WEBVTT - TechStuff Redux: Will NVIDIA Save or Ruin The World? 0:00:00.040 --> 0:00:02.560 Hi. Is Us Valoscian here and Cara Price, and we're 0:00:02.560 --> 0:00:04.800 taking some time off for the holidays. We'll be back 0:00:04.800 --> 0:00:07.960 with new episodes starting in January. In the meantime, instead 0:00:07.960 --> 0:00:10.680 of leaving this feed empty, we wanted to share one 0:00:10.680 --> 0:00:13.600 of my favorite episodes from last year. This week, we're 0:00:13.640 --> 0:00:17.240 rearing my conversation with Stephen Witt from November. He's an 0:00:17.239 --> 0:00:20.759 author and frequent contributor to The New Yorker who wrote 0:00:20.800 --> 0:00:23.400 the book on one of the biggest companies in the world. 0:00:23.680 --> 0:00:26.760 You may have heard of it in video. In this episode, 0:00:26.840 --> 0:00:30.160 we hear how the CEO, Jensen Huang, went from working 0:00:30.160 --> 0:00:34.839 at Denny's to being the world leader manufacturing AI chips. 0:00:35.159 --> 0:00:51.199 Hope you enjoy it and thanks for listening. Welcome to 0:00:51.200 --> 0:00:55.080 Tech Stuff. I'm Us Vlashan here with Cara Price. Hey, Kara, Hi, 0:00:55.200 --> 0:00:58.600 as so years ago around the time that we were 0:00:58.600 --> 0:01:03.960 reporting our first podcast together on the forthcoming AI Revolution 0:01:04.240 --> 0:01:08.920 around no longer forthcoming. Yeah, you invested in in Vidia, 0:01:09.560 --> 0:01:14.320 which is up over one hundred x since then. Congratulations. 0:01:14.520 --> 0:01:17.080 You know, I just felt when we reported on it 0:01:17.120 --> 0:01:18.959 that it was going to be the future, and so 0:01:19.120 --> 0:01:22.600 I did invest in it, and you know, very happily. 0:01:22.680 --> 0:01:25.280 So now, what was it? Was there something that tipped 0:01:25.280 --> 0:01:28.080 you over the edge back then to do it? Well? 0:01:28.120 --> 0:01:30.000 I was just sort of thinking to myself, this is 0:01:30.040 --> 0:01:33.119 the thing that's going to power everything. So obviously it's 0:01:33.200 --> 0:01:36.319 something that people are going to be paying attention to 0:01:36.400 --> 0:01:38.200 and investors are going to be paying attention to, and 0:01:38.240 --> 0:01:40.640 it just it made a lot of sense to me 0:01:40.680 --> 0:01:41.119 at the time. 0:01:41.720 --> 0:01:45.360 We know Warren Buffett is retiring this year, there's a 0:01:45.400 --> 0:01:50.520 slot open. In Vidia is, of course the most valuable 0:01:50.520 --> 0:01:53.520 company in the world today, recently topping five trillion dollars 0:01:53.560 --> 0:01:55.880 in value, and of course a lot of people are 0:01:55.880 --> 0:01:58.960 crying bubble, not just for Nvidia but for the AI 0:01:59.040 --> 0:02:01.520 industry as a whole, actually carrying curiously. 0:02:01.760 --> 0:02:05.639 You held your stock, I held I look, I needed 0:02:05.640 --> 0:02:08.000 some things, so I sold some, but still I still 0:02:08.040 --> 0:02:08.520 have a lot. 0:02:08.919 --> 0:02:10.880 I have enough, And you're gonna hold on. You're not 0:02:10.880 --> 0:02:13.080 worried about the bubble. I think I am gonna hold on. 0:02:13.200 --> 0:02:16.560 But the questions people have are, what if AI doesn't 0:02:16.600 --> 0:02:20.440 get better infinitely as it scales, what if people invent 0:02:20.600 --> 0:02:23.680 new chips that are far more efficient than the Invidia chips, 0:02:24.320 --> 0:02:28.160 and what if the adoption of AI by other companies 0:02:28.520 --> 0:02:31.079 doesn't give them the results that they hope for financially. 0:02:31.680 --> 0:02:33.680 And so I talked about all of that with somebody 0:02:33.720 --> 0:02:37.160 who knows Nvidia better than anyone else. In fact, he 0:02:37.200 --> 0:02:40.360 literally wrote the book on it, The Thinking Machine, Jensen 0:02:40.440 --> 0:02:44.560 Wang in Vidia and the world's most coveted microchip. He 0:02:44.600 --> 0:02:46.520 actually interviewed Wang six times. 