WEBVTT - The Story: Will NVIDIA Save or Ruin The World? 0:00:13.800 --> 0:00:16.960 Welcome to tech stuff. I'm as Vloshan here with Cara Price. 0:00:17.200 --> 0:00:21.200 Hey Kara, Hi, as So years ago, around the time 0:00:21.280 --> 0:00:24.800 that we were reporting our first podcast together on the 0:00:24.840 --> 0:00:26.400 forthcoming AI. 0:00:26.480 --> 0:00:29.200 Revolution around no longer forthcoming. 0:00:30.160 --> 0:00:34.360 You invested in Nvidia, which is up over one hundred 0:00:34.600 --> 0:00:37.159 x since then. Congratulations. 0:00:37.560 --> 0:00:40.120 You know, I just felt when we reported on it 0:00:40.159 --> 0:00:42.000 that it was going to be the future, and so 0:00:42.159 --> 0:00:45.639 I did invest in it, and you know, very happily. 0:00:45.720 --> 0:00:47.440 So now, what was it? 0:00:47.479 --> 0:00:49.480 Was it something that tipped you over the edge back 0:00:49.520 --> 0:00:50.319 then to do it? 0:00:50.920 --> 0:00:52.879 Well, I was just sort of thinking to myself, this 0:00:52.920 --> 0:00:55.880 is the thing that's going to power everything. So obviously 0:00:55.920 --> 0:00:59.200 it's something that people are going to be paying attention 0:00:59.240 --> 0:01:01.080 to and investors are going to be paying attention to, 0:01:01.120 --> 0:01:03.640 and it just made a lot of sense to me 0:01:03.680 --> 0:01:04.120 at the time. 0:01:04.760 --> 0:01:08.399 We know Warren Buffett is retiring this year, there's a 0:01:08.440 --> 0:01:13.560 slot open. In Vidia is, of course the most valuable 0:01:13.560 --> 0:01:16.560 company in the world today, recently topping five trillion dollars 0:01:16.600 --> 0:01:18.920 in value, and of course a lot of people are 0:01:18.920 --> 0:01:22.000 crying bubble, not just for Nvidia but for the AI 0:01:22.080 --> 0:01:25.639 industry as a whole. Actually carrying curiousy you held your stock. 0:01:25.920 --> 0:01:29.360 I held I look, I needed some things, so I 0:01:29.400 --> 0:01:31.559 sold some, but I still have a lot. 0:01:31.959 --> 0:01:33.840 I have enough, and you're going to hold on. You're 0:01:33.840 --> 0:01:34.760 not worried about the bubble. 0:01:34.880 --> 0:01:36.120 I think I am going to hold on. 0:01:36.240 --> 0:01:39.600 But the questions people have are, what if AI doesn't 0:01:39.600 --> 0:01:43.480 get better infinitely as it scales, what if people invent 0:01:43.640 --> 0:01:45.840 new chips that are far more efficient than the in 0:01:45.880 --> 0:01:50.120 Nvidia chips, And what if the adoption of AI by 0:01:50.320 --> 0:01:53.200 other companies doesn't give them the results so they hope 0:01:53.200 --> 0:01:56.240 for financially. And so I talked about all of that 0:01:56.280 --> 0:01:59.840 with somebody who knows Nvidia better than anyone else. In fact, 0:02:00.120 --> 0:02:02.720 he didn't really wrote the book on it, The Thinking Machine, 0:02:03.040 --> 0:02:07.200 Jensen Wang in VideA and the world's most coveted microchip. 0:02:07.520 --> 0:02:09.560 He actually interviewed Wang six times. 0:02:10.040 --> 0:02:13.480 He's moody, and you know, he has what I would 0:02:13.520 --> 0:02:16.839 describe as somewhat self indulgent performances of anger from time 0:02:16.880 --> 0:02:17.240 to time. 0:02:17.720 --> 0:02:20.880 That's Stephen Witt, and he actually got interested in Nvidia 0:02:21.200 --> 0:02:24.320 shortly after chatchipt took the world by storm. 0:02:24.720 --> 0:02:26.960 I had been using CHATGPT and I was like, Wow, 0:02:27.200 --> 0:02:29.400 this thing is amazing. This is like twenty twenty two, 0:02:29.960 --> 0:02:33.400 and I am cooked like, there's not going to be 0:02:33.560 --> 0:02:35.160 room for me as a writer. This thing can already 0:02:35.160 --> 0:02:37.360 write almost as well as I can, and actually writes 0:02:37.400 --> 0:02:38.840 better than I did when I was young. 0:02:39.440 --> 0:02:42.799 As Stephen dug around to understand what was powering this technology, 0:02:43.240 --> 0:02:45.840 he got more and more interested in the company building 0:02:45.919 --> 0:02:47.480 its physical infrastructure. 