The Manosphere Report

0:00Special episode this week. We've preempted the weekly show.

0:04And there's only three people we preempt the show for.

0:07President Trump, Jesus, and Jensen.

0:10And I'll let you pick which order we do that.

0:14But what an amazing run you've had and a great event.

0:18Every industry is here. Every tech company is here.

0:22Every AI company is here.

0:24Incredible. Incredible. Extraordinary. And one of the great announcements of the past year has been

0:29Grok. When you made the purchase of Grok, did you realize how insufferable Chamath would

0:34become? I had an inkling that...

0:38We're his friends. We have to deal with them every week.

0:41I know it. You had to deal with them for the six -week close.

0:44I know it. It's like two weeks.

0:46Two weeks. It's all coming back to me now.

0:48It's making me rather uncomfortable.

0:50The thing is, many of our strategies are presented in broad daylight at GTC years

0:59in advance of when we do it.

1:01Two and a half years ago, I introduced the operating system of the AI factory.

1:07And it's called Dynamo. Dynamo, as you know, is a piece of instrument, a machine

1:12that was created by Siemens to turn, essentially, water into electricity.

1:19And Dynamo powered the factory of the last industrial revolution.

1:23So I thought it was the perfect name for the operating system of the next

1:28industrial revolution, the factory of that.

1:30And so inside Dynamo, the fundamental technology is disaggregated inference.

1:37Jason, I know you're super technical.

1:40Absolutely. I know it. I'll let you take this one.

1:42Go ahead and define it for the audience.

1:43I don't want to step on you.

1:44Yeah, thank you. I knew you wanted to jump in there for a second.

1:48But it's disaggregated inference, which means the pipeline, the processing pipeline of inference is extremely

1:55complicated. And in fact, it is the most complicated computing problem today.

1:59Incredible scale, lots of mathematics of different shapes and sizes.

2:03And we came up with the idea that you would change, you would disaggregate parts

2:09of the processing such that some of it can run on some GPUs.

2:14The rest of it can run on different GPUs.

2:17And that led to us realizing that maybe even disaggregated computing could make sense, that

2:24we could have different heterogeneous nature of computing.

2:27That same sensibility led us to Mellanox.

2:30You know, today, NVIDIA's computing is spread across GPUs, CPUs, switches, scale -up switches, scale

2:38-out switches, networking processors. And now we're going to add Grok to that.

2:42And we're going to put the right workload on the right chips.

2:45You know, we just really evolved from a GPU company to an AI factory company.

2:49I mean, I think that was probably the biggest takeaway that I had.

2:52You're seeing this fundamental disaggregation where we've gone from a GPU and now you have

2:57this complexion of all these different options that will eventually exist.

3:01The thing that you guys said on stage, or you said on stage, was, I

3:05would like the high -value inference people to take a listen to this.

3:10And 25 % of your data center space, you said, should be allocated to this

3:14Grok, LPU, GPU combo. We should add Grok to about 25 % of the Verirubins

3:19in the data center. So can you tell us about how the industry looks at

3:24this idea of now basically creating this next generation form of disaggregated pre -fill, decode,

3:29disag, and how people do you think will react to it?

3:32Yeah. And take a step back.

3:33And at the time that we added this, we went from...

3:38Large language model processing to agentic processing.

3:43Now, when you're running an agent, you're accessing working memory.

3:48You're accessing long -term memory.

3:50You're using tools. You're really beating up on storage really hard.

3:55You have agents working with other agents.

3:57Some of the agents are very large models.

3:59Some of them are smaller models.

4:01Some of them are diffusion models.

4:02Some of them are autoregressive models.

4:05And so there are all kinds of different types of models inside this data center.

4:09We created Vera Rubin to be able to run this extraordinarily diverse workload.

4:14My sense is, and so we added what used to be a one -rack company.

4:19We now added four more racks.

4:21So NVIDIA's TAM, if you will, increased from whatever it was to probably something, call

4:28it 33%, 50 % higher.

4:31Now, part of that 33 % or 50%, a lot of it's going to be

4:35storage processors. It's called Bluefield.

4:38Some of it will be, a lot of it I'm hoping, will be Grok processors.

4:42And some of it will be CPUs.

4:45And a lot of it's going to be networking processors.

4:48And so all of this is going to be running, basically, the computer of the

4:54AI revolution, called Agents, the operating system of modern industry.

5:00What about embedded applications? So, you know, my daughter's teddy bear at home wants to

5:05talk to her. What goes in there?

5:06Is it a custom ASIC?

5:08Or does there end up becoming much more kind of a broader set of TAM

5:12with developing tools that are maybe different for different use cases of the edge and

5:16an embedded application? We think that there's three computers in the problem at the largest

5:22scale when you take a step back.

5:24There's one computer that's really about training the AI model, developing, creating the AI.

5:30Another computer for evaluating it, depending on the type of problem you're having.

5:34Like, for example, you look around, there's all kinds of robots and cars and things

5:38like that. You have to evaluate these robots inside a virtual gym that represents the

5:46physical world. So it has to be software that obeys the laws of physics.

5:50And that's a second computer.

5:52We call that Omniverse. The third computer is the computer at the edge, the robotics

5:57computer. That robotics computer, one of them could be a self -driving car.

6:00Another one's a robot. Another one could be a teddy bear, a little tiny one

6:04for a teddy bear. One of the most important ones is one that we're working

6:07on that basically turns the telecommunications base stations into part of the AI infrastructure.

6:14So now all of the $2 trillion industry, all of that in time will be

6:20transformed into an extension of the AI infrastructure.

6:24And so radios will become edge devices, factories, warehouses, you name it.

6:30And so there are three, these three basic computers, all of them, you know, are

6:35going to be necessary. Jensen, last year, I think you were ahead of the rest

6:41of the world in saying inference isn't going to 1 ,000x.

6:44Just last year. Yes. Brad, you're hurting my feelings.

6:47Is it going to 1 ,000 ,000x is going to 1 ,000 ,000 ,000x?

6:50Yeah. Right? And I think people at the time thought it was pretty hyperbolic because

6:53the world was still focused on pre -scaling, on training.

6:56Here we are. Now inference has exploded.

6:59We're inference constrained. You announced an inference factory that I think is leading edge, that's

7:05going to be 10x better in terms of throughput to the next factory.

7:08But yet if I listen to what the chatter is out there, it's that your

7:13inference factory is going to cost $40 or $50 billion.

7:16The alternatives, the custom ASICs, AMD, others, are going to cost $25 to $30 billion,

7:21and you're going to lose share.

7:23So why don't you talk to us, what are you seeing, how do you think

7:26about share, and does it make sense for all these folks to pay something that's

7:30a 2x premium to what others are marketing?

7:32The big takeaway, the big idea, is that you should not equate the price of

7:42the factory and the price of the tokens, the cost of the tokens.

7:47It is very likely that the $50 billion factory, and in fact, I can prove

7:53it, that the $50 billion factory will generate for you the lowest cost tokens.

7:59And the reason for that is because we produce these tokens at extraordinary efficiency.

8:05Ten times, you know, the difference between $50 billion, now it turns out $20 billion

8:11is just land, power, and shell, right?

8:12And then on top of that, you have storage anyways, networking anyways, you've got CPUs

8:17anyways, you've got servers anyways, you've got cooling anyways.

8:20The difference between that GPU being 1x price or half x price is not between

8:27$50 billion and $30 billion.

8:29Pick your favorite number, but let's say between $50 billion and $40 billion.