0:02:47.000 --> 0:02:50.359 He's moody, and you know, he has what I would 0:02:50.480 --> 0:02:53.799 describe as somewhat self indulgent performances of anger from time 0:02:53.840 --> 0:02:54.240 to time. 0:02:54.680 --> 0:02:58.079 That's Stephen Witt, and he actually got interested in Nvidia 0:02:58.160 --> 0:03:01.280 shortly after chatchipt took the world by storm. 0:03:01.720 --> 0:03:03.960 I had been using chat GBT and I was like, Wow, 0:03:04.200 --> 0:03:06.440 this thing is amazing. This is like twenty twenty two, 0:03:06.960 --> 0:03:10.359 and I am cooked, Like there's not going to be 0:03:10.520 --> 0:03:12.120 room for me as a writer. This thing can already 0:03:12.120 --> 0:03:14.040 write almost as well as I can, and actually it 0:03:14.040 --> 0:03:15.840 writes better than I did when I was young. 0:03:16.400 --> 0:03:19.760 As Stephen dug around to understand what was powering this technology, 0:03:20.240 --> 0:03:22.800 he got more and more interested in the company building 0:03:22.880 --> 0:03:24.440 its physical infrastructure. 0:03:24.960 --> 0:03:26.679 What brought me to a video was I was trying 0:03:26.680 --> 0:03:28.639 to write about open Aye and them, and it was 0:03:28.680 --> 0:03:30.920 just a crowded there, a million journalists swarming around. I 0:03:30.919 --> 0:03:32.320 was like, there's got to be some other story here. 0:03:32.800 --> 0:03:35.280 And what I've done as a journalist is look for 0:03:35.400 --> 0:03:39.960 big movements of money that aren't being covered. And I 0:03:40.000 --> 0:03:42.480 looked at in Video's stock price, and I in my 0:03:42.560 --> 0:03:44.800 mind they were still the gaming company, and I was like, 0:03:44.840 --> 0:03:46.600 what the hell is going on here? This company's worth 0:03:46.600 --> 0:03:49.600 a trillion dollars. And then as I started to investigate it, 0:03:50.040 --> 0:03:52.400 I was like, Oh, wow, they build all the hardware. 0:03:52.560 --> 0:03:54.520 They built all the hardware that makes this stuff go. 0:03:55.040 --> 0:03:58.400 That's fascinating. And then I kind of learned about Jensen 0:03:58.520 --> 0:03:59.760 and I was like, wait, this company has had the 0:03:59.800 --> 0:04:02.920 same CEO through both the gaming and the AI days. 0:04:02.920 --> 0:04:05.120 And not only that, this is the same as the founder. 0:04:05.200 --> 0:04:06.800 He's been the CEO for thirty years. It's the same 0:04:06.840 --> 0:04:07.520 guy all along. 0:04:08.160 --> 0:04:10.320 So Stephen wrote the book on Nvidia, but he also 0:04:10.320 --> 0:04:12.480 wrote a great New York A piece recently about data 0:04:12.520 --> 0:04:15.240 centers and an essay for The New York Times about 0:04:15.280 --> 0:04:18.719 quote the AI prompt that could end the world. So 0:04:19.279 --> 0:04:22.200 he's really a farm to table thinker, from chips to 0:04:22.360 --> 0:04:25.400 data centers to AI to the apocalypse. It was a 0:04:25.400 --> 0:04:26.120 fun conversation. 0:04:26.839 --> 0:04:28.880 Yeah, as a writer, he seems to sort of be 0:04:28.920 --> 0:04:31.680 at the center of everything in a way that I 0:04:31.720 --> 0:04:34.760 find very compelling, And I actually want to know more 0:04:34.800 --> 0:04:37.280 about the AI prompt that could end the world. 0:04:37.480 --> 0:04:38.600 Well, in that case, you have to listen to the 0:04:38.640 --> 0:04:40.480 whole interview because we talk about it right at the end. 0:04:41.240 --> 0:04:45.080 Okay's the conversation with Stephen Witt. For the layman, what 0:04:45.200 --> 0:04:47.680 is in Vidia and how did it become the most 0:04:47.720 --> 0:04:48.920 valuable company in the world. 