0:02:47.960 --> 0:02:49.680 What brought me to a video was I was trying 0:02:49.720 --> 0:02:51.679 to write about Open Eye and them, and it was 0:02:51.720 --> 0:02:53.880 just a crowd of the romillion journalists swarming around. I 0:02:53.960 --> 0:02:55.880 was like, there's gotta be some other story here. And 0:02:56.000 --> 0:02:58.639 what I've done as a journalist is look for big 0:02:58.720 --> 0:03:03.280 movements of money that aren't being covered. And I looked 0:03:03.400 --> 0:03:05.880 at Nvidia's stock price and I in my mind they 0:03:05.919 --> 0:03:07.000 were still the gaming company. 0:03:07.520 --> 0:03:09.000 I was like, what the hell is going on here? 0:03:09.040 --> 0:03:10.560 The company's worth a trillion dollars. 0:03:10.960 --> 0:03:14.040 And then as I started to investigate it, I was like, Oh, wow, 0:03:14.240 --> 0:03:16.560 they build all the hardware. They built all the hardware 0:03:16.560 --> 0:03:20.239 that makes this stuff go. That's fascinating. And then I 0:03:20.360 --> 0:03:22.079 kind of learned about Jensen and I was like, wait, 0:03:22.120 --> 0:03:24.440 this company has had the same CEO through both the 0:03:24.560 --> 0:03:27.000 gaming and the AI days and not only that this 0:03:27.120 --> 0:03:28.560 is the same This is the founder, he's been the 0:03:28.600 --> 0:03:29.400 CEO for thirty years. 0:03:29.440 --> 0:03:30.480 It's the same guy all along. 0:03:31.200 --> 0:03:33.320 So Stephen wrote the book on Nvidia, but he also 0:03:33.360 --> 0:03:35.960 wrote a great in New yorka piece recently about data centers, 0:03:36.560 --> 0:03:38.640 and an essay for The New York Times about quote 0:03:38.960 --> 0:03:42.480 the AI prompt that could end the world. So he's 0:03:42.520 --> 0:03:45.640 really a farm to table thinker, from chips to data 0:03:45.720 --> 0:03:49.080 centers to AI to the apocalypse. It was a fun conversation. 0:03:49.880 --> 0:03:51.880 Yeah, as a writer, he seems to sort of be 0:03:51.960 --> 0:03:54.640 at the center of everything in a way that I 0:03:54.720 --> 0:03:57.800 find very compelling, And I actually want to know more 0:03:57.800 --> 0:04:00.000 about the AI prompt that could end the world. 0:04:00.640 --> 0:04:01.840 In that case, you have to listen to the whole 0:04:01.880 --> 0:04:04.520 interview because we talk about it right at the end. Okay, 0:04:04.520 --> 0:04:08.000 it's the conversation with Stephen Witt for then Layman, what 0:04:08.240 --> 0:04:10.680 is in Vidia and how did it become the most 0:04:10.760 --> 0:04:11.839 valuable company in the world. 0:04:12.480 --> 0:04:15.480 In Vidia is basically a hardware designer. They make a 0:04:15.560 --> 0:04:19.480 special kind of microchip called a graphics processing unit, and 0:04:19.560 --> 0:04:22.239 the initial purpose of this thing was to just render 0:04:22.400 --> 0:04:26.440 graphics in video games. So if you were a video gamer, 0:04:26.520 --> 0:04:28.640 you knew who this company was because you would actually 0:04:28.720 --> 0:04:32.279 build your whole PC just around this in Vidia card, 0:04:32.680 --> 0:04:35.440 so this was the engine that rendered the graphics on 0:04:35.560 --> 0:04:38.640 your screen. Sometime around two thousand and four two thousand 0:04:38.640 --> 0:04:42.120 and five, scientists began to notice how powerful these cards were, 0:04:42.520 --> 0:04:45.200 and they started hacking into the cards, like hacking into 0:04:45.200 --> 0:04:49.400 the circuitry to get to those powerful mathematical functions inside 0:04:49.600 --> 0:04:53.400 the microchip. And Jensen Wong saw this and he said, wait, 0:04:53.520 --> 0:04:55.480 this is a whole new market that I can pursue. 