8:32That is not a large percentage when the $50 billion data center is actually 10

8:39times the throughput. That's the reason why I said that even for most chips, if

8:46you can't keep up with the state of the technology and the pace that we're

8:49running, even when the chips are free, it's not cheap enough.

8:52Can I just ask a general strategy question?

8:55Yeah. I mean, you're running the most valuable company in the world.

8:59This thing is going to do $350 billion plus of revenue next year, $200 billion

9:04of free cash flow. It's compounding at these crazy rates.

9:07How do you decide what to do?

9:09Like, how do you actually get the information?

9:12I mean, it's famous now, these sort of emails that people are meant to send

9:16you. But how do you really decide to get an intuition of how to shape

9:20the market, where to really double down, where to maybe pull back, where to actually

9:23go into a greenfield? How does that information get to you?

9:26How do you decide these things?

9:27In a final analysis, that's the job of the CEO.

9:29Yeah. And our job is to define the strategy, define the vision, define the strategy.

9:34We're informed, of course, by amazing computer scientists, amazing technologists, great people all over the

9:39company. But we have to shape that future.

9:41Well, part of it has to do with, is this something that's insanely hard to

9:45do? If it's not hard to do, we should back away from it.

9:48And the reason for that, if it's easy to do, obviously.

9:52Lots of competitors. Yeah, a lot of competitors.

9:53Is this something that has never been done before that's insanely hard to do?

9:58And that somehow taps into the special superpowers of our company.

10:03And so I have to find this confluence of things that meets the standard.

10:08And in the end, we also know that a lot of pain and suffering is

10:12going to go into it.

10:13There are no great things that are invented because it was just easy to do.

10:17And just like first try, here we are.

10:19And so if it's super hard to do, nobody's ever done it before.

10:22It's very likely that you're going to have a lot of pain and suffering.

10:25And so you better enjoy it.

10:26So can you just look at maybe three or four of the more long -tail

10:29things you announced? And just talk about the long -term viability of whether it's the

10:34data centers in space. Or whether it's what you're trying to do with ADAS in

10:37autos. Or what you're trying to do on the biology side.

10:41Just give us a sense of how you see some of these curves inflecting upwards

10:44in some of these longer -tail businesses.

10:46Excellent. Physical AI. Large category.

10:50We believe, and I just mentioned, we have three computing systems.

10:53All the software platforms on top of it.

10:55All the software platforms on top of it.

10:55Physical AI as a large category, it's technology industry's first opportunity to address a $50

11:05trillion industry that has largely been void of technology until now.

11:10And so we need to invent all of the technology necessary to do that.

11:14I felt that that was a 10 -year journey.

11:16We started 10 years ago.

11:18We're seeing it inflecting now.

11:19It is a multi -billion dollar business for us.

11:22It's close to $10 billion a year now.

11:24And so it's a big business, and it's growing exponentially.

11:27And so that's number one.

11:28I think in the case of digital biology, I think we are literally near the

11:32chat GPT moment of digital biology.

11:34We're about to understand how to represent genes, proteins, cells.

11:39We all already know how to understand chemicals.

11:41And so the ability for us to represent and understand the dynamics of the building

11:48blocks of biology, that's a couple, two, three, five years from now.

11:51In five years' time, I completely believe that the healthcare industry or digital biology is

11:57going to inflect. And so these are a couple of the really great ones, and

12:00you can see they're all around us.

12:02Agriculture. Agriculture. Inflecting now. No question.

12:05Yeah. Jensen, I want to take you from the data center to the desktop.

12:10The company was built in large part on hobbyists, video gamers, and all those graphic

12:15cards in the beginning. And you mentioned in front of, I think, 10 ,000 people

12:19here, just clawed, open claw, clawed code, and what a revolution agents have become.

12:26And specifically, the hobbyists who are really where a lot of energy, we see a

12:33lot of the innovation breaks, want desktops.

12:36You announced one here. I believe it's the Dell 6800.

12:40This is a very powerful workstation to run local models, 750 gigs of RAM.

12:45Obviously, the Mac studio sold out everywhere.

12:48In my company, we're moving to open claw everything.

12:52Freeberg just got claw pilled.

12:53You got claw pilled, I understand, and you're obsessed with these.

12:57What is this from the streets movement of creating open source agents and using open

13:04source on the desktop mean to you?

13:05So great. Where is that going?

13:07Yeah, so great. First of all, let's take a step back.

13:09In the last two years, we saw basically three inflection points.

13:13The first one was generative.

13:18ChatGPT brought AI to the common everybody, to our awareness.

13:23But the fact of the matter is the technology sat in plain sight months before

13:27GPT. It wasn't until ChatGPT put a user interface around it, made it easy for

13:34us to use, that generative AI took off.

13:36Now, generative AI, as you know, generates tokens for internal consumption as well as external

13:42consumption. Internal consumption is thinking, which led to reasoning.

13:4601 and 03 continued that wave of ChatGPT, grounded information, made AI not only answer

13:54questions, but answer questions in a more grounded way useful.

13:57We started seeing the revenues and the economic model of open AI start to inflect.

14:04Then the third one was only inside the industry that we saw, cloud code.

14:10The first agentic system that was very useful.

14:13Really revolutionary stuff. But cloud code was only available for enterprises.

14:18Most people outside never saw anything about cloud code until OpenClaw.

14:25OpenClaw basically put into the popular consciousness what an AI agent can do.

14:32That's true. That's true. That's true.

14:33That's the reason why OpenClaw is so important, from a cultural perspective.

14:37Now, the second reason why it's so important is that OpenClaw is open, but it

14:43formulates, it structures a type of computing model that is basically reinventing computing altogether.

14:53It has a memory system, it's a short -term memory file system.

14:58Skills. It has scales. Did you say skills or scales?

15:02Skills. Oh, skills. They do have scales, theoretically, yeah.

15:05Skills. So the first thing, it has resources, it manages resources, it does scheduling, right?

15:13And it cron jobs, it could spawn off agents, it could decompose a task and

15:20solve problems, it does scheduling.

15:23It has I .O. subsystems, it could input, it has output, it can connect to

15:27WhatsApp. And also, it has a API that allows it to run multiple types of

15:33applications called skills. Yeah. These four elements fundamentally define a computer.

15:39Yeah. And therefore, what do we have?

15:41We have a personal artificial intelligence computer for the very first time.

15:49Open source. It's open source.

15:50It runs literally everywhere. And so this is now the, this is the, this is

15:55basically the blueprint, the operating system of modern computing.

15:59Yeah. And it's going to run literally everywhere.

16:01Now, of course, one of the things that we have to help it do is

16:04whenever you have agentic software, you have to make sure that an agentic software has

16:09access to sensitive information, it can execute code, it can communicate externally.

16:13We have to make sure that all of it has to be governed, all of

16:17it has to be secure, and that we have policies that, that gives these agents

16:22two of the three things, but not all three things at the same time.

16:25And so the governance part of it, we contributed to Peter, Peter Steinberger was here.

16:30And, and so we've got a mound of great engineers working with him to help

16:33secure and keep that thing so that it could protect our privacy, protect our security.

16:37Jensen, that paradigm shift makes some of the AI legislation that has passed around the

16:43country to regulate AI and a lot of the proposed legislation effectively moot, doesn't it?

16:48Can you just comment for a second on how quickly the paradigm shift kind of

16:52obviates a lot of the models for regulatory oversight of AI, which is becoming a

16:57very hot topic in politics right now?

