Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494

0:00 The following is a conversation with Jensen Huang, CEO of NVIDIA,

0:04 one of the most important and influential

0:07 companies in the history of human civilization.

0:11 NVIDIA is the engine powering the AI revolution,

0:13 and a lot of its success can be directly attributed to Jensen's sheer

0:18 force of will and his many brilliant bets and decisions as a leader,

0:23 engineer, and innovator.

0:26 This is Lex Fridman Podcast.

0:28 And now dear friends, here's Jensen Huang.

0:33 You've propelled NVIDIA into a new era in AI,

0:37 moving beyond its focus on chip scale design to now rack scale design.

0:41 And I think it's fair to say that winning for NVIDIA for a long

0:45 time used to be about building the best GPU possible, and you still do,

0:49 but now you've expanded that to extreme co-design of GPU,

0:53 CPU memory, networking, storage, power cooling, software, the rack itself,

0:58 the pod that you've announced, and even the data center.

1:01 So let's talk about extreme co-design.

1:03 What is the hardest part of co-designing system

1:07 with that many complex components and design variables?

1:11 Yeah, thanks for that question.

1:12 So first of all, the reason why extreme co-design is necessary is because

1:16 the problem no longer fits inside one computer to be accelerated by one GPU.

1:24 The problem that you're trying to solve is you would like

1:27 to go faster than the number of computers that you add.

1:31 So you added you know, 10,000 computers,

1:33 but you would like it to go a million times faster.

1:37 Then all of a sudden you have to take the algorithm,

1:43 you have to break up the algorithm,

1:44 you have to refactor it, you have to shard the pipeline,

1:48 you have to shard the data, you have to shard the model.

1:50 Now all of a sudden when you distribute the problem this way,

1:56 not just scaling up the problem,

1:58 but you're distributing the problem, then everything gets in the way.

2:02 This is the Amdahl's Law problem where the amount of speed up you

2:07 have for something depends on how much of the total workload it is.

2:12 And so if computation represents 50% of the problem,

2:17 and I sped up computation infinitely like a million times,

2:23 you know, I only sped up the total workload by a factor of two.

2:27 Now all of a sudden, not only do you have to distribute a computation,

2:31 you have to, you know, shard the pipeline somehow.

2:34 You also have to solve the networking problem because

2:38 you've got all of these computers are all connected together.

2:42 And so distributed computing at the scale that we do, the CPU is a problem,

2:48 the GPU is a problem, the networking is a problem, the switching is a problem.

2:52 And distributing the workload across all these computers is a problem.

2:57 It's just a massively complex computer science problem.

2:59 And so we just gotta bring every technology to bear.

3:04 Otherwise, we scale up linearly or we scale

3:10 up based on the capabilities of Moore's Law,

3:12 which has largely slowed because Dennard scaling has slowed.

3:16 I'm sure there's trade-offs there.

3:18 Plus you have a complete disparate disciplines here.

3:20 I'm sure you have specialists in each one of these high bandwidth memory,

3:24 the network and the NVLink,

3:26 the NICs, the optics and the copper that you're doing,

3:29 the power delivery, the cooling, all of that.

3:30 I mean, there's like world experts in each of those.

3:32 How do you get 'em in a room together

3:34 to figure out-- That's why my staff is so large.

3:37 What's the process—can you take me through

3:39 the process of the specialists and the generalists?

3:41 Like how do you put together the rack when you know the s-

3:45 the set of things you have to shove into a rack together?

3:48 Like what does that process look like of designing it all together?

3:51 Yeah.

3:51 There's the first question, which is: what is extreme co-design?

3:55 You're, you, we're optimizing across the entire

3:57 stack of software from architectures to chips,

4:00 to systems, to system software, to the algorithms, to the applications.

4:04 That's one layer.

4:05 The second thing that you and I just talked about is goes beyond

4:09 CPUs and GPUs and networking chips and scale up switches and scale out switches.

4:15 And then of course, you gotta include power and cooling and all of that because,

4:20 you know, all these computers are extremely, extremely power hungry.

4:24 They do a lot of work and they're very energy efficient,

4:28 but they in aggregate still consume a lot of power.

4:31 And so that's one.

4:32 The first question is, what is it?

4:34 The second question is, why is it, and we just spoke about the reason,

4:38 you know you want to distribute the workload so that you

4:41 can exceed the benefit of just increasing the number of computers.

4:47 And then the third question is, how is it, how do you do it?

4:51 And that's the, that's kind of the miracle of this company.

4:55 You know, when you're designing a computer,

4:57 you have to have operating system of computers.

4:59 When you're designing a company,

5:02 you should first think about what is it that you want the company to produce.

5:05 You know, I see a lot of companies organization charts,

5:07 and they all look the same.

5:09 Hamburger organization charts, soft organization charts,

5:12 and car company organization charts.

5:14 They all look the same.

5:16 And it doesn't make any sense to me.

5:17 You know, the goal of a company is to be the company is to be the machinery,

5:22 the mechanism, the system that produces the output.

5:26 And that output is the product that we like to create.

5:29 It is also designed, the architecture of the company should

5:33 reflect the environment by which it exists.

5:36 It almost indirectly says what you should do with the organization.

5:40 My direct staff is 60 people.

5:43 You know, I don't have one-on-ones with 'em because it's impossible.

5:46 You can't have, you can't have 60 people on your staff if you're,

5:49 you know, gonna get work done and-- So you still have 60 reports.

5:52 You still have across-- More, yeah.

5:54 More.

5:54 And most stars at least have a foot in engineering.

5:59 Almost all of them.

6:01 There's experts in memory, there's experts in CPUs, there's experts in optical.

6:05 All, all—- That's incredible.

6:07 Yeah, GPUs and— Architecture, algorithms, design, um—- So,

6:12 you constantly have an eye on the entire stack, and you're having to, like,

6:15 intense discussions about the designer of the entire stack?

6:18 And no conversation is ever one person.

6:20 That's why I don't do one-on-ones.

6:22 We present a problem and all of us attack it.

6:26 You know, because we're doing extreme co-design.

6:30 And literally, the company is doing extreme co-design all the time.

6:33 So, even if you're talking about a particular component,

6:36 like cooling, networking, everybody's listening in?

6:41 Yeah, exactly.

6:41 And they can contribute, "Well, this doesn't work for the power distribution.

6:44 This doesn't-"- Exactly.

6:46 This doesn't work for the memory.

6:47 This doesn't work for this."- Exactly.

6:50 And whoever wants to tune out, tune out.

6:52 You know what I'm saying?

6:54 And the reason for that is because the people who are on the staff,

6:57 they know when to pay attention.

6:59 There's supposed...

6:59 You know, it's something they could have contributed to, they

7:02 didn't contribute to, "I'm going to call them out." You know?

7:04 And so, "Hey, come on, let's get in here."- So,

7:07 as you mentioned, NVIDIA is this company that's adapting to the environment.

7:11 So, at which point can you say,

7:13 did the environment change and began adapting sort of secretly-...

7:19 in the early days from GPU for gaming,

7:22 maybe the early deep learning revolution to we're now

7:24 going to start thinking of it as an AI factory?

7:27 What does NVIDIA do?

7:28 It produces AI, let's build a factory that makes AI.

7:32 Uh, I could, I c- you,

7:33 you could- I could reason through what just systematically.

7:35 We started out as a, as an accelerator company.

7:39 But the problem with accelerators is that the application domain's too narrow.

7:43 It has the benefit of being incredibly optimized for the job.

7:47 You know, any specialist has that benefit.

7:49 The problem with intense specialization is that, of course,

7:53 your market reach is narrower, but that's, that's even fine.

7:59 The problem is, the market size also dictates your R&D capacity.

8:06 And your R&D capacity ultimately dictates the influence

8:11 and impact that you can possibly have in computing.

8:13 And so, when we first started out as an accelerator, very specific accelerator,

8:19 we always knew that had- that was going to be our first step.

8:22 We had to find a way to become accelerated computing.

8:25 But the problem is, when you become a computing company,

8:29 it's too general purpose and it takes away from your specialization.

8:33 The tur- I connected two words that are actually have fundamental tension.

8:39 The better computing company we become, the worse we became as a specialist.

8:43 The more of a specialist, the less capacity we have to do overall computing.

8:48 And so, that...

8:50 And I connected those two words together on purpose,

8:52 that the company has to find that really narrow path,

8:57 step by step by step, to expand our aperture of computing,

9:02 but not give up on the most important specialization that we had.

9:06 Okay, so the first step that we took beyond acceleration was,

9:10 we invented a programmable pixel shader.

9:13 So, that was the first step towards programmability.

9:15 You know, it was our first journey towards moving into the world of computing.

9:20 The second thing that we did was we created we put FP32 into our shaders.

9:28 That FP32 step, IEEE-compatible FP32,

9:31 was a huge step in the direction of computing.

9:35 It was the reason why all of the people

9:38 who were working on, on stream processors and, you know,

9:43 other types of data flow processors discovered us.

9:45 And they said, "Hey, all of a sudden, you know,

9:48 we might be able to use this GPU that's incredibly computationally intensive,

9:51 and it's now, you know,

9:53 compliant with IEEE." I can take my software that I was writing, you know,

9:58 previously on CPUs, and I can, you know,

10:00 see about, you know, using the GPU for that.

10:04 And which led us to create, put C on top of FP32, what's called, we call Cg.

10:11 The Cg path took us to eventually CUDA.

10:14 CUDA, step by step by step We...

10:18 Well, putting CUDA on GeForce,

10:20 that was a strategic decision that was very, very hard to do,

10:24 because it cost the company enormous amounts of our profits,

10:28 and we couldn't afford it at the time.

10:30 But we did it anyways because we wanted to be a computing company.

10:34 A computing company has a computing architecture.

10:37 A computing architecture has to be compatible

10:40 across all of the chips that we build.

10:42 Can you take me through that decision?

10:44 So, putting CUDA on GeForce, could not afford to do?

10:46 Can you explain that decision?

10:48 Why, why boldly choose to do that anyway?

10:52 Can you explain that decision?

10:53 Yeah, excellent.

10:54 That was, that was the first...

10:55 I would say that that was the first strategic

11:00 decision that is as close to an existential threat.

11:07 For people who don't know, it turned out to be, spoiler alert,

11:10 one of the most incredibly brilliant decisions ever made by a company.

11:15 So, CUDA turned out to be an incredible

11:19 foundation for computation in this AI infrastructure world.

11:23 So-- Thank you-...

11:24 just setting the context.

11:25 It turned out to be a good decision.

11:27 Yeah, it turned out to have been a good decision.

11:29 I think the...

11:30 So, here's the way it went.

11:31 So, we invented this thing called CUDA,

11:34 and It expanded the aperture of applications

11:38 that, that we can accelerate with our accelerator.

11:41 The question is, how do we, how do we attract developers to CUDA?

11:48 Because a computing platform is all about developers.

11:51 And developers don't come to a computing platform just because,

11:57 you know, it could perform something interesting.

11:59 They come to a computing platform because the install base is large.

12:04 Because a developer, like anybody else,

12:06 wants to develop software that reaches a lot of people.

12:09 So, the install base is, in fact,

12:11 the single most important part of an architecture.

12:14 The architecture could attract enormous amounts of criticism.

12:18 For example, no architecture has ever attracted more criticism than the x86....

12:24 you know, as a less than, less than elegant architecture,

12:29 but yet it is the defining architecture of today.

12:32 It gives you an example that in fact

12:35 so many RISC architectures which were beautifully architected,

12:42 incredibly well-designed by some of the brightest

12:44 computer scientists in the world, largely failed.

12:47 And so I've given you two examples where one is, you know, one is elegant,

12:52 the other one's barely aesthetic,

12:54 and so yet x86 survived and the reason for-- Install base is everything.

12:59 Install base defines an architecture.

13:01 Not...

13:02 Everything else is secondary, okay?

13:04 And so there were other architectures at the time.

13:07 CUDA came out, OpenCL was here.

13:09 There were...

13:09 You know, there's several other competing architectures.

13:11 But the thing that...

13:12 The decision that we made that was good was we said,

13:16 "Hey, look, ultimately it's about,

13:19 Install base and what is the best way we could

13:22 get a new computing architecture into the world?" By that timeframe,

13:27 GeForce had become successful.

13:29 We were already selling millions and millions of GeForce GPUs a year,

13:33 and we said, "You know, we ought to put CUDA on GeForce and put it

13:39 into every single PC whether customers use it or not,

13:43 and use it as a starting point of cultivating our install base." Meanwhile,

13:48 we'll go and attract developers, and we went to universities and wrote

13:55 books and taught classes and put CUDA everywhere.

13:58 And eventually people discover...

14:00 And at the time, the PC was the primary computing vehicle.

14:03 There was no cloud, and we could put

14:05 a supercomputer in the hands of every researcher in school,

14:08 every scientist, you know, every engineering school, every...

14:11 or every student in school, and eventually something amazing will happen.

14:15 Well, the problem was CUDA increased our cost of that GPU,

14:20 which is a consumer product, so tremendously, it completely consumed all

14:25 of the company's gross profit dollars.

14:29 And so at the time, the company was probably,

14:31 you know, worth, I don't know, at the time, eight...

14:36 Was it like $8 billion or something?

14:37 Like six, $7 billion or something like that.

14:40 After we launched CUDA, I recognized that it was going to add so much cost,

14:47 but it was something we believed in.

14:50 You know, our market cap went down to like one and a half billion dollars.

14:53 And so we were down,

14:55 we were down there for a while and we clawed our way way back slowly,

15:00 but we carried CUDA on GeForce.

15:02 I always say that NVIDIA is the house that GeForce built,

15:06 because it was GeForce that took CUDA out to everybody.

15:10 Researchers, scientists,

15:11 they discovered CUDA on GeForce because they were all, you know...

15:18 Many of 'em were gamers.

15:19 Many of them built their own PCs anyways.

15:22 In a university lab, many of them built clusters themselves,

15:26 you know, using PC components.

15:28 And, and so that, you know, that's kind of how we got going.

15:31 And then that became the platform

15:33 and the foundation for the deep learning revolution.

15:35 That was also another great, great observation.

15:37 Yeah.

15:38 That existential moment, do you remember...

15:40 Like, what were those meetings like?

15:42 What were those discussions like, deciding as a company, risking everything?

15:48 Well I had to make it clear to the board what we're trying to do,

15:55 and the management team knew our gross margins were gonna get crushed.

16:00 So you could imagine a world where GeForce

16:03 would carry the burden of CUDA and none

16:06 of the gamers would appreciate it and none of the gamers would pay for it.

16:10 You know, they only pay certain price and it doesn't matter what your cost is.

16:14 And so the...

16:15 You know, we, we increased our cost by 50% and that con- consumed...

16:20 And we were a 35% gross margin company, and so it, it was a...

16:25 It was quite a difficult decision to make.

16:27 But you could imagine that someday this would go into workstations

16:30 and it would go into supercomputers and, and in those segments,

16:34 maybe we can capture more margin.

16:37 so you could, you could reason your way

16:39 into being able to afford this, But it still took...

16:43 It took a decade.

16:45 But that, but that's more of, like,

16:46 conversation with the board convincing them, but you psychologically-...

16:49 as NVIDIA's continued to make bold bets that predict the future,

16:57 and in part, especially now, define the future.

17:01 So I'm almost looking for wisdom about how

17:05 you're able to make those decisions, to make leaps-...

17:08 like that as a company.

17:15 Well, first of all I'm informed by a lot of curiosity.

17:21 At some point, there's a reasoning system that, that convinces me,

17:28 so clearly this outcome will happen.

17:31 That this will happen.

17:34 And so I believe it in my mind,

17:36 and when I believe it in my mind, you know, you know how it is.

17:40 You manifest a future and that future is so convincing,

17:44 there's no way it won't happen.

17:48 There's a lot of suffering in between,

17:50 but you've gotta believe what you believe.

17:53 So you envision the future-...

17:56 and you essentially, from a sort of engineering perspective, manifest it?

18:00 Yeah.

18:00 And you, you reason about how to get there.

18:02 You reason about why it, it must exist.

18:06 And, you know, I reason...

18:08 We all reason it here.

18:10 The management team would reason about it.

18:11 All the people that I...

18:13 We spend a lot of time reasoning about it.

18:15 The thing, the thing that...

18:17 The next part of it is probably a skill thing, which is,

18:20 you know, oftentimes in leadership the leadership

18:24 stays quiet or they learn about something,

18:26 and then they do some manifesto, and it's a brand-new year,

18:29 and somehow at the end of the year,

18:31 next year, we're gonna have a brand-new plan.

18:34 Big huge layoff this way,

18:35 big huge organization change this way, new mission statement...

18:39 brand new logos you know, that kind of stuff.

18:43 Um, we've just never, I never do things that way.

18:46 When I learn about something and it's starting to influence how I think,

18:50 I'll make it very clear to everybody near me that, you know,

18:54 this, this is interesting.

