer that should exist in 2026. I wish I had that, where it runs 20 agents simultaneously, helps you do a whole bunch of things. Tell us about your

launch, tell us how it's going, and tie that back to open models. How are

you using them, and how's that helping the ecosystem? Yeah. So Perplexity Computer is an orchestra conductor of everything AI can do today.

It runs in the cloud. And all the musicians in that orchestra is basically all these different frontier models, the connectors, they're playing music that's equivalent to the work that you want AI to do for you on the computer. One other

way to think about it, as a person doing work on your computer, what do you really need? You need a browser, you need access to the internet. So giving

AI's access to the internet was the core idea of perplexity. And then giving AI access to the browser was the core idea of our Comet browser. Now we complete the puzzle where we give them access to a file system and a code execution sandbox and 20 different models, hundreds of connectors, a lot of CLIs. Turns out when you do that, AI can become a director directing all these sub-agents to do all

your workloads for you every day, asynchronously, always on. So that is the core idea of Perplexity Computer. Open source has always accelerated new applications and open models naturally fit into computer. We're already using models from Alibaba, the Quinn series, and we're gonna be using a lot more. So models in our mind and

as far as computer is concerned are just tools, just like the file system or the code sandbox or these connectors or the CLIs, models are just tools. What actually

matters is the orchestra. And how are people orchestrating So many different applications. People are building their own mini versions of Bloomberg Terminal that's connected to their own portfolio that they have access to the Plat connector. People are

building web applications that famously monitor the situation around the world, whatever is going on.

Because you have access to the internet, you can create your own live streams of what's going on. You can build applications that pull internal context. Our entire company, Perplexity, is just running on Slack where this computer is yet another person on Slack that is just orchestrating these workloads. We estimate around $90 million worth of labor for

our max users in the first couple of weeks of launch of the enterprise product.

That is really terrific. There's such massive increase in continuous agent usage that obviously requires compute behind it. Arthur, Mistral has always had models that punch above their weight from an efficiency perspective. We just had a panel talking all about 2026 as the year of efficiency. Where are you seeing

demand for this in your own customer base where people care not just about max capability, but efficiency? I mean, the demand comes from all of our enterprise customers. It

depends a little bit on the vertical, but efficiency matters when you're deploying agents at scale, and then suddenly you care about token cost, and you care about the unit economics, and that's where we build open-weight models so that our customers can have cheaper access to that technology. So it matters for cost. It also matters for the UX and the experience generally, because when you have a small model, it can actually go

faster. And if it goes faster, you can solve your task faster as well. So

faster. And if it goes faster, you can solve your task faster as well. So

that aspect around latency is actually critical. If you lower the latency, you can actually change the kind of applications you're building. And so when our customers are building for their own customers, it typically is a problem, because they want to have good UX.

And good UX means to have specialized models that are small enough. It's a big use case as well that we target quite significantly and that we work with a lot of companies, which is edge deployment. And edge deployment is about mobile deployment, but oftentimes it's also about putting models into robotic systems, into industrialization systems, into factories.

And so you need to have these small models that have audio as input, that have video as input, and that can process that, and that can do it in a way which is fully disconnected. So we're actually announcing today Mistral Small 4, which is really punching above its weight in that respect. And it's a reasoning model that can actually be deployed on the Edge in that respect. Congratulations, that's super exciting. I

think one of the things that is somewhat forgotten when we talk about use of models is you want this iterative loop, either in the models generation and verification capability itself or between the human. I haven't been waiting for computers for, I don't know, a decade and a half before the last few years, and I'm waiting all the time, right? And so I do think efficiency is going to be a huge front

time, right? And so I do think efficiency is going to be a huge front of competition. If you put it, it's super important. And the other thing that is

of competition. If you put it, it's super important. And the other thing that is important is also to be able to modify the systems and to learn from interactions in between user and the models themselves. And that's where having open-weight models and a way to access the parameters and to change them to steer the behavior is actually critical. When you go into the depth and the productivization of the AI systems, you

critical. When you go into the depth and the productivization of the AI systems, you actually need that deep access. Thank you, Arthur, on the importance of latency. Latency is very, very important for the creative world, for the generative AI world. And Robin, for Black Forest Labs, how are you

reducing latency? How are you supporting the creative community in their quest to make better

reducing latency? How are you supporting the creative community in their quest to make better and better content out there? Yeah. I think for the creative community in particular, open models are super important because they are more formable.

We see that actually. I kind of want to disagree with this take of models are just tools. I think models are a super important substance for the creative community. You can fine tune them, you can interact with them in all different kinds

community. You can fine tune them, you can interact with them in all different kinds of ways. It's not just input, output. You can take them apart, you can plug

of ways. It's not just input, output. You can take them apart, you can plug them into workflows, node-based systems, stuff like that. And then they go way beyond creative community. It's like a basis for research, for innovation. I think

community. It's like a basis for research, for innovation. I think

it's fundamental technology and really not just a tool. For latency, yes, 100%.

It's one of the most important aspects of all kinds of models today. We work

on it, like a lot of like, distillation research that we do. And yeah, I think we are pretty good at getting the number of inference steps that you need with a diffusion model or flow matching model down to, let's say, four or something like that, which then really helps in combination with stuff like quantization to make these models really efficient. You mentioned diffusion models. They started the whole wave for creatives and

creative community to build things that they didn't think they could build as efficiently and as powerfully as before. What's next? What's on the horizon? I think

it's really interesting that we see how especially visual models expanding in scope from content creation to simulation to physical AI applications. And it's like the shared technology underneath is like... Or the technology underneath is shared, right? It's like

the same or very related modeling approaches. But we're approaching new scales of data, of models and multi-modality that then enables models to expand into what we find really, really exciting physical AI simulation use cases, real-time video generation.

Aidan, whether we look at these models as just tools or a moldable substance, as Robin was saying, they've been relatively hard to work with.

for most of the industry, even the engineering audience for the past few years. What are you seeing in your enterprise customers and the people you interact with about sophistication in terms of post-training and customization now? It's

increasing dramatically, and I think open source models are contributing to that because it gives the customer the freedom to explore and experiment themselves without having to depend on any other organization. One of the things I'm most excited about with open models is the

other organization. One of the things I'm most excited about with open models is the notion of control. So Cohere does a lot of work with high security settings. Privacy

is not just important, but essential for the customers that we work with inside telco and financial services, et cetera. And what open models do is give the customer the ability to choose where and how they deploy. So when we're delivering that technology, when we're delivering north our agentic platform, We have to make sure that our models

can be deployed on-prem, air-gapped, on-device, anywhere. And open

models enable that. In terms of customization, I think that these models, they're fantastic general reasoners. They can be applied to pretty much any domain, anything.

But there's a production gap that needs to be closed. There's a level of performance that you need when you're deploying into these extremely important, high-stakes scenarios that you're not going to get out of a generic model. So the ability to customize and close that gap is essential to actual deployment and scale of the technology into the economy and the ecosystem. I have a hot take question. There's

this meme in VC that closed is easier to monetize, open is easier to mobilize. Do you agree with that? Is it easier to get distribution? Is

it easier to work with so that you're not locked in because you're having to pay one or two closed models? What do you think?

Well, I think distribution at the end comes down to value you provide. For

example, there were a lot of open models, but you could argue that the pivotal moment in open source among model providers was when DeepSeek surpassed OpenAI 01.

That's when everyone started to go bonkers about it. And I would say the same thing keeps repeating. Kimi, Moonshot, I think even Elon called them out for their work yesterday night. So in general, closed or open, the capabilities are basically what fundamentally define distribution. And that's kind of why we've taken the position that

we would use whatever is the best capability for each model and put it all into one orchestrated system. One advantage obviously in open source is cost. Because you can take the model and do your own inference. you are obviously

cost. Because you can take the model and do your own inference. you are obviously gonna have a massive cost advantage serving that over paying a closed model provider.

But what usually has happened over the last two years is every time there's a great open source model, the closed model providers always lower their costs, which is good for application layer companies like us. And we very much welcome that competition.

In terms of a follow-up to that, Arthur, since costs are coming down for closed models, how do you sort of stay relevant and continue to sort of push open source and be potentially a center for European excellence and sovereign AI for Europe? What we find is that the value is in the integration

in between models we do and the distributed systems that we use to deploy the agents to make sure that execution is durable, observable, to make sure that the models can improve when the users are giving feedback. So at the end of the day, if you only bring an open model to an enterprise, it's not going to know what to do with it. But if you bring it the right platform with the

right sandboxes, execution, with the right inference at scale that is efficient, and the right customization mechanism that they don't need to care about, they don't need They don't want to care about fine tuning, they care about seeing the system improve over time. And

so what we're really bringing to enterprise is the right level of abstraction, that they can work on our open source models, on other open source models if need be, and make sure that they can actually create the business applications that are driving the ROI. And so the motion there, which has been successful for us, is to really

ROI. And so the motion there, which has been successful for us, is to really bring that platform, but also bring the experts so that they work with the subject matter experts and actually customize the systems at a level where the ROI can be found. So that's the gap, bridging the gap between open-weight models that effectively bring control

found. So that's the gap, bridging the gap between open-weight models that effectively bring control and low-cost and strategic autonomy. And the ROI that you need to have if you're running an enterprise or running a line of business is the gap that we solve.