0:04:56.279 --> 0:04:57.200 So he built the. 0:04:57.279 --> 0:05:01.840 Software platform that turns the graphics card into basically a 0:05:02.120 --> 0:05:08.440 low budget supercomputer. Now you may ask who is this for, Well, 0:05:08.600 --> 0:05:11.880 it's not really for established research scientists because they can 0:05:12.000 --> 0:05:16.400 usually afford time on a conventional supercomputer. It's for scientists 0:05:16.440 --> 0:05:19.280 who are sort of marginalized, who can't afford time on 0:05:19.360 --> 0:05:23.400 a supercomputer and whose research is out of favor, So 0:05:23.800 --> 0:05:28.880 it's for mad scientists. It's for scientists who are pursuing 0:05:29.640 --> 0:05:34.359 unpopular or weird or kind of offbeat scientific projects. But ultimately, 0:05:34.640 --> 0:05:38.800 the key use case turned out to be AI, and 0:05:38.960 --> 0:05:42.760 specifically a branch of AI that most AI researchers thought 0:05:43.040 --> 0:05:47.000 was crazy, called neural network technology. And what you're doing 0:05:47.040 --> 0:05:49.720 here is you're building software that kind of resembles the 0:05:49.800 --> 0:05:53.760 connections in the human brain. It's inspired by the biological brain. Actually, 0:05:54.279 --> 0:05:57.839 you build a bunch of synthetic neurons in a little file, 0:05:57.920 --> 0:06:01.320 and then you train them by repeat exposing them to 0:06:01.440 --> 0:06:03.880 training data. So what this could mean, for example, if 0:06:03.880 --> 0:06:06.520 we're trying to build a neural network to recognize objects 0:06:06.640 --> 0:06:09.560 to do computer vision, then we'll show it tens of 0:06:09.600 --> 0:06:13.159 thousands or hundreds of thousands, or ultimately millions of images. 0:06:13.120 --> 0:06:15.760 And slowly rewire. 0:06:15.160 --> 0:06:18.640 Its neurons until it can start to identify things. Now, 0:06:18.800 --> 0:06:20.800 this had been proposed going all the way back to 0:06:20.880 --> 0:06:23.480 the nineteen forties, but nobody had ever been able to 0:06:23.520 --> 0:06:27.159 get it to work. And the missing piece, it turns out, 0:06:27.640 --> 0:06:30.919 is just raw computing power. Jeffrey Hinton, who they call 0:06:30.960 --> 0:06:33.280 him the Godfather of AI, he said, you know, the 0:06:33.360 --> 0:06:36.120 point we never thought to ask was what if we 0:06:36.240 --> 0:06:39.800 just made it go a million times faster? And that's 0:06:39.839 --> 0:06:42.560 what Nvidia's hardware did. It made AI and these neural 0:06:42.600 --> 0:06:46.400 networks in particular, train and learn a million times faster. 0:06:46.720 --> 0:06:49.360 We actually had Hinton on the podcast earlier this year. 0:06:49.800 --> 0:06:52.520 Was he in early one of these mad scientists whose 0:06:52.560 --> 0:06:56.039 research was unpopular and therefore started buying in video chips. 0:06:56.240 --> 0:06:57.840 Yes, very much so, and there were few. 0:06:57.880 --> 0:06:59.080 There was a community of these guys. 0:06:59.160 --> 0:06:59.880 It wasn't just him. 0:07:00.279 --> 0:07:01.840 There were a number of other people doing it, most 0:07:01.880 --> 0:07:04.560 of whose work has now been recognized. But they were 0:07:04.680 --> 0:07:07.640 very much on the margins of computer science. They couldn't 0:07:07.640 --> 0:07:10.360 get five thousand dollars in research funding, but they could 0:07:10.400 --> 0:07:14.400 get enough money to afford two five hundred dollars in 0:07:14.520 --> 0:07:18.640 video retail graphics gaming cards, which they did. And Hinton 0:07:18.680 --> 0:07:21.080 had a graduate student named Alex Krzewski who was just 0:07:21.120 --> 0:07:24.720 an ACE programmer, and he turned the neural net that 0:07:24.760 --> 0:07:27.440 he ran on these cards into something called alex net, 0:07:27.760 --> 0:07:31.520 which then started to recognize images better than any AI 0:07:31.760 --> 0:07:35.600 had ever done before. Like it smashed the paradigm, and 0:07:35.680 --> 0:07:38.160 so that engineered around twenty twelve or twenty thirteen a 0:07:38.240 --> 0:07:42.360 paradigm shift in AI, and since then everything that has 0:07:42.480 --> 0:07:45.480 happened has been a repeated application of the thing. Alex 0:07:45.680 --> 0:07:49.440 discovered that if you took neural nets, if you ran 0:07:49.560 --> 0:07:53.160 them on Nvidia technology, you would have a very powerful result. 