0:04:49.440 --> 0:04:52.839 Invidia is basically a hardware designer. They make a special 0:04:52.960 --> 0:04:56.640 kind of microchip called a graphics processing unit, and the 0:04:56.680 --> 0:04:59.880 initial purpose of this thing was to just render graphics 0:05:00.080 --> 0:05:03.480 in video games. So if you were a video gamer, 0:05:03.520 --> 0:05:05.680 you knew who this company was because you would actually 0:05:05.680 --> 0:05:09.320 build your whole PC just around this in video card. 0:05:09.640 --> 0:05:12.480 So this was the engine that rendered the graphics on 0:05:12.520 --> 0:05:15.280 your screen. Sometime around two thousand and four to two 0:05:15.320 --> 0:05:18.520 thousand and five, scientists began to notice how powerful these 0:05:18.560 --> 0:05:21.560 cards were, and they started hacking into the cards, like 0:05:21.640 --> 0:05:25.400 hacking into the circuitry to get to those powerful mathematical 0:05:25.480 --> 0:05:29.680 functions inside the microchip. And Jensen Wong saw this and 0:05:29.720 --> 0:05:31.760 he said, wait, this is a whole new market that 0:05:31.839 --> 0:05:35.480 I can pursue. So he built the software platform that 0:05:35.640 --> 0:05:41.080 turns the graphics card into basically a low budget supercomputer. 0:05:42.080 --> 0:05:45.760 Now you may ask who is this for, Well, it's 0:05:45.800 --> 0:05:49.359 not really for established research scientists because they can usually 0:05:49.400 --> 0:05:53.560 afford time on a conventional supercomputer. It's for scientists who 0:05:53.600 --> 0:05:56.400 are sort of marginalized, who can't afford time on a 0:05:56.440 --> 0:05:58.600 supercomputer and whose research is. 0:05:58.560 --> 0:05:59.240 Out of favor. 0:06:00.320 --> 0:06:05.400 So it's for mad scientists. It's for scientists who are 0:06:05.480 --> 0:06:10.400 pursuing unpopular or weird or kind of offbeat scientific projects. 0:06:10.440 --> 0:06:15.120 But ultimately, the key use case turned out to be AI, 0:06:15.760 --> 0:06:19.479 and specifically a branch of AI that most AI researchers 0:06:19.520 --> 0:06:23.800 thought was crazy called neural network technology. And what you're 0:06:23.800 --> 0:06:26.600 doing here is you're building software that kind of resembles 0:06:26.640 --> 0:06:29.279 the connections in the human brain. It's inspired by the 0:06:29.279 --> 0:06:33.599 biological brain. Actually, you build a bunch of synthetic neurons 0:06:34.000 --> 0:06:36.440 in a little file, and then you train them by 0:06:36.480 --> 0:06:40.320 repeatedly exposing them to training data. So what this could mean, 0:06:40.320 --> 0:06:42.200 for example, if we're trying to build a neural network 0:06:42.240 --> 0:06:45.800 to recognize objects to do computer vision, then we'll show 0:06:45.839 --> 0:06:48.600 it tens of thousands or hundreds of thousands, or ultimately 0:06:48.680 --> 0:06:53.200 millions of images and slowly rewire its neurons until it 0:06:53.200 --> 0:06:56.760 can start to identify things. Now, this had been proposed 0:06:56.760 --> 0:06:58.880 going all the way back to the nineteen forties, but 0:06:59.080 --> 0:07:02.279 nobody had ever able to get it to work. And 0:07:02.320 --> 0:07:06.360 the missing piece, it turns out, is just raw computing power. 0:07:06.720 --> 0:07:09.320 Jeffrey Hinton, who they call him the godfather of AI, 0:07:09.600 --> 0:07:11.400 he said, you know, the point we never thought to 0:07:11.520 --> 0:07:13.880 ask was what if we just made it go a 0:07:14.000 --> 0:07:18.000 million times faster? And that's what in Nvidia's hardware did. 0:07:18.000 --> 0:07:20.920 It made AI and these neural networks in particular, train 0:07:21.080 --> 0:07:23.400 and learn a million times faster. 0:07:23.720 --> 0:07:26.440 We actually had Hinton on the podcast earlier this year. 0:07:26.760 --> 0:07:29.520 Was he an early one of these mad scientists whose 0:07:29.520 --> 0:07:33.080 research was unpopular and therefore started buying in video chips. 