And it's the gap where we build our business model. Robin, as one of the creators of Stable Diffusion, one of the things that you had foresight into was this idea that's really attractive to me of latent demand that we can serve with these new tools. There is clearly latent demand for expressing

creativity from the community that's begun to use these models in the last few years.

Where else do you think people don't realize how much latent demand there is for creative tools? Like what else should we imagine is gonna come where people wanna do

creative tools? Like what else should we imagine is gonna come where people wanna do it all the time, all day, I don't think everybody predicted that there were going to be that many people recreationally making images or videos. Yeah, 100%.

And I think making images or making videos is just one application for these models, as I said. They are becoming more general. So I think a lot of the demand is actually not latent. It's very obvious. If you look at the frontier models today, having that capability available open source, it's just like, a

no-brainer, you have to work towards that if you want to maximize distribution. So I

think that's to answer your question of distribution versus monetization. I think

if you want to maximize distribution, just make the most capable model and get that out there. And again, it's like a basis for so much more than just image

out there. And again, it's like a basis for so much more than just image generation or video generation. And I think that when you have an open model available and these use cases are getting explored on top of the open model, that really helps to unlock new use cases that then other companies implement on top of these models. And I think one pretty important aspect, especially when you look at agentic applications,

models. And I think one pretty important aspect, especially when you look at agentic applications, is that there's a lot of model in the loop kind of applications that will be enabled in the future or that people are already working on where image generation is a great example, right? Like that's one that is being productionized already where low latency really helps you to make repeated calls to the

model and then embed that into your product. For stuff like coding agents currently, I think there's like a lot of like, you know, the work is being done to implement the code and then you make the product out of it. But actually going towards more efficiency and having like repeated model calls enabled all the time is I think something that needs to happen and I think where open source can also really

help. Let's take us a little step back. In, you were

help. Let's take us a little step back. In, you were one of the co-authors of the seminal 2017 paper about Transformers, All You Need Is Attention. A lot has happened since 2017 in AI.

Attention. A lot has happened since 2017 in AI.

What has surprised you the most? That's a really hard, I feel like it's a surprise every three, four months that comes out. The extent to which we've been able to create a general platform for intelligence that can be applied in so many different domains is a huge shock. When we were building the Transformer, it was built for

Google Translate. That was the objective. That's what we were hill climbing on. I don't

Google Translate. That was the objective. That's what we were hill climbing on. I don't

think any of us who were involved in creating the Transformer had any foresight into what might come and how people might run with it. And I don't think we can take any credit for it either. I think it's really the community that picked up the Transformer as a platform and carried it forward. into all of these different domains. And what's the blocker now for more and more adoption? Is it the infrastructure?

domains. And what's the blocker now for more and more adoption? Is it the infrastructure?

Is it trust? Is it open-weight models being more available, easier to use? What do

you think? MARK PENETRATION ON THE CONSUMER SIDE OF HOUSE, IN TERMS OF THE AVERAGE PERSON USING THESE MODELS, SEEMS TO BE WELL ON ITS WAY TO EVERYONE WILL BE ABLE TO USE THIS MODEL IN THE SAME WAY THAT THEY USED GOOGLE FOR SEARCH.

THAT SEEMS TO BE QUITE safe and steady and on a track towards success. On

the enterprise side of house, in terms of deploying this technology into global economies, there are huge barriers. There's the trust barrier, there's the privacy barrier, trying to bring these systems into an organization, give them the access and authority to take actions that previously we would only be able to consider a human for. Those sorts of trust

barriers, I think, are the ones that need to overcome. Privacy is a huge component of that. Sovereignty is a huge component of that. And it's something that, you know,

of that. Sovereignty is a huge component of that. And it's something that, you know, at Cohere, we're super excited to make progress on. Great. One question,

Ervin, that I have for you is you expose the ability for end consumers to choose models in all your surfaces, right? The conventional wisdom, at least in venture, would be like the users kind of don't care what's underneath the surface. Is that But because I asked Robin, this is surprising.

surface. Is that But because I asked Robin, this is surprising.

I'm like Aiden, I'm surprised by everything in AI. I try really hard to lean into capability. I'm still like, wow, it's a shock every day. Do you feel like

into capability. I'm still like, wow, it's a shock every day. Do you feel like users care or why do they care? What are you learning from that? I'll tell

you an interesting stat. So first of all, ProPlexi started out as a consumer company.

But an interesting emerging outcome is our enterprise adoption is growing fast even if we don't want to work on it. One thing that our enterprise users are doing, more than 50% of every user, not just the organization, is paying different models every day they use the product. They love... Well, that's really powerful. They're

actively making that choice. Most people are. Exactly, yeah. So at the beginning of 2025, my worldview was models are going to get commoditized. And towards the end of the year, and I think... different frontier lab CEOs are saying this too, that models are specializing. Even with encoding, people would say plot code

and codex are good at different things. And our engineers use both of them. So

what's happening right now is models are specializing in different things. And so

that makes the need for selecting models for different use cases very important. Of course,

we're going to do that orchestration for you. But if you have a certain opinion yourself that this model suits my use case the best, you can go and pick it. And it's happening. The user behavior indicates it's happening. And it's one of the

it. And it's happening. The user behavior indicates it's happening. And it's one of the biggest value propositions of our app and our platform is that you don't need to feel locked into one vendor. And by the way, we are supporting both open and closed models here. Like on Friday, we brought up NemoTron, the latest open source model from NVIDIA. And they want to keep improving it. And we're going to keep putting

from NVIDIA. And they want to keep improving it. And we're going to keep putting all the newer versions on our app every time that happens. And that also unlocks a different kind of application, like people are talking about sovereignty and AI. You obviously

need a model, but you also need the orchestration, the actual product that connects your internal data, web data, proprietary data, all these different tools, run security on sandboxes. That's

the work we've already done, except we can do it all with open source models too now. And that'll unlock enterprise on-prem or sovereign applications for us.

too now. And that'll unlock enterprise on-prem or sovereign applications for us.

So what do you hope in five years, if we're sitting here, what do you hope will the open AI ecosystem will do for the AI economy?

Sorry, the open space. The open AI space, yes. So I'd say the reality is that enterprises, they don't really want to care about what's underlying.

What they care about is indeed controlling the systems, knowing that nobody can turn it off. and they care about the things improving over time. And so

off. and they care about the things improving over time. And so

to Aravin's point, today some of them still care about what model to pick, but eventually that's kind of a bug. It's just a sign of immaturity of the space because what they want to have is really business applications that are solving their problems. And so when we use Mistral Vibe to actually do this kind of business applications and work with our customers to really create the software they need to operate their

processes, they don't really care about it being open or it being... because they care about having control, they care about that if we are to disappear, they actually can maintain business continuity, et cetera, so that's the sovereignty part of it. So eventually, I think in five years, the technology and abstractions that are needed to operate agents at scale, reliably, in a way that is improving, et cetera, this is going to happen.

And I think there is kind of two worlds, either a world where you have a couple of players that are locking the space from model to orchestration to applications or a world where you have a choice in between the different bricks of the platform that you can use. So you can choose your own infrastructure provider, you can choose your own models, you can actually see the code of the orchestration side and

have enough transparency and control over the entire stack. And so in five years I would expect that we'll come up with the right abstractions, that's going to be the Linux of AI where you have open source inference, you have open source models, you have open source orchestration, open source coding tools. Everything is

actually going to be quite cheap because when it's open source you can deploy it yourself so it's hard to make margins. But the integration part is where the value is going to be created. Just like, I guess, Redapp for instance has a pretty good business model there. So the bet that we're taking is invest significantly on making open source win, and then build the orchestration and the help that enterprises needs to

stitch all of these bricks together. That's great. Well, at least there would be some money to be made with integration for the rest of us investing in companies.

Robin, underneath Black Forest Labs, what open models are you using? Are they all?

from China, Chinese models, or are they all over the world? We are making our own models, so we are primarily using them. We've actually been working with Mistral for our latest Flux2 model. So there's a BLM component that is able to process the inputs, and that's the Mistral 3 model. And then Flux2 is our own from

scratch trained model architecture. Why did you decide to make your own model?

since there's so many out there? I think our models are more efficient and better. So I think that's, I don't know, like our mission is really here

and better. So I think that's, I don't know, like our mission is really here to push the frontier of visual intelligence and these visual generative models. And I think to your point of like, how do you monetize these open models, right? There's a

lot of talk about like open source, but then also about like open weight. And

I think there's a case to be made for new business license models around open source or open-weight models in AI because it is so resource intensive, obviously, that you have to find some way. For us, it's actually been interesting. We have a lot of open-weight models that we put out there, and they are contributing the majority to our revenue. It's more than our API that we host. And yeah, we keep

iterating on it, but I think there is a real opportunity to figure out something new. Thank you guys so much. This is a

new. Thank you guys so much. This is a great panel and core to the ecosystem here at NVIDIA. I did not realize that the commitment to open ecosystem came with a commitment to cool, dark sunglasses.

Alfred and I will sign up next time. You've got to send a text, guys.

Come on. It's about the community working on things together. Thank you, guys. Really

appreciate it. Thank you. Time to check back in with Tiff out on Arena Green.