0:07:53.680 --> 0:07:56.600 So this is all fascinating, but to some it may 0:07:56.760 --> 0:08:00.480 sound a big geeky and like inside baseball, which is 0:08:00.520 --> 0:08:02.960 why I was very attracted to a quote in your 0:08:03.000 --> 0:08:07.440 recent Yoka piece which said, if Americas want to retire comfortably, 0:08:07.960 --> 0:08:11.480 in Vidia has to succeed. Yes, And I'm curious what 0:08:11.560 --> 0:08:13.080 your conversations with the edits around that. 0:08:13.400 --> 0:08:13.720 Around that. 0:08:13.920 --> 0:08:16.680 Oh no, that's that's straightforward. That one actually sailed right 0:08:16.680 --> 0:08:17.280 through fact checking. 0:08:17.280 --> 0:08:18.040 There were no questions. 0:08:18.240 --> 0:08:20.840 What has happened since Alex invented this thing in his 0:08:20.920 --> 0:08:23.880 bedroom is that we scaled it up from two graphics 0:08:23.960 --> 0:08:26.760 cards to two hundred thousand or more, and we have 0:08:26.840 --> 0:08:28.800 plans to scale them up to two or two million 0:08:28.840 --> 0:08:31.360 and then twenty million. Right So this is the data 0:08:31.440 --> 0:08:34.040 center boom which we're going through right now. It's a 0:08:34.080 --> 0:08:39.920 new industrial revolution. We're basically building these giant barns full 0:08:40.080 --> 0:08:44.120 of in Vidia microchips to run calculations to build better 0:08:44.240 --> 0:08:46.880 AI twenty four to seven around the clock, and it's 0:08:46.880 --> 0:08:50.040 one of the largest apployments of capital in human history. 0:08:50.880 --> 0:08:55.199 This has made in Nvidia the most valuable company in 0:08:55.280 --> 0:08:58.959 the world, and it has created a situation where in 0:08:59.040 --> 0:09:02.079 Vidia stock is more concentrated in the S and P 0:09:02.200 --> 0:09:04.880 five hundred than any stock since they started keeping track. 0:09:05.400 --> 0:09:08.520 And actually Microsoft, who is the second biggest, has that 0:09:08.679 --> 0:09:11.840 valuation largely because they're building these sheds and renting out 0:09:11.840 --> 0:09:12.520 in video equipment. 0:09:12.559 --> 0:09:13.360 So that's linked too. 0:09:14.240 --> 0:09:17.280 So think about that fifteen percent of the stock market 0:09:17.400 --> 0:09:21.120 is these two stocks, right, they have to succeed. Americans 0:09:21.160 --> 0:09:25.240 in particular, are usually invested passively through index funds in 0:09:25.440 --> 0:09:27.959 something that looks exactly like the S and P five hundred, 0:09:28.480 --> 0:09:31.400 So you know, if in Vidio crashes, it's going to 0:09:31.480 --> 0:09:33.200 create a lot of paint throughout the economy. 0:09:33.480 --> 0:09:35.319 I want to talk about the data centers, but before that, 0:09:35.360 --> 0:09:37.800 I want to talk about the man who founded in 0:09:37.920 --> 0:09:40.920 Vidio and is a CEO today at Jensen Huang. He 0:09:40.960 --> 0:09:43.400 seems to pop up everywhere, but he also seems to 0:09:43.480 --> 0:09:46.760 be more inscrutable. I mean, who is he? And do 0:09:46.840 --> 0:09:49.319 you see him as different from Zuckerberg and Altman and 0:09:49.440 --> 0:09:51.000 Bezos In some significant way? 0:09:51.679 --> 0:09:55.360 Jensen most resembles of all executives Elon Musk because he 0:09:55.520 --> 0:09:59.000 is an engineering wizard. Bezos is smart as hell, and 0:09:59.080 --> 0:10:01.920 so Zuckerberg. But timately they're kind of software guides. You 0:10:02.000 --> 0:10:03.839 know they're coming at the computer from the keyboard in 0:10:03.920 --> 0:10:07.200 the terminal. Jensen is totally different. He approaches computing from 0:10:07.200 --> 0:10:10.440 the circuit up. He's a degree not in computer science originally, 0:10:10.480 --> 0:10:14.559 but actually in electrical engineering. Okay, So for Jensen, the 0:10:14.880 --> 0:10:18.760 computer is a piece of hardware that runs calculations in 0:10:19.320 --> 0:10:23.000 a microchip, and he literally designed those microchips on paper 0:10:23.240 --> 0:10:24.319 at the beginning of his career. 