18:57 This is going to make a difference.

18:59 This is going to impact that.

19:01 And I reason about things step by step by step.

19:04 Oftentimes, I've already made up my mind,

19:06 but I'll take every possible opportunity,

19:09 external information, new insights, new discoveries, New engineering,

19:14 you know, revelations, new milestones developed,

19:17 I'll take those opportunities and I'll use

19:20 it to shape everybody else's belief system.

19:25 And I'm doing that literally every single day.

19:28 I'm doing that with my board,

19:30 I'm doing that with my management team, I'm doing that with my employees.

19:33 I'm trying to shape their belief systems such that when I come the day I say,

19:40 "Hey, let's buy Mellanox," it's completely

19:44 obvious to everybody that we absolutely should.

19:48 On the day that I said, "Hey guys,

19:51 let's go all in on deep learning," and let me tell you why.

19:55 I've already been laying down the bricks

19:58 to different organizations inside the company.

20:01 Every organization and everybody,

20:05 many of the people might have heard everything.

20:08 Most of the company hears, of course, pieces of it.

20:11 And on the day that I announce it,

20:14 everybody's kind of bought in to many pieces of it.

20:19 And in a lot of ways, I like to announce these things,

20:23 and I imagine, that the employees are kind of saying,

20:27 "You know, Jensen, what took you so long?" And in fact,

20:31 I've been shaping their belief system for some time, and therefore leadership.

20:35 Sometimes it looks like you're leading from behind,

20:39 but you've been shaping their, you know,

20:41 to the point where on the day that I declared it, 100% buy-in.

20:44 But that's what you want.

20:45 You want to bring everybody along.

20:47 You know, otherwise,

20:48 we announce something about deep learning and everybody goes,

20:51 "What are you talking about?" You know,

20:52 you announce something about let's go all in on this thing,

20:56 and your management team, your board, your employees, your customers,

21:00 they're kind of like, "Where's this coming from?

21:02 You know, this is insane." And so, so GTC effect,

21:06 if you go back in time, you look at the keynotes,

21:11 I'm also shaping the belief system of my partners

21:14 in the industry and I'm using that to shape,

21:17 you know, the belief system of my own employees.

21:20 And so by the time that I announce something,

21:24 like for example, we just announced Grok.

21:27 We've been late...

21:28 I've been talking about the stepping stones for two and a half years.

21:33 You just go back and go, "Oh my gosh,

21:35 they've been talking about it for two and a half years."

21:37 And so I've been laying the foundation step by step by step,

21:41 so when the time comes you announce it, everybody's saying, "You know,

21:43 what took you so long?"- But it's not just inside the company.

21:46 You're shaping the landscape, the broader global landscape of innovation.

21:49 Like, putting those ideas out there, you really are manifesting reality.

21:53 We don't build computers.

21:54 We actually don't build clouds.

21:56 We don't...

21:56 As it turns out, we're a computing platform company.

21:59 And so nobody can buy anything from us.

22:02 That's the weird thing.

22:03 You know, we vertically design, vertically integrate to design and optimize,

22:09 but then we open up the entire platform

22:12 at every single layer to be integrated into other

22:16 companies' products and services and clouds and supercomputers

22:20 and OEM computers and so the amazing thing is,

22:23 I can't do what I do without having convinced them first.

22:27 And so most of GTC is about manifesting a future that by the time that we...

22:34 My product is ready, they're going, "What took you so long?"- Yeah.

22:39 So one of the things you've been a believer

22:42 for a long time is scaling laws, broadly defined.

22:46 So are you still a believer in the scaling laws?

22:49 Yeah, yeah.

22:50 Yeah, we have more scaling laws now.

22:51 So I think you've outlined four of them with pre-training,

22:54 post-training, test time, and agentic scaling.

22:56 What do you think, when you think about the future,

23:01 deep future and the near-term future,

23:03 what are the blockers that you're most concerned about that keep you

23:08 up at night that you have to overcome in order to keep scaling?

23:13 Well, we can go back and reflect on what people thought were blockers.

23:16 So in the beginning, we were the first...

23:19 The pre-training scaling law.

23:20 You know, people thought well rightfully so,

23:24 that the amount of data that we have,

23:26 high-quality data that we have will limit the intelligence that we achieve.

23:30 And that scaling law was an important, very important scaling law.

23:34 The larger the model, the correspondingly more data results in a better...

23:38 With a results in a smarter AI.

23:41 And so that was pre-training.

23:43 And Ilya Sutskever, Ilya said, "We're out of data," or something like that.

23:47 "Pre-training is over," or something like that.

23:49 The industry panicked, you know, that this is the end of AI.

23:54 And of course, of course that, that's obviously not true.

23:56 We're gonna keep on scaling the amount of data that we have to, to train with.

24:01 A lot of that data is probably gonna be synthetic,

24:04 and that also confused people, you know?

24:06 And what people don't realize is they've kind of forgotten that most of the data

24:12 that, that we are training that we teach

24:14 each other with, inform each other with, is synthetic.

24:17 You know, I...

24:19 It's synthetic because it didn't come out of nature.

24:23 You created it.

24:25 I'm consuming it.

24:26 I modify it, augment it, I regenerate it, somebody else consumes it.

24:32 And so, so we've now reached a level where

24:37 AI is able to take ground truth, augment it......

24:43 Enhance it, synthetically generate an enormous amount of data.

24:47 And that part of post-training continues to scale,

24:50 and so the amount of data that we

24:52 could use that is human generated will be smaller, and smaller, and smaller.

24:57 The amount of data that we use to train model,

25:02 is going to continue to scale to the point where we're no longer limited...

25:07 Training is no longer limited by...

25:08 Data is now limited by compute.

25:10 And the reason for that is most of the data is synthetic.

25:13 Then the next phase is test time,

25:17 and I still remember people telling me that, "Inference?

25:21 Oh, yeah, that's easy.

25:23 Pre-training, that's hard." These are giant

25:25 systems that people are talking about.

25:27 Inference must be easy.

25:28 And so inference chips are gonna be little tiny chips, and...

25:32 you know, they're not, they're not like NVIDIA's chips.

25:34 Oh, those are gonna be complicated and expensive,

25:36 and, you know, we could make...

25:38 And this is- in the future, inference is gonna be the biggest market,

25:41 and it's gonna be easy, and we're gonna commoditize it.

25:44 You know, everybody can build their own chips.

25:46 And, and that was always illogical to me because inference is thinking,

25:53 and I think thinking is hard.

25:56 Thinking is way harder than reading.

25:59 You know, pre-training is just memorization and generalization,

26:02 you know, and looking for patterns in relationships.

26:05 You're reading and reading, versus thinking, reasoning, solving problems,

26:12 taking unexplored experiences, new experiences, and breaking it down into...

26:20 Decomposing it into, you know, solvable pieces that we then go off,

26:25 either through first principle reasoning,

26:27 or, you know, through previous examples, prior experiences.

26:31 You know, or just uh,

26:34 exploration and search and, you know, trying different things.

26:38 And that whole process of, of test time scaling,

26:44 Inference, is really about thinking.

26:46 And it's about reasoning, it's about planning, it's about search, it's about...

26:50 And so how could that possibly be compute light?

26:53 And we were absolutely right about that.

26:56 You know, so test time scaling is intensely compute intensive.

27:01 Then the question is, okay,

27:02 now we're at inference and we're at test time scaling, what's beyond that?

27:06 Well, obviously, we have now created, you know, one agentic person,

27:12 and that one agentic person has a large language model that we've now now,

27:17 you know, developed.

27:18 But during test time,

27:19 that agentic system goes off and does research and bangs on databases,

27:24 and it goes out and, you know,

27:27 uses tools, and one of the most important things it does

27:30 is spins off and spawns off a whole bunch of sub-agents.

27:32 Which means we're now creating large teams.

27:36 It's so much easier to scale NVIDIA by hiring

27:40 more employees than it is to scale myself.

27:44 And so the next scaling law is the agentic scaling law.

27:46 It's kind of like multiplying AI.

27:51 Multiplying AI, we could spin off agents as fast as you want to spin off agents.

27:55 And so, you know, I...

27:57 You know, I have four scaling laws.

27:59 And as we use the agentic systems,

28:02 they're gonna create a lot more data, they're gonna create a lot of experiences.

28:06 Some of it we're gonna say, "Wow, this is really good.

28:09 We ought to memorize this." That data set

28:12 then comes all the way back to pre-training.

28:15 We memorize and generalize it.

28:16 We then refine it and fine-tune it back into post-training.

28:21 Then we enhance it even more with test time, you know,

28:25 and the agents, agentic systems, you know, put it out to the industry.

28:30 And so this loop, this cycle, is gonna go on and on and on.

28:34 It kinda comes down to basically intelligence is

28:38 gonna scale by one thing, and that's compute.

28:42 But there's a tricky thing there that you have to anticipate and predict,

28:45 which is some of these components,

28:48 it requires different kind of hardware to really do it optimally.

28:52 So you have to anticipate where the AI innovation's going to lead.

28:56 For example, a mixture of-- Perfect-...

28:58 experts with sparsity.

28:59 Perfect.

29:00 With hardware, you can't just pivot on a week's notice.

29:03 You have to anticipate what that's going to look like.

29:05 It has some-- So good-...

29:07 that's so scary and difficult to do, right?

29:09 For example, These AI model architectures are

29:12 being invented about once every six months.

29:15 Right?

29:16 And system architectures and hardware architectures kind of every three years.

29:26 And so you need to anticipate what likely is going to happen,

29:30 you know, two, three years from now.

29:33 And there's a couple ways that you could do that.

29:34 First of all, we could do research internally ourselves,

29:36 and that's one of the reasons why we have basic research,

29:38 we have applied research.

29:40 We create our own models.

29:41 And so we have hands-on life experience right here.

29:44 This is part of the co-design that I'm talking about.

29:48 We're also the only AI company in the world

29:49 that works with literally every AI company in the world.

29:51 And so to the extent that we can,

29:54 we try to get a sense of what are the challenges that people are experiencing.

29:59 So you're listening to the whispers across the industry, the AI labs.

30:02 That's right.

30:03 You got to listen and learn from everybody.

30:05 And have a...

30:06 And then the last part is to have an architecture that's flexible,

30:10 that can adapt and move with the wind.

30:12 And one of the benefits of, of CUDA is that it's,

30:16 you know, on the one hand, an incredible accelerator.

30:19 On the other hand, it's really flexible.

30:22 And so that balance, incredible balance between specialization,

30:26 otherwise we can't accelerate the CPU,

30:31 versus generalization, so that we can adapt with changing algorithms,

30:35 that's really, really important.

30:37 That's the reason why CUDA has been so resilient on the one hand,

30:41 and yet we continue to enhance it.

30:44 We're at CUDA 13.2, and so we're evolving the architecture

30:48 so fast that we can stay with, you know, with the modern algorithms.

30:55 For example...

30:56 When mixtures of experts came out,

30:59 That's the reason why we had NVLink 72 instead of NVLink eight.

31:04 We could now take an entire four trillion,

31:06 10 trillion parameter model and put it in one

31:09 computing domain as if it's running on one GPU.

31:13 Um, people probably didn't notice, I said it,

31:18 but if you look at the architecture of the Grace Blackwell racks,

31:24 it was completely focused on doing one thing, processing the LLM.

31:30 All of a sudden, one year later, you're looking at a Vera Rubin rack.

31:34 It has storage accelerators.

31:36 It has this incredible new CPU called Vera.

31:41 It has Vera Rubin and NVLink 72 to run the LLMs.

31:45 It also has this new additional rack called Rock.

31:49 And so this entire rack system is completely different than the previous one,

31:55 and it's got all these new components in it.

31:58 And the reason for that is because the last

32:00 one was designed to run MoE large language models, inference.

32:04 And this one is to run agents and agents bang on tools, and-- Obviously,

32:11 the design of the system had to have

32:15 been done before Claude Code, Codex, OpenClaw.

32:18 So you were anticipating the future, essentially.

32:21 And that comes from what?

32:23 From the whispers, from understanding what all the state-- No-...

32:26 of the art is about?

32:26 No, it's easier than that.

32:28 You just reason about it.

32:31 First of all, you just reason.

32:34 no matter, no matter what happens,

32:37 at some point in order for that large language model to be a digital worker...

32:42 Let's just, let's just use that metaphor.

32:44 Let's say that we want the LLM to be a digital worker.

32:47 What does that have to do?

32:49 It has to access ground truth.

32:51 That's our file system.

32:52 It has to be able to do research.

32:54 It doesn't know everything.

32:55 We don't have...

32:56 And I don't wanna wait until this AI becomes, you know,

33:00 universally smart about everything, past, present,

33:03 and future before I make it useful.

33:05 And so therefore, I might as well let it go do research.

33:09 It's obviously, if it wants to help me, it's gotta use my tools.

33:13 You know, a lot of people would say,

33:15 "You know AI is gonna completely destroy software.

33:17 We don't need software anymore.

33:18 We don't even need tools anymore." That's ridiculous.

33:20 Let's use the...

33:22 Let's use a thought experiment.

33:25 And you could just sit there, enjoy a glass of whiskey,

33:29 and, and think about all these things, and it would become completely obvious.

33:32 Like, if I were to create the most amazing- the most

33:37 amazing agent that we can imagine in the next 10 years.

33:41 Let's say it'd be a humanoid robot.

33:43 If that humanoid robot were to be created,

33:46 is it more likely that the humanoid robot comes into my house and uses

33:50 the tools that I have to do the work that it needs to do?

33:54 Or does this hand turns into a 10-pound hammer in one instance,

33:59 turns into a scalpel in another instance,

34:02 and in order to boil water, it beams, you know, microwaves out of its fingers?

34:07 You know, or is it more likely just to use a microwave, you know?

34:10 And the first time it goes up to the microwave,

34:13 it probably doesn't know how to use it.

34:15 But that's okay.

34:16 It's connected to the internet.

34:17 It reads the manual of this microwave, reads it, instantly becomes an expert.

34:24 And so it uses it.

34:26 And so I think the...

34:27 I just described, in fact, almost all of the properties of OpenClaw.

34:35 You know, that it's gonna use tools,

34:36 that it's gonna access files, it's gonna be able to do research.

34:39 It has I/O subsystem.

34:41 And when you're done reasoning through it,

34:43 reasoning about it in that way, Then you say, "Oh, my gosh,

34:49 the impact to the future of computing is

34:52 deeply profound." And the reason for that is,

34:55 I think we've just reinvented the computer.

34:57 And then now you say, "Okay, when did we reason about that?

35:00 When did we reason about OpenClaw?" If you

35:03 take the OpenClaw schematic that I used at GTC, you'll find it two years ago.

35:11 Literally, two years ago at GTC,

35:13 I was talking about agentic systems that exactly reflect OpenClaw today.

35:21 And, of course, the confluence of many things had to happen.

35:27 First of all, we needed Claude and GPT and, you know,

35:30 all of these models to reach a level of capability.

35:33 So their innovation and their breakthroughs

35:35 and their continued advances was really important.

35:39 And then, of course, somebody had to create an open source,

35:42 you know project that was sufficiently robust,

35:46 you know, and sufficiently complete and that we can all put to work.

35:53 And I think OpenClaw did for, did

35:56 for agentic systems what ChatGPT did for generative systems.

35:59 And I just think it's a very big deal.

36:02 Yeah, it's a really special moment.

36:04 I'm not exactly sure why it captured so much of the world's attention,

36:08 but it did, more than Claude Code and Codex and so on.

36:12 Because consumers could reach it.

36:14 Sure, yeah.

36:14 But there's also so much of this is vibes.

36:17 And Peter, I had a podcast with him, he's a wonderful human being.

36:22 So part of it is also the humans that represent the thing.

36:25 Yeah, no doubt.

36:25 Part of it is memes and the— 'Cause we're all trying to figure it out.

36:28 There's really serious and complicated security concerns

36:31 about when you have such powerful technology,

36:33 how do you hand over your data so they can do useful stuff?

36:37 But then there's scary things associated with that.

36:39 And we, as a civilization, as individual people and as a civilization,

36:42 figuring out how to find that right balance.

36:44 Yeah, we jumped on it right away and we

36:47 sent a bunch of security experts this way.

36:49 And we did this thing called OpenShell.

36:52 It's already been integrated into, into OpenClaw.

36:56 And NVIDIA put forward NemoClaw.

36:58 Yep, exactly.

36:59 They install super easy.

37:01 It makes sure that it's secure.

37:03 We give you two out of three rights.

37:06 Agentic systems can access sensitive information,

37:08 it can execute code, and it can communicate externally.

37:11 We could keep things safe if we gave you two out

37:16 of those three capabilities at any time, but not all three.

37:21 And out of those two out of three capabilities,

37:23 we also give you access control based

37:25 on whatever rights that you're given by enterprise.

37:29 And then we connect it to a policy

37:31 engine that all these enterprises already have.