Thanks, Sarah. I'm here with Thuhin Srivastava, who is the co-founder and CEO of Base 10. Thuhin, how are you this morning? I'm great, Tiffany. Thanks so much for having

10. Thuhin, how are you this morning? I'm great, Tiffany. Thanks so much for having me. Thanks for being here. Thuhin, I need to know, what does it really, what

me. Thanks for being here. Thuhin, I need to know, what does it really, what does it actually take to run models reliably at scale? Yeah, absolutely.

Look, you need to be able to have a lot of compute to start with.

You need to be able to have a software layer that optimizes for both speed, cost, and reliability, and be able to scale with the customer as they grow things up. So we work with great customers like Abridge and OpenEvidence and Cursor and Lovable,

up. So we work with great customers like Abridge and OpenEvidence and Cursor and Lovable, and these are customers that are growing tremendously fast. And for us, what matters to them is a software layer, a support layer, and a compute layer that can support their growth. Absolutely. You just listed some huge customers. Are there any stories or examples

their growth. Absolutely. You just listed some huge customers. Are there any stories or examples that you can share from one of them? Yeah, absolutely. Look, one of my favorite customers has come to customer code, Open Evidence. Open Evidence powers a question answering service for doctors. It's clinical decision support. They've been growing incredibly fast. And what

does Open Evidence need? Well, they're being used bedside by doctors with patients. And what

they care most about is reliability. So it never goes down and speed. And how

quickly can customers get their answers, or doctors and nurses get their answers at that site. I think that's such a great example. I always lean on real world examples

site. I think that's such a great example. I always lean on real world examples because it really hits home then as to how this technology and what Base 10 is doing impacts everyone. And I think it's so great when we focus on health care and how that is bringing so much benefit to so many people. Yeah, absolutely.

And we're seeing that across everything from coding to health care sales and go to market. Literally, we're seeing an explosion of AI in every application. Absolutely. You mentioned earlier

market. Literally, we're seeing an explosion of AI in every application. Absolutely. You mentioned earlier speed, reliability, and cost. Where is the industry headed on all three?

Yeah, so look, on speed, what we are seeing is especially for these authentic use cases or real-time use cases, latency needs to be as low as possible. So we

do a lot of work with NVIDIA to make sure these models run very fast.

As an example, we use libraries like Dynamo and TRTLM to get these running, but speed has to be as low as possible. Reliability is, we can't actually get these models integrated into real-world use cases unless they're working most of the time, if not all the time. And so reliability has to be very high. And cost is, look as the cost of AI goes down we're just gonna see a lot more AI

and that's what we've been seeing is that you know as new tips go come up we're getting a lot more efficient as software gets better we're getting a lot more efficient and that's just driving down the cost which at the end is driving more adoption of AI. Absolutely let's talk about the open model explosion we've been seeing recently here how has that changed what your customers are needing from you? Yeah, look,

you know, there's great closed models and there's great open models. And we serve both customers serving closed source models and customers trying to use open source models in their applications. What we see with open source models is that, you know, customers can own

applications. What we see with open source models is that, you know, customers can own the end experience a lot more. So developers can build, you know, go more verticalized with their customers. And so they can not only use how the models work, trained and the weights but also how they're optimized to be used and you know how

um how task specific they are so rather than have one model that rules them all there are a bunch of small models that i'd be using for smaller tasks i love that very task specific owning the full vertical it's key especially when you're thinking about growth and longevity Tuhin, thank you so much for your time.

I'm really excited about what the future with Open Models holds and how Base 10 is playing such a pivotal role. I'm looking forward to more conversations with you, but for now, I'm going to throw it back to the stage.

Thanks, Tiff. Before we move on to the next chapter, a quick reminder for the attendees here. If you have a ticket for the keynote, head on inside and get

attendees here. If you have a ticket for the keynote, head on inside and get your seat. You can still watch the rest of the program in there. For the

your seat. You can still watch the rest of the program in there. For the

past few years, most people have experienced AI through chatbots. But the next phase of AI is something entirely different. The reason the term agentic AI is going to be on everyone's lips is that we're moving from systems that simply respond to systems that can reason, plan, use tools, and take action. AI agents are capable

of carrying out complex, multi-step work on their own today, and that changes everything. Frontier

everything. Frontier agentic systems have reached an inflection point for the usefulness of agents across the world. and enterprises everywhere. There's one

billion knowledge workers in the world. There are probably going to be 10 billion digital workers working with us side by side. You're seeing incredible compute demand because of it. In this new world of AI, compute is revenues. 30

million software engineers around the world will be AI assisted. the industry of applications that AI has created.

It's a revolutionary technology. Healthcare,

IT, manufacturing, every industry will be revolutionized. Agentec

revolutionized. Agentec AI is where AI begins to do real work and the builders pushing the frontier are here to discuss this with us today. Please welcome Peter Steinberg, founder of OpenClaw, Harrison Chase, co-founder and CEO of Langchain, Vincent Vicer,

co-founder and CEO of Prime Intellect, and Sam Rodriguez, founder and CEO of Edison Scientific. We have to start with the man with the claws growing out of

Edison Scientific. We have to start with the man with the claws growing out of his head. You have created an absolutely viral phenomenon. Can you walk us through

his head. You have created an absolutely viral phenomenon. Can you walk us through just a little bit of the story of what were you trying to solve and why do you think it resonated so much? Good morning. I was trying to solve...

You know, sometimes you just want to talk to your computer. Like, I walk to the kitchen and I'm like, I need to do this little thing. Hey, Jensen. Hey,

my God. Welcome to GTC. Good morning. Jensen, welcome to your own party.

Good to see you. Hi. Good to see you. Wow, what an appearance. What's going

on out here? Look at the set. We're warming this up for you. This is

incredible. Well, you guys don't get to watch it from the inside. I was just on the inside. Was it good? It's incredible, you guys. All these people are watching you guys. There's probably already 15,000 people in there. Wow. They're really here to see

you guys. There's probably already 15,000 people in there. Wow. They're really here to see you. No, they're here for the pregame show. This is just like a Super Bowl.

you. No, they're here for the pregame show. This is just like a Super Bowl.

The pregame show is better than the game. We're at least louder. We dance a bunch, though. I'm sure you've got some surprises for everybody. Well, tell you what, I

bunch, though. I'm sure you've got some surprises for everybody. Well, tell you what, I got a lot to say. I've got to figure out how to get it out, all out. So I'm going to go prepare. But Peter, it's great to have you

all out. So I'm going to go prepare. But Peter, it's great to have you here. You've started quite a movement. Yeah, it's getting me a lot of sleepless

here. You've started quite a movement. Yeah, it's getting me a lot of sleepless nights. And the NVIDIA team working with you? Yeah. They're doing a good job? We

nights. And the NVIDIA team working with you? Yeah. They're doing a good job? We

were cooking last night. That's cool. That's cool. I'm glad to see you all dressed up and cleaned up though. You were working all night. Well, you know, sometimes you gotta do what you gotta do. Yeah, well, you should have, next time have your claw do the work. We're working on that. You know, somehow, somehow the faster and the smarter the claws are, the harder we work. Do you, have you figured

that out? We work hard. We're in the critical path now all the time. Still,

that out? We work hard. We're in the critical path now all the time. Still,

like back in the old days, I'll send off a project and I have to wait for people to work on it for like a week. Now I send off a project and they're back within 15 minutes. And so you're in a critical path again. I have a feeling, just like the internet made us all busier, I have

again. I have a feeling, just like the internet made us all busier, I have a feeling AI is going to make us all busier. You are making us busier.

Well, I'm definitely making you busier. You're building AI factories that all these great people are building companies on top of. Well, I appreciate that. This is the rock star team. All right, I'll let you guys at it. I'll see you guys on the

team. All right, I'll let you guys at it. I'll see you guys on the inside. Thank you, Jensen. Hey, guys. Hello. Wow. Check it out.

inside. Thank you, Jensen. Hey, guys. Hello. Wow. Check it out.

This is just like the Super Bowl. Thanks, guys. Thank you. Go stretch.

There might be more people here to see you than the Super Bowl. Where were

we? I'm going to ask this question again, Peter. OK, you woke up, and you're like, I got to talk to my computer. And then. We wanted this since May, and to me it was so obvious that all the big labs would be building it. And then months passed and it was November, and there was still nothing. And

it. And then months passed and it was November, and there was still nothing. And

I'm like, do I have to do anything myself? And I just vibed it into existence, basically. And then ever since, it's been a rocket ship. Clearly,

existence, basically. And then ever since, it's been a rocket ship. Clearly,

people want AI that actually does things. Like, you have your own identity.