16:59Well, this is, this is the part that, that we just, with policymakers, we need

17:04to, we need to always get in front of them and Brad, you do a

17:07great job doing this. We had to get in front of them and inform them

17:10about the state of the technology, what it is, what it is not.

17:13It is not a biological being.

17:17It is not alien. It is not conscious.

17:23It is computer software. Yeah, exactly.

17:26And, and it is not something that we say things like we don't understand it

17:32at all. It is not true.

17:34We don't understand it all.

17:35We understand a lot of things about this technology.

17:37And, and so, so I think one, we have to make sure that we continue

17:40to inform the policymakers and not affect, not allow doomerism and extremism to affect how

17:48policymakers think and understand about this technology.

17:51However, however, we still have to recognize this technology is moving really fast and don't

17:55get policy ahead of the technology too quickly.

17:58And the risk that we, we run as a nation, our greatest source of national

18:04security concern with respect to AI is that other countries adopt this technology while we

18:10are so angry. at it or afraid of it or somehow paranoid of it that

18:15our industries, our society don't take advantage of AI.

18:19And so I'm just mostly worried about the diffusion of AI here in the United

18:23States. Can you just double click if you were in the seat in the boardroom

18:27of Anthropic over that whole scuttlebutt with the Department of War, it sort of builds

18:32on this idea of people didn't know what to think.

18:35It's sort of added to this layer of either resentment or fear or just general

18:40mistrust that people have sometimes at the software levels of AI.

18:44What do you think you would have told Dario and that team to do maybe

18:47differently to try to change some of this outcome and some of this perception?

18:51The first thing that I would say about Anthropic is, first of all, the technology

18:55is incredible. We are a large consumer of Anthropic technology.

18:58Really admire their focus on security, really admires their focus on safety.

19:04The culture by which they went about it, the technology excellence by which they went

19:10about it, really fantastic. I would say that the desire to warn people about the

19:19capability, the technology is also really terrific.

19:22We just have to make sure that we understand that the world has a spectrum

19:26and that warning is good, scaring is less good.

19:30Right. And because this technology is too important to us.

19:35Right. And I think that it is fine to predict the future, but we need

19:42to be a little bit more circumspect.

19:44We need to have a little bit more humility that, in fact, we can't completely

19:49predict the future. And to say things that are quite extreme, quite catastrophic, that there's

19:57no evidence of it happening, could be more damaging than people think.

20:01And of course, we are technology leaders.

20:05There was a time when nobody listened to us.

20:08Yeah. But now, because technology is so important in the social fabric, such an important

20:14industry, so important to national security, our words do matter.

20:18And I think we have to be much more circumspect.

20:20We have to be more moderate.

20:22We have to be more balanced.

20:23We have to be more thoughtful.

20:24Well, I would nominate you.

20:27I think the industry's got to get together.

20:3017 % popularity of AI in the United States.

20:33I mean, we see what happened to nuclear, right?

20:36We basically shut down the entire nuclear industry.

20:38And now we have 100 fission reactors being built in China and zero in the

20:41United States. We hear about moratoriums on data centers.

20:45So I think we have to be a lot more proactive about that.

20:47But I want to go back to this agentic explosion that you're seeing inside your

20:52company, the efficiencies, the productivity gains inside your company.

20:55There's a lot of debate whether or not we're seeing ROI, right?

20:59And you and I entering into this year, the big question was, are the revenues

21:03going to show up? Are the revenues going to scale like intelligence?

21:07And then we had this kind of Oppenheimer moment of $5, $6 billion a month

21:11by Anthropic in February. Do you think as you look ahead, you announced a trillion

21:17dollar, you know, visibility into a trillion dollars of just Blackwell and Vera Rubin over

21:22the course of the next couple of years, when you see this happening at Anthropic

21:25and OpenAI, do you think we're on that curve now where we're going to see

21:29revenues scale in the way that intelligence is scaling?

21:32When you look around, when you, I'll answer this a couple of different ways.

21:35When you look around this audience, you will see that Anthropic and OpenAI is represented

21:39here. But in fact, every but 99 % of everything that is here is all

21:44AI and it's not Anthropic and OpenAI.

21:46Right, right. And the reason for that is because AI is very diverse.

21:51I would say that the second most popular model as a category is open models.

21:58Open source. Yeah, open source.

22:00Open ways, open source. Open AI is number one.

22:03Open source is number two.

22:05Very distant third is anthropic.

22:07And that tells you something about the scale of all of the AI companies that

22:11are here. And so it's important to recognize that.

22:16Let me come back and say a couple of things.

22:18One, when we went from generative to reasoning, the amount of computation we needed was

22:24about 100 times. When we went from reasoning to agentic, the computation is probably another

22:31100 times. Now we're looking at in just two years, computation went up by a

22:37fact 10 ,000x. Meanwhile, people pay for information, but people mostly pay for work.

22:50Talking to a chatbot and getting an answer is super great.

22:54Helping me do some research, unbelievable.

22:57But getting work done, I'll pay for.

22:59And so that's where we are.

23:01Agentic systems get work done.

23:03They're helping our software engineers get work done.

23:05And so then you take that.

23:08You got 10 ,000x more compute.

23:11You get probably, at this point, 100x more consumption now.

23:14Yes. Yeah. And we haven't even started scaling yet.

23:18We are absolutely at a millionx.

23:20Which is, I think, a great place to talk about the number of engineers you

23:24have. 20 ,000, 30 ,000 at the company.

23:26We have 43 ,000 employees.

23:28I would say 38 ,000 are engineers.

23:32The conversation we've had on the pod a number of times is, oh my God,

23:36look at the token usage in our companies.

23:38It is growing massively. And some people are asking, hey, when I join a company,

23:43how many tokens do I get?

23:44Because I want to be an effective employee.

23:46And you postulated, I believe, during your two -and -a -half -hour keynote, pretty long

23:52keynote, well done, that you were spending...

23:55If it was well done, it would be shorter.

23:57Yeah. You didn't have time to do...

23:59You didn't have time to write it for an hour and 45.

24:02So you guys know there is no practice.

24:06And so it's a gripping and ripping.

24:07Grip and ripping. Yeah, yeah.

24:08I love it. So I just want to let you know I was writing the

24:11speech while I was giving the speech.

24:12Okay, so... You never know.

24:14But does that mean if we do back...

24:17I apologize. Back in the envelope, man, $75 ,000 in tokens for each engineer or

24:21something like that. So are you spending, in NVIDIA, $1 billion, $2 billion on tokens

24:26for your engineering team right now?

24:28We're trying to. Let me give you the thought experiment.

24:30Let's say you have a software engineer or AI researcher and you pay them $500

24:34,000 a year. We do that all the time.

24:37Okay, this is happening all of the time.

24:39That $500 ,000 engineer, at the end of the year, I'm going to ask them,

24:44how much did you spend in tokens?

24:46And that person said, $5 ,000.

24:48I will go ape something else.

24:50Yes. Right. If that $500 ,000 engineer did not consume at least $250 ,000 worth

24:57of tokens, I am going to be deeply alarmed.

25:01Okay? And this is no different than one of our chip designers who says, guess

25:06what? I'm just going to use paper and pencil.

25:08I don't think I'm going to need any CAD tools.

25:11Right, right, right. This is a real paradigm shift to start thinking about these all

25:14-star employees. It almost reminds me of what we learned in the NBA when LeBron

25:19James started spending a million dollars a year just on his health of his body

25:22and maintaining it. That's right.