0:10:24.520 --> 0:10:25.680 And that's all he's ever done. 0:10:26.679 --> 0:10:28.199 And this is a little bit why even though he 0:10:28.280 --> 0:10:29.959 runs the most valuable company in the world, it's a 0:10:30.000 --> 0:10:33.800 little baffling to to kind of people. Nothing Jensen makes 0:10:33.880 --> 0:10:37.120 is really that accessible. It's all deep inside the computer. 0:10:37.440 --> 0:10:39.000 As a quote in your piece that I liked what 0:10:39.160 --> 0:10:42.120 he said, I find that I'm best when I'm under adversity. 0:10:42.720 --> 0:10:45.439 My heart rate actually goes down. Anyone who's dealt with 0:10:45.559 --> 0:10:47.640 RUSSIAU in a restaurant knows what I'm talking about. 0:10:48.040 --> 0:10:51.040 Yeah, yeah, I mean he started out at Denny's, so 0:10:51.160 --> 0:10:53.679 his first job was basically I think he was a 0:10:53.760 --> 0:10:56.480 bus boy at first, and then graduated to dishwasher and 0:10:56.559 --> 0:10:59.440 ultimately became a server. And I was talking to someone 0:10:59.480 --> 0:11:01.800 in the company and she's like, you know what Jensen 0:11:01.920 --> 0:11:05.040 is actually a lot calmer and more compassionate with his 0:11:05.120 --> 0:11:08.640 employees when things are going wrong. It's when the company's 0:11:08.640 --> 0:11:10.400 stock price is way up but it looks like everything's 0:11:10.440 --> 0:11:13.400 going great that he really becomes much more cruel, like 0:11:13.520 --> 0:11:17.160 much much meaner to everybody. So he is actually in 0:11:17.280 --> 0:11:20.000 some ways a nicer person when things are going wrong. 0:11:20.120 --> 0:11:21.680 When he succeeds, it makes him nervous. 0:11:22.120 --> 0:11:24.400 One of his colleagues are described working with him as 0:11:24.600 --> 0:11:27.200 kind of like sticking your finger in the electric socket. 0:11:28.600 --> 0:11:29.600 That's quite the metaphor. 0:11:30.000 --> 0:11:32.680 It's one hundred percent accurate. I mean, I've interacted with Jensen. 0:11:32.720 --> 0:11:34.559 It is like sticking your finger in the electric socket. 0:11:34.679 --> 0:11:38.960 He's so tightly wound. He expects so much to happen 0:11:39.440 --> 0:11:42.640 in every conversation. Just to even start talking to him, 0:11:42.640 --> 0:11:44.160 you have to be totally up to speed. He's not 0:11:44.160 --> 0:11:46.000 going to waste any times. He's not going to suffer fools. 0:11:46.559 --> 0:11:49.199 And he's also really intense and unpredictable, and you just 0:11:49.240 --> 0:11:50.080 don't know where he's. 0:11:49.920 --> 0:11:51.040 Going to go in any conversation. 0:11:51.640 --> 0:11:54.720 And you know, he has what I would describe as 0:11:54.800 --> 0:11:57.800 somewhat self indulgent performances of anger from time to time, 0:11:58.080 --> 0:11:59.959 and That's especially true if you're one of his executive 0:12:00.360 --> 0:12:02.680 If you're not delivering, he's going to stand you up 0:12:02.720 --> 0:12:04.280 in front of an audience of people and just start 0:12:04.440 --> 0:12:08.679 screaming at you. But really, I mean yelling, and it's 0:12:08.720 --> 0:12:10.559 not fun, and he will humiliate you in front of 0:12:10.600 --> 0:12:13.000 an audience. I think people at Nvidia have to develop 0:12:13.080 --> 0:12:15.240 very thick skins. He actually did this to me at 0:12:15.280 --> 0:12:19.840 one point, so I kind of know exactly. Oh yeah, yeah, Well. 0:12:20.200 --> 0:12:22.880 I kept asking him about the future. Jensen does not 0:12:23.040 --> 0:12:26.520 like to speculate. He doesn't have actually a science fiction 0:12:26.640 --> 0:12:28.000 vision of what the future is going to look like. 0:12:28.600 --> 0:12:31.679 