And now we're heading into a future where we make this possible. CHRISTOPHER BUSH-

Harrison, you've been building LangChain for a while to help developers so that they can deploy, build, and test agents so they can build reliably. We've had certain situations where they build very reliably and other places where it's

reliably. We've had certain situations where they build very reliably and other places where it's not building as reliably. You termed the phrase, Context engineering. Tell us a little bit more about that and how you're making Langtrain

Context engineering. Tell us a little bit more about that and how you're making Langtrain make agents much more reliable for all the rest of us. Yeah, absolutely. Excited to

be here. And we did not coin the phrase, but I like to think we were pretty early on board that ship. We can take credit. No,

no, no. When agents mess up, they mess up usually because the LLM inside messes up, and the LLM inside messes up usually because it doesn't have the right context. Sometimes it's just not smart enough as well, and hopefully everyone here who's helping

context. Sometimes it's just not smart enough as well, and hopefully everyone here who's helping build models will help make them better. But oftentimes they mess up just because they don't have the right context. And so a big part of getting agents or agentic systems or even just LLM systems to work is giving them the right context. And

the interesting thing is we've seen kind of a transition from manual context engineering where you pipe certain data to the LLM to letting the LLM do more and more of the context engineering for itself. So letting it call tools, letting it decide what to remember, what to pull back from memory. And so everything's just context engineering at the end of the day, but the exact shape that it's taken has changed over

the last few years. How has developers sort of changed over the last year or two as agents have become much more capable? What are they building on top of Langchain? What are they trying to do to sort of make sure that they have the right harness to build upon? So everyone always wanted to build this idea of an LLM running in a loop and doing things. And if you remember

like auto GPT from like three years ago, that's what it was. And the issue was the models just weren't good enough at the time to do anything reliable. And

so we built a bunch of kind of like scaffolding around it and from when LangChain started to maybe about a year ago, a lot of our focus was on things like LangGraph, which were these more deterministic kind of scaffolding around the model. Recently,

we've seen that with Claude code and OpenClaw and things like that, the models are now just good enough to run in a loop and do things. And so a lot of it has become what we call kind of like harness engineering. How do

you build the harness around the model? Because it's not like every, not everything's gonna collapse into the model, but the model will do more and more, and so the harness and the environment that it interacts with is really important, and so we've seen a big shift towards harness engineering in the past few months. Vincent, you're helping a lot of people improve quality where the agents are still not yet sufficiently good.

How do you decide with a customer where it's the harness or where it's the model itself? That's a great question. Yes, so for the context, like with Prime Intellect,

model itself? That's a great question. Yes, so for the context, like with Prime Intellect, we train agentic models and help customers to do so. And I think from context engineering, I think there's a lot of work needed to basically scale reinforcement learning to make the models really good on the specific harness, on the specific use case. So

I think the cloud codes of the world are great examples where ultimately having the harness and the shape of the terminal and scaling RL within that and then rolling it out to the users and using those production traces to further improve the model.

I think it's like kind of like how you really scale agenda capabilities of models.

And I think for different customers, like they care about different things. So we have like some customers who train also very general purpose agenda models. For example, one of our customers trained Trinity, which actually is like the, I think, second most used open claw model on open rotter. So those models, for example, are intended to have very wide use cases. But in others, a very narrow use case of, for example, automating

spreadsheets and accounting. And then they just create the RL environment for that and scale within that, but then also by rolling it out and having the users interact with it, there's more opportunities to basically scale their capabilities much further. One of the big questions about extending agents as they get more long horizon and in different domains is

verifiability. It's sort of pretty intuitive that in math and code, even in perhaps

verifiability. It's sort of pretty intuitive that in math and code, even in perhaps parts of science where you can do experimentation, which we also want to talk about, you know what the answer is. How do you think about the next domains where there's some intuition on how to verify? I think actually a lot of domains have

some measure of verifiability. I think actually one of the ones I'm most excited about is sort of like automating AI research itself, which I think is also a a segue to Sam's work on automating science. And I think that will really also take agents, I think, to the next level. It's like I think having agents on the fly self-improve to figure out their specific environment and their context. I think to some

extent, I think models are still stuck in this static model era. And they're not continuously learning yet. They're not continuously improving yet. And I think that will be one of the most exciting areas. And I think autonomous AI research, I think, will to a point when one can like synthetically generate our environments and synthetically scale RL for example. But I think over time, I think we'll probably like automate all of AI

example. But I think over time, I think we'll probably like automate all of AI research and I think that will probably really like get a genetic AI capabilities to the next level. This isn't quite research, but I have a job running right now, and we hope it's done by the end of the pregame. That is, it's an SFT job, and I'm like, well, if it doesn't work, just try again, figure out

what's wrong. And so I think to your point of putting it in a loop,

what's wrong. And so I think to your point of putting it in a loop, we're going to get very far across customization with a bunch of more sophisticated techniques.

Yeah, I think there was this sort of moment, like after Omklaw that Kapathi spawned with an auto-research project, and I think It kind of like very quickly showed us, I think, how easily you can actually improve model capabilities using agentic AI.

Huffred? Sam, so you're building AI for scientists in biology and chemistry. Most of these guys are doing things that at least are, as Sarah

and chemistry. Most of these guys are doing things that at least are, as Sarah mentioned, verifiable or they're zero one, they're binary. You can say yes or no to something. You're building something in the world of chemistry and biology, which is I would say maybe just probabilistic, distributed in some

way. It's not easy to do that. How are you solving that? Well, so,

way. It's not easy to do that. How are you solving that? Well, so,

yeah, so I run Edison Scientific. We're building AI agents for science. We're focused in particular on drug discovery and drug development. And it's actually, you know, science is at the end of the day verifiable, right? You run experiments when you're basically asking nature what is the answer and nature tells you. So, you know, in some

senses, we have some very good properties there, right? Unfortunately, the experiments that really matter when you're thinking about discovering new drugs are clinical trials. And clinical trials are, yeah, you can get a result, but it'll take you three years or something, right? Or maybe one year or two years, depending. And so you really just have

right? Or maybe one year or two years, depending. And so you really just have to find proxies along the way where there are subtasks that are verifiable, where you can get evaluations from human experts and then kind of bootstrap that way towards something that is capable of doing science at a

human level, at the same level as expert humans, or maybe better than expert humans, but maybe not like full superhuman. And then ultimately by putting it into a loop, which will take much longer, but by putting it into a loop, you can maybe get to the full superhuman level. But I think this is just one of the problems, the more general problem here, is just when you want to interact with

the physical world, with the real world, things in the real world take time. And

that's going to be a huge limitation for agents and one of the things we're going to have to figure out. You've mentioned in the past that right now the models are B-level scientists, and they don't have, maybe to Sarah's point, the intuition, the understanding, and maybe sort of the bridge between the digital world and the human world.

How far away are we from getting models to be A-level intelligence and science? I

think not far, actually. So I forget exactly when I said that. Now I would probably say they're like A-minus. Probably, right? So you really want to be at like A-plus. But... It's coming along very quickly and I think the critical thing to

A-plus. But... It's coming along very quickly and I think the critical thing to understand is that you get a huge amount of mileage from the fact that you can do the same things that the humans would do much faster, right? So even

if you know an agent on an individual run produces something that's like a minus level, but it can do it you know at scale times faster and at scale then on the back end you can just kind of screen for So which of these outputs is passable? Which of these outputs is usable? And you still end up

getting the benefit, which is that the thing is done in 1,100 of the time.

And this is where we get into this idea of the agent swarms. The fact that you can run agents at scale can make up for the fact that any individual agent run may be less good than the best human in that space.

Right? So Sam, what are you pointing as the sort of problem that you're trying to solve with Edison? Well, what I tell people is science requires three things. It

requires capital, logistics, and talent. Capital scales, logistics scales, talent does not scale. And so fundamentally, if we want to scale, we need to be able

not scale. And so fundamentally, if we want to scale, we need to be able to figure out how to remove talent as a bottleneck. And that's what we're focused on doing. And so I tell people, people talk a lot about lab in the

on doing. And so I tell people, people talk a lot about lab in the loop, which means that you want to have the agents able to come up with experiments and then interact with robots to run those experiments. I actually think that's insufficiently ambitious. What we really want is a clinic in the loop, where the agents are

ambitious. What we really want is a clinic in the loop, where the agents are able to do everything that needs to happen in order to get the drug to the point where it can be tested in humans. And then you coordinate with, medical centers to run those experiments and then iterate. But we should really, in a matter of a year or two, be at the point where we can just scale discovery

of new medicines to a level that has not been seen before. And we, humanity, as a species, we don't have to live with disease. And I think that we will not soon. That's a great vision for the future where we don't have to live with disease and everything is happy and sunny like we are here today.

I think AI for science and in particular medicine is great as a field to think about not only for its impact on us all, but also because it helps us imagine the level of ambition we should have in every domain.

I think intuitively, Sam's like, well, we can solve these problems. We need a lot more science and people understand that. We just don't have the capacity. Peter, you... We're

hopefully living in a future, a year or a few years ahead of the rest of us. Everybody wants to know, well, are you working less? You kind of sound

of us. Everybody wants to know, well, are you working less? You kind of sound like you're working more. What are your claws doing for you? What is it going to be like for all of us when we've really optimized our claws set up or whatever's next? Yeah, I don't think we're going to be working less. I'm sorry.

I'm sorry. The future is just too exciting. I think we can work on more interesting things. maybe model side A-minus,

interesting things. maybe model side A-minus, but they can really do a lot of the things that are a little bit more tedious. And then at least when I look at the field of how software

more tedious. And then at least when I look at the field of how software is being built in the future, because OpenClaw is also very much a prototype on how to build software much faster than in the classical ways, all these boring things that would have taken me so long, the agents can do now. And ultimately,

I have to do more of the hard thinking, but I can move so much faster. We can move so much faster and keep it more interesting. What's a boring

faster. We can move so much faster and keep it more interesting. What's a boring thing you no longer do, or you do much less of? Writing code.

Now I think about the shape the code should have or the features I want.

It's actually now much more of a challenge of saying no, because things are so easy to be prompted into existence. you still have to think about all the ramification. Like, where do you actually want to go? Like, people talk about techniques. I

ramification. Like, where do you actually want to go? Like, people talk about techniques. I

think that's even underrated in software to a degree. But there's so much more still that's hard, where agents are just starting to also become useful. Like, when I use Codex, I don't just do software. I also do all my knowledge work now.