25:24Here he is at age 41 still playing.

25:26It really is, hey, these are incredible.

25:28Thank you. Knowledge workers, why wouldn't we give them superhuman abilities?

25:33That's exactly right. Where does that go?

25:35If we extrapolate out two or three years from now, what is the efficiency of

25:40that all -star at NVIDIA and what they're able to accomplish?

25:43What do they look like?

25:44Well, first of all, things that, wow, this is too hard.

25:49That thought is gone. This is going to take a long time.

25:52That thought is gone. We're going to need a lot of people.

25:55That thought is gone. This is no different than in the last industrial revolution, somebody

26:01goes, boy, that building really looks heavy.

26:03Nobody says that. Wow, that mountain looks too big.

26:07Nobody says that. Everything that's too big, too heavy, takes too long, those ideas are

26:14all gone. You're reduced to creativity.

26:15That's right. What can you come up with?

26:16Exactly. Now, the question is, how do you work with these agents?

26:21Well, it's just a new way of doing computer programming.

26:24In the past, we code.

26:26In the future, we're going to write ideas, architectures, specifications.

26:31We're going to organize teams.

26:34We're going to help them define how to evaluate the definition of good versus bad.

26:39What does it look like when something is a great outcome?

26:42How to iterate with you?

26:44How to brainstorm? That's really what you're looking for.

26:46And I think that every engineer is going to have 100, 100 agents.

26:51Back to the PR problem the industry has right now.

26:54You have executives like David Freeberg with Ohalo, who's looking at literally taking through the

27:02use of technology, your technology and AI, the number of calories produced and making high

27:07quality calories. What is the factor you think you can bring the cost down, Freeberg?

27:13And what impact does this vision have for what you're doing?

27:17Zero shot genomic modeling. And it works.

27:20Yeah. And you have that moment and you're like, holy shit.

27:23Honestly, like, and that's after people are replacing entire enterprise software stacks in a night.

27:30I did something in 90 minutes.

27:31I was telling the guys about replaced the whole software stack and, like, a whole

27:35bunch of workload. 90 minutes on Claude ran this agentic system, built the whole thing,

27:39deployed it, and we got it.

27:40On a Sunday night. On a Sunday night, 10 p .m.

27:42I was done at 11 .30.

27:43I went to bed. As the CEO, you replaced software.

27:45Yeah. And everyone on my management team had to do a similar exercise over the

27:50weekend. What we saw on Monday, I was like, it's over.

27:54But the technical stuff, the science stuff, we did something in 30 minutes using auto

27:59research. And I'd love your view on auto research and what that tells us about

28:02how far we still have to go in terms of efficiency.

28:04But using auto research and a chunk of data, something was published internally that we

28:09said, oh, my God. And that would normally be a PhD thesis that would take

28:13seven years. It would be one of the most celebrated PhD pieces we've ever seen

28:16in this field. And it would be in the journal of science.

28:19And it was done in 30 minutes on a desktop computer running on auto research

28:23with all the data we just ingested.

28:24We got it on Friday.

28:25We're like, hey, let's try it.

28:27Boot it up, go into GitHub, download it, auto research, and ran it.

28:30And you see everyone's face just go like.

28:33And then the potential of what this is unlocking for us is like the kind

28:37of thing that would take seven years.

28:38And it happened in 30 minutes.

28:40And we're experiencing it in genomics.

28:42And we're like, this is unbelievable.

28:44So I think the acceleration is widening the aperture for everyone in a way that

28:50you didn't imagine a few years ago.

28:52But just going back to the auto research point, can you just comment on what

28:56you think about the fact that this thing got published with 600 lines of code

28:59in a weekend and the capacity that it has to run locally and achieve what

29:03it can achieve with all of these diverse data sets?

29:05And what that tells us about.

29:07at the early stages we are in terms of optimization on algorithms and hardware.

29:11The fundamental reason why OpenClaw is so incredible, number one, is its confluence, its timing

29:19with the breakthroughs in large language model.

29:23Its timing was perfect. It was impeccable.

29:25Now, in a lot of ways, Peter wouldn't have come up with it, probably, if

29:29not for the fact that Claude and GPT and ChatGPT have reached a level that

29:34is really very good. It is also a new capability that allows these models to

29:40tool use. The tools that we've created over time, web browsers and Excel spreadsheets and

29:47in the case of chip design, Synopsys and Cadence and Omniverse and Blender and Autodesk,

29:55all of these tools are going to continue to be used.

29:58Some people say that the enterprise IT software industry is going to get destroyed.

30:06Let me give you the alternative view.

30:08The enterprise software industry is limited by butts and seats.

30:12It's about to get a hundred times more agents banging on those tools.

30:17They're going to be agents banging on SQL.

30:19They're going to be agents banging on vector databases, agents banging on Blender, agents banging

30:23on Photoshop. And the reason for that is because those tools are, first of all,

30:27do a very good job.

30:28Second, those tools are the conduit between us.

30:33In the final analysis, when the work is done, it has to be represented back

30:37to me in a way that I can control.

30:39And I know how to control those tools.

30:42And so I need everything to be put back into Synopsys.

30:45I want everything to be put back into Cadence because that's how I control it.

30:49That's how I ground truth.

30:50Let me ask you a question about open source.

30:52So we have these closed source models.

30:54They're excellent. We have these open weight models.

30:57Many of the Chinese models are incredible.

30:59Absolutely incredible. Two days ago, you may not have seen this because you were busy

31:03on stage, but there was a training run that happened in this crypto project called

31:08BitTensor. Subnet three, they managed to train a four billion parameter LAMA model, totally distributed

31:14with a bunch of people contributing excess compute.

31:19But they were able to do it statefully and manage a training run, which I

31:23thought was like a pretty crazy technical accomplishment.

31:26Yeah. Because it's like random people and each person gets a little share.

31:29Our modern version of folding at home.

31:31Exactly. Yeah. So what do you think about the end state of open source?

31:36Do you see this decentralization of architecture as well and decentralization of compute to support

31:42open weights and a totally open source approach to making sure AI is broadly available?

31:48I believe we fundamentally need models as a first class product, proprietary product, as well

31:56as models as open source.

31:58These two things are not A or B, it's A and B.

32:02There's no question about it.

32:03And the reason for that is because models is a technology, not a product.

32:08Models is a technology, not a service.

32:10For the vast majority of consumers, the horizontal layer, the general intelligence, I would really,

32:16really love not to go fine tune my own.

32:19I would really love to keep using ChatGPT.

32:22I love to use Cloud.

32:23I love to use Gemini.

32:24I love to use X.

32:26And they all have their own personalities, as you know, which is kind of depends

32:29on my mood and depends on what problem I'm trying to solve.

32:32I might do it on X or I might do it on ChatGPT.

32:35And so that segment of the industry is thriving.

32:39It's going to be great.

32:40However, all these industries, their domain expertise, Their specialization has to be channeled, has to

32:48be captured in a way that they can control.

32:52And that it can only come from open models.

32:54The open model industry we're contributing tremendously to.

32:57It is near the frontier.

33:00And quite frankly, even if it reaches the frontier, I think that products as a

33:07service, world -class models as a product is going to continue to thrive.

33:12Every startup we're investing in now is open source first, and then going to the

33:18proprietary model. Yeah, and the beautiful thing is because you have a great router you

33:22connect it to, on first day, every single day, you're going to have access to

33:27the world's best model. And then it gives you time to cost reduce and fine

33:32-tune and specialize. So you're going to have world -class capabilities out to shoot every

33:36single time. Jensen, can I say, of course.

