Skip to content

Podcast

Jensen Huang LIVE- Nvidia's Future, Physical AI, Rise of the Agent, Inference Explosion, AI PR Crisis

All-In with Chamath, Jason, Sacks & Friedberg

Source ↗ ← All highlights
  • NVIDIA Became An AI Factory
    • NVIDIA evolved from a GPU maker into an AI factory by disaggregating inference and matching workloads to heterogeneous chips.
    • Jensen described Dynamo, Verirubin, Grok and Bluefield as parts of an OS that routes prefill, decode and specialized inference across GPUs, LPUs and networking processors. Transcript: Jensen Huang The thing is, many of our strategies are presented in broad daylight at GTC years in advance of when we do it. Of the AI factory and it’s called Dynamo. Dynamo, as you know, is a piece of instrument, a machine that was created by Siemens to turn essentially water into electricity. And Dynamo powered the factory of the last industrial revolution. So I thought it was the perfect name for the operating system of the next industrial revolution, the factory of that. And so inside Dynamo, the fundamental technology is disaggregated inference. Jason, I know you’re super technical. Absolutely. I’ll let you take this one. Jason Calacanis Go ahead and define it for the audience. I don’t want to step on you. Jensen Huang Yeah, thank you. I knew you wanted to jump in there for a second. But it’s disaggregated inference, which means the pipeline, the processing pipeline of inference is extremely complicated. In fact, it is the most complicated computing problem today. Incredible scale, lots of mathematics of different shapes and sizes. That you would change, you would disaggregate parts of the processing such that some of it can run on some GPUs, rest of it can run on different GPUs, and that led to us realizing that maybe Even disaggregated computing could make sense, that we could have different heterogeneous nature of computing. That same sensibility led us to Mellanox. You know, today, NVIDIA’s computing is spread across GPUs, CPUs, switches, scale-up switches, scale-out switches, networking processors, and now we’re going to add Grok to that, And we’re going to put the right workload on the right chips. (Time 0:01:24)
  • Agents Create Heterogeneous Data Center Workloads
    • Agentic workloads multiply diversity and storage pressure: large, small, diffusion and autoregressive models plus memory and tools.
    • Jensen says Vera Rubin was designed to run this heterogeneous agent ecosystem efficiently. Transcript: Jensen Huang Yeah, and take a step back. At the time that we added this, we went from large language model processing to agentic processing. Now, when you’re running an agent, you’re accessing working memory, you’re accessing long-term memory, you’re using tools, you’re really beating up on storage really hard. You have agents working with other agents. Some of the agents are very large models. Some of them are smaller models. Some of them are diffusion models. Some of them are autoregressive models. And so there’s all kinds of different types of models inside this data center. We created Verirubin to be able to run this extraordinarily diverse workload. My sense is, and so we added, we used to be a one rack company. We now add a four more racks. So NVIDIA’s TAM, if you will, increased from whatever it was to probably something, call it, you know, 33%, 50% higher. Now, part of that 33% or 50%, a lot of it’s going to be storage processors. It’s called Bluefield. Some of it will be, a lot of it, I’m hoping, will be Grok processors. And some of it will be CPUs. And a lot of it’s going to be networking processors. And so all of this is going to be running basically the computer of the AI revolution called Agents. (Time 0:04:06)
  • Physical AI Targets A $50 Trillion Industry
    • Physical AI targets a massive $50 trillion market by digitizing industries that lacked technology.
    • Jensen says NVIDIA started this 10-year journey and already sees ~ $10B annual revenue from physical AI, growing exponentially. Transcript: Jensen Huang Businesses. Excellent. Physical AI, large category. We believe, and I just mentioned, we have three computing systems, all the software platforms on top of it. Physical AI as a large category. It’s technology industry’s first opportunity to address a $50 trillion industry that has largely been void of technology until now. And so we need to invent all of the technology necessary to do that. I felt that that was a 10-year journey. We started 10 years ago. We’re seeing it inflecting now. It is a multi-billion business for us. It’s close to $10 billion a year now. And so it’s a big business, and it’s growing exponentially. (Time 0:11:20)
  • Limit Agent Capabilities For Safer Deployment
    • Secure and govern open agent platforms by limiting simultaneous access to sensitive capabilities.