0:07:33.200 --> 0:07:35.640 Yes, very much so. And there was a community of 0:07:35.640 --> 0:07:37.800 these guys. It wasn't just him. There were a number 0:07:37.800 --> 0:07:39.600 of other people doing it, most of whose work has 0:07:39.600 --> 0:07:42.440 now been recognized. But they were very much on the 0:07:42.480 --> 0:07:45.559 margins of computer science. They couldn't get five thousand dollars 0:07:45.560 --> 0:07:48.720 in research funding, but they could get enough money to 0:07:48.760 --> 0:07:53.720 afford two five hundred dollars in video retail graphics gaming cards, 0:07:53.800 --> 0:07:56.800 which they did, and Hinton had a graduate student named 0:07:56.800 --> 0:08:00.320 Alex Krzewski who was just an ACE programmer, and he 0:08:00.560 --> 0:08:02.800 turned the neural net that he ran on these cards 0:08:03.080 --> 0:08:06.640 into something called alex net, which then started to recognize 0:08:06.680 --> 0:08:10.240 images better than any AI had ever done before. Like 0:08:10.320 --> 0:08:13.880 it smashed the paradigm. And so that engineered around twenty 0:08:13.880 --> 0:08:16.960 twelve or twenty thirteen a paradigm shift in AI. And 0:08:17.080 --> 0:08:21.040 since then everything that has happened has been a repeated 0:08:21.080 --> 0:08:24.560 application of the thing. Alex discovered that if you took 0:08:24.600 --> 0:08:28.160 neural nets, if you ran them on Nvidia technology, you 0:08:28.200 --> 0:08:30.160 would have a very powerful result. 0:08:30.680 --> 0:08:33.640 So this is all fascinating, but to some it may 0:08:33.720 --> 0:08:37.439 sound a big geeky and like inside Baseball, which is 0:08:37.480 --> 0:08:39.959 why I was very attracted to a quote in your 0:08:40.000 --> 0:08:43.360 recent ne Yoka piece which said, if Americans want to 0:08:43.360 --> 0:08:47.880 retire comfortably, in Vidia has to succeed. Yes, And I 0:08:47.960 --> 0:08:50.120 was curious what your conversations with the edits around that, 0:08:50.360 --> 0:08:50.719 around that. 0:08:50.880 --> 0:08:53.640 Oh no, that's that's straightforward. That one actually sailed right 0:08:53.679 --> 0:08:56.200 through fact checking. There were no questions what has happened 0:08:56.240 --> 0:08:58.680 since Alex invented this thing in his bedroom. Is that 0:08:58.720 --> 0:09:01.520 we scaled it up from two graphics cards to two 0:09:01.640 --> 0:09:04.480 hundred thousand or more, and we have plans to scale 0:09:04.520 --> 0:09:06.920 them up to two or two million than twenty million, right, 0:09:07.440 --> 0:09:09.920 So this is the data center boom which we're going 0:09:09.920 --> 0:09:14.000 through right now. It's a new industrial revolution. We're basically 0:09:14.040 --> 0:09:18.480 building these giant barns full of in Vidia microchips to 0:09:18.679 --> 0:09:22.600 run calculations to build better AI twenty four to seven 0:09:22.640 --> 0:09:25.240 around the clock, and it's one of the largest deployments 0:09:25.280 --> 0:09:29.280 of capital in human history. This has made in Vidia 0:09:29.640 --> 0:09:34.280 the most valuable company in the world, and it has 0:09:34.360 --> 0:09:38.280 created a situation where in Vidia stock is more concentrated 0:09:38.480 --> 0:09:40.480 in the S and P five hundred than any stock 0:09:40.559 --> 0:09:43.920 since they started keeping track. And actually Microsoft, who is 0:09:43.960 --> 0:09:47.800 the second biggest, has that valuation largely because they're building 0:09:47.840 --> 0:09:49.840 these sheds and renting out in video equipment. So that's 0:09:49.840 --> 0:09:53.600 linked to So think about that fifteen percent of the 0:09:53.640 --> 0:09:57.120 stock market is these two stocks, right, they have to succeed. 0:09:57.600 --> 0:10:01.880 Americans in particular, are usually invested massively through index funds 0:10:02.160 --> 0:10:04.280 and something that looks exactly like the S and P 0:10:04.400 --> 0:10:08.120 five hundred, so you know, if in Vidio crashes, it's 0:10:08.160 --> 0:10:10.240 going to create a lot of paint throughout the economy. 0:10:10.480 --> 0:10:12.320 I want to talk about the data centers, but before that, 0:10:12.360 --> 0:10:14.760 I want to talk about the man who founded in 0:10:14.800 --> 0:10:17.920 Nvidia and is a CEO today at Jensen Huang. He 0:10:17.920 --> 0:10:20.440 seems to pop up everywhere, but he also seems to 0:10:20.480 --> 0:10:23.800 be more inscrutable. I mean, who is he and do 0:10:23.800 --> 0:10:26.360 you see him as different from Zuckerberg and Oltman and 0:10:26.400 --> 0:10:28.000 Bezos in some significant way? 0:10:28.640 --> 0:10:32.400 Jensen most resembles of all executives Elon Musk because he 0:10:32.480 --> 0:10:36.040 is an engineering wizard. Bezos is smart as hell, and 0:10:36.040 --> 0:10:39.040 so Zuckerberg. But ultimately they're kind of software guides. You know, 0:10:39.080 --> 0:10:41.400 they're coming at the computer from the keyboard in the terminal. 