33:39Nobody wants the U .S.

33:41to win the global AI race more than you, right?

33:44But a year ago, the Biden -era diffusion rule really was an anti -American diffusion

33:51of AI around the world.

33:52So here we are a year into the new administration.

33:55Give us a grade. Where are we in terms of global diffusion and the rate

34:01at which we're spreading U .S.

34:03AI technology around the world?

34:04Are we an A? Are we a B?

34:06Are we a C? What's working?

34:08What's not working? Well, first of all, President Trump wants American industry to lead.

34:15He wants American technology industry to lead.

34:17He wants American technology industry to win.

34:20He wants us to spread American technology around the world.

34:24He wants the United States to be the wealthiest country in the world.

34:27He wants all of that.

34:28But at the current moment as we speak, NVIDIA gave up a 95 % market

34:35share in the second largest market in the world.

34:38And we're at zero percent.

34:40I know. That's right. President Trump wants us to get back in there.

34:45And the first thing is to get licensed for the companies that we're going to

34:51be able to sell to.

34:52We've got many companies who have requested for licenses.

34:56We've applied for licenses for them.

34:58And we've got approved licenses from Secretary Lutnik.

35:02Now we've informed the Chinese companies and many of them have given us purchase orders.

35:07And so we're in the process of cranking up our supply chain again to go

35:12ship. I think at the highest level, Brad, I think one of the things that

35:17we should acknowledge is this.

35:18Our national security is diminished when we don't have access to miniature motors, rare earth

35:28minerals. It's diminished when we don't control our telecommunications networks.

35:33It's diminished when we can't provide for sustainable energy for our country.

35:37It is fundamentally diminished. Every single one of these industries is an example of what

35:43I don't want the AI industry to be.

35:46When we look forward in time and we say, what do we want?

35:50What does it look like when American technology industry, American AI industry leads the world?

35:57We can all acknowledge that there is no way that AI models is one universally.

36:04We can all acknowledge that that is an outcome that makes no sense.

36:08However, we can all imagine that the American tech stack from chips to computing systems

36:17to the platforms are used broadly by the world where they build their own AI.

36:23They use popular culture. AI, they use private AI, whatever, and they can build their

36:27applications in their society. I would love that the American tech stack is 90 %

36:32of the world. I would love that.

36:34The alternative, if it looks like solar, rare earth, magnets, motors, telecommunications, I consider that

36:44a very bad outcome for national security.

36:47How much are you monitoring the situation with the conflicts around the world right now?

36:53And how much does it worry you, Jensen?

36:54So China and Taiwan, and then helium availability coming out of the Middle East, I

36:58understand, can be a supply chain risk to semiconductor manufacturing.

37:02How much do these situations worry you?

37:04How much are you spending on them?

37:06Well, first of all, I think in Middle East, we have 6 ,000 families there.

37:11We have a lot of Iranians at NVIDIA, and their families are still in Iran.

37:15And so we have a lot of families there.

37:18The first thing is they're quite anxious, they're quite concerned, quite scared.

37:22We're thinking about them all the time.

37:24We're monitoring and keeping an eye on them all the time.

37:26They have 100 % of our support.

37:28I've been asked several times, are we still considering being in Israel?

37:32We are 100 % in Israel.

37:35We are 100 % behind the families there.

37:38We are 100 % in the Middle East.

37:39I was also asked, given what's happening in the Middle East, is that an area

37:45where we believe that we can expand artificial intelligence too?

37:50I believe that there's a reason we went to war, and I believe at the

37:54end of the war, Middle East will be more stable than before.

37:58And so if we were there, if we're considering it before, we should absolutely be

38:01considering it after. And so I'm 100 % in on that.

38:05With respect to Taiwan, Taiwan, we have to do three things.

38:11One, we have to make sure that we re -industrialize the United States as fast

38:15as we can. And whether it's the chip manufacturing plants, the computer manufacturing plants, or

38:20the AI factories. How are we doing on that?

38:22We're doing excellent. By gaining the strategic support, by gaining the friendship of the supply

38:30chain of Taiwan, by gaining their friendship, by gaining their support, we were able to

38:37build Arizona and Texas, California, at incredible rates.

38:43They are genuinely a strategic partner.

38:45We really, they deserve our support.

38:50They deserve our friendship. They deserve our generosity.

38:54And they're doing everything they can to accelerate the manufacturing process for us.

38:59And so I think that's number one.

39:01Number two, we ought to diversify the manufacturing supply chain.

39:05And whether it's South Korea, whether it's Japan, it's Europe, we ought to diversify the

39:11supply chain, make it more resilient.

39:12And number three, let's be, let's demonstrate restraint.

39:17And while we're reducing, increasing our diversity and resilience, let's not press, push, you know.

39:28Unnecessarily. Unnecessarily. We need to be patient.

39:30It's thoughtful. Is helium a problem?

39:32A lot of reports. You know, I think helium could be a problem, but it's

39:36also the case that the supply chain probably has a lot of buffer in it.

39:39Yeah. These kinds of things tend to have a lot of buffer.

39:42But, but, you know. Yeah.

39:45You've made massive progress in self -driving.

39:49You made a big announcement.

39:51You've added many more partners, including BYD.

39:54There was just a video of you driving around in a Mercedes and a huge

39:58announcement with Uber that you're going to have a number of cars.

40:02Thank you, David. Be sure sister.

40:02on the road from many different manufacturers.

40:04Your bet is, I believe that there's going to be an Android type open source

40:11platform that you're going to play a major part in with dozens of car providers.

40:17And then maybe on the other side, there could be an iOS with Tesla or

40:20Waymo. What's your strategy thinking there and how that chessboard emerges?

40:25Because it feels like you have a pretty deep stack and in some ways you're

40:31competing and in other places you're collaborative.

40:34Yeah. It's taking a step back.

40:38We believe that everything that moves will be autonomous completely or partly someday.

40:45Number one. Number two, we don't want to build self -driving cars, but we want

40:48to enable every car company in the world to build self -driving cars.

40:51And so we built all three computers, the training computer, the simulation computer, the evaluation

40:56computer, as well as the car computer.

40:59We developed the world's safest driving operating system.

41:03We also created the world's first reasoning autonomous vehicle so that it could decompose complicated

41:11scenarios into simpler scenarios that it knows how to navigate through, just like us, reasoning

41:17systems. And so that reasoning system called Alpamayo has enabled us to achieve incredible results.

41:25We open this, we vertical optimization, we horizontally innovate, and we let everybody decide, do

41:33you want to buy one computer from us?

41:35In the case of Elon and Tesla, they buy our training computers.

41:38Do they want to buy our training computer and our simulation computers?

41:41Or do you want to let us work with us to do all three and

41:45even put the car computer in your car?

41:46So we, you know, our attitude is we want to solve the problem.

41:52We're not the solution provider.

41:54And we're delighted, however you work with us.

41:57Let me build on this question because I think it's like, it's so fascinating.

42:00You actually do create this platform.

42:03A thousand flowers are blooming.

42:05But it's also true that some of those flowers want to now go back down

42:09in the stack and try to compete with you a little bit.

42:12Google has TPU. Amazon has Inferentia and Tranium.

42:15You know, everybody's sort of spinning up their own version of, I think I can

42:19out NVIDIA, NVIDIA, even though they also tend to be huge customers.