He has a data driven vision from engineering principles of 0:12:31.760 --> 0:12:33.920 where he thinks technology is going to go. But if 0:12:33.960 --> 0:12:36.679 he can't see beyond that, he won't speculate. But I 0:12:37.280 --> 0:12:39.079 noticed that other people at this firm would talk about it, 0:12:39.120 --> 0:12:42.319 and I really wanted to get into his imagination. I 0:12:42.360 --> 0:12:44.199 guess I would say, of where he thinks all this 0:12:44.320 --> 0:12:46.800 can go. So I presented him with a clip from 0:12:47.000 --> 0:12:49.760 Arthur C. Clark discussing the future of computers, and this 0:12:49.840 --> 0:12:52.040 is back from nineteen sixty four, but it was kind 0:12:52.080 --> 0:12:54.679 of anticipating the current reality we were in where we 0:12:54.720 --> 0:12:57.720 would start training mechanical brains, and those brains would train 0:12:57.800 --> 0:13:01.760 faster than biological brains and eventually proceed biological prints. And 0:13:01.760 --> 0:13:03.160 so I show this clip to some other people in 0:13:03.200 --> 0:13:05.439 in video, and they've gotten very They kind of like 0:13:06.160 --> 0:13:08.880 smelled up and started giving these grand soliloquies about the 0:13:08.920 --> 0:13:10.760 future that were like very beautiful and articulate. 0:13:10.800 --> 0:13:13.080 And I was hoping to get that response from Jensen. 0:13:14.120 --> 0:13:16.240 Instead, he just starts screaming at me, how about how 0:13:16.320 --> 0:13:18.480 stupid the clip was, how he didn't give a shit 0:13:18.520 --> 0:13:19.000 about Arthur C. 0:13:19.120 --> 0:13:19.360 Clarke. 0:13:19.400 --> 0:13:20.880 He never read one of his books, he didn't read 0:13:20.920 --> 0:13:23.520 science fiction, and he thought the whole line of questioning 0:13:23.600 --> 0:13:26.080 was pedestrian and that I was letting him down by 0:13:26.120 --> 0:13:29.160 asking I was wasting his time. Despite having written his biography, 0:13:29.280 --> 0:13:31.800 Jensen remains a little bit of a puzzle and just 0:13:31.840 --> 0:13:34.200 that I cannot tell you what's going on inside his brain. 0:13:34.320 --> 0:13:38.040 So well, but I will say this, he's extremely neurotic, 0:13:39.679 --> 0:13:41.439 by which I mean I don't even mean this in 0:13:41.480 --> 0:13:42.200 a clinical sense. 0:13:42.240 --> 0:13:44.160 I just mean that, by his own admission, he's totally 0:13:44.240 --> 0:13:48.280 driven by negative emotions. So even though he's on top 0:13:48.320 --> 0:13:51.480 of the world, I think his mind is telling him constantly, 0:13:51.520 --> 0:13:54.079 you're going to fail. This is a temporary thing, and 0:13:54.120 --> 0:13:56.319 the video is going to go back down again, you know, 0:13:56.480 --> 0:13:59.520 twice in his tenure as CEO in video stock price 0:13:59.559 --> 0:14:01.240 has a retreat by almost ninety percent. 0:14:01.600 --> 0:14:03.560 What could make it happen now? What keeps me up 0:14:03.559 --> 0:14:04.360 at night today? 0:14:04.720 --> 0:14:06.600 What could happen today? Anything? Any number of things. 0:14:07.000 --> 0:14:09.440 This would not be comprehensive. But there's three big risks. 0:14:10.040 --> 0:14:13.719 The first is just competition. Nvidia is making so much 0:14:13.800 --> 0:14:17.679 money and everyone's seeing that, and this attracts competition in 0:14:17.720 --> 0:14:20.560 the same manner that chum attracts sharks. Right, It's like 0:14:20.640 --> 0:14:23.280 throwing blood in the water for other microchip designers to 0:14:23.320 --> 0:14:26.200 earn a seventy percent eighty percent gross margin, which is 0:14:26.240 --> 0:14:26.720 what they do on. 0:14:26.800 --> 0:14:27.480 Some of these chips. 