37:33 And so we're going to try to do our best to help OpenClaw become a better claw.

37:40 So you eloquently explained how we have a long history

37:43 of blockers that we thought were going to be blockers, and we overcame them.

37:47 But now looking into the future, what do you think might be the blockers

37:50 now that it's clear that agents will be everywhere?

37:53 So obviously we're going to need compute.

37:55 So what is going to be the blocker for that scaling?

38:00 Power is a concern, but it's not the only concern.

38:02 But that's the reason why we're pushing so hard on extreme co-design,

38:07 so that we can improve the tokens per

38:12 second per watt orders of magnitude every single year.

38:16 And so in the last 10 years, Moore's Law would have progressed computing about

38:21 100 times in the last 10 years.

38:24 We progressed and scaled up computing by a million times in the last 10 years.

38:29 And so we're gonna keep on doing that through extreme co-design.

38:32 So energy efficiency, perf per watt,

38:36 completely affects the revenues of a company.

38:39 It affects the revenues of a factory.

38:42 And we're just going to push that to the limit so that we

38:46 can keep on driving token costs down as fast as we can.

38:50 You know, the our computer price is going up,

38:54 but our token generation effectiveness is going up

38:57 so much faster that token cost is coming down.

39:00 It's just coming down an order of magnitude every year.

39:04 So power, that's an interesting one.

39:06 So the way to try to get around the power blocker

39:09 is to try to, with the tokens per second per watt,

39:12 try to make it more and more efficient.

39:14 Of course, there's the question of how do we get more power.

39:16 We should also get more power.

39:17 That's a really complicated one.

39:19 You've talked about small modular nuclear power plants.

39:21 There's all kinds of ideas for energy.

39:23 How much does it keep you up at night?

39:26 The bottlenecks in the supply chain of AI,

39:29 like ASML with EUV lithography machines,

39:32 TSMC with advanced packaging like CoWoS,

39:34 and SK Hynix with the high bandwidth memory?

39:39 All the time, and we're working on it all the time.

39:42 No company in history has ever grown

39:45 at a scale that we're growing while accelerating that growth.

39:49 It's incredible.

39:50 And it's hard for people to even understand this.

39:53 In the overall world of AI computing, we're increasing share.

39:57 And so supply chain, upstream and downstream, are really important to us.

40:04 I spend a lot of time informing all the CEOs

40:08 that I work with, what are the dynamics that's going to cause,

40:12 The growth to continue or even accelerate?

40:15 It's part of the reasons why to the entire right-hand side of me were

40:20 CEOs of practically the entire IT industry

40:25 upstream and practically the entire infrastructure industry downstream.

40:32 And they were all...

40:33 There were several hundred CEOs.

40:36 And I don't think there's ever been keynotes where several hundred CEOs show up.

40:40 And part of it is, I'm telling them about our business condition now.

40:46 I'm telling them about the growth drivers

40:48 in the very near future and what's happening.

40:50 And I'm also describing where are we going to go next so that they could use all

40:55 of this information and all of the dynamics

40:57 that are here to inform how they want to invest.

41:01 And so I inform them that way like I inform my own employees.

41:06 And then of course, then I make trips out to them and make sure that, "Hey,

41:10 listen, I want you to know this quarter, this coming year, this next year,

41:14 these things are going to happen." And if

41:17 you look at the CEOs of the DRAM industry,

41:21 The number one DRAM in the world was DDR memory for CPUs in data centers.

41:30 About three years ago, I was able to convince several of the CEOs that even

41:36 though at the time HBM memory was used quite scarcely,

41:39 you know, and, and barely by supercomputers that this was going

41:43 to be a mainstream memory for data centers in the future.

41:46 And at first it sounded ridiculous, but several of the CEOs believed me

41:51 and decided to invest in building HBM memories.

41:54 Another memory was rather odd to put into a data center,

41:59 is the low power memories that we use for cell phones.

42:03 And we wanted them to adapt them for supercomputers in the data center.

42:07 And they go, "Cell phone memory

42:10 for supercomputers?" And I explained to them why.

42:12 Well, look at these two memories, LPDDR5, HBM4.

42:16 The volumes are so incredible.

42:19 All three of them had record years in history,

42:22 and these are, these are 45-year companies.

42:24 And so, you know, I...

42:28 That's part of my job, is to inform and shape, inspire, you know.

42:36 So you're not just manifesting the future and maybe inspiring NVIDIA,

42:40 the different engineers of the company, you're,

42:44 you're manifesting the supply chain of the future.

42:46 So you're having conversations with TSMC, with ASML.

42:50 Upstream, downstream.

42:51 Upstream, downstream.

42:52 So that's the thing.

42:54 GEV, Caterpillar.

42:55 Yeah, that's downstream from us.

42:57 Yeah, yeah, there you go.

42:59 Yeah, the whole thing.

43:00 I mean, but that's so...

43:02 There's so much incredibly difficult engineering

43:05 that happens in the entire semiconductor industry,

43:09 and it just feels scary how intricate the supply chain is,

43:15 how many components there are, but it works somehow.

43:19 Exactly, the deep science.

43:20 The deep engineering, the incredible manufacturing,

43:22 and so much of the manufacturing is already robotics,

43:26 but we have a couple of hundred suppliers that contribute

43:30 the technology that goes into our 1.3 million component rack.

43:36 Each rack is 1.3, one and a half million components.

43:41 There are 200 suppliers across the Vera Rubin rack.

43:45 So it's interesting that you don't list that as the thing

43:47 that keeps you up at night in the list of blockers.

43:49 But I'm doing, I'm doing all the things necessary to-- Okay-...

43:53 yeah, see?

43:54 I can go to sleep because I checked it off.

43:56 I said, okay, you know, I go, I yeah,

43:58 I can go to sleep and I go, well, let's see, what re- let's reason about this.

44:02 What's important for us?

44:03 Um, because okay, let's reason about this.

44:07 Because we changed the system architecture from the original

44:11 DGX-I that you remembered to NVLink-72 rack scale computing-...

44:16 what's gonna...

44:16 What does that, what does that mean?

44:19 What does that mean to software?

44:20 What does that mean to engineering?

44:22 What does that mean to how we design and test?

44:24 How, and what does that mean to the supply chain?

44:27 Well, one of the things that it meant was we moved supercomputer,

44:33 supercomputer integration at the data center

44:37 into supercomputer manufacturing in the supply chain.

44:41 If you're doing that, you also have to recognize you're gonna move one...

44:47 And if, if you're, if you're, you know,

44:50 total footprint of whatever data center you're gonna build,

44:54 let's say you would like to have, you know,

44:58 50 gigawatts of supercomputers that are running simultaneously,

45:02 and it takes one week to manufacture that 50 gigawatts of supercomputers,

45:08 then each week in the supply chain,

45:11 the supercomputers are gonna need a gigawatt of power.

45:13 And so, so we're gonna need the supply chain

45:16 to increase the amount of power it has to build,

45:19 test, to build and test the supercomputers in the supply chain before I ship it.

45:25 Well, NVLink-72 literally builds supercomputers

45:26 in the supply chain and ships 'em two, three tons at a time per rack.

45:32 It used to be—they used to come in parts

45:35 and we used to assemble 'em inside the data center.

45:37 But that's impossible now because NVLink-72 is so dense.

45:41 And so that's an example.

45:42 And I would have to go and to, you know, I've...

45:45 Fly into the supply chain,

45:46 go meet my partners saying, "Hey," I said, "guess what?

45:49 So here's what I'm going to do with...

45:50 This is the way we used to build our DGXs.

45:53 We're gonna build them this way.

45:54 This is gonna be so much better because we're going to need

45:57 'em for inference." The market for inference is, you know, coming.

46:02 The inflection point for inference is coming.

46:04 It's gonna be a big market.

46:05 And so I first explain to them what's going on, why it's gonna happen,

46:09 and then I ask 'em to make several billion dollars of capital investments each.

46:16 And because they, you know, they trust me and I'm very respectful of 'em,

46:22 and I give 'em every opportunity to question me and I

46:25 spend time to explain things to people and I reason about it.

46:28 I draw on pictures and I reason about it in first principles.

46:32 And by the time I'm done with them, they know what to do.

46:35 So it's a lot of it is about

46:38 relationships and building a shared view of the future.

46:41 Uh, but do you worry about certain bottlenecks?

46:43 I mean, what are the biggest bottlenecks in the supply chain?

46:47 Are you, are you worried about ASML's EUV tooling?

46:49 Are you, are you worried about the packaging,

46:52 CoWoS packaging of TSMC, about how fast it could scale?

46:55 Like you said, you're not only growing incredibly fast,

46:59 you're accelerating your growth.

47:00 So it feels like everybody in the supply chain,

47:03 and those are certainly bottlenecks, would have to scale up.

47:07 Are you having conversations with them, like, how can you scale up faster?

47:11 Do you worry about it?

47:13 No.

47:14 Okay.

47:14 Because, because I told 'em what I needed.

47:17 They understood what I need.

47:18 They told me what they're gonna go do,

47:20 and I believe them what they're going to do.

47:22 Interesting.

47:22 That's great to hear.

47:23 So maybe if we can just linger on the power for a little bit.

47:26 What are your hopes for how to solve the energy problem?

47:31 One of the areas, Lex, that I'm that I would love,

47:34 I would love us to talk about and just get the message out,

47:38 you know our power grid is designed

47:44 for the worst case condition with some margin.

47:48 Well, 99% of the time we're nowhere near the worst case condition

47:52 because the worst case condition is a few days in the winter,

47:55 a few days in the summer, and extreme weather.

47:58 Most of the time we're nowhere near

48:01 the worst case condition and we're probably running around, call it 60% of peak.

48:07 And so 99% of the time, our power grid has excess power,

48:14 and they're just sitting idle,

48:16 but they have to be there sitting idle because just in case,

48:19 when the time comes, hospitals have to be powered and, you know,

48:22 infrastructure has to be powered and airports have

48:24 to run and so on and so forth.

48:25 And so the question that I have is whether we could go

48:30 and, Help them understand and create

48:34 contractual agreements and design computer architecture systems,

48:37 data centers, such that when they need,

48:42 The maximum power for infrastructure in society,

48:46 that the data centers would get less.

48:49 But that's in a very rare instance anyways.

48:51 And during that time,

48:52 we either have a backup generator for that little part of it,

48:55 or we just have our computers shift the workload somewhere else,

48:57 or we have the computers just run slower.

49:00 You know, we could degrade our performance,

49:03 reduce our power consumption and provide for a, you know,

49:08 slightly longer latency response, you know,

49:10 when somebody asks for, you know, asks for an answer.

49:13 And so I think that that way of using computers,

49:16 of building data centers, instead of expecting 100% uptime-...

49:22 and these contracts that are really, really quite rigorous,

49:26 it's putting a lot of pressure on the grid to be able to...

49:29 Now, they're gonna have to increase from their maximum.

49:32 I just wanna use their excess.

49:35 It's just sitting there.

49:37 Yeah, that's not talked about enough.

49:38 So what's stopping there?

49:40 Is it regulation?

49:41 Is it bureaucracy?

49:43 I think it's a three-way problem.

49:45 It starts with the end customer.

49:47 The end customer puts requirements on the data

49:51 centers that they can never not be available, okay?

49:56 So that the end customer expects perfection.

50:00 Now, in order to deliver that perfection,

50:02 you need a combination of backup generators

50:05 and your grid power supplier to deliver on perfection.

50:09 And so everybody's gotta have six nines.

50:13 Well, I think first of all, right now, we ought to have everybody understand

50:17 that when the customer asks for these things,

50:19 you have somebody in your data center operations team disconnected from the CEO.

50:24 I bet the CEO doesn't know this.

50:26 I'm gonna talk to all the CEOs.

50:29 The CEOs are probably not paying any

50:31 attention to the contracts that are being signed,

50:34 and so everybody wants to sign the best contract, of course.

50:37 And they go down to cloud service providers, and the contract, the...

50:41 The two contract negotiators that are...

50:43 You, I could just see them now.

50:46 You know, negotiating these multi-year contracts.

50:48 Both sides want, you know, the best contract.

50:52 As a result, the CSPs then have to go down to the utilities,

50:57 and they expect the nine, the six nines.

50:59 And so I think the first thing is just make sure that all of the customers,

51:05 the CEOs and the customers realize what they're asking for.

51:09 Now, the second thing is we have to build data centers that gracefully degrade.

51:13 And so if the power, if the utility, if the grid tells us, "Listen,

51:16 we're gonna have to back you down to about 80%," we're gonna say,

51:20 "That's no problem at all." We're just gonna move our workload around.

51:23 We're gonna make sure that data's never lost,

51:26 but we can reduce the computing rate and use less energy.

51:31 The quality of service degrades a little bit.

51:33 For the critical workloads,

51:34 I shift that somewhere else right away so I don't have that problem,

51:40 and so, you know, whoever,

51:41 whichever data center still has 100% uptime, and so...

51:44 How difficult of an engineering problem is that, that smart,

51:47 dynamic allocation of power in a data center?

51:49 As soon as you could specify, you could engineer it.

51:52 U- beautifully put.

51:54 So long as it obeys the laws of physics on first principles, I think we're good.

51:58 What was the third thing you were mentioning?

52:00 Um...

52:01 So the second thing is the data centers.

52:03 And the third thing is we need the utilities

52:06 to also recognize that this is an opportunity-...

52:11 and instead of saying,

52:12 "Look it's gonna take me five years to increase my grid capability," uh,

52:18 if you, if you have, if you're willing to take power of this level of guarantee,

52:25 I can make them available for you next month and at this price.

52:29 And so if utilities also offered more segments of power delivery promises,

52:36 then I think everybody will figure out what to do with it.

52:39 Yeah, but there's just way too much waste in the grid right now.

52:42 We should go after it.

52:44 Uh, you've highly lauded Elon and xAI's accomplishment in Memphis,

52:49 in building Colossus supercomputer, probably in record time in just four months.

52:55 It's now at 200,000 GPUs and growing very quickly.

52:58 Is there something that you could speak

53:01 to the understand about his approach that's instructive

53:04 to, broadly to all the data center

53:06 creators that's that enabled that kind of accomplishment?

53:10 His approach to engineering,

53:11 his approach to the whole management of construction, everything?

53:16 First of all, Elon is deep in so many different topics.

53:19 Um, Yet he's also a really good systems thinker.

53:24 And so he's able to think through multiple disciplines,

53:28 and, and he obviously pushes things, questions everything,

53:35 where they're, number one, is it necessary?

53:39 Number two, does it have to be done this way?

53:41 And then numbers, you know, does it have, does it have to take this long?

53:45 And, and so, so he, he has,

53:47 he has the a- he has the ability to question everything,

53:53 To the point where everything is down to its minimal amount that's necessary,

53:58 you can't take anything else out.

54:02 And yet the, the necessary capabilities of the product remains, you know?

54:08 And so he's, he is as minimalist as you could possibly imagine,

54:12 and he does it at a system scale.

54:14 I think...

54:15 I also love the fact that he is he is represented.

54:20 He is present at the point of action.

54:25 You know, he'll just go there.

54:27 If there's a problem, he'll just go there and then,

54:30 "Show me the problem." You know, when you do all of this in combination,

54:34 you overcome a lot of previous,

54:36 "This is just the way we do it." "Um, you know, I'm waiting for them.

54:42 Uh," You know, I mean, it's just, everybody has a lot of excuses.

54:46 And so, and then the last thing is when you act personally

54:49 with so much urgency it causes everybody else to act with urgency, you know?

54:55 And every supplier has a lot of customers going on.

54:58 Every supplier has a lot of projects going on, and he made it his...

55:02 He makes it his business that he's

55:04 the top priority of everybody else's, you know, projects.

55:07 And so he does that by demonstrating it.

55:10 Yeah, I've been in a bunch of those meetings.

55:11 It's just, it's fun to watch, 'cause really,

55:13 not enough people ask the question like,

55:15 "Okay, so, can this be done a lot faster, and how?

55:20 Why does it have to take this long?"- Yeah, right.

55:22 And then in the...

55:23 That becomes an engineering question often.

55:24 And yes, I think when you get the ground truth of actually...

55:28 I remember, one of the times I was hanging out with him,

55:31 he literally is going through the entire process

55:33 of how to plug in cables into a rack.

55:36 He's, he's working with an engineer on the ground that's doing that task,

55:40 and he's just trying to understand what does

55:42 that process look like so it can be less error-prone.

55:44 And just building up that intuition from every single

55:48 task involved in, putting together a data center— ...you start

55:53 to immediately get a sense at the detailed scale

55:56 and at the broad systems scale of where the inefficiencies are,

56:01 and so you can make it more and more and more efficient.

56:04 Plus you have the big hammer of being able to say,

56:07 "Let's do it totally different-"- Yeah.

56:08 That's right.

56:09 and remove all possible blockers."- That's right.