42:24How do you navigate that?

42:25And what do you think happens over time?

42:28And where do those things play in the complexion of this kind of vision?

42:32Yeah, really great. You know, first of all, we're the only AI company.

42:36We're an AI company. We build foundation models.

42:39We're at the frontier in many different domains.

42:41We build every single layer, every single stack.

42:45We're the only AI company in the world that works with every AI company in

42:48the world. They never show me what they're building, and I always show them exactly

42:52what I'm building. Right. And so the confidence comes from this.

42:57One, we are delighted to compete on what is the best technology.

43:04And to the extent that we can continue to run fast, I believe that buying

43:09from NVIDIA still is one of the most economic things they could do.

43:12And there's just incredible confidence there.

43:14Number one. Number two. We're the only architecture that could be in every cloud.

43:18And that gives us some fundamental advantages.

43:20We're the only architecture you could take from a cloud and put into on -prem,

43:25in the car, in any region.

43:26In space. That's right. In space.

43:28And so there's a whole part of our market, about 40 % of our business,

43:33most people don't realize this, 40 % of our business, unless you have the CUDA

43:37stack, unless you can build an entire AI factory, you have the customers don't.

43:41know what to do with you they're not trying to build chips they're not trying

43:44to buy chips they're trying to build ai infrastructure and so they want you to

43:48come in with the full stack and we've got the whole stack and so surprisingly

43:52nvidia is gaining market share if you look at where we are today we're gaining

43:57share do you think what happens is these guys try and they realize oh my

44:00god it's too much and then they come back is that why this share grows

44:03well we're gaining share for several reasons one um our velocity has gone we help

44:10people realize it's not about building the chip it's about building the system and that

44:14system is really hard to build uh and and so their their their business with

44:18us is increasing in the case of aws i think they just announced i think

44:22it was yesterday that they're going to buy a million chips uh in the next

44:27couple years i mean that's a lot of chips from from aws and that's on

44:31top of all the chips they've already bought and so we're delighted to do that

44:34but number one we're gaining share this last couple years because we now have anthropic

44:40coming to nvidia meta sl is coming to nvidia and the growth of open models

44:47is incredible and that's all on video and so we're growing in share because of

44:52the number of models we're also growing in share because outs all of these companies

44:57are outside the cloud and they're growing regionally in enterprise and industries at the edge

45:04and that entire segment of growth is you know really hard to do if it's

45:08just building an asic brad related to that um and not to get in the

45:13weeds on the numbers but analysts don't seem to believe right so if you look

45:17at the consensus forecast you said compute could 1 million x right and yet they

45:23have you growing next year at 30 percent the year after that at 20 percent

45:26and in 2029 which is supposed to be a monster year at seven percent right

45:31so if you just if you take your tam and you apply their growth numbers

45:36it suggests that your share will plummet do you see anything in your future order

45:41book that would make that correct yeah first of all they just don't understand the

45:46scale and the breadth of ai yes yeah yeah i think that's most people think

45:52that ai is in the top five hyperscalers right that's right there's also an orthodoxy

45:58around these law of large numbers where you know they have to go back to

46:01their investment banking risk committee and show some model they're not going to believe in

46:06their minds that five trillion goes to 15 trillion they're like it can go to

46:11seven or they do not have a 10 trillion dollar company it's all just cya

46:15stuff but it's never happened before so you can't say well and and because because

46:19you have to redefine what it is that you do there was somebody who made

46:22an observation recently that nvidia jensen how can you be larger than intel in servers

46:30and the reason for that is because the cpu market of the entire data center

46:35was about 25 billion dollars a year right we do 25 billion dollars a year

46:39as you guys know in a very in the time that we were sitting here

46:42right and so obviously obviously obviously that was a joke no it's but it's all

46:49in podcast don't worry everything on the show is right god don't worry about it

46:53it's all in wow okay anyways that was not guidance but anyhow anyhow it the

47:00the point is how big you can be yeah depends on what is it that

47:04you make right nvidia is not making chips number one making chips does not help

47:09you solve the ai infrastructure problem anymore it's too complicated number three most people think

47:15that ai is narrowly and the things that they talk about and hear and see

47:20AI is much, open AI is incredible, they're going to be enormous.

47:24Anthropic is incredible, they're going to be enormous.

47:27But AI is going to be much, much bigger than that.

47:30And we addressed that segment.

47:32Tell us about data centers in space for a second.

47:34Yeah, we're already in space.

47:37How should the layman think about what that business is versus when you hear about

47:43these big data center build -outs that's happening on the ground?

47:46Well, we should definitely work on the ground first because we're already here, and number

47:51one. Number two, we should prepare to be out in space, and obviously there's a

47:54lot of energy in space.

47:56The challenge, of course, is that cooling, you can't take advantage of conduction and convection.

48:03And so you can only use radiation.

48:05And radiation requires very large surfaces.

48:08Now, that's not an impossible thing to solve, and there's a lot of space in

48:12space. But nonetheless, the expense is still quite there.

48:17We're going to go explore it.

48:19We're already there. We're already radiation -hardened.

48:21We have CUDA in satellites around the world.

48:26They're doing imaging, image processing, AI imaging.

48:30And that kind of stuff ought to be done in space instead of sending all

48:33the data back here and do imaging down here.

48:35We ought to just do imaging out in space.

48:37And so there's a lot of things that we ought to do in space.

48:40And in the meantime, we're going to explore what is the architecture of data centers

48:44look like in space. And it'll take years.

48:47It's okay. I got plenty of time.

48:49I wanted to double -click on healthcare.

48:51I know you've got a big effort there.

48:52We're all of a certain age where we're thinking about lifespan, health span.

48:57I mean, we all look great, I think.

48:59Some better than others. I think some better than others.

49:01I don't know what your secret is, Jensen.

49:03I mean, what are you taking?

49:06What's off the menu? You've got to talk to me when we're backstage.

49:09I want to know in the green room what you've got going on.

49:11Squats and push -ups and sit -ups.

49:13Okay. But what you know in terms of the build -out in healthcare, where is

49:22that going? And what kind of progress are we making?

49:25I was just using Claude to do some analysis and saying, like, where are all

49:28these billing codes? We spend twice as much money in the U .S.

49:31We seem to get half as much.

49:33It seemed like 15 % to 25 % of the dollar spent were on these

49:38first GP visits. And I think we all know, like, ChatGPT and a large language

49:44model does a better job more consistently today at a first visit.

49:48So what has to happen there to kind of break through all that regulation and

49:52have AI have a true impact on the healthcare system?

49:55There's several areas that we're involved in healthcare.

49:59One is AI physics, or AI biology.

50:08Using AI to understand, represent, predict biology behavior, biological behavior.

50:13And so that's one. That's very important in drug discovery.

50:16There's second, which is AI agents.

50:19And that's where the assistance in helping diagnosis and things like that.

50:23Open evidence is a really good example.

50:25Hippocratic is a really good example.

50:26Love working with those companies.

50:27I really think that this is an area where agentic technology is going to revolutionize

50:32how we interact with doctors and how do we interact for healthcare.

50:37The third part that we're involved in is physical AI.

50:40The first one is AI physics, using AI to predict physics.

50:42The second one is physical AI.

50:45AI that understands the properties of the laws of physics.

50:48And that's used for robotic surgery.

50:52Huge amounts of activities there.

50:53Every single instrument. Every single instrument.