    • Jensen: give agents two of three abilities (sensitive info, execution, external comms) but not all three to reduce risk. Transcript: Jensen Huang Yeah, so great. First of all, let’s take a step back. In the last two years, we saw basically three inflection points. The first one was generative. ChatGPT brought AI to the common everybody, to our awareness. But the fact of the matter is the technology set in plain sight months before GPT. It wasn’t until ChatGPT put a user interface around it, made it easy for us to use, that generative AI took off. Now, generative AI, as you know, generates tokens for internal consumption as well as external consumption. Internal consumption is thinking, which led to reasoning. 01 and 03 continued that wave of chat GPT, grounded information, made AI not only answer questions, but answer questions in a more grounded way useful. We started seeing the revenues and the economic model of open AI start to inflect. Then the third one was only inside the industry that we saw. Clock code. The first agentic system that was very useful. Really revolutionary stuff. But Claw Code was only available for enterprises. Most people outside never saw anything about Claw Code until Open Claw. Open Claw basically put into the popular consciousness what an AI agent can do. That’s the reason why OpenClaw is so important from a cultural perspective. Now the second reason why it’s so important is that OpenClaw is open, but it formulates, it structures a type of computing model that is basically reinventing computer altogether. It has a memory system, it’s a short-term memory file system. It has scales. Did you say skills or scales? Skills. Jason Calacanis Oh, skills. They do have scales, theoretically. Skills. Jensen Huang So the first thing, it has resources, it manages resources, does scheduling. And it cron jobs. It could spawn off agents. It could decompose a task and solve problems. It does scheduling. It has I.O. Subsystems. It could input. It has output and connect to WhatsApp. And also, it has an API that allows it to run multiple types of applications called skills. These four elements fundamentally define a computer. And therefore, what do we have? We have a personal artificial intelligence computer for the very first time. Open source. It’s open source. It runs literally everywhere. And so this is basically the blueprint, the operating system of modern computing. Yeah. And it’s going to run literally everywhere. Now, of course, one of the things that we have to help it do is whenever you have agentic software, you have to make sure that an agentic software has access to sensitive information, It can execute code, it can communicate externally. (Time 0:13:41)
  • Fear-Based Messaging Harms AI Adoption
    • Public warnings about AI risks can backfire if framed catastrophically; tech leaders must be circumspect.
    • Jensen praised Anthropic’s tech but said scaring policymakers without evidence harms diffusion and national competitiveness. Transcript: Chamath Palihapitiya It’s sort of added to this layer of either resentment or fear or just general mistrust that people have sometimes at the software levels of AI. What do you think you would have told Dario and that team to do maybe differently to try to change some of this outcome and some of this perception? Jensen Huang The first thing that I would say about Anthropic is first of all the technology is incredible. We are a large consumer of Anthropic technology. Really admire their focus on security, really admires their focus on safety. The culture by which they went about it, the technology excellence by which they went about it, really fantastic. I would say that the desire to warn people about the capability, the technology is also really terrific. We just have to make sure that we understand that the world has a spectrum and that warning is good, scaring is less good. Right. And because this technology is too important to us right and and i think that it is fine to uh predict the future but we need to be a little bit more circumspect we need to have a little bit More humility that in fact we can’t completely predict the future and the ability and to say things that that are quite extreme quite catastrophic that there’s no evidence of it happening Um could be more damaging than people think and and of course we are technology leaders uh there were there was a time when nobody listened to us yeah um now, because technology is so important In the social fabric, such an important industry, so important to national security, our words do matter. And I think we have to be much more circumspect. We have to be more moderate. We have to be more balanced. (Time 0:19:09)
  • Give Engineers Generous Token Budgets
    • Invest in token access for top engineers proportional to their value.