56:11 Is there parallels in the NVIDIA Extreme Systems co-design approach

56:14 that you see in the way Elon approaches systems engineering?

56:19 Well, first of all, co-design is an ultimate systems engineering problem.

56:22 And so we approach, we approach the work that we do from that first,

56:25 from that The other thing that we do and this is,

56:30 this is a philosophy, a thought, a state of mind, I guess, a method,

56:39 That I started 30 years ago, and it's called the speed of light.

56:43 The speed of light is not just about the speed.

56:45 The speed of light is,

56:46 is my shorthand for what's what's the limit of what physics can do.

56:53 And so every single, everything,

56:55 everything that we do is compared against the speed of light.

56:58 Memory speed, Math speed, power, cost, time,

57:03 effort, number of people, manufacturing cycle time.

57:08 And when you think about latency versus throughput,

57:12 When you think about cost versus throughput, cost versus capacity,

57:17 all of these things, You test against the speed of light

57:22 to achieve all of these different constraints separately.

57:27 And then when you consider it together,

57:30 you know you have to make compromises because

57:33 a system that achieves extremely low latency versus a cheap,

57:36 a system that achieves very

57:38 high throughput are architected fundamentally differently.

57:42 But you want to know what's the speed

57:44 of light of a system that achieves high throughput,

57:46 what's the speed of light of a system that achieves low latency?

57:52 And then when you think about the total system, you can make trade-offs.

57:55 And so I force everybody to think about what's

57:58 the, what the first- the first principles, the limits-...

58:01 the physical limits, For everything before we, you know, before we do anything.

58:08 And we test everything against that.

58:10 And so that's a good frame of mind.

58:13 I don't love the other methods, which is continuous improvement.

58:19 The problem with continuous improvement, it...

58:23 First of all, you should engineer something from first principles at the speed,

58:27 you know, with speed of light thinking.

58:29 Limit it only by physical limits, and physics limits.

58:33 And after that, of course you would improve it over time.

58:38 Um, but I don't like going into a problem and somebody says,

58:42 "Hey, you know, it takes 74 days to do this today-""...

58:45 Right now.

58:46 And we can do it for you in 72 days." You know,

58:50 I'd rather strip it all back to zero-...

58:52 and say, "First of all, explain to me why 74 days in the first place.

58:56 And let's note, let's think about what's possible today.

58:59 And if I were to build it completely from scratch,

59:02 you know, how long would it take?" Oftentimes, you'd be surprised.

59:06 It might come to six days.

59:08 Now, the rest of the six days,

59:11 the 74, could be very well-reasoned and compromises,

59:14 and, you know, cost reductions, and all kinds of different things.

59:17 But at least you know what they are.

59:20 And then now that you know that six days is possible,

59:24 then the conversation from 74 to six, surprisingly much more effective.

59:30 In such incredibly complex systems that you're working

59:32 with, is simplicity sometimes a good heuristic to reach for?

59:36 I mean, if I can just...

59:40 I mean, the pod, the Vera Rubin pod that you announced is just incredible.

59:44 We're talking about seven chips, seven chip types,

59:47 five purpose-built rack types, 40 racks, 1.2 quadrillion transistors,

59:52 nearly 20,000 NVIDIA dies, over 1,100 Rubin GPUs,

59:57 60 exaflops, 10 petabytes per second of scale bandwidth.

1:00:01 That's all just one...

1:00:03 That's just one pod.

1:00:04 That's just one pod.

1:00:06 Yeah, that's just one pod.

1:00:07 I mean, in-...

1:00:08 so you have the...

1:00:09 And then even the NVL 72 rack alone is 1.3 million components,

1:00:15 1300 chips, 4,000 pods crammed into a single 19-inch wide rack.

1:00:19 And Lex, we're probably gonna have to crank out about 200 of these pods a week,

1:00:24 just to put it in perspective.

1:00:26 The amount of different components, I suppose simplicity is impossible,

1:00:30 but is that a metric that you kind of reach for in trying to design things?

1:00:35 You know, the phrase, the phrase that I use most often is,

1:00:39 we- we need things to be as complex as necessary, but as simple as possible.

1:00:43 And so the question is, is all that complexity there necessary?

1:00:48 And we ought to test for that.

1:00:50 And we got to challenge that.

1:00:51 And then after that, everything else above it, you know, is gratuitous.

1:00:56 But it's still almost incredible.

1:00:58 Semiconductor industry broadly, but what NVIDIA is doing,

1:01:03 some of the greatest engineering in history.

1:01:05 So these systems are just truly, truly marvels of engineering.

1:01:10 It is the most complex computer the world has ever made.

1:01:13 Yeah, the engineering teams, I mean-...

1:01:14 I don't, it's not a competition, but I don't know.

1:01:16 If it was like an Olympics of engineering teams,

1:01:19 I mean, TSMC does incredible engineering.

1:01:21 Like I said, ASML at every scale,

1:01:23 but NVIDIA is gonna give them a run for their money.

1:01:27 Just incredible, incredible teams.

1:01:29 Well, it's gold medalists in every single,

1:01:31 in every single sport, all assembled right here.

1:01:34 And have to work together.

1:01:35 And report directly to you.

1:01:37 This is wonderful.

1:01:38 You recently traveled to China.

1:01:41 so it's interesting to ask you China's been

1:01:44 incredibly successful in building up its technology sector.

1:01:47 What do you understand about how China's able to, over the past 10 years,

1:01:54 build so many incredible world-class companies,

1:01:57 world-class engineering teams, and just this technology ecosystem-...

1:02:01 that produces so many incredible products?

1:02:05 A whole bunch of reasons for, well, first of all, let's start with some facts.

1:02:09 50% of the world's AI researchers are Chinese,

1:02:13 plus or minus, and they're mostly in China still.

1:02:18 We have many of them here, but there's amazing researchers still in China.

1:02:23 They, their tech industry showed up at precisely the right time.

1:02:30 At the time of the mobile cloud era their way of contributing was software,

1:02:34 and so this is a country's incredible science and math.

1:02:39 Really well-educated kids.

1:02:42 Their tech industry was created during the era of software.

1:02:48 They're very comfortable with modern software.

1:02:52 China is not one giant economic country.

1:02:56 It's got many provinces and cities with mayors all competing with each other.

1:03:01 That's the reason why there's so many EV companies.

1:03:03 That's the reason why there's so many AI companies.

1:03:05 That's the reason why there's so many,

1:03:07 every company you could imagine, they all create some of them.

1:03:13 And as a result, they have insane competition internally.

1:03:19 And, you know, what remains is an incredible company.

1:03:24 They also have a social culture where it's family first,

1:03:30 friends second, and company third.

1:03:34 And so, the amount of conversation that goes back and forth between...

1:03:43 They're essentially open source all the time.

1:03:47 So the fact that they contribute more

1:03:49 to open source is so sensible because they're probably,

1:03:52 "What are we protecting?" You know,

1:03:54 my engineers, their brothers are in that company,

1:03:57 their friends are in that company, and they're all schoolmates.

1:04:00 You know, the schoolmate concept.

1:04:01 There's a, you know, one schoolmate, you're brother for life.

1:04:06 And so they, they share knowledge very, very quickly.

1:04:12 And so there's no sense keeping technology hidden.

1:04:14 You might as well put it on open source.

1:04:17 And so the open source community then amplifies,

1:04:20 accelerates the innovation process.

1:04:22 So you get this rapid, incredible great talent,

1:04:26 rapid innovation because of open source and just,

1:04:29 you know, the nature of friends, and insane competition.

1:04:34 Among comp- among the company, what emerges is incredible stuff.

1:04:39 And so this is the fastest innovating country in the world today,

1:04:45 and this is something that has everything that, everything that I've

1:04:47 just said is fundamental to just how the kids were grown,

1:04:52 the fact that they have excellent education, the fact that they,

1:04:55 parents want them to do well in school,

1:04:57 the fact that they, their culture is that way.

1:04:59 These are, you know, these are just the thing about their country,

1:05:04 and they showed up at precisely the time

1:05:06 when technology is going through that exponential.

1:05:09 Plus culturally, it's pretty cool to be an engineer.

1:05:12 It connects to all the components that you're mentioning...

1:05:16 It's a builder nation.

1:05:18 It's a builder nation.

1:05:19 Yeah, it's a builder nation.

1:05:21 Our country's leaders, incredible, but they're mostly lawyers.

1:05:24 Their country's leaders, and because we're, they're trying to keep us safe,

1:05:29 rule of law, governing, their country was built out of poverty.

1:05:35 And so most of their leaders are incredible engineers.

1:05:40 Some of the brightest minds.

1:05:43 To take a small tangent, because you mentioned open source,

1:05:46 I have to go to Perplexity here, who you have been a fan of a long time.

1:05:51 Love it, yeah.

1:05:52 And thank you for releasing open source Nemotron 3 Super,

1:05:56 which you can also use inside Perplexity to look stuff up.

1:05:59 Now, which is 120 billion parameter open weight MoE model.

1:06:05 What's your vision with open source?

1:06:09 So you mentioned China with DeepSeek and MiniMax,

1:06:13 with all these companies really pushing forward the open source AI movement,

1:06:19 and NVIDIA is really leading the way

1:06:22 in close to state-of-the-art open source LLMs.

1:06:26 What's your vision there?

1:06:29 First, if we're gonna be a great AI computing company,

1:06:32 we have to understand how AI models are evolving.

1:06:36 One of the things that I love about Nemotron

1:06:39 3 is it's not a, just a pure transformer model, it's transformer and SSMs.

1:06:45 And we were early in, Developing the conditional GANs,

1:06:51 which, that progressive GANs, which led step-by-step to diffusion.

1:06:55 And so, The fact that we're doing basic research in model

1:07:00 architecture and in different domains gives us visibility into, you know,

1:07:05 what kind of computing systems would do a good job for future models.

1:07:09 And so it is part of our extreme co-design strategy.

1:07:14 Second, um, I think we rightfully recognize that on the one hand,

1:07:21 we want world-class models as products, and they should be proprietary.

1:07:27 On the other hand, we also want AI

1:07:30 to diffuse into every industry and every country,

1:07:34 every researcher, every student.

1:07:37 And if everything is proprietary,

1:07:39 it's hard to do research and it's hard to innovate on top of, around, with.

1:07:46 And so...Open source is fundamentally necessary

1:07:49 for many industries to join the AI revolution.

1:07:53 NVIDIA has the scale and we have the motives to not only skills, scale,

1:08:01 and motivation to build and continue to build

1:08:06 these AI models for as long as we shall live.

1:08:10 And so therefore, we ought to do that.

1:08:11 We can open up, we can activate every industry, every researcher,

1:08:17 you know, every country to be able to join the AI revolution.

1:08:21 There's the third reason,

1:08:22 which is from that, to recognizing that AI is not just language.

1:08:28 These AIs will likely use tools and models

1:08:33 and sub-agents that were trained on other modalities of information.

1:08:39 Maybe it's biology or chemistry or you know, laws of physics, or you know,

1:08:46 fluids and thermodynamics, and not all of it is in language structure.

1:08:50 And so somebody has to go make sure that weather prediction,

1:08:55 biology, AI, AI for biology, physical AI, all of that stuff stays,

1:09:02 can be pushed to the limits and pushed to the frontier.

1:09:05 We don't build cars,

1:09:06 but we wanna make sure every car company has access to great models.

1:09:10 We don't, discover drugs,

1:09:11 but I wanna make sure that Lilly has the world's best biology AI systems,

1:09:16 so that they can go use it for discovering drugs.

1:09:19 And so these three fundamental reasons,

1:09:22 both in, recognizing that AI is not just language, that AI is really broad,

1:09:27 that we wanna engage everybody into the world of AI,

1:09:30 and then also co-design of AI.

1:09:32 Well, I have to say, once again,

1:09:34 thank you for open sourcing, really truly open sourcing Nemotron 3 and...

1:09:39 Yeah, I appreciate you were saying that.

1:09:40 We open sourced the models, we open sourced the weights,

1:09:42 we open sourced the data, we open sourced how we created it.

1:09:46 Yeah, it's pretty amazing.

1:09:48 Uh, it's really, it's really incredible.

1:09:51 You're originally from Taiwan and have a close relationship with TSMC.

1:09:56 So I have to ask, TSMC I think also

1:10:00 is a legendary company in terms of the engineering teams,

1:10:02 in terms of the incredible engineering work that they do.

1:10:06 What what do you understand about TSMC

1:10:09 culture and their approach that explains how

1:10:11 they're able to achieve this singular unmatched

1:10:15 success in everything they're doing with semiconductors?

1:10:20 You know, first of all, the deepest misunderstanding about TSMC is

1:10:26 that their technology is all they have.

1:10:32 That somehow they have a really great transistor,

1:10:37 and if somebody shows up another transistor, game over.

1:10:41 It's the technology and of course, you know,

1:10:45 I don't mean just the transistor, the metallization systems,

1:10:47 the packaging, the 3D packaging, the silicon photonics,

1:10:51 the, you know, all of the technology that they have.

1:10:54 That technology is really what makes the company special.

1:10:56 Their technology makes the company special.

1:10:59 But their ability to orchestrate the demands,

1:11:05 the dynamic demands of hundreds of companies in the world as they're moving up,

1:11:12 shifting out, you know, increasing, decreasing, push, pushing out,

1:11:17 pulling in changing from customer to customer, Wafer starting,

1:11:23 wafer stopping, Emergency wafer starts, you know,

1:11:28 all of this dynamics of the world's complexity

1:11:32 as the world is shape-shifting all the time,

1:11:36 and somehow they're running a factory with high throughput,

1:11:41 high yields, really great costs, excellent customer service.

1:11:46 They take their work, they take their promises seriously.

1:11:49 They, when your wafer,

1:11:50 because they know that they're helping you run your company, when the wafers,

1:11:54 when the wafers were promised to show up, the wafers show up,

1:11:57 you know, so that you could run your company appropriately.

1:12:00 And so their system,

1:12:01 their manufacturing system is completely miraculous, I would say.

1:12:05 Then the second thing is their culture.

1:12:08 This culture is simultaneously, Technology focused on one hand,

1:12:12 advancing technology,

1:12:14 simultaneously customer service oriented on the other hand.

1:12:17 A lot of companies are very customer service oriented,

1:12:21 but they're not very technology excellent.

1:12:23 They're not at the bleeding edge of technology.

1:12:27 There are a lot of companies who are tech, at the bleeding edge of technology,

1:12:29 but they're not the best customer service oriented company.

1:12:32 And so it just depends on somehow they've,

1:12:34 they've balanced these two and they're world-class at both.

1:12:39 And then probably the third thing is the technology that I

1:12:42 most value in them that they created this, you know,

1:12:46 this, Intangible called trust.

1:12:49 I trust them to put my company on top of them.

1:12:53 That's a very big deal.

1:12:55 When they trust, I mean,

1:12:56 there's a really close relationship there that you've established,

1:12:58 and that trust is established based on many years of performance,

1:13:02 but there's human relationships involved there as well.

1:13:05 Three decades, I don't know how many tens,

1:13:07 hundreds of billions of dollars of business we've done through them,

1:13:11 and we don't have a contract.

1:13:14 That's pretty great.

1:13:15 Amazing.

1:13:15 Okay, there's this story...

1:13:19 That in 2013, the founders of TSMC,

1:13:21 Morris Chang offered you the chance to become TSMC's chief executive,

1:13:26 And you said you already had a job.

1:13:28 Is this story true?

1:13:30 Story is true.

1:13:31 I didn't dismiss it.

1:13:33 Um but I was deeply honored and, and of course, of course uh,

1:13:38 I knew then as I know now,

1:13:40 TSMC is one of the most consequential companies in history.

1:13:45 And, and Morris is one of the highest regarded executive

1:13:49 and, and business and personal friend that I've had in my life.

1:13:55 And, um ...Uh, for him to ask is, uh um, I was humbled and, and really honored.

1:14:05 But, but the work that I'm doing here is really important,

1:14:08 and I've seen, you know, in my mind's eye, what NVIDIA was going

1:14:13 to be and what the impact that we could have.

1:14:17 And uh, it was really important work.

1:14:20 And it's my responsibility, you know,

1:14:22 my sole responsibility to make this happen.

1:14:25 And so I uh, I declined it, You know,

1:14:31 not, not because it wasn't an incredible offer.

1:14:34 It's an unbelievable offer but, but I simply couldn't take it.

1:14:39 I think NVIDIA, both NVIDIA and TSMC are two

1:14:42 of the greatest companies in the history of human civilization.

1:14:45 And running either one, I'm sure, is incredibly complicated effort and takes...

1:14:50 You have to truly be all in.

1:14:52 Uh, everybody at every scale, not just at the CEO level.

1:14:55 Everybody is really truly all in-- Yeah.

1:14:58 Yeah, no doubt-...

1:14:59 To, to accomplish this kind of complexity.

1:15:00 So now I can help both companies.

1:15:02 Exactly.

1:15:03 So NVIDIA is now the most valuable company in the world.