Like, I don't click through Slack anymore. My agent clicks through Slack, finds all the threads. correlates it with my email, maybe some issue tracker

threads. correlates it with my email, maybe some issue tracker GitHub, and then tells me exactly what I need to do. And I can run five of these in parallel, and then just focus on the things that are really interesting. So the whole field of knowledge work is completely being transformed. And right now,

interesting. So the whole field of knowledge work is completely being transformed. And right now, you see there's companies who really get it, and there's so many companies where the gap between what we could do and what is there is wider than ever.

One of the things I saw with OpenClaw is how, you know, with software, people get it, but knowledge work still has this gap, and now suddenly people just figure out, oh, it can do my emails. It can check. It can actually go on this website and does all these things where ultimately we solve problems, right? And this

domain of writing code generalizes into, oh, it can solve any kind of problem to me. So that's super exciting. on writing code. Harrison, we know

me. So that's super exciting. on writing code. Harrison, we know you to be a very good coder from your previous life. Are you coding? My

previous life, yeah. Are you coding anymore? That answers the question.

I'm not coding anymore, but even within our company, we're thinking a lot about what EPD, engineering, product, and design, looks like in this new world. A lot of what you said really resonates. I think we're seeing that system thinking, whether at the engineering level or the product level or design, That was always the most important thing, but the amount of time you could spend on it was kind of dwarfed by the

amount of time you had to write code or write a PRD or do a Figma or something like that. Now we're seeing that all that stuff is cheap. It's

not perfect, but it's cheaper and freer. A lot of the bottleneck is choosing the right problems to work on. I think everyone has to have some sense of product and some sense of understanding, but then also having the right system thinking to say no, to point in the right direction, to make sure that everything that these agents are doing for you is building towards a cohesive narrative. I think we're seeing this

out mostly in the engineering space because the models are really good at coding. And

also the environment for coding is pretty buttoned up. You've got your GitHub repo. You

can pull it down. It's all there. If you think about knowledge work, there's so many different systems you have to integrate with. It's less clear what that environment looks like. But we're seeing a massive shift in EPD. How do you hire differently now

like. But we're seeing a massive shift in EPD. How do you hire differently now that that shift has happened? think we've seen like two profiles of people be be really attractive and I'm curious for others thoughts as well but we see kind of like that there's this like generalist, scrappy AI power user that

has product sense, has some kind of engineering chops and can take a problem, wrestle with it, and just go really far and really high agency to do that. And

then we still absolutely see that these senior architects in product and engineering design to help bring some of that systems thinking and get that initial over the line into something that's production ready. Those are the two archetypes that I think are having the most success with. I'd be curious for others' thoughts. We're still

trying to think through this a lot. Yeah, I think I would share this. I

think to some extent you need to hire, I think, for the really AI native people. One of my first interview questions is how people use AI.

people. One of my first interview questions is how people use AI.

It's usually a bad sign if they don't properly use it. For example, if you interview a developer and they're not properly using AI, to develop and I think there's like unique sort of like young talent that I think also has the like the opportunity to almost like leapfrog across the industries by becoming like early adopters of AI right like I think it's not just for developers I think it goes across

every knowledge worker I think like the the talent that will be very AI native I think will be much more interesting I think to all the companies out there I've heard people talk about the fact that there is no front-end engineer, back-end engineer.

You don't really need to be a specialist as a designer or PM. There's just

the concept of builders. And then there's build systems where you have great harnesses and then live ops because you ship it and then you understand what's going on in the world. I speak Rust now, it's terrifying. Nobody wants that. Sam, I have to

the world. I speak Rust now, it's terrifying. Nobody wants that. Sam, I have to ask you before we run out of time, when you think about this scalable A minus scientist, and then the need to still physically do experiments and get data that we don't have yet, what is the role of scientists

in your organization or across the industry over the next few years? How is it changing? Am I just physically running the lab and setting a problem until

changing? Am I just physically running the lab and setting a problem until we can automate the lab? Yeah, no, it's a great question. But I think like, I agree basically with the sentiment that people are just going to be busier, right?

So fundamentally, there are a few things that humans kind of have to do. The

first one is like defining the objective function, right? Or like deciding resource allocation, like decide what is important. Because the agent can come back and the agent can say, you know, if the goal is X, then is what we should do if the goal is Y this is what we should do if the goal is Z this is what we should do you still have to decide what the goal is okay

and you have to decide that by understanding all the various trade-offs and and that's really hard to outsource actually right so so that's that's one thing also just like it is much less the case as we've discussed before it's like much less the case in science that we are, I expected that it will be much

less the case that we're gonna like suddenly get super intelligence, right? Because we are limited so much by what is known. by what experiments have been run and so on. And so I do think that for some significant amount of time in science,

on. And so I do think that for some significant amount of time in science, it's gonna remain more of like a kind of co-worker scenario, whereas maybe in software engineering in two or three years, humans are just not doing it at all. Can

I ask for your opinion on, I would posit that because we know more about physics and chemistry, we'll see more advances here in automation than in biology, where we know very little. Oh yeah, and specifically in math, for example. in a

couple of years should just be completely automated and solved. Because it's axiomatic, right? Like

you don't have, you don't need to run experiments. You have all the information you need in order to figure things out, right? In physics, you're ultimately bottlenecked at some point by what experiments you can run, right? But also physics is to a large extent like mostly math. In biology, you are very bottlenecked by experiments. And so

what the models can do is they can help you to decide what experiments to run. Anyone who thinks right now that we are running the best experiments that we

run. Anyone who thinks right now that we are running the best experiments that we could possibly be running has never done biology research before.

The way that we allocate resources is definitely not optimal. However, even once we have optimal allocation of resources, the experiments will still take time to run. I

really resonate with that. And generally, I just think that every problem has a solution, but every solution creates more problems. And we're going to discover new problems. You guys are solving so many different things in your respective companies. And you're going to continue to solve things. None of us are thinking that we're going to work any less because there's just another problem to solve, another problem to solve, another problem to solve.

Thank you guys so much. This has been a great discussion. I do feel an unfortunate sinking sensation where I'm like, I have token anxiety, my throughput is not high enough. I'm gonna be asked to do more intellectual labor, but it's okay. It's balanced

enough. I'm gonna be asked to do more intellectual labor, but it's okay. It's balanced

out by the fact that we're all gonna have great software and solve medicine, but it's a mix. It's not a, I guess, a hang out on the beach mix.

Peter, I will ask one last question, because I think there's some really interesting announcements and work you've done with NVIDIA. How do you think about the pervasiveness of the adoption of these individual claw setups? Companies in China saying, come into our shop, we'll set it up for

claw setups? Companies in China saying, come into our shop, we'll set it up for you. Do you imagine that's because of the privacy, because of the tool access,

you. Do you imagine that's because of the privacy, because of the tool access, because people want to talk to their computers and their own data? What's the driving factor? I think it's like I try to make it fun and weird, and

factor? I think it's like I try to make it fun and weird, and people are curious, and then people try it out, and most haven't really experienced AI in that way. And suddenly you have a friend on Telegram or WhatsApp that... actually

do things, that it changes how people think about AI. And I think this is what excites me the most because in this world where we're having these rapid changes that are coming upon us, more people need to play with AI. It needs to be something fun and something that I can use to help me because we're still so early. There's still so much we can learn about how to

automate all those things, how we can transform companies. So that's what excites me the most. Thank you guys for such a wonderful, exciting discussion. We

most. Thank you guys for such a wonderful, exciting discussion. We

unfortunately have to wrap it up here, but thank you. Thank you. Thank you. Thank

you. I love that the answer to a practical question where I asked about the utility was like, actually, it's great to tinker with. It's really about the claw energy.

Let's check back in with Tiffany. Tiff. I'm here with Bill McDermott, who is the Chairman and CEO of ServiceNow. Bill, the first thing for me that comes to mind when I think about ServiceNow is around agentic AI. And I want to get your take. How have you seen this shift from conversational to agentic AI happen, especially

take. How have you seen this shift from conversational to agentic AI happen, especially when it comes to enterprises? Well, Tiffany, first I have to give so much of the credit to Jensen, the one and only Jensen, the GOAT. It's an honor to be here celebrating this pre-Super Bowl party of technology. So ServiceNow

could never have been the company we are without Jensen and without NVIDIA. We started

building models together seven years ago and we're busy at it. Whether

it's the GPU factory that's running businesses all over the world are now really moving into inferencing because as you said, the agents are playing such an important role in running companies. Models are super important and powerful, but they're just

running companies. Models are super important and powerful, but they're just expensive advice if you don't put them into action. And that's where the workflow combined with the power of NVIDIA really comes in to transforming companies. Absolutely. The key

words is putting them into action. What are some examples of that, some examples of AI agents that you've really seen either help your own workflow or maybe some stories you've heard? Yeah, sure, in our company, for example, 90% of the tasks that used to be done by humans in customer service or HR service delivery are

now being done agents. We have prioritized making sure people stay gainfully employed and retraining, retooling, and reskilling them so they've done even more important work. But no one's denying the power of the autonomous enterprise. Anything that's non-differentiated