0:10:41.640 --> 0:10:45.040 Jensen is totally different. He approaches computing from the circuit up. 0:10:45.400 --> 0:10:47.960 He's a degree not in computer science originally, but actually 0:10:47.960 --> 0:10:52.319 in electrical engineering. Okay, so for Jensen, the computer is 0:10:52.320 --> 0:10:57.000 a piece of hardware that runs calculations in a microchip, 0:10:57.040 --> 0:11:00.360 and he literally designed those microchips on paper at the 0:11:00.440 --> 0:11:02.640 beginning of his career, and that's all he's ever done. 0:11:03.640 --> 0:11:05.160 And this is a little bit why even though he 0:11:05.240 --> 0:11:06.959 runs the most valuable company in the world. It's a 0:11:06.960 --> 0:11:10.800 little baffling to to kind of people. Nothing Jensen makes 0:11:10.840 --> 0:11:14.120 is really that accessible. It's all deep inside the computer. 0:11:14.400 --> 0:11:15.959 As a quote in your piece that I liked what 0:11:16.120 --> 0:11:19.160 he said, I find that I'm best when I'm under adversity. 0:11:19.679 --> 0:11:22.439 My heart rate actually goes down. Anyone who's dealt with 0:11:22.520 --> 0:11:24.640 RUSSIAU in a restaurant knows what I'm talking about. 0:11:25.040 --> 0:11:28.040 Yeah, yeah, I mean he started out at Denny's, so 0:11:28.120 --> 0:11:30.720 his first job was basically I think he was a 0:11:30.720 --> 0:11:33.480 bus boy at first, and then graduated to dishwasher and 0:11:33.520 --> 0:11:36.440 ultimately became a server. And I was talking to someone 0:11:36.440 --> 0:11:38.800 in the company and she's like, you know what, Jensen 0:11:38.880 --> 0:11:42.040 is actually a lot calmer and more compassionate with his 0:11:42.080 --> 0:11:45.520 employees when things are going wrong. It's when the company 0:11:45.600 --> 0:11:47.400 stock price is way up but it looks like everything's 0:11:47.400 --> 0:11:50.400 going great that he really becomes much more cruel, like 0:11:50.480 --> 0:11:54.200 much much meaner to everybody. So he is actually, in 0:11:54.240 --> 0:11:57.079 some ways a nicer person when things are going wrong. 0:11:57.120 --> 0:11:58.720 When he succeeds, it makes him nervous. 0:11:59.080 --> 0:12:01.760 One of his colleagues describe working with him as kind 0:12:01.760 --> 0:12:05.760 of like sticking your finger in the electric socket. That's 0:12:05.840 --> 0:12:06.600 quite the metaphor. 0:12:06.960 --> 0:12:09.679 It's one hundred percent accurate. I mean I've interacted with Jensen. 0:12:09.720 --> 0:12:11.600 It is like sticking your finger in the electric socket. 0:12:11.640 --> 0:12:15.960 He's so tightly wound. He expects so much to happen 0:12:16.400 --> 0:12:19.640 in every conversation. Just to even start talking to him, 0:12:19.640 --> 0:12:21.120 you have to be totally up to speed. He's not 0:12:21.160 --> 0:12:23.000 gonna waste any time. He's not going to suffer fools. 0:12:23.520 --> 0:12:26.160 And he's also really intense and unpredictable, and you just 0:12:26.200 --> 0:12:28.079 don't know where he's going to go in any conversation. 0:12:28.640 --> 0:12:31.720 And you know, he has what I would describe as 0:12:31.760 --> 0:12:34.840 somewhat self indulgent performances of anger from time to time. 0:12:35.040 --> 0:12:37.199 And that's especially true if you're one of his executives. 