0:14:28.000 --> 0:14:30.880 So Google has built a whole alternative stack for AI 0:14:31.000 --> 0:14:33.320 computing around their own, their own kind of platform, and 0:14:33.360 --> 0:14:35.400 they're starting to lease that out to new customers. That's 0:14:35.400 --> 0:14:38.160 a big risk. There's a big risk that Chinese companies 0:14:38.360 --> 0:14:42.840 build alternative, cheaper stacks to what Nvidiat does. Intel had 0:14:43.080 --> 0:14:45.600 ninety ninety five percent of the CPU market at one 0:14:45.640 --> 0:14:49.440 point in this country. Now they're falling apart. Conquering one 0:14:50.040 --> 0:14:53.400 cycle in microchips is no guarantee that you will conquer 0:14:53.480 --> 0:14:56.520 the next one, and history demonstrates that quite clearly. 0:14:56.840 --> 0:15:00.800 So that could happen. B Basically, what hap happens. 0:15:01.000 --> 0:15:04.280 In the data center is we're doing a mathematical operation 0:15:04.440 --> 0:15:09.840 called a matrix multiplication, and it's extremely computationally expensive to 0:15:10.000 --> 0:15:13.400 do this, so without getting two technical basically, to train 0:15:13.480 --> 0:15:17.160 an AI right now, we have to do ten trillion 0:15:17.600 --> 0:15:21.240 trillion individual computations, which is more than the number of 0:15:21.320 --> 0:15:25.840 observable stars in the universe. However, maybe it's possible that 0:15:25.960 --> 0:15:28.960 we find some more efficient way of doing that. Maybe 0:15:29.000 --> 0:15:32.120 there's a way that requires only ten billion trillion or 0:15:32.120 --> 0:15:34.960 even ten hundred trillion. Right Nvidios stock price would go 0:15:35.040 --> 0:15:37.400 down because we wouldn't have to build so many data centers, right, 0:15:37.400 --> 0:15:39.760 we'd have a more efficient training solution. All of this 0:15:39.920 --> 0:15:41.920 is a more complex way of saying, maybe there's a 0:15:42.000 --> 0:15:45.360 technological solution where we you know, right now we're route 0:15:45.440 --> 0:15:48.480 forcing our way to AI. It's a heavy industrial problem. 0:15:48.680 --> 0:15:52.240 We're talking about building nuclear power plants to bring these 0:15:52.280 --> 0:15:56.280 things online. I think maybe it's possible that there's a 0:15:56.320 --> 0:16:00.080 technological solution that trains these things faster, and if we 0:16:00.160 --> 0:16:02.160 discovered it, we wouldn't have to buy so many in 0:16:02.280 --> 0:16:04.760 Vidio microchips that would also make their stock price go down. 0:16:05.720 --> 0:16:09.240 But the third thing is basically right now for the 0:16:09.360 --> 0:16:13.880 last thirteen or fourteen years, the more microchips we stuff 0:16:13.920 --> 0:16:16.680 into the barn, Okay, the more microchips we throw at 0:16:16.720 --> 0:16:18.680 this problem, the better AI we. 0:16:18.800 --> 0:16:21.760 Gnd this is the scaling law in quotes. 0:16:22.320 --> 0:16:24.920 Okay, is not a law in the universe that this 0:16:25.080 --> 0:16:25.720 has to happen. 0:16:26.440 --> 0:16:30.360 It's not some immutable, physical proven thing from first principles 0:16:30.400 --> 0:16:32.800 of physics that the more microchips we have, the better 0:16:32.880 --> 0:16:35.200 AI we have. In fact, no one is entirely sure 0:16:35.200 --> 0:16:40.360 why this works. Presumably, like most other forces in the universe, 0:16:41.280 --> 0:16:43.560 this will hit some kind of s curve. It'll start 0:16:43.600 --> 0:16:46.360 to plateau or level off at some point. We're not 0:16:46.440 --> 0:16:49.480 there yet. But if we did hit a plateau, if 0:16:49.520 --> 0:16:54.080 stuffing more microchips into the barn only resulted in marginally 0:16:54.160 --> 0:16:56.880 better AI or didn't improve it at all, I think 0:16:56.920 --> 0:16:59.000 in videos stock price will go down a lot, and 0:16:59.080 --> 0:17:01.320 I think it would look make this whole era look 0:17:01.400 --> 0:17:03.080 kind of like a bubble if that were to happen. 0:17:03.600 --> 0:17:06.280 Now, is this why in Vidia has kind of become 0:17:07.040 --> 0:17:10.400 the bank of the air evolution In a sense, they're 0:17:11.240 --> 0:17:14.720 wanting to lend money and lock other companies into the 0:17:14.800 --> 0:17:19.359 current paradigm of AI, maybe even hoping to defensively prevent 0:17:19.560 --> 0:17:23.399 other more economical approaches from emerging and consolidating video's position. 