work. But no one's denying the power of the autonomous enterprise. Anything that's non-differentiated work will get done by agents, and that's just a fact. So companies have to get in front of it because it's coming really fast. If you look at companies like Rolls-Royce reinventing manufacturing, or For example, Bell Canada rethinking

customer support with agents across their 22 million customer base. It's all

now being done by autonomous agents. Adobe, we could talk about so many companies, including the great NVIDIA, the most valuable publicly traded company itself. We're so honored to work with NVIDIA and have our agents busy at work helping them be an even greater company. Absolutely, and I want to re-emphasize that number. You said 90% internally

company. Absolutely, and I want to re-emphasize that number. You said 90% internally when it comes to certain tasks. That's incredible. And I love that you really, with that, as you mentioned, prioritized then upskilling and focusing on different ways individuals can be innovative and what's really coming next. Yeah, it's so important because if AI

isn't in service to people, what good is AI? So we have to really prioritize the impact it's having on people to give people a better life. And it's all possible, but we have to prepare for it, we have to plan for it, and really get busy on it right now. As an example, in our org charts, everybody has an AI agent in their pocket working for them. So whether they're on the

move or they're in the office, they have an AI agent. We're now building AI specialists. So you could have an AI specialist, an HR specialist, a security specialist, and

specialists. So you could have an AI specialist, an HR specialist, a security specialist, and they're being orged into the org chart and the fabric of the company itself. Not

as something to look down upon, but something to really lift us up, make us more productive, help those customers win. And when you do that, your company is going to be thriving, growing, prospering, and creating jobs and opportunities for everyone.

Absolutely. I couldn't agree more. I love the general theme of our conversation is upskilling, using AI to... find the best ways it can be productive for your entire team, for across the entire org? Sure. Well, in our case, we're pretty lucky.

We just believe in NVIDIA and in Jensen. So when they're on a move, whether it was training or inferencing or building the whole factory to reinvent every single data center around the world, we just leaned into that and believed in it. And I

believe in him and NVIDIA now more than ever. So I'm just psyched to be a part of this program. Well, Bill, I think that's a great note to end the conversation on. Thank you for your time. Thank you. Stay tuned. We have one more conversation coming up, but first, let's take it back to the stage.

Thank you, Tiffany.

And just a reminder to those of you who have tickets to the keynote, you should head inside. You can watch the rest of the pregame and the arena is filling up. AI has gotten really good at understanding language and images. Now it's learning

filling up. AI has gotten really good at understanding language and images. Now it's learning how the physical world works. Chatbots had their moment. Now it's physical AI's turn, enabling new possibilities for robotics, autonomous vehicles, intelligent machines, medicine, areas of all kinds. This is AI's breakthrough moment into the real world.

kinds. This is AI's breakthrough moment into the real world.

The ChatGPT moment for general robotics is just around the corner.

Physical AI. AI that understands the world.

Everything that moves will be autonomous.

It starts with the ability to learn to be a robot. You could teach them.

They'll learn from you. build digital twins to develop, test, and optimize robot fleets at scale before deploying into the real world. And self-driving cars.

world. And self-driving cars.

The standard chassis has now become a computing platform on wheels.

We can deploy these amazing robots in exactly the world that we built for ourselves.

This will be the largest technology industry the world's ever seen.

And here to discuss how AI enters the physical world. Raquel Erdison, founder and CEO Wabi. Giacomo Corbo, founder and CEO of PhysXX. Deepak Pathak, founder and

Wabi. Giacomo Corbo, founder and CEO of PhysXX. Deepak Pathak, founder and CEO of Skilled AI. And Daniel Nadler, founder and CEO of Open Evidence.

Raquel, I love what you're doing. I grew up in New York City and I don't drive. What? I've had to learn how to drive coming out here, but my

don't drive. What? I've had to learn how to drive coming out here, but my brother still doesn't know how to drive. And you're building autonomous vehicles and you're doing it not just doing the brute force way, but you're doing it with simulation as well. Tell us a little bit more about what you're doing at Wabi. Yeah, so

well. Tell us a little bit more about what you're doing at Wabi. Yeah, so

at Wabi, basically we have, you know, we are pioneering technology for verifiable in-to-end systems and simulation first, where you can really build a solution that truly scales. and that you can actually also prove the safety of the systems that you can deploy in the real world. We started with self-driving trucks, which are

really highly performant on the road and doing commercial operations. And we recently announced that we are also entering the robot taxi market with a very different technological approach where a single brain can actually really drive all these different four factors. So

it's a pretty exciting time for us. What do you see as the frontier problems that you're still solving on autonomy today? So if you look at really what you want is a physical AI platform that can truly generalize because these are physical AI robots that, no matter what happens on the road, really need to

work and do the right thing. So what we are really doing now is just scaling the technology, deploying commercially driverless with our OEM partner, Volvo, which is what you're going to see next, one of the next things from the company. And really you will see our beautiful new robotaxis soon. And we

have a partnership with Uber where we are going to deploy a minimum of 25,000 robotaxis. So this is really the next level of scale that you're going to see

robotaxis. So this is really the next level of scale that you're going to see over the next couple of years. What, 25,000 across the world in certain areas? What

parts of the world? Yeah, great question. So we haven't unveiled yet what the first market is. But definitely we plan to be a global company.

market is. But definitely we plan to be a global company.

And now we will start with the markets that makes sense, right? But yeah, you will see at some point technology from Wabi even beyond, I would say, cell driving and transportation, which is, thanks to this physical AI platform, a really exciting time to bring robotics everywhere. Great. Thank you. Giacomo, PhysXX works with

customers in a bunch of different very real-world domains, from oil and gas to semis companies. think about the models we have today that are quite

companies. think about the models we have today that are quite by nature approximate like what's the room for that and our foundation models today and traditional physics-based simulation um well you know maybe to start with like we're we're building a new engineering simulation software stack one that you know is trying

to move the world of a lot of industrials and the engineering manufacturing production that happens within these these companies to the So, you know, to your point, you know, today, a lot of engineering and, you know, being able to predict physics and behavior runs through, you know, numerical simulations. They are

approximate. You know, we only understand so well the phenomena that are being predicted. And,

you know, better improving those runs through first principle scientific discovery.

But they're also incredibly slow. And so being able to move that to inference means that we can do prediction anywhere between 10,000 to close to a million times faster.

But we can also sort of embed that very fast prediction into places where you can't do high fidelity simulation. So in control applications and the rate of improvement now for us to be able to do simulation runs through data generation rather than scientific discovery. Like the thing that is sort of powered, whether it's LLMs and

agents, everything around robotics as well and embodiment-free control is scaling laws. And one of the things that we've established over the course last 24 months is scaling laws holding in this domain as well. And I think that changes everything for industrials. It's

super exciting. I think something that is not intuitive to folks who work outside of the hard engineering domains is that actually it is a computationally bound field.

Because I think of AI as being incredibly compute intensive, but it's actually much more efficient and much faster than some of the approaches that actually dominate today. And it's

worse than that. It's because you're generating that can take multiple days. If I want to simulate the combustion inside a large jet engine, it can be 500 hours plus.

If I want to be able to now understand what combustion looks like under leaner conditions, 1% leaner conditions, you can blow that simulation time up by an order of magnitude. And it becomes completely intractable to design things when things are so compute

magnitude. And it becomes completely intractable to design things when things are so compute intensive, when they take so long. What's worse is that there is no such thing really as reuse for numerical simulation data. And one of the things that's changing here is the generation of numerical simulation data to generate the training corpus that I need

to be able to bring, to have AI models that understand physics and chemistry at the fidelity required for a lot of these applications. These things can be fine-tuned and steered into new domains and applications, but that paradigm shift of having a training corpus to train models is something that's entirely different from what engineering

sits on today. Genuine curiosity question. Did your customers keep all of those simulation results in reserve, and is it just about accessing them, or did they throw them away? The cost of storage being non-zero means that that data is systematically

them away? The cost of storage being non-zero means that that data is systematically purged, yes. Oh, man. Data being systematically purged. Deepak, you...

purged, yes. Oh, man. Data being systematically purged. Deepak, you...

So some of the things that you've shown us at Skilled, where you can train a new robot that you've never trained on before and make it work, and do things that it never was taught to do. Is it really, are robotics, is it generalizing? Are your models generalizing? Is it true that it's generalizing? Yes. So at

generalizing? Are your models generalizing? Is it true that it's generalizing? Yes. So at

Skilled, we are building a general purpose brain for robots, which means that any robot, any task, one brain. But as you mentioned, Alfred, this is absolutely general.

Just to put in words, the same brain controlling a humanoid in a kitchen environment, the same brain controlling a quadruped dog going around in busy areas, the same brain making a robotic arm cook an egg or omelet. Now, it is generalizing. And as

you can see in the results we have released, you can put a new robot, you can either change the shape of the robot by removing limbs, it can adapt on the fly. But the key question is why? Why are we going so general?

Because it's so much beyond even humans. Even humans control only one body. And the

reason for that is data. I very much agree in the previous points that data is scarce. In language, vision, speech, we had the luxury of having lots of data

is scarce. In language, vision, speech, we had the luxury of having lots of data and being able to just apply the wave of AI and all these models to the next field, next field, next field. But in robotics, data itself is a problem.

by having this brain so general, we can improve it with data from any kind of robot, no matter, we don't differentiate, we don't distinguish, we are not partial.