50:55Whether it's ultrasound or, you know, C - The second one is eating, or the

50:58second one is counting the dance department.

50:58Or when people have to look at ourفسiempoary this past year.

50:58or whatever instrument we interact with in a hospital in the future will be agentic.

51:03Open claw in a safe version will be inside every single instrument.

51:07And so in a lot of ways, that instrument is gonna be interacting with patients

51:11and nurses and doctors in a very unique way.

51:13Yeah, we're seeing so much investment in AI weapons.

51:15It'd be wonderful to see some investment in AI EMTs and paramedics and saving lives,

51:22not just taking them, which I think is a great segue into robotics.

51:25You've got dozens of partners.

51:27We had this very weird, I don't know what I want to call a lost

51:31decade or 20 years of Boston Dynamics.

51:34Google bought a bunch of companies.

51:36They then wound up selling them and spinning them out where people just thought robotics

51:40is just not ready for prime time.

51:42And now here we have the world's greatest entrepreneur at this time tied with you,

51:47Elon Musk doing, well, that was a good save, I hope.

51:51Optimus, pretty impressive. And then other companies in China.

51:54How close is that to actually being in our lives where we might see a

52:01chef, a robotic chef, a robotic nurse, a robotic housekeeper, this humanoid factor actually working

52:09in the real world, knowing what you know with those partners and the fidelity, especially

52:13in China where they seem to be doing as good a job as we're doing

52:16here, or maybe better? We invented the industry largely.

52:22America invented, you could argue we got into it too soon.

52:25Yeah. And we got exhausted, we got tired about five years before the enabling technology

52:34appeared. Yes, the brain. Yeah, yeah.

52:35And we just got tired of it just a little too soon, okay?

52:40That's number one. But it's here now.

52:41Now, the question is how much longer?

52:43From the point of high -functioning existence proof, high -functioning existence proof to reasonable products,

52:54technology never takes more than a couple, two, three cycles.

52:57And so a couple, two, three cycles would basically be somewhere around three years to

53:01five years. That's it. Three years to five years, we're going to have robots all

53:04over the place. I think China is formidable.

53:09And the reason for that is because their microelectronics, their motors, their rare earth, their

53:15magnets, which is foundational to robotics, they are the world's best.

53:19And so in a lot of ways, our robotics industry relies deeply on their ecosystem

53:25and their supply chain. And they're, you know, obviously moving very quickly.

53:31We're going to, you know, our robotics industry will have to rely a lot on

53:34it. The world's robotics industry will have to rely on a lot on it.

53:37And so I think you're going to see some fast, fast movements here.

53:42Ultimately, one for one, Elon seems to think we're going to have one robot for

53:46every human, seven billion for seven billion, eight billion for eight billion.

53:49Well, I'm hoping more. Yeah, I'm hoping more.

53:51Yeah. Well, first of all, there's a whole bunch of robots that are going to

53:55be in factories working around the clock.

53:57There's going to be a whole bunch of robots that don't move.

54:00They move a little bit.

54:02Almost everything will be robotic.

54:03What does the world look like?

54:05Sorry, let me just say, I think, like, this is one of the, the robotics

54:07for me is one of the pieces that I think unlocks economic mobility opportunities for

54:12every individual. Everyone now, like, when everyone got a car, they could now go and

54:17do a lot of different jobs.

54:18When everyone gets a robot, their robot can do a lot of work for them.

54:22They can stand up an Etsy store or a Shopify store.

54:25They can create anything they want with their robot.

54:28They could do things that they independently cannot do.

54:30I think the robot is going to end up being the greatest unlock for prosperity

54:34for more people on Earth than we can.

54:36you've ever seen with any technology before.

54:38Yeah, no doubt. I mean, just the simple math at the moment is we're millions

54:43of people short in labor today.

54:44Right? We're actually really desperate in need of robotics.

54:49And so that all of these companies could grow more if they had more labor.

54:55I mean, number one. Some of the things that you mentioned are super fun.

54:59I mean, because of robots, we'll have virtual presence.

55:02You know, I'll be able to go into the robot of my house and virtually

55:08operate it. I'm on a business trip.

55:11Right, teleoperator. Walk around the house.

55:12And walk the dog. Yeah, walk the dog.

55:14Break the leaves. Yeah, exactly.

55:16Break out the dog. Maybe not quite that, but just, you know, just wander around

55:20and just see what's going on in the house, you know, chat with the dogs,

55:23chat with the kids. Yeah.

55:25Time travel is also, we're going to be able to travel at the speed of

55:28light, you know? And so, you know, clearly we're going to send our robots ahead

55:32of us. Yeah. I'm not going to send myself.

55:34I'm going to send a robot, you know?

55:36Check it out. Yeah, yeah.

55:37And then I'm going to upload my AI.

55:39Well, it's inevitable. It unlocks the moon and it unlocks Mars as targets for colonization,

55:44which gives us infinite resources.

55:46Getting back from the moon is effectively zero energy cost to move material back because

55:50you can use solar and accelerate.

55:52So you could have factories that make everything the world needs on the moon and

55:56the robots are going to be the unlock for enabling that.

55:58That's right, this distance no longer matters.

56:00Distance doesn't matter. Yeah. The more revenue we get out of models and agents, the

56:05more we can invest in building the infrastructure, which then unlocks more capabilities on models

56:09and agents. Dario on Dworkesh's podcast recently said, by 27, 28, we'll have hundreds of

56:16billions of dollars of revenue out of the model companies and the agent companies.

56:19And he forecast a trillion dollars by 2030, right?

56:23This is non -infrastructure AI revenue.

56:26I think he's being very conservative.

56:30I believe Dario and Anthropic is going to do way better than that.

56:33Wow. Way better than that.

56:35Wow. So from 30 billion to a trillion.

56:37Yep. And the reason for that is the one part that he hasn't considered is

56:41that I believe every single enterprise software company will also be a reseller, value -added

56:49reseller of Anthropic code, Anthropics tokens.

56:52Value -added reseller of OpenAI.

56:55Have to be. That's right.

56:56Yeah. And they're going to, that part of their - Look at this logarithmic expansion.

57:01Yes. Yeah. Their go -to market is going to expand tremendously this year.

57:06What do you think in that world is the moat?

57:08What's left over? I mean, you have some moats that are, frankly, I think, as

57:12this scales, almost insurmountable. The best one that nobody talks about is probably CUDA, which

57:18is just like an incredible strategic advantage.

57:21But in the future, if a model can be used to create something incredible, then

57:27the next spin of a model can be used to maybe disrupt it.

57:30Sort of in your mind, what do you think for these companies that are building

57:33at that application layer, what's their moat?

57:35Like, how do they differentiate themselves?

57:37Deep specialization. Deep specialization. I believe that these models, they're going to have general models

57:45that are connected into the software company's agentic system Right.

57:51Many of those models are cloud models and proprietary models, but many of those models

57:57are specialized sub -agents that they've trained on their own.

58:03Right. So the call to arms for you, for entrepreneurs, is, look, know your vertical.

58:07That's right. Know it as deep and as better than everybody else.

58:11That's right. And then wait for these tools because they're catching up to you, and

58:13now you can imbue it with your...

58:15knowledge. That's right. The sooner you connect your agent, the sooner you connect your agent

58:19with customers, that flywheel is going to cause your agent to get...

58:23It very much is an inversion of what we do today because today we build

58:27a piece of software and we say, what generalizes?

58:30And then let's try to sell it as broadly as possible and then sell the

58:32customization around it. And in fact, exactly right.