0:12:37.360 --> 0:12:39.680 If you're not delivering, he's going to stand you up 0:12:39.720 --> 0:12:41.240 in front of an audience of people and just start 0:12:41.400 --> 0:12:45.679 screaming at you. But really I mean yelling, and it's 0:12:45.679 --> 0:12:47.559 not fun, and he will humiliate you in front of 0:12:47.559 --> 0:12:50.000 an audience. I think people at Nvidia have to develop 0:12:50.080 --> 0:12:52.240 very thick skins. He actually did this to me at 0:12:52.240 --> 0:12:56.960 one point, so I kind of know exactly. Oh yeah, yeah, Well, 0:12:57.200 --> 0:12:59.840 I kept asking him about the future. Jensen does not 0:12:59.920 --> 0:13:03.520 like to speculate. He doesn't have actually a science fiction 0:13:03.640 --> 0:13:05.080 vision of what the future is going to look like. 0:13:05.559 --> 0:13:08.679 He has a data driven vision from engineering principles of 0:13:08.720 --> 0:13:10.920 where he thinks technology is going to go. But if 0:13:10.920 --> 0:13:13.679 he can't see beyond that, he won't speculate. But I 0:13:14.240 --> 0:13:16.040 noticed that other people at this firm would talk about it, 0:13:16.080 --> 0:13:19.319 and I really wanted to get into his imagination. I 0:13:19.320 --> 0:13:21.240 guess I would say of where he thinks all this 0:13:21.320 --> 0:13:23.840 can go. So I presented him with a clip from 0:13:23.960 --> 0:13:26.760 Arthur C. Clark discussing the future of computers, and this 0:13:26.800 --> 0:13:29.040 is back from nineteen sixty four, but it was kind 0:13:29.040 --> 0:13:31.640 of anticipating the current reality we were in where we 0:13:31.679 --> 0:13:34.679 would start training mechanical brains and those brains would train 0:13:34.760 --> 0:13:38.679 faster than biological brains and eventually would supersede biological brains. 0:13:38.679 --> 0:13:40.080 And so I'd show this clip to some other people 0:13:40.080 --> 0:13:42.320 in the video, and they've gotten very They kind of 0:13:42.360 --> 0:13:45.800 like smelled up and started giving these grand soliloquies about 0:13:45.800 --> 0:13:47.880 the future that were like very beautiful and articulate, And 0:13:48.120 --> 0:13:51.480 I was hoping to get that response from Jensen. Instead, 0:13:51.520 --> 0:13:53.720 he just starts screaming at me, how about how stupid 0:13:53.720 --> 0:13:55.720 the clip was, how he didn't give a shit about 0:13:55.800 --> 0:13:57.360 Arthur C. Clarke, He never read one of his books, 0:13:57.400 --> 0:13:59.640 he didn't read science fiction, and he thought the whole 0:13:59.679 --> 0:14:02.440 line of questioning was pedestrian, and that I was letting 0:14:02.520 --> 0:14:05.040 him down by asking I was wasting his time. Despite 0:14:05.080 --> 0:14:07.960 having written his biography, Jensen remains a little bit of 0:14:08.000 --> 0:14:10.160 a puzzle, and just that I cannot tell you what's 0:14:10.160 --> 0:14:12.840 going on inside his brain. So well, but I will 0:14:12.840 --> 0:14:17.720 say this, he's extremely neurotic, by which I mean, I 0:14:17.720 --> 0:14:19.480 don't even mean this in a clinical sense. I just 0:14:19.480 --> 0:14:21.680 mean that, by his own admission, he's totally driven by 0:14:22.080 --> 0:14:25.880 negative emotions. So even though he's on top of the world, 0:14:26.320 --> 0:14:28.800 I think his mind is telling him constantly, you're going 0:14:28.880 --> 0:14:31.600 to fail. This is a temporary thing, and Vidio is 0:14:31.600 --> 0:14:34.000 going to go back down again, you know, twice in 0:14:34.040 --> 0:14:37.160 his tenure as CEO, in Vidia's stock price has retreated 0:14:37.160 --> 0:14:38.280 by almost ninety percent. 0:14:38.560 --> 0:14:40.520 What could make it happen now? What keeps me up 0:14:40.560 --> 0:14:42.520 at night today? What could happen today? 0:14:42.520 --> 0:14:45.320 Anything? Any number of things. This would not be comprehensive, 0:14:45.320 --> 0:14:48.400 But there's three big risks. The first is just competition, 0:14:49.080 --> 0:14:52.960 and Vidia is making so much money and everyone's seeing that, 0:14:53.080 --> 0:14:55.840 and this attracts competition in the same maner that chum 0:14:55.920 --> 0:14:58.840 attracts sharks. Right, it's like throwing blood in the water 0:14:58.920 --> 0:15:01.680 for other microchip design to earn a seventy percent eighty 0:15:01.680 --> 0:15:03.880 percent gross margin, which is what they do on. 0:15:03.760 --> 0:15:04.560 Some of these chips. 