1:15:08 I have to ask, what is the NVIDIA's biggest moat,

1:15:13 as the folks in the tech sector say?

1:15:15 The edge you have that protects you from the competition.

1:15:21 Our single most important property as a company

1:15:27 is the install base of our computing platform.

1:15:31 Our single most important thing today is our, is the install base of CUDA.

1:15:37 Now, the reason why, uh, 20 years ago, of course, there was no install base.

1:15:46 But what makes...

1:15:47 And if somebody came up with a GUDA

1:15:50 or TUDA it wouldn't make any difference at all.

1:15:53 And the reason for that is because it's never been just about the technology.

1:15:59 The technology, of course, was incredible visionary.

1:16:01 But it's the fact that the company was dedicated to it,

1:16:07 stuck with it, expanded its reach.

1:16:09 Um, it wasn't three people that made CUDA successful.

1:16:13 It was 43,000 people that made CUDA successful.

1:16:16 And the several million developers that believed in us,

1:16:21 That trusted that we were going to continue to make CUDA 1, 2,

1:16:25 3, 13, that they decided to port and dedicate their software on top of it,

1:16:30 their mountain of software on top of it.

1:16:33 And so the install base is the number one most important advantage.

1:16:38 That install base, when you amplify it with the velocity

1:16:41 of our execution at the scale that we're talking about,

1:16:45 no company in history had ever built systems of this complexity, period.

1:16:50 And then to build it once a year is impossible.

1:16:56 And that velocity combined with the install base,

1:17:02 in the developer's mind, you just go now, take the developer's mind.

1:17:06 From the developer's perspective, if I support CUDA,

1:17:11 tomorrow it'll be 10 times better.

1:17:13 I just have to wait six months on average.

1:17:16 Not only that, if I develop it on CUDA,

1:17:19 I reach a few hundred million people, computers.

1:17:23 I'm in every cloud, I'm in every computer company,

1:17:26 I'm in every single industry, I'm in every single country.

1:17:31 So if I create an open source package and I put it on CUDA first,

1:17:36 I get these both attributes simultaneously.

1:17:39 And not only that, I trust 100% that NVIDIA

1:17:45 is going to keep CUDA around and maintain it and improve it and keep optimizing

1:17:51 the libraries for as long as they shall live.

1:17:57 You could take that to the bank, and that last part, trust.

1:18:00 You put all that stuff together,

1:18:02 if I were a developer today, I would target CUDA first.

1:18:06 I would target CUDA most.

1:18:09 And that's the reason that I think in the final analysis is our first,

1:18:14 that's even our first-- core advantage.

1:18:17 Our second one is our ecosystem.

1:18:21 The fact that we vertically integrated this incredibly complex system,

1:18:25 but we integrate it horizontally into every single company's computers.

1:18:31 We're into Google Cloud, we're into Amazon, we're in Azure.

1:18:33 You know, we're ramping up AWS like crazy right now.

1:18:36 We're in new companies like CoreWeave and Nscale.

1:18:40 We're in supercomputers at Lilly.

1:18:43 We're in enterprise computers.

1:18:44 We're at the edge in radio base stations.

1:18:47 You know, I mean, it's just crazy.

1:18:49 One architecture is in all these different systems.

1:18:51 We're in cars, we're in robots, we're in satellites, we're out in space.

1:18:54 And so the fact that you have

1:18:57 this one architecture and the ecosystem is so broad,

1:18:59 it basically covers every single industry in the world.

1:19:03 Well, how does the, how does the CUDA install base

1:19:07 evolve into the future with AI factories as a moat?

1:19:10 What do you...

1:19:11 Do you think it's possible that NVIDIA

1:19:13 of the future is all about the AI factory?

1:19:16 Well, the unit of computing used to be GPU to us.

1:19:20 Then it became a computer, then it became a cluster.

1:19:23 Now it's an entire AI factory.

1:19:25 When I see a computer, when I see what NVIDIA builds,

1:19:28 in the old days, I would, you know, I visualize the chip.

1:19:32 And then when I announced the new product, new generation, like,

1:19:36 "Ladies and gentlemen, we're announcing Ampere today," I'd pick up the chip.

1:19:40 That was my mental model-...

1:19:42 of what I was building.

1:19:43 Today, I wouldn't...

1:19:44 Picking up the chip is kind of still adorable.

1:19:48 But it's adorable.

1:19:49 It's not my mental model of what I'm doing.

1:19:53 My mental model is this giant gigawatt thing

1:19:57 that has power generations connected to the grid.

1:20:00 It's got cooling systems and networking of incredible monstrosity, you know.

1:20:05 10,000 people are in there trying to install it,

1:20:09 hundreds of networking engineers in there,

1:20:10 thousands of engineers behind it trying to power it up.

1:20:14 You know, powering up one of those factories,

1:20:16 as you know, it's not somebody going,

1:20:18 "It's on now." It takes thousands of people to bring it up.

1:20:22 So mentally, you're actually...

1:20:23 When you're thinking about a single unit of compute, you're like literally,

1:20:27 when you go to bed at night,

1:20:29 you're thinking now about collection of racks, so pods, not individual chips.

1:20:33 Entire infrastructure.

1:20:34 And I'm hoping my next click is when I'm thinking about building computers,

1:20:37 it's, you know, planetary scale.

1:20:39 That'll be the next click.

1:20:42 Well, what do you think about the space angle that Elon has talked about,

1:20:46 doing compute in space for solving some of the...

1:20:50 It makes some of the energy issues in terms of scaling energy easier.

1:20:57 Cooling issues is not easy.

1:20:58 Yeah.

1:20:59 Cooling.

1:20:59 Well, there's a large number of engineering complexities involved with that.

1:21:03 So, so what...

1:21:04 You know, NVIDIA has also announced that you're already thinking about that.

1:21:09 Yeah, we're already there.

1:21:10 NVIDIA GPUs are the first GPUs in space.

1:21:14 And I didn't realize it, it was so interesting to...

1:21:18 I would have declared it maybe.

1:21:20 We're in space.

1:21:21 You know, little, little astronaut suit on one of our GPUs.

1:21:27 But we've been in space.

1:21:28 It's the right place to do a lot of imaging.

1:21:32 You know, because those satellites have,

1:21:33 have really high resolution imaging systems,

1:21:36 and they're sweeping the Earth, you know, continuously now.

1:21:39 And, um, you want, you know, centimeter scale,

1:21:43 imaging imaging that is done continuously for the world,

1:21:46 so that, you know, you'll basically have real time telemetry of everything.

1:21:51 You don't wanna beam that back down to earth.

1:21:54 It's just, you know, petabytes and petabytes of data.

1:21:58 You gotta just do AI right there at the edge,

1:22:01 throw away everything you don't need, you've seen before, didn't change,

1:22:04 and then just keep the stuff that, that you need.

1:22:06 And so AI had to be done at the edge.

1:22:09 Obviously we have, we have a 24/7 solar, if we put it at the polars.

1:22:16 And uh, but, you know, there's no conduction, no convection.

1:22:23 And so, you know, you're pretty much just radiation.

1:22:26 And uh, but, you know, space is big.

1:22:28 I guess, you know, we're just gonna put big, giant radiators out there.

1:22:32 How crazy of an idea do you think it is?

1:22:34 Like is this, is this five years out, 10 years out, 20 years out?

1:22:37 So we're talking about blockers for AI scaling.

1:22:41 You know, I'm just so much more practical.

1:22:44 I look for where, where, um my next, next bucket of opportunities are first.

1:22:51 Meanwhile, I'm cultivating space.

1:22:53 And so I send engineers to go work on the problem.

1:22:57 We're starting to...

1:22:58 We're learning a lot about it.

1:23:00 How do we deal with radiation?

1:23:01 How do we deal with degrading performance?

1:23:03 How do we deal with a continuous, Testing and attestation of, of de- defects?

1:23:10 And you know, how do we deal with redundancy?

1:23:13 And how do we degrade gracefully and things like that?

1:23:16 And so we could, we could do a...

1:23:18 What about software?

1:23:19 How do you think about software and redundancy and performance out in space?

1:23:24 Make it so that the computer never breaks, it just gets slower, you know.

1:23:31 And I...

1:23:32 So we could start doing a lot of engineer exploration upfront.

1:23:35 But in the meantime, my, my favorite answer is ge- eliminate waste.

1:23:41 You know, we've got all that idle power,

1:23:43 I want to evacuate it as fast as possible.

1:23:47 Yeah.

1:23:48 There's a lot of low-hanging fruit here on Earth-...

1:23:51 That we can utilize for the AI scaling.

1:23:54 Quick pause.

1:23:56 Quick 30-second thank you to our sponsors.

1:23:58 Check them out in the description.

1:24:00 It really is the best way to support this podcast.

1:24:04 Go to lexfridman.com/sponsors.

1:24:06 We got Perplexity for curiosity-driven knowledge exploration,

1:24:10 Shopify for selling stuff online, LMNT for electrolytes,

1:24:15 Fin for customer service AI agents, and Quo for a phone system,

1:24:20 like calls, texts, contacts, for your business.

1:24:23 Choose wisely, my friends.

1:24:25 And now, back to my conversation with Jensen Huang.

1:24:30 Do you think NVIDIA may be worth 10 trillion at some point?

1:24:36 Let's ask it this way.

1:24:37 What does the future of the world look like where that's true?

1:24:45 I think that NVIDIA's growth is Extremely likely, and in my mind, inevitable.

1:24:54 And let me explain why.

1:24:56 We're the largest computer company in history.

1:25:00 That alone should beg the question, why?

1:25:04 And the reason of course...

1:25:06 Two reasons.

1:25:07 First, two foundational technical reasons.

1:25:10 The first reason is that computing went from being a retrieval-based,

1:25:14 file retrieval system.

1:25:15 Almost everything is a file...

1:25:18 We pre-write something, we pre-record something.

1:25:21 You know, we draw something, we put it on the web, we put it in a file.

1:25:25 And we use a recommender system, some smart filter,

1:25:28 to figure out what to retrieve for you.

1:25:30 And so we were a pre-recording, human pre-recording, and file retrieving system.

1:25:36 That's what a computer is, largely.

1:25:39 To now, AI computers are contextually aware,

1:25:43 which means that it has to process and generate tokens in real time.

1:25:47 So we went from a retrieval-based

1:25:49 computing system to a generative-based computing system.

1:25:53 We're gonna need a lot more processing in this new world than in the old world.

1:25:57 We need a lot of storage in the old world.

1:26:00 We need a lot of computation in this new world.

1:26:03 And so, so that's the first part of it.

1:26:08 We fundamentally changed computing and the way how computing is done.

1:26:11 The only thing that would cause it to go back......

1:26:14 is if this way of computation,

1:26:18 this way of computing generating information that's contextually relevant,

1:26:22 situationally aware,

1:26:23 that is grounded on new insight before it generates information,

1:26:29 this computation-intensive way of doing computing would

1:26:33 only go back if it's not effective.

1:26:36 So if...

1:26:36 For the last 10, 15 years while working on deep learning,

1:26:40 if at any single moment I would have

1:26:44 come to the conclusion that, "You know what?

1:26:46 This is not gonna work out.

1:26:49 I think this is a dead end." Or, "It's not gonna scale,

1:26:51 it's not gonna solve this modality,

1:26:52 not gonna be used in this application." Then, of course,

1:26:55 I would feel very differently about it,

1:26:57 but I think the last five years has given me

1:27:00 more confidence than the last ten year, previous ten years.

1:27:04 The second idea, is computers,

1:27:07 because it was a storage system, it was largely a warehouse.

1:27:14 We're now building factories.

1:27:16 Warehouses don't make much money.

1:27:20 Factories directly correlates with the company's revenues.

1:27:24 And so, the computer did two things.

1:27:29 Not only did it change the way it did it, its purpose in the world changed.

1:27:36 It's no longer a computer, it's a factory.

1:27:39 It's a factory, it's used for generation of revenues.

1:27:45 We're now seeing not only is this factory generating products,

1:27:49 commodities that people want to consume,

1:27:53 we're seeing that the commodities are so interesting,

1:27:56 so valuable, so, to so many different audiences

1:28:00 that the tokens are starting to segment, like iPhones.

1:28:03 Mm-hmm.

1:28:04 You have free tokens, you have premium tokens,

1:28:07 and you have several tokens in the middle.

1:28:10 Yeah.

1:28:10 And so intelligence, as it turns out, you know, it's a scalable product.

1:28:15 There's extremely high intelligence products, tokens that you could...

1:28:18 that are used for specialized things, people be willing to pay.

1:28:22 You know, the idea that somebody's willing to pay

1:28:25 $1000 per million tokens is just around the corner.

1:28:28 It's not if, it's only when.

1:28:32 And so now we're seeing that the commodity

1:28:36 that this factory makes is actually valuable,

1:28:38 and is revenue generating and profit generating.

1:28:41 How, now the question is how many of these factories can, does the world need?

1:28:48 How much, how many tokens does the world need?

1:28:52 And how much is society willing to pay for these tokens?

1:28:58 And what would happen to the world's economy

1:29:02 if the productivity were to improve so substantially?

1:29:07 What would happen...

1:29:08 Are we, are we gonna discover new drugs, new products, new services?

1:29:12 And so when you take these things in combination,

1:29:15 I am absolutely certain that the world's GDP is going to accelerate in growth.

1:29:22 I'm absolutely certain the percentage of that GDP that will be

1:29:27 used for computation will be 100 times more than the past...

1:29:32 because it's no longer a storage unit.

1:29:36 It's a product generation unit.

1:29:39 And so when you look at it in that context

1:29:42 and then you back into what is NVIDIA's,

1:29:45 what does NVIDIA does NVIDIA do and how much of that new economics,

1:29:51 new industry would we have to benefit to address,

1:29:55 I think we're gonna be a lot, lot bigger.

1:29:58 And then the rest of it, to me, is is it possible for NVIDIA to be a, you know,

1:30:05 $3 trillion revenues company in the near future?

1:30:08 The answer is of course yes.

1:30:09 And the reason for that is because it's not limited by any physical limits.

1:30:15 There's nothing that I see that says,

1:30:17 you know, gosh, $3 trillion is not possible.

1:30:22 And as it turns out, NVIDIA's supply chain is,

1:30:27 the burden is shared by 200 companies.

1:30:31 And the fact that we scale out

1:30:34 on the backs with the partnership of this ecosystem,

1:30:38 the question is do we have the energy to do so?

1:30:41 And surely we will have the energy to do so.

1:30:45 And so all of these things combined, that number is just a number, you know?

1:30:51 And I still remember,

1:30:53 NVIDIA was a, NVIDIA was a, the first time we crossed a billion dollars,

1:30:58 I was reminded of a CEO who told me, "You know, Jensen,

1:31:01 it's theoretically impossible for a fabless semiconductor company to exceed

1:31:05 a billion dollars." And I won't bore you with why,

1:31:10 but the end, of course it's illogical and there's a lot of evidence we're not.

1:31:14 And then, somebody told me, "You know, Jensen,

1:31:17 you'll never be more than $25 billion because of some

1:31:22 other company." Somebody told me that, "You'll never be,

1:31:24 you know, because..." And so the, those aren't first principled thinking.

1:31:32 And the simple way to think about that is what is it

1:31:36 that we make and how large is the opportunity that we can create?

1:31:42 Now, NVIDIA is not in the market share business.

1:31:44 Almost everything that I just talked about don't exist.

1:31:47 That's the part that's hard.

1:31:51 You know, if NVIDIA was a $10 billion company trying to take market share,

1:31:57 then it's easy to see for shareholders that, oh, yeah,

1:32:01 if they could just take 10% share, they could be this much larger.

1:32:06 But it's hard for people to imagine how large we

1:32:10 could be because there's nobody I could take share from.

1:32:12 You know?

1:32:13 And so I think that that's one of the challenges....

1:32:16 for the world is, is the imagination of the future.

1:32:20 But I got plenty of time, and I'll keep reasoning about it,

1:32:23 and I'll keep talking about it,

1:32:24 and every single GTC will become more and more real.

1:32:27 You know, and then more and more people will talk about it,

1:32:30 and one of these days, you know, we'll get there.

1:32:32 But I'm 100% we'll get there.

1:32:34 Yeah, this view of you know, token factories essentially,

1:32:38 this token per second per watt, and every token having value.

1:32:42 Like it's an actual thing that brings value,

1:32:45 and it brings different kinds of value,

1:32:47 different amounts of value to different people with value.

1:32:49 That's the actual product, is really could be loosely thought of as the token.

1:32:53 And so you have a bunch of token factories.

1:32:54 And then it's very easy, first principles, to imagine a future,

1:32:58 given all the potential things that AI can solve,

1:33:00 that you're going to need an exponential number more of token factories.

1:33:06 And what's really interesting, the reason why I was so excited about it,

1:33:09 the iPhone of tokens arrived.

1:33:11 What do you call it?