0:17:23.440 --> 0:17:24.760 I mean, how much of a chess game is this 0:17:24.880 --> 0:17:26.920 in terms of thinking about the future of computing for 0:17:27.000 --> 0:17:27.680 Jensen and others. 0:17:27.880 --> 0:17:30.720 Oh yeah, it's chess, and Jensen is an I mean 0:17:30.800 --> 0:17:32.760 expert chess player at this kind of chess. 0:17:32.800 --> 0:17:35.320 He's really good at thinking about the. 0:17:35.359 --> 0:17:38.280 Competitive positioning of where he is and where other people are. 0:17:38.640 --> 0:17:41.119 You know, in vidiots early days, the graphics the GPU 0:17:41.200 --> 0:17:43.320 market back in the video game days was really crowded. 0:17:43.680 --> 0:17:45.760 At one point there were fifty or sixty participants in 0:17:45.840 --> 0:17:48.200 this market. I talked to David Kirk, who was the 0:17:48.280 --> 0:17:51.560 chief scientist in Vidio during this time. Jensen would go 0:17:51.600 --> 0:17:53.639 into his office and a whiteboard and you would have 0:17:53.680 --> 0:17:56.879 a list of all his competitors up there, and not 0:17:56.960 --> 0:17:58.600 only that, they would have a list of who the 0:17:58.680 --> 0:18:02.879 best engineers working at those competitors work and then they 0:18:02.920 --> 0:18:06.159 would come up with plans to poach those engineers and 0:18:06.280 --> 0:18:08.480 get them to come work for a video so that 0:18:08.560 --> 0:18:10.960 they would drain the brain power of their competitors and 0:18:11.080 --> 0:18:15.280 force them to collapse. I've compared the early graphics days 0:18:15.359 --> 0:18:17.359 to the movie Battle Royale, whe all the kids are 0:18:17.359 --> 0:18:19.040 on the island, they have to kill each other. It 0:18:19.160 --> 0:18:21.400 was like that. There were like forty competitors and only 0:18:21.440 --> 0:18:25.359 one could survive. Jensen one. He won the Battle Royale. 0:18:25.440 --> 0:18:26.719 He was the last guy standing. 0:18:26.760 --> 0:18:27.920 I mean, he won the knife fight. 0:18:28.400 --> 0:18:34.000 So he is unbelievably ruthless and unbelievably good at identifying 0:18:34.000 --> 0:18:36.560 where the competition is and what he could do, not 0:18:36.760 --> 0:18:38.440 just to beat them in the marketplace, but actually to 0:18:38.520 --> 0:18:40.000 hollow out their engineering talent. 0:18:41.320 --> 0:18:44.959 Who are in Video's biggest customers and what are they 0:18:45.040 --> 0:18:46.080 buying the chips for? 0:18:46.880 --> 0:18:50.200 Okay, it's a bit complex. The biggest customers, they don't 0:18:50.200 --> 0:18:54.400 disclose it. Almost certainly it's Microsoft and then probably Amazon. 0:18:54.800 --> 0:18:57.680 What these companies do is they trained some AI on 0:18:57.800 --> 0:19:00.159 their own, but what they're really doing is putting They're 0:19:00.160 --> 0:19:02.200 the ones building the sheds, they're the ones building the 0:19:02.320 --> 0:19:05.159 data centers. So in video sells them the microchips, and 0:19:05.200 --> 0:19:08.440 then kind of the ultimate end user is a frontier 0:19:08.520 --> 0:19:12.080 AI lab, so that could be something like Anthropic or 0:19:12.160 --> 0:19:15.000 open AI. So essentially the way to think about this 0:19:15.200 --> 0:19:18.720 is in Vidia sells the microchips to Microsoft or Amazon 0:19:18.840 --> 0:19:23.120 or maybe Oracle. Oracle builds and operates a gigantic data 0:19:23.200 --> 0:19:25.720 center with one hundred thousand microchips in it that takes 0:19:25.720 --> 0:19:28.640 as much power as like a small city, and then 0:19:28.840 --> 0:19:32.320 clients like open ai come and lease it out from them. 0:19:39.320 --> 0:19:42.920 After the break Why data centers are worried about break ins? 0:19:43.320 --> 0:20:03.879 Stay with us. Let's talk about data centers. There's something 0:20:03.960 --> 0:20:07.199