0:15:04.960 --> 0:15:07.920 So Google has built a whole alternative stack for AI 0:15:08.000 --> 0:15:10.320 computing around their own, their own kind of platform, and 0:15:10.320 --> 0:15:12.360 they're starting to lease that out to new customers. That's 0:15:12.400 --> 0:15:15.160 a big risk. There's a big risk that Chinese companies 0:15:15.320 --> 0:15:19.840 built alternative, cheaper stacks to what Nvidia does. Intel had 0:15:20.040 --> 0:15:22.600 ninety ninety five percent of the CPU market at one 0:15:22.600 --> 0:15:26.560 point in this country. Now they're falling apart. Conquering one 0:15:27.000 --> 0:15:30.440 cycle in microchips is no guarantee that you will conquer 0:15:30.480 --> 0:15:33.920 the next one, and history demonstrates that quite clearly, so 0:15:33.960 --> 0:15:38.640 that could happen. B Basically, what happens in the data 0:15:38.680 --> 0:15:42.880 center is we're doing a mathematical operation called a matrix multiplication, 0:15:43.680 --> 0:15:48.400 and it's extremely computationally expensive to do this. So without 0:15:48.440 --> 0:15:51.640 getting two technical basically to train an AI right now, 0:15:52.000 --> 0:15:57.160 we have to do ten trillion trillion individual computations, which 0:15:57.200 --> 0:16:01.400 is more than the number of observable stars in the universe. However, 0:16:01.760 --> 0:16:05.120 maybe it's possible that we find some more efficient way 0:16:05.160 --> 0:16:07.640 of doing that. Maybe there's a way that requires only 0:16:07.920 --> 0:16:10.800 ten billion trillion or even ten hundred trillion, right, and 0:16:10.920 --> 0:16:13.120 video stock price would go down because we wouldn't have 0:16:13.160 --> 0:16:14.680 to build so many data centers. Right, we'd have a 0:16:14.680 --> 0:16:17.200 more efficient training solution. All of this is a more 0:16:17.240 --> 0:16:20.600 complex way of saying, maybe there's a technological solution where 0:16:20.600 --> 0:16:23.120 we you know, right now, we're brute forcing our way 0:16:23.160 --> 0:16:26.320 to AI. It's a heavy industrial problem. We're talking about 0:16:26.400 --> 0:16:30.480 building nuclear power plants to bring these things online. I 0:16:31.120 --> 0:16:34.920 think maybe it's possible that there's a technological solution that 0:16:35.000 --> 0:16:37.920 trains these things faster, and if we discovered it, we 0:16:37.960 --> 0:16:40.080 wouldn't have to buy so many in video microchips. That 0:16:40.080 --> 0:16:42.960 would also make their stock price got down. But the 0:16:43.000 --> 0:16:47.440 third thing is basically right now, for the last thirteen 0:16:47.520 --> 0:16:52.280 or fourteen years, the more microchips we stuff into the barn. Okay, 0:16:52.320 --> 0:16:55.440 the more microchips we throw at this problem, the better. 0:16:55.160 --> 0:16:58.840 AI we GND. This is the scaling law law in quotes. 0:16:59.320 --> 0:17:01.960 Okay, is not a law in the universe that this 0:17:02.040 --> 0:17:05.880 has to happen. It's not some immutable, physical proven thing 0:17:06.320 --> 0:17:09.160 from first principles of physics that the more microchips we have, 0:17:09.480 --> 0:17:11.520 the better AI we have. In fact, no one is 0:17:11.640 --> 0:17:16.640 entirely sure why this works. Presumably, like most other forces 0:17:16.680 --> 0:17:20.159 in the universe, this will hit some kind of escurve. 0:17:20.160 --> 0:17:22.840 It'll start to plateau or level off at some point. 0:17:23.080 --> 0:17:25.760 We're not there yet. But if we did hit a plateau, 0:17:26.320 --> 0:17:30.239 if stuffing more microchips into the barn only resulted in 0:17:30.520 --> 0:17:33.720 marginally better AI or didn't improve it at all, I 0:17:33.720 --> 0:17:35.560 think in video stock price will go down a lat 0:17:35.920 --> 0:17:38.040 and I think it would look make this whole era 0:17:38.119 --> 0:17:40.080 look kind of like a bubble if that were to happen. 0:17:40.600 --> 0:17:43.320 Now, is this why in Nvidia has kind of become 0:17:44.000 --> 0:17:47.400 the bank of the air evolution In a sense, they're 0:17:48.200 --> 0:17:51.760 wanting to lend money and lock other companies into the 0:17:51.800 --> 0:17:56.360 current paradigm of AI, maybe even hoping to defensively prevent 0:17:56.520 --> 0:18:00.399 other more economical approaches from emerging and consolidating videos position. 