1:33:13 Wait, are you saying OpenClaw's iPhone?

1:33:15 That's interesting.

1:33:16 Agents.

1:33:17 Yeah, agents.

1:33:17 True.

1:33:18 Agents in general.

1:33:19 The iPhone of tokens arrived.

1:33:21 It is the fastest-growing application in history.

1:33:23 It went straight up.

1:33:25 Went straight up.

1:33:26 That says something.

1:33:28 Yep, there's no question OpenClaw is the iPhone of tokens.

1:33:32 Is there something truly, as you know,

1:33:34 something truly special happening from about December,

1:33:38 where people have really woke up to the power

1:33:41 of Claude Code of Codex, of OpenClaw.

1:33:44 Um, I mean, I'm embarrassed to admit

1:33:47 that on the way here in the airport, I've...

1:33:52 It's the first time I've done this in public.

1:33:55 I was programming, quote unquote, by talking to my laptop.

1:33:59 And I was embarrassed because I was

1:34:01 pretending like I'm talking to a human colleague.

1:34:04 Uh, I'm not sure how I feel about the future where everybody-...

1:34:07 is walking around talking to their AI,

1:34:09 but it's such an efficient way to get stuff done.

1:34:13 And it's more likely that your AI is bothering you all the time.

1:34:18 And the reason for that is because it's getting stuff done so fast.

1:34:22 It's reporting back to you, "I got that done." "You know,

1:34:24 what do you want me to do next?" You know, it...

1:34:27 That's the part that I think-...

1:34:28 most people don't realize is they're

1:34:30 The person who's gonna be chatting with them,

1:34:32 texting them most, is their claws or lobster.

1:34:38 What an incredible future.

1:34:39 Uh, I read that you attribute a lot of your success

1:34:42 to your ability to work harder than

1:34:43 anyone and withstand more suffering than anyone.

1:34:46 So we can list many of the things that entails.

1:34:50 I mean, dealing with failure,

1:34:53 the cost and engineering problems we've talked about.

1:34:55 The human problems, uncertainty, responsibility, exhaustion, embarrassment,

1:35:01 the near-death company moments that you've mentioned, But also the pressure.

1:35:06 Now, as the CEO of this company that economies

1:35:13 and nations strategize around plan their, Financial allocations around,

1:35:20 plan their AI infrastructure around, how do you deal with this much pressure?

1:35:26 What gives you strength, given how many nations and peoples depend on you?

1:35:38 I'm conscious about the fact that, NVIDIA's

1:35:43 success is very important to United States.

1:35:47 We generate enormous amounts of tax revenues.

1:35:51 We established technology leadership for our nation.

1:35:54 Technology leadership is important for national security.

1:35:57 National security not just in one aspect of national security,

1:36:00 all aspects of national security.

1:36:02 When our country's more prosperous,

1:36:05 we could do a better job with domestic policies and helping social benefits.

1:36:11 Because we're generating so much re-industrialization in the United States,

1:36:15 we're creating mountains of jobs.

1:36:18 We're helping shift, how we,

1:36:23 how we build things back to the United States in so many different plants,

1:36:28 chips, computers, and of course, these AI factories.

1:36:31 I'm completely aware that, that...

1:36:35 And I have the benefit,

1:36:37 and this is a real a real gift with mainstream investors,

1:36:45 teachers, policemen who have somehow, for whatever reason,

1:36:51 invested in NVIDIA or because they watched Jim

1:36:54 Cramer bought some stock and now are millionaires.

1:36:58 And I am completely aware of that circumstance.

1:37:03 I'm aware of the circumstance that NVIDIA, is central to a very large network

1:37:12 of ecosystem partners behind us and downstream from us.

1:37:16 And so the way I deal with that is exactly what I just did.

1:37:20 I reason about what is...

1:37:23 what is it that we're doing?

1:37:26 What is it causing?

1:37:27 What's the impact that has on other people benefit, you know,

1:37:31 positively or even, even through great burden, for example, to supply chain?

1:37:38 And the question is, therefore, what are you gonna do about it?

1:37:44 In almost everything that I feel, I break it down,

1:37:48 I reason about, "Okay, "what's the circumstance?

1:37:51 What has changed?

1:37:52 What's hard?

1:37:54 And what am I gonna do about it?" And I break it down, decompose the problem,

1:38:00 and the decomposition of these circumstances turns

1:38:07 it into manageable things that I can do.

1:38:09 And the only thing after that I could do is, "Did you do it?

1:38:13 Did you either do it or did you get somebody else to do it?

1:38:16 And if you didn't do it,

1:38:18 you reasoned that you need to do it, and you didn't do it,

1:38:21 and you didn't get anybody else to do it, then stop crying about it." You know?

1:38:27 And so-...

1:38:28 so I'm fairly Tough on myself.

1:38:32 And, but I also break things down so that, so that I don't panic.

1:38:37 I can go to sleep because I've made the list of things that needed to be done,

1:38:43 and I've made sure that everything that could put our company in harm's way,

1:38:47 could put my partners in harm's way,

1:38:49 put our industry in harm's way, I've told somebody.

1:38:52 Everything that I feel could put anybody in harm's way, I've told someone.

1:38:59 And I've told that someone who could do something about it.

1:39:02 And so I've gotten it off my chest or I'm doing something about it.

1:39:07 And so after that, Lex, what else can you do?

1:39:11 So given all the insane,

1:39:12 intense amount of suffering on the journey of building up NVIDIA,

1:39:18 have you hit low points psychologically?

1:39:22 Oh, yeah.

1:39:23 Oh, yeah.

1:39:24 Sure.

1:39:24 All the time.

1:39:25 All the time.

1:39:27 And there—- All the time.

1:39:28 you just break down the problem into pieces?

1:39:31 Yeah.

1:39:31 Yeah.

1:39:32 See what you could do about it?

1:39:34 And, and part of, and, you know, Lex, part of it, part of it is forgetting.

1:39:39 One of the most important attributes of AI learning, as you know, is, right?

1:39:43 Systematic forgetting.

1:39:43 You, you need to know when to forget some things.

1:39:47 You can't memorize everything.

1:39:48 You can't keep everything and, and, you know, you,

1:39:51 you want to— you don't want to carry everything.

1:39:53 One of the things that I do very quickly is decompose the problem,

1:39:56 I reason about the problem, and I share the load with it.

1:39:59 When I say I tell everybody, I'm essentially sharing that burden.

1:40:03 As quickly as possible.

1:40:06 Whatever worries me, tell somebody else.

1:40:08 Don't just keep it.

1:40:09 You know, don't freak them out.

1:40:12 Decompose the problem into smaller parts and get people to, so,

1:40:17 and, and inspire them to be able to go do something about it.

1:40:20 But part of it is just, just forgetting.

1:40:22 You know, like, a lot of it is you gotta be tough on yourself.

1:40:27 You know, just come on, stop crying about it.

1:40:29 Let's get going.

1:40:29 You know?

1:40:30 And, and then you get out of bed.

1:40:31 And then the other part is, is you, you're attracted to the next shiny light,

1:40:37 the next future, you know, the next opportunity,

1:40:39 the next, "Okay, that's behind us.

1:40:41 Let— what's next?" It's a lot, I think,

1:40:44 you know, you watch this with great athletes.

1:40:46 They, they just worry about the next point.

1:40:50 The last point is behind them.

1:40:53 The embarrassment, the, you know— the setback.

1:40:56 You know, and, and then, and because I do so much of my job publicly, you know?

1:41:01 Lex, you do a fair amount of your job publicly too.

1:41:03 And so, so I do a lot of my job publicly.

1:41:06 And so you know, I say a lot of things

1:41:09 that, that seem sensible at the time or funny at the time,

1:41:13 mostly it's just because it's funny to me at the time.

1:41:15 And then, you know, you reflect on it, it's less funny, but, but...

1:41:20 Yeah.

1:41:20 No, trust me, I know.

1:41:22 But you basically allow yourself to be pulled by the light of the future.

1:41:26 Forget the past and just keep-- That's right.

1:41:29 keep working towards that.

1:41:29 I mean, you did say,

1:41:31 there's this kind of famous thing you said that if you knew how hard

1:41:36 it would be to build NVIDIA it turned out to be, what is it?

1:41:41 A million times more hard than you anticipated that you wouldn't do it.

1:41:47 Yeah, right.

1:41:47 Um, but isn't...

1:41:48 You know, when I hear that, that's

1:41:51 probably true about everything worth doing, right?

1:41:54 Exactly.

1:41:54 That is, by the way, what I was trying to explain,

1:41:57 is that there's a, there's a incredible superpower of being,

1:42:02 being have a, the mind of a child.

1:42:07 You know?

1:42:08 And I say to myself oftentimes when I look at something, and almost,

1:42:12 almost everything, My first thought is, "How hard can it be?" You know?

1:42:20 And so you get yourself into that mode, how hard could it be?

1:42:24 And, and nobody's ever done it.

1:42:26 It looks gigantic.

1:42:28 It's gonna cost hundreds of billions of dollars.

1:42:31 It's gonna take, you know, all this...

1:42:32 And you just go, "Yeah, but how hard could it be?" You know?

1:42:35 How hard could it be?

1:42:37 And so you gotta get yourself into that state of mind.

1:42:40 You don't wanna, you don't wanna actually over simulate everything and all

1:42:46 the setbacks and all the trials and tribulations and all the disappointments.

1:42:49 You don't wanna simulate all that in advance.

1:42:51 You don't wanna know that.

1:42:53 You don't, you wanna go into a new experience thinking it's gonna be perfect,

1:42:56 it's gonna be great, it's gonna be incredibly fun.

1:42:59 And then while you're there, you know,

1:43:01 you need to have, you need to have endurance,

1:43:04 you need to have grit, so that when the setbacks actually happened,

1:43:07 and those setbacks are gonna surprise you,

1:43:10 the disappointments are gonna surprise you,

1:43:12 you know, the embarrassments are gonna surprise you,

1:43:15 the humiliations are gonna surprise you.

1:43:17 You just can't let...

1:43:18 Now you just gotta turn on the other bit, which is just forget about it.

1:43:22 Move on, keep moving.

1:43:23 And, and to the extent that, to the extent that my assumptions

1:43:31 about the future and why the future is gonna manifest,

1:43:36 so long as those assumptions and that input

1:43:40 doesn't change or didn't change materially,

1:43:42 then I should expect that the output won't change.

1:43:45 And so my simulated output of the future is still gonna happen.

1:43:50 And if it's still gonna happen, I'm still gonna go after it.

1:43:54 I believe it's gonna, you know,

1:43:56 and so there's a combination of two or three human characteristics,

1:44:01 the ability to go into a, into an experience fresh-minded,

1:44:05 the ability to forget the setbacks, the ability to believe in yourself,

1:44:11 you know, to believe what you believe and stay, stay true to that belief.

1:44:16 But you're constantly reevaluating.

1:44:18 This combination of three, four,

1:44:21 five things I think is, is really important for resilience.

1:44:27 And, and...

1:44:28 and, you know, I'm fortunate that, that whatever life experiences

1:44:33 led to this, I've got kind of those four, five things.

1:44:37 You know, I'm always curious, always learning.

1:44:40 I'm always learning from everybody, you know?

1:44:42 I'm always asking my...

1:44:44 And because I'm humble about everything,

1:44:46 I'm always thinking, "Gosh, they did that so nicely.

1:44:50 They did that so wonderfully." You know, I wonder what they're thinking through.

1:44:54 How do they...

1:44:55 You know, so I'm simulating everybody.

1:44:57 In a lot of ways, you know, I'm emulating almost everybody I watch, right?

1:45:02 You're empathetic towards everything that they

1:45:04 do that you're observing and respect.

1:45:07 And, and so you're constantly learning and, you know.

1:45:11 You're now one of the wealthiest people on Earth.

1:45:14 One of the most successful humans on Earth.

1:45:18 Is it harder to be humble and to be able to...

1:45:20 Do you feel the effect of money and power and fame in making

1:45:25 it harder for you to sort of be wrong in your own head?

1:45:32 Enough to hear out an opinion of somebody else

1:45:35 when they disagree with you and learn from them?

1:45:37 Those kinds of things.

1:45:42 Um, surprisingly, no.

1:45:42 And I would, I would actually go the other way.

1:45:46 Because I do so much of my work publicly,

1:45:49 when I'm wrong, pretty much everybody sees it.

1:45:54 You get humbled.

1:45:54 Fair enough.

1:45:55 And when I'm wrong, when I'm wrong or it didn't turn out that way or you know,

1:46:01 I mean, most of the things that, that I say outside I'm fairly certain about.

1:46:07 And the reason for that is because,

1:46:09 because it's gonna impact somebody else and I

1:46:12 want to be quite concerned about that and quite, circumspect about that.

1:46:15 For stuff that, that I'm reasoning about inside a meeting,

1:46:20 you know a lot of things could turn out differently.

1:46:23 And so, but it doesn't ever stop me from reasoning.

1:46:25 The way that the way that I manage and lead,

1:46:28 I, you know, I'm constantly reasoning in front of people.

1:46:31 And even when I'm talking to you,

1:46:33 you can kind of see me kind of reasoning through things.

1:46:35 And I want to make sure that you

1:46:37 understand what I'm saying not because I told you-...

1:46:40 because I'm so humble about what I'm about to tell you.

1:46:43 I kind of show you the steps that I got there.

1:46:46 And then you can decide whether you believe what I said in the end.

1:46:49 And so I'm doing that all day long in meetings.

1:46:52 With all of my employees,

1:46:53 I'm constantly reasoning through, "Let me tell you what,

1:46:55 how I see it." And then I reason through it.

1:46:58 It gives everybody the opportunity to intercept and say,

1:47:01 "I disagree with that part." The nice

1:47:05 thing about reasoning through things and letting

1:47:07 people interact with it is that they don't have to disagree with your outcome.

1:47:12 They can disagree with your reasoning steps.

1:47:15 And they could pull me in different directions, and then we can reason forward.

1:47:19 And so we're kind of, you know, collective path searching method.

1:47:26 And it's really fantastic.

1:47:29 Yeah, you have this way about you of...

1:47:32 When you're explaining stuff, I can feel you actually reasoning on the spot

1:47:36 about it with a constant open-mindedness where you could...

1:47:40 I could feel like I could steer your thinking.

1:47:44 And that's a, that's really beautiful that you've been able

1:47:46 to maintain that after so many years of success, and pain.

1:47:50 I think sometimes pain makes you close, closes you down a bit.

1:47:55 Mm-hmm.

1:47:55 Yeah.

1:47:56 And I think to maintain-- Yeah.

1:47:57 Tolerance for embarrassment, I think is...

1:48:00 Yes, that's...

1:48:00 The tolerance...

1:48:01 I mean, that's a real thing.

1:48:03 Is many years of embarrassing yourself.

1:48:05 Even those meetings knowing that there's people around you where you

1:48:09 declared one idea and it was shown that that idea was wrong-...

1:48:13 and be able to admit that and to grow from that.

1:48:15 That's not, that's very difficult on a human level.

1:48:18 Yeah.

1:48:18 Well, you know.

1:48:19 They knew that recently my first job was, you know, cleaning toilets, so.

1:48:25 I'm glad you maintained that same spirit of Denny's the, the work.

1:48:30 I mean, that, that was beautiful.

1:48:31 Your whole journey from, starting from Denny's is a beautiful one.

1:48:34 Let me ask you about video games.

1:48:37 So I'm a big gaming fan.

1:48:41 So I have to say thank you to NVIDIA for many years of incredible graphics.

1:48:47 By the way, GeForce is our still, to this day-...

1:48:52 our number one marketing strategy.

1:48:55 Right.

1:48:55 People learn about NVIDIA while they're in their teenage years.

1:49:00 And then they go to college and they know

1:49:02 who NVIDIA is and then in beginning is just,

1:49:04 you know, playing Call of Duty, you know?

1:49:06 You know, Fortnite.

1:49:07 And then later they're using CUDA,

1:49:09 and then later they're using NVIDIA and, you know,

1:49:12 Blender and Dassault and Autodesk.

1:49:16 Yeah.

1:49:16 I mean, I should say I mentioned to a friend that I'm talking with you.

1:49:21 He said, "Oh, they make great gaming GPUs."- Yeah, exactly.

1:49:26 It's like-- Exactly.

1:49:27 You know, there's more to it, but, yeah, yeah, people really love the...

1:49:32 It really brought a lot of joy to a lot of people.

1:49:34 The, the, the hardware really brings these worlds to life.

1:49:38 There was some controversy around this with DLSS 5.

1:49:44 Can you explain to me the drama around this?

1:49:46 Uh, I guess people, the gamers online were

1:49:50 concerned that it makes games look like AI slop.

1:49:54 Uh, what do you think of this drama?

1:49:56 Yeah.

1:49:57 Uh, I think their perspective makes sense and I could see

1:50:02 where they're coming from, because I don't love AI slop myself.