We can use data from a humanoid acting in a kitchen, even a quadruped going on a road learning something helps other robots do new things in the background. And

that is the main thesis behind this approach. And through your approach, are you able to explain how you can rip off a limb and it still adapts. Can you

explain how that happens? Or is it just the model is just doing it? So

the key idea, the key difference in some ways is, so if you have heard the term in context learning, in language models, right? Now this term in itself is only about three, four years old. Like in the whole history of AI, you would never, and Raquel here has also been a professor, so she would say this term was not existent. We always were training models to be efficient at learning. But

when we made these models learn at big scale, learning efficiently emerged as a property. So you can learn by showing you a few examples. And that is what

property. So you can learn by showing you a few examples. And that is what is happening here. So these general brain models, they are not, it's not like they have learned across every single body ahead of time. They are adapting on the fly.

So when you remove a limb or something, what really happens is, that new knowledge of history that I have one less motor to power goes in the context and it learns from that context. So this idea of in-context learning had only been shown in language or vision where there is ample data and not outside the field, not even outside language pretty much. This is the first time it is shown in the

physical world and thanks to simulation and other approaches which you can leverage here.

Now some of the models are in the open source world, they figure out how to distill them pretty cheaply. It's pretty hard to distill a robotics model, from what I understand. It is pretty hard to distill a robotics model. First, the obvious reason,

I understand. It is pretty hard to distill a robotics model. First, the obvious reason, even the pre-training models don't have as much data. So we are already solving that problem. Second, when you deploy a model on a robot, you, over time, want that

problem. Second, when you deploy a model on a robot, you, over time, want that model to become more specialized. And as we are also finding with language models, like if you are using a model for coding, don't care at the time how good is it in writing a prose. But the key is you cannot directly jump for

that specialty. You have to first go general and then go specialized. Because if you

that specialty. You have to first go general and then go specialized. Because if you go specialized, you will not have enough data and you will over fit. And that's

the same philosophy. So in robotics, you don't have access to the whole original brain. You have access to a distilled version on the device. So

original brain. You have access to a distilled version on the device. So

protects against that. But it's also challenging technically because now you have to fit everything on a device and run in a very fast-paced manner. Danielle, at Open Evidence, you guys have made the decision to train your own models, to get to specialized models, and then you'll describe it as medical superintelligence. Why do that versus use the general models? Well, we also use general models, but in medicine,

models? Well, we also use general models, but in medicine, When we're talking about physical world intelligence, the ultimate feedback loop from the physical world that we care about as humans is the human body. And I think it's important to distinguish the efforts in medical AI from some of the other efforts in AI.

Efforts in coding, for example, are extremely valuable. We use it internally in the company.

But if AI generates the wrong line of code, you can delete that, you could try it out, you could that it's not working and you can try something else if AI generates the wrong treatment recommendation you're going to get a feedback loop from the physical world from the human body that you're not so happy to just try again or delete because it's going to it's going to hurt a patient or it's

going to at best not not not cure what you're trying to cure so Like many of my colleagues on this stage, you have to go into the physical world. You have to understand what are the feedback loops that come from the human

world. You have to understand what are the feedback loops that come from the human brain, which is what a neurologist does. What are the feedback loops that come from the human heart, which is what a cardiologist does. And medicine more than other fields doesn't need to be explained to what specialization is. In many other fields, this idea in AI of specialized intelligence, specialized agents, and so on,

is newer and it's less intuitive to the people working in the fields. In medicine,

they've been specializing since medical school and then beyond. And so this is sort of running with the theme of this panel, which is why I'm on this panel and not on one of the other panels, which is you have to be grounded in the physical world. You have to be grounded in what the physical world dictates and allows. And in the physical world of medicine, in biophysical medicine and the human body,

allows. And in the physical world of medicine, in biophysical medicine and the human body, the brain and the heart and all of the other organs only allow so much.

So then you have to recreate and mirror the type of specialization in knowledge work that the medical community has already self-organized around. And that's why as we train AI, it's in our opinion not going to look like one specialized general intelligence, but many sub-specialized sub-intelligences that collaborate and cooperate

together. And each physician in this future is going to have the benefit

together. And each physician in this future is going to have the benefit of a digital twin of a cardiologist, a digital twin of a neurologist, a digital twin of every specialist and subspecialist that they can elastically and dynamically call up to help them on a patient's case. It seems obvious why consumers should want

trustworthy medical AI, but I think what you're describing is actually a model of medicine that is, to me, idealized, which kind of actually looks like the way models work today. You have a router, you go to an expert model that is better

work today. You have a router, you go to an expert model that is better at this particular task, higher performance. Maybe you can just restate why you think these clinicians and doctors who did go to medical school and specialized need AI at all.

They have a bunch of this knowledge themselves. That's a great question. So I would actually argue that in the 1920s, when the tools of medicine were gauze and scissors, they probably didn't need AI. What's happened in the last 20 or 30 years is this golden age of biotechnology, where we've developed all these These drugs which

are tools for physicians that extend their toolkit beyond the gauze and scissors that you might see in a World War One film. But there are too many of these tools now. If you speak to any physician, they'll say there's 10 to 20,000 FDA

tools now. If you speak to any physician, they'll say there's 10 to 20,000 FDA approved drug compounds. They all interact in ways sometimes which are benign, but in other which are very harmful to the patient. The combinatorial of 10,000 or let alone 20,000 FDA approved drug compounds is nearly infinite. Medical school as a vehicle, which is a

four year program, doesn't allow anymore physicians, human physicians, limited by the wet brains that we have in our skulls to cover the enormous surface area of biomedical knowledge. So it's this sort of tension that, on the one hand, it's this phenomenal accomplishment of the human species that we've developed all these drugs, that there are two new biomedical papers being published every minute, 24 hours a day,

that there's 40 million biomedical papers published in the world medical literature. And it's only going to accelerate. It's accelerating, it's doubling every five years, even by the most conservative estimates, by liberal estimates, it's doubling every 73 days. No matter what it is, even if it's every five years, they can't just stay in medical school. They can't, even if it were every five years, they'd have to just stay in medical school, which

is a four-year program for the rest of their life. That's not feasible. And even

people went to medical school 20 or 30 years ago, they've all had this moment of studying 12 hours a day, they're tired, it's the end of the day, and they wish they could just have a twin. People have had this thought long before Jensen was talking about digital twins, which is why Jensen is so right that the future is digital twins. I think we've all had this thought in grad school, if

I could just have bunch of twins of me that could learn this material. Grad

school today. Yeah, what's grad school today? But for those of you who remember what grad school was, we've all had that experience. I wish I could just have a copy of me that could learn this material and know it, who I could sort of elastically call up the next morning when I have the test. Well,

that's the experience of physicians. The next morning they have the test. The test is the patient. It's a patient who exists in a body. The body exists in the

the patient. It's a patient who exists in a body. The body exists in the physical world. You can't just delete the generated line of code that doesn't work. The

physical world. You can't just delete the generated line of code that doesn't work. The

feedback loop might be catastrophic. So we need to help physicians help patients by giving them digital twins that can help them learn all this material. Can I follow up, Daniel, on one thing that you've done twice? You've done this twice. This is

your second company in AI. This is AI for medicine. Your previous company was Kensho, with AI for finance. How are you building these two companies differently, Open Evidence and Kensho? Well, one of the gentlemen on the stage here prior was with me at

Kensho? Well, one of the gentlemen on the stage here prior was with me at Kensho. So first of all, it's great to see all these Kensho alumni starting their

Kensho. So first of all, it's great to see all these Kensho alumni starting their own companies. Harrison from Langchain. And Langchain, and then Mikey Shulman at Suno, and Martin,

own companies. Harrison from Langchain. And Langchain, and then Mikey Shulman at Suno, and Martin, all these guys. I think the world has changed so much that it would almost be easier to enumerate the overlap than to enumerate the differences. The overlap

involves, you know, there's still a venture capital landscape, you need capital.

is the end of the overlap. Everything is so completely different today. I've learned nothing.

I've really had to start from scratch all over again. The technology that exists that you can presuppose today as your starting level, I think you're going to, and this is maybe still, I don't know if this is a heterodox statement or it's not, but I think are going to have some of the most valuable companies in the world 10 or 15 years from now have sub 100

employees. And I think the world's not prepared for that.

employees. And I think the world's not prepared for that.

Take open evidence, we have sub 100 employees and 300 million Americans this year are gonna be treated by a doctor who used open evidence in the loop. So we're

talking about each employee, indirectly supporting six million patients. I mean, the scale is unfathomable. That's directly a result of what Jensen and NVIDIA and these tools and the people who develop on top of that technology have enabled as the new starting point. And I think the world economy and certainly the tech economy It's

going to look unrecognizable. You're already seeing it in some of the companies on this stage and some of the companies that were on the previous stage, where you have some of the most impactful companies in the... Certainly some of the most impactful companies in the world have sub 2,000 people, sub 3,000 people, and some of the most impactful companies in the world, including Open Evidence, have sub 100 people supporting 300 million

patient interactions with doctors in the United States alone. So that's a new tech economy.