58:36We create a horizontal, but notice there are all these GSIs and all of these

58:43consultants who are specialists, who then take your horizontal platform and specializes it into.

58:49Exactly. And that's arguably a five or six time bigger industry is the customization.

58:53It is. Absolutely. Very much is.

58:55That's right. So I think that these platform companies have an opportunity to become that

59:01specialist, to become that vertical domain expert.

59:04I just want to give you your flowers.

59:05I think it was three years ago you said, you're not going to lose your

59:08job to AI. You're going to lose your job to somebody using AI.

59:12And here we are, the entire conversation has revolved around this concept of agents making

59:17people superhuman and the business opportunity expanding and entrepreneurship expanding.

59:22You actually saw it pretty clearly.

59:24That's right. Have you changed your view?

59:26Well, I go... This is Doomer Dan.

59:29I'm not Doomer. Doomer Dan.

59:30I do have... Doomer. No, you can hold space for, I think, two ideas.

59:34One is there are going to be a large...

59:35That's Viral J. Calico. No, no.

59:37But that's just because he doesn't hang out with me enough.

59:40But we, I mean, we fog a little bit.

59:42Be careful what you're talking about.

59:44We will show you your breakfast table.

59:45He'll follow you around. I'm not asking for it.

59:47I'm just saying... He'll follow you around.

59:49I'm not asking for it.

59:50You can come with me and Tucker.

59:51We ski in Japan every January.

59:53Really? I love it. We and Tucker will go road trip.

59:55Wow. Okay. There is going to be job displacement.

59:58And then the question becomes, you know, do those people have the fortitude, the resolve

1:00:04to then go embrace these, you know, technologies?

1:00:07We're going to see 100 % of driving go away by humans.

1:00:10That's just, that's a beautiful thing in the lives saved.

1:00:14But we have to recognize that's 15 million people in the United States, 10 to

1:00:1715 million who are employed in that way.

1:00:20And so that is going to happen.

1:00:21Yes? I think that jobs will change.

1:00:24For example, there are many chauffeurs today who drives the car.

1:00:29I believe that many of those chauffeurs will actually be in the car sitting behind

1:00:35the steering wheel while the car is driving by itself.

1:00:38And the reason for that is because remember what a chauffeur does.

1:00:42In the end, these chauffeurs, they're helping you.

1:00:44They're your assistants. They're helping you with your luggage.

1:00:47They're helping you. I mean, they're helping you with a lot of things.

1:00:49And so I wouldn't be surprised, actually, if the chauffeurs of the future become your

1:00:55mobility assistant and they are helping you do on a whole bunch of other stuff.

1:00:59Check into the hotel. Yeah, and the car is driving by itself.

1:01:02The autopilot in planes created a lot more pilots and didn't take any of the

1:01:07pilots out of the cockpit, even though the autopilot is flying the plane 90 %

1:01:11of the time. And by the way, while that car is driving itself, that chauffeur

1:01:14is going to be doing a bunch of other work on his phone and he's

1:01:17going to be arranging, for example.

1:01:18Doing other stuff. Coordinating a bunch of things for you.

1:01:21Getting, you know. Yeah. The pie just grows in a way that.

1:01:24One of the things that, yes, every job will be transformed.

1:01:29Some jobs will be eliminated.

1:01:31However, we also know that many, many jobs will be created.

1:01:35The one thing that I will say to young people who are coming out of

1:01:37school who are anxious about AI, be the expert of using AI.

1:01:43Yes. How much, look, we all want our employees to be expert at using AI

1:01:48and it's not trivial. Not trivial.

1:01:52Yeah. And so. Knowing how to specify, not to over -prescribe, leaving enough room for

1:01:59the AI to innovate and create while we guide it to the outcome we want,

1:02:04all of that requires artistry.

1:02:07You had this great advice to when you were at Stanford, I think it was,

1:02:10which is, I wish to you pain and suffering.

1:02:12Do you remember that? Yeah.

1:02:13Fantastic. What's your advice to young people around what they should be studying?

1:02:17So if they're sort of about to leave high school, because now those are the

1:02:22kids that are at this like really native, they haven't made a decision about college,

1:02:25what to study, if at all, go to college.

1:02:28How do you guide those kids?

1:02:29What would you tell them?

1:02:31I still believe that deep science, deep math, language skills, you know, as you know,

1:02:39language is the programming language of AI.

1:02:42The ultimate programming language. And so as it turns out, it could be that the

1:02:46English major could be the most successful.

1:02:48And so I think I would just advise whatever education you get, just make sure

1:02:55that you're deeply, deeply expert in using AIs.

1:02:59One of the things that I wanted to say with respect to jobs, and I

1:03:02want everybody to hear it, that in fact, at the beginning of the deep learning

1:03:06revolution, one of the finest computer scientists in the world, deeply, deeply, I deeply, deeply

1:03:12respect, predicted that computer vision will completely eliminate radiologists.

1:03:19And that the one field he advises everybody to not go into is radiology.

1:03:26Ten years later, his prediction was at 100 % right.

1:03:31Computer vision has been integrated into all of the radiology technologies and radiology platforms in

1:03:37the world 100%. The surprising outcome is the number of radiologists actually went up and

1:03:43the demand for radiologists is skyrocketed.

1:03:46And the reason for that is because everybody's job has a purpose and its task.

1:03:54The task that you do is studying the scans.

1:03:58But your purpose is to help the doctors, helping the patient diagnose disease.

1:04:04And so what's surprising is, because the scans are now being done so quickly, they

1:04:09could do more scans, improving health care.

1:04:12Yes. But doing more scans more quickly allows patients to be onboarded a lot more

1:04:20quickly. Treated a lot more quickly.

1:04:22And as it turns out, because hospitals enjoy making money too.

1:04:26Yeah. Right. They're doing more scans.

1:04:29They're treating more customers. Early detection becomes...

1:04:32The revenues go up. And guess what?

1:04:34Perfect detail. Need more... Yeah.

1:04:35And a country that grows faster, productivity increases, a wealthier country can put more teachers

1:04:41in the classroom, not less teachers in the classroom.

1:04:43That's right. You just give every one of those teachers a personalized curriculum for every

1:04:47student in the room. It makes them all bionic and leads to a lot more.

1:04:51Every single student will be assisted by AI, but every single student will need great

1:04:56teachers. Yeah. Amazing. Jensen, congratulations on all your success.

1:05:00And really, this is an incredibly positive, uplifting discussion.

1:05:03We really appreciate you taking the time for us.

1:05:06He is the steward we need.

1:05:07You are. You are the steward we need.

1:05:09I think you need to be more vocal.

1:05:10No, I'm being very, very honest.

1:05:11Be more vocal about the positive side of it.

1:05:13I think there's so much humorism.

1:05:14But I also think it takes the humility to have this level of success and

1:05:17be humble about, we're making software, guys.

1:05:20Yeah. And I think that that's actually really healthy for people to hear.

1:05:25We have done this before.

1:05:26We have invented categories and industries before.

1:05:29Yes. We don't need to go to this skewer.

1:05:31fear mongering place it does nothing and we get to choose right we have autonomy

1:05:36and agency we get to pick how to we sure do play this okay everybody

1:05:40we'll see you next time thank you on the all -in interview okay well done

1:05:45brother thanks man good job thank you sir that was awesome good good you guys

1:05:50are awesome look at this look at this big crowd behind you guys man i

1:05:53think they're here for you

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