    • Jensen’s thought experiment: a $500k engineer should consume far more than $5k in tokens—ideally ~$250k—to be superhumanly productive. Transcript: Jensen Huang Let’s say you have a software engineer or AI researcher, and you pay them $500,000 a year. We do that all the time. Okay, this is happening all of the time. That $500,000 engineer, at the end of the year, I’m going to ask them, how much did you spend in tokens? And that person said, $5,000. I will go ape something else. Yes. David Sacks Right. Jensen Huang If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed. (Time 0:25:04)
  • AutoResearch Delivered A PhD Result Overnight
    • Chamath ran an AutoResearch experiment that turned a potential seven-year PhD result into a 30-minute desktop run.
    • He described replacing an enterprise software stack in 90 minutes using an agentic Claude workflow on a Sunday night. Transcript: David Friedberg And I’d love your view on auto research and what that tells us about how far we still have to go in terms of efficiency. Research and a chunk of data, something was published internally that we said, oh my god. And that would normally be a PhD thesis that would take seven years. It would be one of the most celebrated PhD thesis we’ve ever seen in this field. And it would be in the journal Science. And it was done in 30 minutes on a desktop computer running on auto research with all the data we just ingested. We got it on Friday. We’re like, hey, let’s try it. Boot it up, go into GitHub, download it on a research, and ran it. And you see everyone’s face just go like. And then the potential of what this is unlocking for us is the kind of thing that would take seven years. And it happened in 30 minutes. And we’re experiencing it in genomics. And we’re like, this is unbelievable. So I think the acceleration is widening the aperture for everyone in a way that you didn’t imagine a few years ago. (Time 0:28:33)
  • OpenClaw Is A New Personal AI OS
    • OpenClaw (open agent framework) is a new computing model: memory, skills, resources, scheduling and I/O form a personal AI computer.
    • Jensen calls OpenClaw the blueprint OS that will run everywhere and connect to tools like Blender and Cadence. Transcript: Jensen Huang Why OpenClaw is so incredible, number one, is its confluence, its timing, with the breakthroughs in large language model. Its timing was perfect. It was impeccable. Now, in a lot of ways, Peter wouldn’t have come up with it, probably, if not for the fact that Claude and GPT and ChatGPT have reached a level that is really very good. It is also a new capability that allows these models to tool use. And Excel spreadsheets and, you know, in the case of chip design, Synopsys and Cadence and Omniverse and Blender and Autodesk. And all of these tools are going to continue to be used. Some people say that the enterprise IT software industry is going to get destroyed. Let me give you the alternative view. The enterprise software industry is limited by butts and seats. It’s about to get 100 times more agents banging on those tools. There are going to be agents banging on SQL. There are going to be agents banging on vector databases, agents banging on Blender, agents banging on Photoshop. And the reason for that is because those tools, first of all, do a very good job. (Time 0:29:47)
  • Use Both Open Source And Proprietary Models
    • Combine proprietary and open models—both are necessary and complementary.
    • Jensen: consumers will prefer managed services like ChatGPT while industries will need open weights to capture domain expertise and control. Transcript: Jensen Huang We fundamentally need models as a first-class product, proprietary product, as well as models as open source. These two things are not A or B, it’s A and B. There’s no question about it. And the reason for that is because models is a technology, not a product. Models is a technology, not a service. For the vast majority of consumers, the horizontal layer, the general intelligence, I would really, really love not to go fine tune my own. I would really love to keep using ChatGPT. I love to use Cloud. I love to use Gemini. I love to use X. And they all have their own personalities, as you know, which just kind of depends on my mood and depends on what problem I’m trying to solve. You know, I might do it on X or I might do it on ChatGPT. And so that segment of the industry is thriving. It’s going to be great. However, all these industries, their domain expertise, their specialization has to be channeled, has to be captured in a way that they can control. And that can only come from open models. The open model industry we’re contributing tremendously to. It is near the frontier. (Time 0:32:23)
  • Full Stack And Portability Drive NVIDIA’s Moat
    • NVIDIA plays both vendor and partner, supplying full stacks to clouds, enterprises and edge, which increases share.