1:50:05 You know, all of the AI generated content increasingly,

1:50:10 um, looks similar and they're all beautiful, and I can...

1:50:15 So I can...

1:50:16 I'm empathetic towards what they're thinking.

1:50:18 That's just not what DLSS 5 is trying to do.

1:50:21 I showed several examples of it.

1:50:24 But DLSS 5 is 3D conditioned, 3D guided.

1:50:29 It's ground truth structure data guided.

1:50:32 And so the artist determined the geometry.

1:50:36 We are completely truthful....

1:50:38 to the geometry maintains in every single frame.

1:50:43 It's conditioned by the textures, the artistry of the artist.

1:50:48 And so every single frame, it enhances but it doesn't change anything.

1:50:54 Now, the question is, the question about enhancing,

1:50:59 DLSS 5 also lets, because it's, the system is open,

1:51:03 you could train your own models to determine,

1:51:06 and you could even in the future prompt it.

1:51:08 You know, I want it to be a toon shader.

1:51:10 I want it to look like this kinda, you know, so you can give it even an example.

1:51:15 And it would generate in the style of that, all consistent with the artistry,

1:51:21 you know, the style, the intent of the artist.

1:51:25 And so all of that is done for the artist,

1:51:30 so that they can create something that is more beautiful,

1:51:34 But still in the style that they want.

1:51:37 I think that they got the impression that the games are gonna

1:51:42 come out the way the games are shipped the way they do,

1:51:46 and then we're gonna post-process it.

1:51:48 That's not what DLSS is intended to do.

1:51:50 DLSS is integrated with the artist, and so it's,

1:51:54 it's about giving the artist the tool of AI, the tool of generative AI.

1:51:59 They could decide not to use it, you know?

1:52:01 I think people are very sensitive to human faces.

1:52:04 And we're now living in this moment, which I think is a beautiful one,

1:52:08 which is people are sensitive to AI slop.

1:52:11 It puts a mirror to ourselves to help

1:52:13 us realize that what we seek is imperfections.

1:52:16 What we seek is sometimes not perfect graphics.

1:52:20 It helps us understand what we find compelling in the worlds we create.

1:52:25 And that's beautiful.

1:52:25 And as long as it's tools that help

1:52:27 us create those worlds-- Yeah, that's right-...

1:52:29 it's wonderful.

1:52:30 That's right.

1:52:30 Yet, yet another tool, and they want the generative,

1:52:33 models to generate the opposite of photo real.

1:52:39 Yeah, it'll do that too.

1:52:40 And so it's just yet another tool.

1:52:42 I think the the gamers might also appreciate

1:52:45 that, that in the last couple of years,

1:52:50 we introduced skin shaders to the game developers.

1:52:56 And many of those games have skin shaders

1:52:59 that include subsurface scattering that make skin look more skin-like.

1:53:05 And so the industries, you know, game developers are looking for more and more

1:53:09 and more tools to express their art.

1:53:12 And so this is just yet more, one more tool, and they get to decide what to use.

1:53:17 Ridiculous question.

1:53:17 What do you think is the greatest or most influential game ever made?

1:53:22 Maybe from NVIDIA's perspective?

1:53:24 Doom.

1:53:25 Doom, unquestionably.

1:53:26 That was the start of the 3D.

1:53:28 I would say Doom, from an art,

1:53:31 the intersection of the cultural implication as well as the industry,

1:53:35 turning a PC into a gaming device.

1:53:38 That was a very important moment.

1:53:40 Now, now of course, flight simulation companies were before it.

1:53:43 And but they just didn't have the popularity

1:53:46 that Doom did to have made the industry turn

1:53:49 the PC from an office automation tool into a personal

1:53:54 computer for families and gamers and things like that.

1:53:57 And so Doom was really impactful there.

1:53:59 From an actual game technology perspective, I would say Virtua Fighter.

1:54:03 And so we're great friends with both of them, you know?

1:54:07 And then there's games more recently,

1:54:10 I mean, Cyberpunk 2077, really nice GPU-accelerated graphics.

1:54:15 Like-- Fully ray traced.

1:54:17 Fully ray traced.

1:54:18 Also, I like, I personally, I'm a huge fan of Skyrim,

1:54:22 uh, Elder Scrolls, and the, you know,

1:54:24 it's, it's been released a long, long time ago,

1:54:27 but people release mods and-- We love mods-...

1:54:30 they create these, these inc- I mean, it-...

1:54:32 it's like a different game and it just allows

1:54:34 me to replay the game over and over and get

1:54:36 i- It makes you realize that you can re- experience

1:54:40 in a totally new way the world you already love.

1:54:44 So-- That's right-...

1:54:45 I do that all the time.

1:54:46 One of my favorite things is just walk across Skyrim.

1:54:48 Uh, we created this thing called RTX Mod.

1:54:51 Yeah, it's a modding tool.

1:54:53 Awesome.

1:54:53 It allows the community to inject the latest technology into an old game.

1:55:00 Of course, like what makes a great video game is not just graphics,

1:55:03 it's also story and character development, but-- That's right-...

1:55:07 beautiful graphics can add to the immersion.

1:55:10 The feeling like it's another place you're transported to.

1:55:15 Ah, what you said, I think accurately,

1:55:18 that the AGI timeline question rests on your definition of AGI.

1:55:26 So, let me ask you about possible timelines here.

1:55:31 Let's, this ridiculous definition perhaps of what AGI is,

1:55:35 but an AI system that's able to essentially do your job.

1:55:42 So, run, no, start, grow,

1:55:46 and run a successful technology company that's worth-- A good one or a one?

1:55:54 No.

1:55:54 It has to- It has to be worth more than a billion,

1:55:58 more, more than a billion dollars.

1:56:01 So, you know, you know how hard it is to do all those components.

1:56:06 So, how far are we away from that?

1:56:09 So, we're talking about OpenClaw that does all the incredibly

1:56:15 complex stuff that are required to to, first of all, innovate,

1:56:20 to find customers, to sell to them, to manage,

1:56:23 to build a team of some agents, some humans, all that kind of stuff.

1:56:28 Is this five, 10, 15, 20 years away?

1:56:31 I think it's now.

1:56:32 I think we've achieved AGI.

1:56:35 Do you think you could have a company run by an AI system like this?

1:56:37 Possible, and the reason for that is this.

1:56:39 You said a billion, and you didn't say forever.

1:56:42 And so for example, uh...

1:56:45 It is not out of the question that a Claw was able to create a web service,

1:56:55 some interesting little app that all of a sudden,

1:57:01 you know, a few billion people used for 50 cents,

1:57:05 and then it went out of business again shortly after.

1:57:09 Now, we saw a whole bunch of those type of companies during the internet era,

1:57:12 and most of those websites were not anything

1:57:16 more sophisticated than what OpenClaw could generate today.

1:57:21 Interesting.

1:57:21 Achieve virality and monetize that virality.

1:57:24 Yeah.

1:57:24 It's just that I don't know what it is, but I couldn't have predicted any

1:57:27 of those companies at the time either, you know?

1:57:29 And-- You're gonna get a lot of people excited with that statement.

1:57:33 It's like, what do you mean?

1:57:34 I can- I can just, uh-...

1:57:36 launch an agent and make a lot of money.

1:57:38 Well, by the way, it's happening right now, right?

1:57:40 You know that when, when you go to China you're gonna see,

1:57:43 you're gonna see a whole bunch of people teaching their, getting their Claws

1:57:48 to try to go out and look for jobs and, you know, do work, make money.

1:57:53 And I'm not, I'm not actually...

1:57:55 I wouldn't be surprised if some social thing

1:57:58 happened or somebody created a, a digital influencer,

1:58:02 super, super cute or some social application that, you know,

1:58:07 feeds your little Tamagotchi or something like that, and, and it

1:58:10 become an out of the blue an instant success.

1:58:13 A lot of people use it for a couple of months and it kind of dies away.

1:58:19 Now, the odds of, you know,

1:58:23 100,000 of those agents, Building NVIDIA is zero percent.

1:58:28 And then the one part that I will, I won't do, And I,

1:58:33 I want to make sure we all do,

1:58:35 is to recognize that people are really worried about their jobs.

1:58:40 And I just want to remind them that the purpose of your job and the tasks

1:58:48 and tools that you use to do your job are related, not the same.

1:58:52 I've been doing my job for 33 years.

1:58:55 I'm the longest running tech CEO in the world, 34 years.

1:58:58 And the tools that I've used to do

1:59:00 my job have changed continuously in the last 34 years,

1:59:05 and sometimes quite dramatically, you know,

1:59:08 over the course of a couple, two, three years.

1:59:10 And the one story that I really wanna make sure that everybody

1:59:15 hears is the story that the first job that computer scientists said,

1:59:21 AI researchers said was gonna go away was radiology.

1:59:25 Because computer vision was going to achieve superhuman levels, and it did.

1:59:30 CV...

1:59:31 Computer vision was superhuman in 2019, 20, maybe a little bit later, 2020?

1:59:39 Okay?

1:59:39 And so it's been a long time since computer vision has been superhuman.

1:59:42 And so the prediction was radiologists would go away

1:59:46 because studying radiology scans was a thing of the past.

1:59:49 AI will do that.

1:59:50 Well, they were absolutely right.

1:59:54 Computer vision is completely superhuman.

1:59:57 Every radiology platform and package today is driven by AI,

2:00:02 and yet the number of radiologists grew.

2:00:05 And so the question is why?

2:00:07 And we now have a shortage of radiologists in the world.

2:00:11 And so, one, the alarmist warning went too far and it

2:00:17 scared people from doing this profession that is so important to society.

2:00:23 And so it did harm.

2:00:25 Now, why was it wrong?

2:00:26 The reason why is because the purpose of a radiologist,

2:00:31 the purpose is to diagnose disease

2:00:33 and help patients and doctors diagnose disease.

2:00:38 And because we're able to study scans at so much faster now,

2:00:42 you could study more scans, you could diagnose better, you could,

2:00:47 you could inpatient faster, you can see people more.

2:00:52 The hospitals are making more money.

2:00:53 You have more patients in the hospital.

2:00:56 You need more radiologists.

2:00:57 I mean, the amazing thing is, it's so obvious this was gonna happen.

2:01:03 The number of software engineers at NVIDIA is gonna grow, not decline.

2:01:08 And the reason for that is because the purpose of a software engineer

2:01:12 and the task of a software engineer coding are related, not the same.

2:01:16 I wanted my software engineers to solve problems.

2:01:18 I didn't care how many lines of code they wrote, you know?

2:01:21 But their job, their purpose of their job didn't change.

2:01:25 Solving problems, working as a team, diagnosing problems, evaluating the result,

2:01:32 looking for new problems to solve, innovation, connecting dots.

2:01:36 You know, none of that stuff is gonna go away.

2:01:39 Do you think it's possible that...

2:01:41 Let's even take coding.

2:01:41 Do you think the number of programmers

2:01:43 in the world might increase, not decrease?

2:01:45 Yes.

2:01:46 And the reason for that is this.

2:01:48 What is the definition of coding?

2:01:52 I believe it is...

2:01:53 The definition of coding, as of today, is simply specifying,

2:01:58 specification, and maybe if you want to be rather directive,

2:02:02 you could even give it an architecture of the software that you wanted to write.

2:02:07 So the question is, how many people could do that?

2:02:11 Describe a specification for a computer to go...

2:02:14 Telling the computer what to go build.

2:02:16 How many people?

2:02:17 I think we just went from 30 million to probably 1 billion.

2:02:21 And so every carpenter in the future will be a coder,

2:02:27 except a carpenter with AI is also an architect.

2:02:33 They've just increased the value that they could deliver to the customer.

2:02:36 Their, their artistry just elevated tremendously.

2:02:43 I believe that every accountant is, you know,

2:02:45 also your financial analyst, also your financial advisor.

2:02:49 So, all of these professions have just been elevated....

2:02:52 and, and if I were a carpenter,

2:02:54 I sees a, I see AI, I would just completely go berserk.

2:02:58 You know, the services I can bring to my clients if I were a plumber,

2:03:03 completely go berserk.

2:03:04 And the people that are currently programmers and software engineers,

2:03:07 I think they're at the cutting

2:03:10 edge of understanding intuitively how to communicate

2:03:15 with the agents using natural language in order

2:03:17 to design the best kind of software.

2:03:20 That's right, exactly.

2:03:21 So over time they'll converge, but I think there's still value in getting,

2:03:26 I think learning how to program, like learning what programming languages are.

2:03:30 The old, the old kind of programming,

2:03:33 what are good practices for programming languages,

2:03:37 what are design principles for programming-- That's right-...

2:03:41 Languages for large software systems?

2:03:43 And the reason for that, Lex, and you know, as you're saying for the audience,

2:03:47 I think the goal of, the goal of specification,

2:03:52 the artistry of specification, the goal and the artistry of it,

2:03:56 Is going to depend on what problem you're trying to solve.

2:04:01 When I'm thinking, when I'm thinking about giving the company

2:04:05 strategies and formulating corporate directions and things that we should do,

2:04:11 I describe it at a level that is sufficiently

2:04:16 specific that people generally understand the direction and it's actionable.

2:04:22 It's specific enough that they can take action on it,

2:04:26 but I under specify it on purpose,

2:04:29 so that enable 43,000 amazing people to make it even better than I imagined.

2:04:36 And so when I'm working with engineers and when I'm working with people,

2:04:41 I think about who, what problem am I trying to solve?

2:04:44 Who am I working with?

2:04:47 And the level of specification,

2:04:50 the level of architecture definition relates to that.

2:04:55 And, and so everybody's going to have to learn how,

2:05:01 where in the spectrum of coding they want to be.

2:05:03 Writing a specification is coding.

2:05:06 And so you might decide to be quite prescriptive

2:05:09 because there's a very specific outcome you're looking for.

2:05:12 You might decide that, you know,

2:05:14 this is an area you want to be much more exploratory,

2:05:17 and so you might under specify and enable you to go back

2:05:20 and forth with the AI to even push your own boundaries of creativity.

2:05:25 And so this artistry of where you are in the spectrum,

2:05:28 this is the future of coding.

2:05:31 But just to linger on it outside of coding,

2:05:33 I think a lot of people, rightfully so, are worried about their jobs,

2:05:38 have a lot of anxiety about their jobs, especially in the white-collar sector.

2:05:42 I don't think any of us know what to do, With tumultuous times that always come

2:05:52 when automations and new technology arrives.

2:05:54 And I just...

2:05:56 First of all, I think we all need

2:06:01 to have compassion and the responsibility to feel sort

2:06:04 of the burden of what the actual suffering feels

2:06:08 like for individual people and families that lose their job.

2:06:10 I think whenever you have transformative

2:06:13 technology like that's coming with artificial intelligence,

2:06:15 there's going to be a lot of pain, and I don't know what to do about that pain.

2:06:21 Hopefully, it creates much more opportunities for those same people,

2:06:25 for the same kind of job as the tooling evolves

2:06:30 and makes them more productive and makes them more fun,

2:06:34 hopefully, as it does in the programming.

2:06:36 I've been having so much fun programming, I have to say.

2:06:39 Like, I've never had this much fun.

2:06:40 So hopefully it makes their job,

2:06:42 automates the boring parts and makes the creative parts,

2:06:46 the ones that the human beings are responsible for.

2:06:48 But still there's going to be a lot of pain and suffering.

2:06:52 So my first recommendation before...

2:06:53 And this is now how I deal with anxiety.

2:06:56 In fact, we just talked about it earlier.

2:06:59 Enormous anxiety about the future, enormous anxiety about the pressure,

2:07:02 enormous anxiety about uncertainty, I first break it down,

2:07:06 and then I'm gonna tell myself, "Okay,

2:07:10 there are some things you can do something about,

2:07:12 there's some things you can't do anything about.

2:07:14 But for the stuff that you can do something about,

2:07:17 let's reason about it and let's go do it."

2:07:20 If we were to hire a new college graduate today,

2:07:23 and I have a choice between two, one that have,

2:07:27 that is no clue what AI is and one that is expert in using AI,

2:07:33 I would hire the one who's expert in using AI.

2:07:38 If I had an accountant, a marketing person,

2:07:42 the one that is expert in using AI, supply chain,

2:07:45 customer service, a salesperson, business development, a lawyer,

2:07:52 I would hire the one who is expert in using AI.

2:07:55 And so I would advise that every college student,

2:07:59 every teacher should encourage their student to be, to go use AI.

2:08:04 Every college student should graduate and be an expert in AI.

2:08:08 And everybody, if you're a carpenter,

2:08:10 if you're, you know, electrician, go use AI.

2:08:15 Go see what it can do to transform your current job, elevate yourself.

2:08:20 If I were a farmer, I would absolutely use AI.

2:08:24 If I were a pharmacist, I would use AI.

2:08:26 I wanna see how, what it could do to elevate my job

2:08:30 so that I could be the innovator to revolutionize this industry myself.