0:18:00.440 --> 0:18:01.800 I mean, how much of a chess game is this 0:18:01.880 --> 0:18:03.879 in terms of thinking about the future of computing. For 0:18:03.960 --> 0:18:04.359 Jensen and. 0:18:04.400 --> 0:18:08.119 Others, Oh yeah, it's chess, and Jensen is an expert 0:18:08.200 --> 0:18:10.720 chess player. At this kind of chess, He's really good 0:18:10.800 --> 0:18:14.359 at thinking about the competitive positioning of where he is 0:18:14.400 --> 0:18:16.840 and where other people are. You know, in Vidiot's early days, 0:18:16.920 --> 0:18:19.600 the GPU market, back in the video game days was 0:18:19.640 --> 0:18:22.080 really crowded. At one point there were fifty or sixty 0:18:22.080 --> 0:18:24.959 participants in this market. I talked to David Kirk, who 0:18:25.040 --> 0:18:28.240 was the chief scientist in VIDIA during this time. Jensen 0:18:28.240 --> 0:18:30.359 would go into his office and a whiteboard and you 0:18:30.359 --> 0:18:32.960 would have a list of all his competitors up there. 0:18:33.680 --> 0:18:35.119 And not only that, they would have a list of 0:18:35.240 --> 0:18:39.600 who the best engineers working at those competitors were, and 0:18:39.640 --> 0:18:42.280 then they would come up with plans to poach those 0:18:42.320 --> 0:18:45.040 engineers and get them to come work for a VIDIA 0:18:45.280 --> 0:18:47.320 so that they would drain the brain power of their 0:18:47.320 --> 0:18:51.520 competitors and force them to collapse. I've compared the early 0:18:51.560 --> 0:18:54.000 graphics days to the movie Battle Royale where all the 0:18:54.040 --> 0:18:55.240 kids are on the island and they have to kill 0:18:55.240 --> 0:18:57.359 each other. It was like that there were like forty 0:18:57.359 --> 0:18:59.119 competitors and only one could survive. 0:19:00.000 --> 0:19:00.560 Here's one. 0:19:01.320 --> 0:19:03.719 He won the Battle Royale. He was the last guy standing. 0:19:03.760 --> 0:19:05.960 I mean he won the knife fight. So he is 0:19:06.200 --> 0:19:11.879 unbelievably ruthless and unbelievably good at identifying where the competition 0:19:12.040 --> 0:19:14.160 is and what he could do, not just to beat 0:19:14.160 --> 0:19:16.119 them in the marketplace, but actually to hollow out their 0:19:16.160 --> 0:19:17.000 engineering talent. 0:19:18.320 --> 0:19:22.000 Who are in Video's biggest customers and what are they 0:19:22.040 --> 0:19:23.080 buying the chips for. 0:19:23.880 --> 0:19:27.160 Okay, it's a bit complex. The biggest customers, they don't 0:19:27.160 --> 0:19:31.440 disclose it. Almost certainly it's it's Microsoft and then probably Amazon. 0:19:31.760 --> 0:19:34.720 What these companies do is they train some AI on 0:19:34.760 --> 0:19:37.040 their own, but what they're really doing is putting. They're 0:19:37.040 --> 0:19:39.200 the ones building the sheds, they're the ones building the 0:19:39.280 --> 0:19:42.040 data centers. So in Video sells them the micro chips, 0:19:42.080 --> 0:19:44.840 and then kind of the ultimate end user is a 0:19:44.880 --> 0:19:48.760 frontier AI lab, so that could be something like Anthropic 0:19:49.000 --> 0:19:51.920 or open AI. So essentially the way to think about 0:19:51.960 --> 0:19:55.280 this is in Vidia sells the microchips to Microsoft or 0:19:55.320 --> 0:19:59.720 Amazon or maybe Oracle. Oracle builds and operates a gigantic 0:19:59.800 --> 0:20:02.480 day center with one hundred thousand microchips in it that 0:20:02.520 --> 0:20:05.440 takes as much power as like a small city, and 0:20:05.480 --> 0:20:09.160 then clients like open ai come and lease it out from. 0:20:08.960 --> 0:20:19.320 Them after the break Why data centers are worried about 0:20:19.320 --> 0:20:40.320 break ins? Stay with us. Let's talk about data centers. 0:20:40.359 --> 0:20:43.800 There's something weird about data centers because on the one hand, 0:20:44.080 --> 0:20:46.760