    • Jensen highlights CUDA portability as a moat enabling on-prem, cloud and edge deployments including space and automotive. Transcript: Jason Calacanis Because it feels like you have a pretty deep stack, and in some ways you’re competing, and in other places you’re collaborative. Yeah. Jensen Huang It’s taking a step back. We believe that everything that moves will be autonomous completely or partly someday, number one. Number two, we don’t want to build self-driving cars, but we want to enable every car company in the world to build self-driving cars. And so we built all three computers, the training computer, the simulation computer, the evaluation computer, as well as the car computer. We developed the world’s safest driving operating system. We also created the world’s first reasoning autonomous vehicle so that it could decompose complicated scenarios into simpler scenarios that it knows how to navigate through, just Like us, reasoning systems. And so that reasoning system called Alpamayo has enabled us to achieve incredible results. We open this, we vertical optimization, we horizontally innovate and we let everybody decide do you want to buy one computer from us in the case of Elon and Tesla they buy our training Computers do they want to buy our training computer in our simulation computers or do you want to let us work with us to do all three and even put the car computer in your car so we you know Our attitude is we want to solve the problem. We’re not the solution provider and we’re delighted however you work with us. Chamath Palihapitiya Let me build on this question because I think it’s like it’s so fascinating. You actually do create this platform. A thousand flowers are blooming, but it’s also true that some of those flowers want to now go back down in the stack and try to compete with you a little bit. Google has TPU. Amazon has Inferentia and Tranium. You know, everybody’s sort of spinning up their own version of, I think I can out NVIDIA NVIDIA, even though they also tend to be huge customers. How do you navigate that? And what do you think happens over time? And where do those things play in the complexion of this kind of vision? Yeah, really great. Jensen Huang You know, first of all, we’re the only AI company. We’re an AI company. We build foundation models. We’re at the frontier in many different domains. We build every single layer, every single stack. We’re the only AI company in the world that works with every AI company in the world. They never show me what they’re building, and I always show them exactly what I’m building. Right. Yeah. And so the confidence comes from this. One, we are delighted to compete on what is the best technology. And to the extent that we can continue to run fast, I believe that buying from NVIDIA still is one of the most economic things they could do. And there’s just incredible confidence there. Number one, number two, we’re the only architecture that could be in every cloud. And that gives us some fundamental advantages. We’re the only architecture you could take from a cloud and put into on-prem, in the car, in any region. In space. That’s right, in space. And so there’s a whole part of our market, about 40% of our business, most people don’t realize this, 40% of our business, unless you have the CUDA stack, unless you can build an entire AI factory, the customers don’t know what to do with you. They’re not trying to build chips. They’re not trying to buy chips. They’re trying to build AI infrastructure. And so they want you to come in with a full stack, and we’ve got the whole stack. (Time 0:41:00)
  • Space Data Centers Are Possible But Thermally Hard
    • Space data centers are feasible for targeted workloads like imaging but face thermal and surface-area challenges.
    • Jensen notes satellites already run CUDA for imaging and radiation cooling requires large radiators, making space data centers long-term experiments. Transcript: Chamath Palihapitiya Yeah. We’re already in space. How should the layman think about what that business is versus when you hear about these big data center build-outs that’s happening on the ground? Jensen Huang Well, we should definitely work on the ground first because we’re already here and number one number two we should prepare to be out in space and obviously there’s a lot of energy in space Um the challenge of course is that cooling you can’t take advantage of conduction and convection exactly and so you can only use radiation and radiation requires very large surfaces And so now that’s not an impossible thing to solve. And there’s a lot of space in space. But nonetheless, the expense is still quite there. We’re going to go explore it. We’re already there. We’re already radiation-hardened. We have CUDA in satellites around the world. They’re doing imaging, image processing, AI imaging. And that kind of stuff ought to be done in space instead of sending all the data back here and do imaging down here. We ought to just do imaging out in space. And so there’s a lot of things that we ought to do in space. And in the meantime, we’re going to explore, what does the architecture of data centers look like in space? And it’ll take years. It’s OK. (Time 0:48:08)
  • Win By Deep Vertical Specialization
    • Build deep vertical specialization to maintain a moat at the application layer.