That's a new reality, and I've had to learn all that from scratch. Kel, you've

been working on autonomy for a long time. What lesson is there from how long self-driving is taking and why you think this is the year of scale that we should take to other physical AI domains? I think cell driving is ahead in many ways of some of the other physical domains like humanoid, et

cetera. So I think there is quite a bit to learn. One is that cell

cetera. So I think there is quite a bit to learn. One is that cell driving started with a roboticist approach, which was try to build a very overly complex software stack where you basically hand code everything. And one of the things that, one of the reasons I started Wabi five years ago was because I didn't believe that that was gonna get us to a point where a scale could ever

happen, right? And with a mindset that we should rethink the world with an AI

happen, right? And with a mindset that we should rethink the world with an AI first approach, right? With a simulation first approach, which we, you know, I think in this panel, we all believe that that's the way where you can really for physical AI and safety critical applications is the only way to really expose the system, test the system to things that otherwise will have tremendous consequences consequences, right? And one of

the things that I think will be very important to take from cell driving is how we approach safety. because you know as in your domain consequences of my mistake are actually tremendous right so safety cannot be an afterthought so one of the things that we pioneer is in can we build simulation systems and even you know simulation systems that work in the physical world as you drive like means

reality testing I went to reality where you know you can really prove for the first time the safety of the driving systems at this scale and this new way of doing end-to-end very defiable with reasoning, this way of testing with simulation first really is going to bring robotics to the next level. And I

think that you will see more and more companies in other physical AI applications really adopting that type of technology. Is physics

helping us completely shorten the development cycle of physics? and engineering

and taking into account all the safety measures that you need to take because if a jet engine explodes, that would not be a good thing. Yeah, it's, so yes, it wouldn't be a good thing, right? So I think in a whole lot of industrial applications, you can't really tolerate hallucinations. You know, we want to be able to, models that can reason and predict physics at the fidelity required for a lot

of the applications, whether it's designing things like a jet engine or it's designing a piece of semiconductor manufacturing equipment. But in designing something, we don't have the same, let's say, safety issues. Because what we can do is we can use the AI to be able to explore this huge combinatorial space.

We can deploy a whole lot more algorithmic sophistication to how we do that now that the simulation step is incredibly fast. But we can then go back to our certified numerical tools in order to be able to validate that we got things right. We can go to a wind tunnel or an engine dyno or the wet lab to be able to validate

that things work well. But that's only on the design side. The moment that you go into operational control, like deploying AI into a smelter, on a mining site to do copper heat bleaching, on a vehicle that is going to do active cooling for the

battery itself, or on a piece of semiconductor manufacturing equipment deployed into a fab, In all those situations, you wanna make sure that you understand what the uncertainties are going to be. And there it's a question of having sort of safety guardrails. There

are ways to be able to do that. I think that the thing that's incredibly exciting is that we're getting into a paradigm where the same AI that is being used to design the thing is being used across the whole product development lifecycle. So

on the manufacturing side, in control applications, and that's different and exciting.

Bak, everybody wants to know if you look forward, is it next year? Is it

the year after? Is it 2030? This is like unimaginable timelines by AI standards. When

do we actually see robots with more general capability and which applications first? So I

think robotics is the longest existing field of AI and the last one to arrive.

And there is a reason for it. Because it is not like chat GPT, where you can just build a completely intelligent agent and then have access to millions of people on the first week or day one. Here, physical rollout takes time. Physical

AI is the biggest market. Now, the question is which area? Because

will I have a robot in my home this year cooking food for me? That's

what it should look like if you look at the internet. But it would be very weird you have a humanoid robot in your home this year, which can do all the cooking, all the work for you, and you still don't see robots around you in factories, grocery stores, hospitals, and everything else is just done by manual. Even

though the numbers say otherwise, there are already million more jobs available than people are able to fill them. So there is a huge demand. In my opinion, the answer is from today all the way to 2030, like whatever year you put in. But

the area of application will differ. Today, I cannot reveal details, but we have a big announcement coming that we are going to do big deployments in enterprise in one shot by partnering with the key players. This is where you start. Enterprise,

because over there, you actually have people who would pay for a robot, will not sue you if something goes wrong. And you can use that data to go to a bit more unstructured, let's say, hospitals, grocery stores, then you use that data to go even more unstructured like homes. You solve problems where you increase

complexity of the generality over time, and you use the data from the previous iteration for the next iteration. This is what we call a data flywheel. Now, data flywheel is the reason self-driving is even possible today. Our goal is to establish data flywheel step by step. not just jump to the last thing. Some

people will get robots at home, but the mass adoption will happen when you begin seeing robots around you everywhere else in uninstructed areas. Sounds like a really fun future. Thank you guys for taking AI to the physical world. I can't wait for

future. Thank you guys for taking AI to the physical world. I can't wait for all of the problems to be solved in the physical world. Jensen's keynote is coming right up. those of you who are still hanging out with us in Arena Green,

right up. those of you who are still hanging out with us in Arena Green, all are welcome to join us inside, so make your way into the keynote now.

Tiffany has been out and about meeting with some of our guests. Tiffany, how's it going out there? It's going great, thanks, Alfred. I am here with Alex Kendall, who is the co-founder and CEO of Wave. autonomous driving is really on everyone's minds. It's the buzz around here, which is so exciting. I think another thing that

minds. It's the buzz around here, which is so exciting. I think another thing that coincides with that is around the safety when it comes to AI and operating in the real world. How do you think about that? Well, isn't it fantastic to see the automotive industry really galvanizing around end-to-end AI for autonomous driving? When we

pioneered this work a decade ago, this is quite a contrarian idea. But what we've seen is recently, There's so many demos now possible showing that this can actually become a feasible product. But we had a demo five years ago, and what we've spent the last five years is showing how this can be a product at

global automotive scale. So we've now done zero shot testing in over 500 cities through Europe, Asia, and North America. And we've shown how we can build a world model that can really understand the risk and anticipate the busy dynamics of the roads that we live in. This means that we're now in a position to not just show

demos, but bring embodied AI into consumer products and robo-taxis at truly world scale. Absolutely. And you mentioned, you look back, you know, five years to where

world scale. Absolutely. And you mentioned, you look back, you know, five years to where you are today. It's incredible the speed which you are moving. I want to look ahead a little bit. What does the next decade look like as AI enters the physical world? Oh, look. Every vehicle is going to be autonomous in the future, and

physical world? Oh, look. Every vehicle is going to be autonomous in the future, and I think automotive is going to be the first large-scale demonstration of embodied AI. But

of course, it's just the beginning. But we're really excited to see some major steps forward this year. Of course, we're partnering with great companies like Uber, Stellantis, Mercedes, and of course, Nissan, where this week we're unveiling a new robo-taxi. It's on here display at GTC. We announced that we're going to be launching it and bringing it

into trials in over 10 cities around the world with Uber. So we're really excited to launch that and see our AI come into consumer vehicles and robo taxis at scale very soon. That is so exciting. And you mentioned it's on display this week.

Yeah, come take a ride. I took a ride as a lot of guests here at the event this week, where they can ride in our robotaxi and consumer vehicles, experience autonomy. It's really designed to be natural, human-like. So it

experience autonomy. It's really designed to be natural, human-like. So it

drives with the hardware It's already built into millions of vehicles today, so no bespoke hardware or retrofit. It can work with the existing vehicles hardware that's being produced at millions of scale volume and doesn't need a high definition map. So it doesn't need to drive in a pre-mapped area, but can truly drive at a global scale. We're

the only company now to drive zero shot in over 500 cities, and we've seen the system drive all around the world. It's quite remarkable. Please, please go have a drive. I definitely am going to need to do that. That's incredible. And you said,

drive. I definitely am going to need to do that. That's incredible. And you said, repeat that again, that stat at the end you said, you're the only company, finish that sentence, to- To have driven zero shot. driving in safety is all about the ability to generalize and driving zero shot means driving somewhere where you've never

had training data from that location before and this is really important because the world's too large to collect data from every single experience so you need to be able to learn general purpose behaviors and that's exactly what we've developed with our world model the ability to anticipate risk understand the dynamic and unexpected movement of people all around the world and to take this behavior and generalize it to new

scenes, to new vehicles, to new countries, cities, weather. That's the core problem of embodied AI and what we've been able to show. Wow, that's truly incredible. I am so excited for this ride. You will definitely see me taking one. Alex, thank you so much for your time. Awesome. Thank you. So, We have covered a lot today, a

lot of different technologies from agentic AI, physical AI. I mean, we even spoke about energy and everything in between. It is truly an exciting time. Now, if you're not in your seat already, you better get there quickly because the keynote is about to start. Thank you so much for following along today, joining these conversations with me. I

start. Thank you so much for following along today, joining these conversations with me. I

mean, I don't know about you, but they were... So inspiring, so much fun. Now,

before we go, I'm going to throw it back to the stage.

Thanks, Tiff. Great work out there. So that's a wrap for the pregame. In just

moments, NVIDIA CEO Jensen Huang will take the stage with his vision for how NVIDIA is building the future we've been all talking about on the stage. We'd like to thank our many incredible guests, top AI leaders who are bringing about the next industrial revolution. I also want to thank my co-hosts, Gavin Baker and Alfred Lin, Tiffany Janssen,

revolution. I also want to thank my co-hosts, Gavin Baker and Alfred Lin, Tiffany Janssen, and the entire NVIDIA team for an unforgettable morning of insights and inspiring ideas. Thank

you, Sarah. Our day is not done yet. Hopefully they saved some seats for us because we're going to have to run over there for Jensen's keynote. I wouldn't want to miss it. Me neither. From the SAP Center in San Jose, California, thanks for watching. Bye.

watching. Bye.