2:08:36 And so that would be the first thing that I would do.

2:08:38 And then I would also, I would also help them...

2:08:44 it is the case that the technology

2:08:47 will dislocate and will eliminate many tasks if...

2:08:52 And because it will automate it, if your job is the If your job is the task,

2:08:58 then you're very highly going to be disrupted.

2:09:02 If your job's purpose includes you, certain tasks-...

2:09:08 then it's vital that you go learn how to use AI to automate those tasks.

2:09:12 And then there's the world of spectrum in between.

2:09:15 And by the way, the beautiful thing about AI,

2:09:18 so the chatbot versions, is you can break down...

2:09:23 You have anxiety and you can break down the problem by talking to it.

2:09:28 Like, I've recently...

2:09:29 It's really just incredible how much you

2:09:31 can think through your life's problems, and through...

2:09:33 And I don't mean, like, therapy problems.

2:09:35 I mean, like, very practically, "Okay,

2:09:37 I'm worried about my..." Literally, "I'm worried about my job.

2:09:40 What are the skills?

2:09:41 What are the steps I need to take?

2:09:42 How do I get better at AI?" Everything you just said,

2:09:44 you could literally ask and it's going to give you-...

2:09:47 a point-by-point plan.

2:09:48 I mean, it's just a great life coach, period.

2:09:51 This-- I don't know how to use AI,

2:09:53 and the AI goes, "Well, let me show you."- Exactly.

2:09:55 It's very meta, but it's- It's kind of incredible.

2:09:58 So people definitely should-- You can't walk up to Excel and say,

2:10:01 "I don't know how to use Excel." You're done.

2:10:03 I mean, that's really what AI has done for me in all walks of life,

2:10:07 is that initial friction of being a beginner

2:10:09 of using a thing for the first time.

2:10:11 I can literally ask about any single thing,

2:10:15 "What are the first steps I need to take?"- That's right.

2:10:17 And that handholding that it does,

2:10:19 removing the friction of all the experiences that the world offers is...

2:10:24 You know, like I mentioned to you offline,

2:10:27 you mentioned, "I'm going to China and Taiwan."- So awesome.

2:10:31 Just ask, "Where do I-"- So excited for you.

2:10:32 "Where do I—what do—" "You know, where do I go?

2:10:34 How do I..." All of those questions—...

2:10:36 immediately answered, and it's beautiful.

2:10:37 Well, when you, when you go to Taiwan, just ask AI-...

2:10:40 "What are Jensen's favorite restaurants in Taiwan?"

2:10:43 And it'll actually-- You don't know?

2:10:45 Oh, yeah.

2:10:46 Is it accurate?

2:10:46 Okay.

2:10:47 Yeah.

2:10:47 All right.

2:10:48 It's all over Taiwan.

2:10:50 Well, you're a rockstar over there.

2:10:52 And like we also mentioned offline, maybe our paths will cross,

2:10:55 which would be really wonderful in computing.

2:10:58 COMPUTEX.

2:10:58 NVIDIA GTC Taiwan.

2:11:01 Uh, do you think there's some things about human nature,

2:11:06 about human consciousness that is fundamentally non-computational?

2:11:10 Maybe something a chip, no matter how powerful, can never replicate?

2:11:18 I don't know if the chip will ever get nervous.

2:11:20 And that's the, you know, of course,

2:11:22 the conditions by which that causes anxiety or nervousness or whatever emotion.

2:11:29 Um, I believe that AI will be able to recognize those and understand those.

2:11:37 I don't think my chips will feel those.

2:11:41 And therefore, the...

2:11:42 How that anxiety, how that feeling,

2:11:45 how that excitement, how that, how that, you know...

2:11:50 All of those feelings manifest in human performance.

2:11:53 For example, extremely amazing human performance, athletic performance,

2:11:57 you know, average or lesser than average.

2:12:01 That entire spectrum of human performance that comes

2:12:05 out of exactly the same circumstances for different people,

2:12:10 manifesting a different outcome, manifesting a different performance.

2:12:15 I don't think there's anything about anything

2:12:19 that we're building that would suggest that two

2:12:23 different computers being presented with all of exactly

2:12:26 the same context would perfo- Of course,

2:12:29 it would produce statistically different outcomes,

2:12:31 but it's not because it felt different.

2:12:34 Yeah, the subjective...

2:12:36 Boy, there's something truly special about

2:12:38 the subjective experience that we humans feel.

2:12:42 Like I mentioned to you, I was pretty nervous talking to you.

2:12:47 Like I mentioned to you, that, the hope, the fear,

2:12:50 the anxiety, and just life itself, the richness of life.

2:12:53 How amazing everything is.

2:12:55 How deeply we fall in love, how deeply our hearts get broken,

2:12:59 how afraid we are of death and how much

2:13:01 pain we feel when our loved ones pass away.

2:13:04 All of that, the whole thing.

2:13:07 I know it's very hard to-...

2:13:09 think AI being able to...

2:13:11 A computational device being able to do that.

2:13:13 But there's so many mysteries about this whole thing that we're yet to uncover,

2:13:17 that I am open to be surprised.

2:13:20 I've been surprised a lot over the past-...

2:13:23 few months and few years.

2:13:25 Scaling can create some incredible miracles in the space of intelligence.

2:13:30 Has been truly marvelous to watch, so I'm open to surprise.

2:13:34 And it's just really important to break down what is intelligence.

2:13:38 You know, the word, that word we use all the time, it's not a mysterious word.

2:13:43 Intelligence has a meaning, you know?

2:13:46 And it's a system that...

2:13:48 You know, it's something that we do that includes perception

2:13:52 and understanding and reasoning and the ability to do plan.

2:13:56 And, you know, that, that loop, that loop, is the...

2:14:01 Fundamentally what intelligence is.

2:14:04 Intelligence is not one word that is exactly equal to humanity.

2:14:11 And that's, I think it's really important to separate the two.

2:14:13 We have two words for that.

2:14:15 I'm not...

2:14:16 I don't over-fantasize about, and I don't over-romanticize about intelligence.

2:14:22 Intelligence is...

2:14:24 And people have heard me say it before,

2:14:26 I actually think intelligence is a commodity.

2:14:29 I'm surrounded by intelligent people.

2:14:33 And I'm surrounded by intelligent people more intelligent than I

2:14:35 am in each one of the spaces that they're in.

2:14:39 And yet, I have a role in that circle.

2:14:42 It's actually kind of interesting.

2:14:45 They're more educated than I am.

2:14:47 They went to better schools than I did.

2:14:51 They're deeper than, in any of the fields that they're in.

2:14:54 All of 'em.

2:14:55 I have 60 of 'em.

2:14:57 They're all superhuman to me.

2:14:59 And somehow, I'm sitting in the middle orchestrating all 60 of 'em.

2:15:03 And so you gotta ask yourself...

2:15:05 Uh, what is, what is it about a dishwasher

2:15:09 that allows that dishwasher to sit in the middle of superhumans?

2:15:13 Does that make sense?

2:15:15 And so, but that's my point.

2:15:17 My point is intelligence is a functional thing.

2:15:22 Humanity is not a, not specified functionally.

2:15:25 It's a much, much bigger word.

2:15:29 And, and our life experience, our tolerance for pain,

2:15:33 our determination, those are, those are different words than intelligence.

2:15:39 And so the thing that I wanna help the audience understand,

2:15:43 if I could give them one thing,

2:15:45 is, is intelligence is a word that we've elevated to a very high form over time.

2:15:51 The, the word we should really elevate is humanity.

2:15:54 Character, humanity.

2:15:55 All those things.

2:15:56 All of those things.

2:15:58 Compassion, generosity, all of the things that you said just now,

2:16:03 I believe those are superhuman powers.

2:16:06 And that now intelligence is gonna be commoditized.

2:16:09 Because we've spoken about it, the most important thing is your education.

2:16:13 The most...

2:16:13 Now, even, even when they said the most important thing is your education,

2:16:17 when you went to school, there's more than just knowledge that you gained.

2:16:22 And so, but unfortunately, our society has put everything into one single word,

2:16:28 and life is more than one word.

2:16:30 And I'm just telling you,

2:16:32 my life would suggest that being lower on the intelligence curve

2:16:38 than everybody around me doesn't change the fact I'm the most successful.

2:16:43 And so, and I think, I think that kind of is I'm trying to hopefully

2:16:48 inspire everybody else that don't let this democratization of intelligence,

2:16:53 this commoditization of intelligence, you know, cause you anxiety.

2:16:58 You should be inspired by that.

2:17:00 Yeah.

2:17:01 I think AI will help us celebrate humans more.

2:17:05 And certainly humanity and human first, and I,

2:17:10 I think what makes this world incredible is humans forever will be so,

2:17:15 and just AI is this incredible tool that makes us-- That's exactly right.

2:17:19 humans more powerful.

2:17:20 That's exactly right.

2:17:21 Uh, so much of the success of NVIDIA and the lives

2:17:26 of millions of people that I mentioned depend on you.

2:17:31 But you're just one human, like we mentioned, a mortal like all of us.

2:17:36 Do you think about your mortality?

2:17:38 Are you afraid of death?

2:17:42 I really don't wanna die.

2:17:44 Um, I have a great life.

2:17:45 I have a great family.

2:17:48 Uh, I have really important work.

2:17:51 Uh, this is, this is not a once in a, once

2:17:58 in a lifetime experience suggests that it has been experienced by many people,

2:18:04 just not one person.

2:18:06 This is a once in a humanity experience, what I'm going through.

2:18:11 Uh, NVIDIA is one of the most consequential technology companies in history.

2:18:15 We're doing very important work.

2:18:17 I take it very seriously.

2:18:19 Um, And so some of the, some of the things that, that of course are,

2:18:24 are practical things, like how do we think about succession planning?

2:18:30 And, I'm famous in saying that I don't believe in succession planning.

2:18:36 Man.

2:18:37 And the reason, the reason for that, the reason

2:18:39 for that isn't because I'm immortal.

2:18:41 The reason for that is because if

2:18:43 you're worried about succession planning, if you're worried,

2:18:49 all that anxiety of succession planning, then what should you do about it?

2:18:52 Then you break it all the way back down.

2:18:54 The most important thing you should do today,

2:18:55 if you care about the future of your company,

2:18:58 post you, is to pass on knowledge, information,

2:19:03 insight, skills, experience as often and continuously as you can,

2:19:07 which is the reason why I continuously

2:19:09 reason about everything in front of my team.

2:19:13 Every single meeting is about a reasoning meeting.

2:19:16 Every moment I spend inside a company,

2:19:19 outside a company is about passing on knowledge to people as fast as I can.

2:19:23 Nothing I learn ever sits on my desk longer than,

2:19:27 you know, a fraction of a second.

2:19:29 I'm passing that information, that knowl- oh my gosh, this is cool.

2:19:32 Before I even finish learning all of it myself,

2:19:35 I'm already pointing it to somebody else.

2:19:37 "Get on this.

2:19:38 This is so cool.

2:19:39 You're gonna wanna, you're gonna wanna learn

2:19:41 this." And so I'm constantly passing knowledge,

2:19:44 empowering people, elevating the capability of everybody around me,

2:19:49 so that the outcome that I, that I seek,

2:19:55 that I hope for, is that I die on the job, you know?

2:19:58 And, and hopefully I die on the job instantaneously, you know?

2:20:03 And there's no long periods of suffering, you know?

2:20:05 It's, uh—- Well, from a fan perspective given your, your extremely, um,

2:20:13 your enormous positive impact on civilization, of course, I hope you keep going.

2:20:18 But also it's just fun to watch what NVIDIA is doing, you know.

2:20:22 It's just the rate of innovation.

2:20:24 And I'm a huge fan of engineering.

2:20:26 There's so much incredible engineering being continuously being done by NVIDIA.

2:20:30 It's just fun to watch.

2:20:31 It's a celebration of humanity,

2:20:33 a celebration of great builders, a celebration of great engineering.

2:20:36 So, it represents something special.

2:20:38 So I hope you and NVIDIA keep going.

2:20:41 What gives you hope about this whole thing we got going on, about humanity,

2:20:46 about the future of humanity?

2:20:48 When you look out, when you think about the future quite a bit,

2:20:51 when you look out 10, 20, 50, 100 years from now, what gives you hope?

2:20:57 I've always had a great confidence in the kindness, uh, the generosity, uh...

2:21:09 the compassion, the human capacity.

2:21:13 I've always been extremely confident of that.

2:21:19 Sometimes more so than I should.

2:21:25 And, and I get taken advantage of, but it doesn't,

2:21:28 it doesn't ever cause me not to.

2:21:31 I start with, always, That people want to do good.

2:21:38 People want to, um help others.

2:21:42 And, vastly, I am proven right.

2:21:47 Constantly proven right.

2:21:50 And, and often it exceeds my expectations.

2:21:56 And, and so I have complete confidence in the human capacity.

2:22:01 I think the, the thing that, the things that give me incredible hope,

2:22:07 Is what I see as, as I extrapolate, as I,

2:22:11 what I see now is possible, and as I extrapolate,

2:22:15 Based on the things that we're doing, what will very likely happen.

2:22:22 And, and that there's so many things that we wanna solve.

2:22:27 There's so many problems we wanna solve.

2:22:29 There's so many things that we wanna build.

2:22:32 There's so many good things that we wanna do that are now within our reach,

2:22:37 and within the reach of my, my lifetime.

2:22:40 You just can't possibly not be romantic about that.

2:22:44 You know what I'm saying?

2:22:46 What an exciting time to be alive.

2:22:48 Like, truly-- How can--...

2:22:50 truly so.

2:22:50 How can you not be romantic about, about that?

2:22:53 The, the fact that, that there is a, there,

2:22:58 it's a reasonable thing to expect the end of disease.

2:23:01 It's a reasonable thing to expect.

2:23:03 It's a reasonable thing to expect that pollution will be drastically reduced.

2:23:09 It's a reasonable thing to expect that traveling

2:23:12 at the speed of light is actually in our future.

2:23:17 And then, you know, not, not for long distances, but short distances.

2:23:21 You know, and people ask me how.

2:23:23 Well, first of all, very soon, I'm gonna put a humanoid on a spaceship,

2:23:26 and it's gonna be, you know, my humanoid,

2:23:29 and, and we're gonna send it out as soon, you know, as soon as possible,

2:23:33 and it's gonna keep improving and enhancing along the flight.

2:23:36 And then when it's time, all of the, all of my consciousness has already been,

2:23:41 you know so much of my life has been uploaded in the internet.

2:23:44 Take all my inbox, take everything that I've done, everything I've said.

2:23:47 You know, it's been collected and becoming my AI.

2:23:51 And I'm just, you know, when the time comes, you know,

2:23:54 we'll just send that at the speed of light, catch up with my robot.

2:24:00 Oh, that's brilliant.

2:24:01 I mean, but for me, that's sorta application-focused.

2:24:04 But also, for me, the curiosity-...

2:24:07 Maxing perspective, I just, all of those mysteries.

2:24:11 There's so much-...

2:24:12 fascinating scientific questions there.

2:24:15 Understanding the biological machine is right around the corner.

2:24:18 It's, it's not 10 years.

2:24:19 It's five years probably.

2:24:20 And then your biological machine,

2:24:22 the human mind and cracking physics, theoretical physics open.

2:24:25 It's so exciting.

2:24:26 Explaining consciousness, that one would be awesome.

2:24:29 And it's all within our reach.

2:24:31 Jensen, thank you so much for everything you've done over the years.

2:24:33 Thank you for everything you're doing for the world.

2:24:36 Thank you for being who you are.

2:24:38 I can tell you're a great human being,

2:24:42 and I wish you incredible success this year.

2:24:45 I can't wait.

2:24:46 As a fan, I can't wait to see what you do next,

2:24:48 and hopefully I'll see you in Taiwan and thank you so much for talking today.

2:24:52 Thank you, Lex.

2:24:53 I had a great time.

2:24:54 And also, if I could just say one more thing.

2:24:56 Yes.

2:24:57 And thank you for all the interviews that you do, the depth,

2:25:01 the respect that you go through with and the research that you do to reveal,

2:25:06 you know, for all of us,

2:25:09 The amazing people that you've interviewed over the years.

2:25:12 I've enjoyed I've enjoyed them immensely.

2:25:15 And as an innovator, to have created this long form,

2:25:21 unbelievable, and yet, you know, it's just captivating.

2:25:24 So anyways, thank you for everything you do.

2:25:26 It means the world.

2:25:26 Thank you, Jensen.

2:25:27 Thank you, Lex.

2:25:29 Thank you for listening to this conversation with Jensen Huang.

2:25:32 To support this podcast, please check out our sponsors in the description,

2:25:35 where you can also find links to contact me,

2:25:38 ask questions, give feedback, and so on.

2:25:42 And now, let me leave you with some words from Alan Kay.

2:25:46 "The best way to predict the future is to invent it." Thank you for listening,

2:25:52 and hope to see you next time.