    • Jensen advises entrepreneurs to embed domain expertise into agents and connect them to customers to create a flywheel. Transcript: Jensen Huang Like, how do they differentiate themselves? Deep specialization. Deep specialization. I believe that these models, they’re going to have general models that are connected into the software company’s agentic system. Right. Many of those models are cloud models and proprietary models, but many of those models are specialized sub-agents that they’ve trained on their own. Right. Chamath Palihapitiya So the call to arms for you, for entrepreneurs is look, know your vertical. Jensen Huang That’s right. Chamath Palihapitiya Know it as deep and as better than everybody else. That’s right. And then wait for these tools because they’re catching up to you, and now you can imbue it with your knowledge. That’s right. Jensen Huang And the sooner you connect your agent, the sooner you connect your agent with customers, that flywheel is going to cause your agent to get hyper… (Time 0:58:10)
  • Automation Often Expands Jobs Not Just Displaces Them
    • Jobs will transform not vanish; automation often increases demand as capacity and services expand.
    • Jensen cites radiology: computer vision integrated into tools increased scans and demand for radiologists, raising revenues and jobs. Transcript: Jason Calacanis This is Doomer Dan. I’m not Doomer. Doomer Dan. I do have… Doomer. No, you can hold space for, think, two ideas. One is there are going to be a large… That’s viral J-Cal. Jensen Huang But that’s just because he doesn’t hang out with me enough. Jason Calacanis We fuck a little bit. Be careful. We don’t talk about it. He will show you your breakfast table. He’ll follow you around. I’m not asking for it. I’m just saying. He’ll follow you around. I’m not asking for it. You can come with me and Tucker. We ski in Japan every January. Love it. Me and Tucker will go road trip. There is going to be job displacement. And then the question becomes, do those people have the fortitude, the resolve to then go embrace these technologies? We’re going to see 100% of driving go away by humans. That’s a beautiful thing in the lives saved, but we have to recognize that’s 15 million people in the United States, 10 to 15 million who are employed in that way. And so that is going to happen, yes? Jensen Huang I think that jobs will change. For example, there are many chauffeurs today who drives the car. I believe that many of those chauffeurs will actually be in the car, sitting behind the steering wheel while the car is driving by itself. And the reason for that is because remember what a chauffeur does. In the end, these chauffeurs, they’re helping you. They’re your assistants. They’re helping you with your luggage. They’re helping you with a lot of things. And so I wouldn’t be surprised, actually, if the chauffeurs of the future become your mobility assistant and they are helping you do a whole bunch of other stuff. Check into the hotel. And the car is driving by itself. David Sacks The autopilot in planes created a lot more pilots and didn’t take any of the pilots out of the cockpit. Even though the autopilot is flying the plane 90% of the time. And by the way, while that car is driving itself, that chauffeur is going to be doing a bunch of other work on his phone and he’s going to be making money doing arranging, for example, coordinating Jensen Huang A bunch of things for you. David Friedberg The pie just grows in a way. Jensen Huang One of the things that, yes, every job will be transformed. Some jobs will be eliminated. However, we also know that many, many jobs will be created. (Time 1:00:01)
  • Study Science Math And Language And Master AI
    • Young people should study deep science, math and language and become expert at using AI.
    • Jensen: language is the ultimate programming language, so even humanities majors who master AI can thrive. Transcript: Jensen Huang Artistry. Chamath Palihapitiya You had this great advice to when you were at Stanford, I think it was, which is, I wish to you pain and suffering. Do you remember that? Yeah. Fantastic. What’s your advice to young people around what they should be studying? So if they’re sort of about to leave high school, because now those are the kids that are at this like really native, they haven’t made a decision about college, what to study, if at all, Go to college. How do you guide those kids? What would you tell them? Jensen Huang I still believe that deep science, deep math, language skills. As you know, language is the programming language of AI. The ultimate programming language. And so as it turns out, it could be that the English major could be the most successful. And so I think I would just advise whatever education you get, just make sure that you’re deeply, deeply expert in using AIs. (Time 1:02:40)