The Thinking Machine | FT Business Book of the Year Winner

Well, first let me say that was a wonderful intro. Thank you.

I did indeed write it. You can have it.

Well, let me answer the first question, which is, you know, why did I write this book? It's the same reason I basically do everything, which is the fear of the consequences of what would happen if I didn't do it. In this case, I started using ChatGPT in late November 2022, right after it launched. I mean, the first day, and was terrified. I immediately said to myself, 'Wow, this thing's extraordinary. It is such a good writer. I have six months to remain employed, and then my job will go obsolete.' So, given that I'm going down with this ship anyways, I might as well write about the thing that took me down. And really, I thought this was genuinely what I thought, that my career was coming to an end, that I had to find something else to do. So what made that happen? You know, OpenAI was kind of the obvious story, but that was very crowded. So I pivoted to Nvidia, which at that time was, you know, still a very important company but hadn't quite achieved the level of fame and notoriety even that they have now. So I got very fortunate in this respect. When I started writing this book, Nvidia was the seventh or eighth largest company in the world by market capitalization. In the course of writing it, it became the most valuable company in the world, and by the time I published it, by most measures, seems to be in fact or have been the most valuable company in human history. So that helped a lot. I got very lucky in that regard. I didn't know that was going to happen. I mean, I could tell that Jensen Huang, the CEO, is an extraordinary individual, and I could tell that this company had made extraordinary technical breakthroughs which were difficult to understand even. I mean, a large part of the book is simply trying to explain that in layman's terms, you know. And then I had a couple headwinds at my back. Suddenly people who thought of Nvidia as like the video game company were like, 'Oh, what's this AI thing?' And then a much larger group of people had never heard of this company, and suddenly it was their single most valuable stockholding in their portfolio. Nobody knew what they made or why they were so valuable. You know, they didn't have much of a public-facing arm, especially not in AI. It's mostly a business-to-business transaction when you build a data center. So people were very confused, and having someone to sort of gently explain who it was was helpful. And then the final piece of the puzzle, and I mean this is, I don't know why I got lucky like this, is I got to talk to Jensen a lot, more than any other journalist. And not only that, but Jensen is just fascinating. He is just such an incredible character. And that usually does not happen [laughter] when you're writing business stories. You know, Jensen has been the founder, CEO, and main character of this story since 1993. I got lucky there.

Some of it's a trade secret, some of it's trade secrets of journalism that I can't completely...

Ultimately, but that was not how I got access. You know, after 15 years of corporate investigative journalism and failing a lot at getting access to individuals, I've gotten better about understanding what the gatekeepers of corporate executives, particularly here I'm referring to the PR executives that surround the executive, want and need and expect. So playing that game a little bit was something that I could do. You know, I think the other thing that is true is I was just obsessed with Jensen, more so I think than any other journalist who approached him. You know, I genuinely knew so much about him before anyone else. Other people were coming to him and they had done their research. I did something like 200 hours of research before I even reached out to Jensen. You know, I was digging through old YouTube videos of Jensen, and I made this comprehensive list. In fact, I read all 700 articles that was ever published in The New York Times about Nvidia, and I made a list of every question any journalist had ever asked him. And then I made my list of questions, which had nothing in common with that list. It was all new material. So, for example, I'd found this old YouTube video of Jensen on a Taiwanese news program in 2009. This is back in the dark ages of like video game technology, where he mentioned in passing that he'd almost died in a car accident once. But the interviewer hadn't followed up, and I couldn't find any other information about it. So actually, that was the first question I asked Jensen when I got him on the phone, or when I got him in person, actually: 'Tell me about the time you almost died in a car accident.' And he just kind of perked up and looked at me. He's like, 'Oh, so you heard about that, huh?' And then told me this incredible story of when he was 21. He'd gotten engaged that night. He was with his fiancée, and they were driving home to his parents — her parents, sorry — to break the good news, and they hit a patch of black ice and rolled the sports car. Nearly died. They had to get cut out of the car. And so I was like, 'All right, I'm getting material that is not just like, oh, tell me how, you know, parallel computing processor works. This is like the good stuff, you know, and that's what you focus on as a journalist.' So, you know, Jensen is a complex character, and I think this emerges in the book. He's not the person that you would present or see if you're just watching YouTube videos of Jensen, if you're just watching him speak at GTC or even in interviews on TV.

Yeah. What were the first two things you said?

Okay. So this is a character that Jensen has created because he actually hates doing media and he hates doing public appearances. And I know this because I stood next to him once when he was giving a meaningless public appearance. Meaningless. And he was so nervous. Like he was rocking back and forth on his feet. He looked like a kid who had to go out and perform in the school play and just did not want to be there. And then he went out and killed it. I mean, just did an incredible job. He's so good with media, but he actually hates it. Have you ever seen the movie Whiplash?

In that movie, there is an abusive music instructor who presents this very friendly kind of face to people, but in private hurls abuse at his students and throws things at them. That's the real Jensen. He is very hard to work for. He is extremely mean at times to his employees. He's extraordinarily demanding. He'll sometimes say in a joking way in public interviews, 'I'm going to torture my employees into greatness,' but it's not a joke. It's really what he does. So there's the real Jensen, which is a very different kind of person than the one that you see in public appearances or the one that he normally presents to journalists.

Absolutely. I think Jensen screams more than Jobs did, actually. I mean, Jensen is a yeller. He will yell for an hour and a half at somebody. You know, the difference is that Jensen is weirdly very kind of abusive, but also like very loving as well. So everybody — it's sort of a cult leader thing where you're very desperate to gain the approval of the boss and very afraid of gaining the disfavor of the boss. There's wonderful stories about Jensen that all of his employees tell about him bailing them out in difficult situations, about him pushing them, encouraging them to go places they wouldn't want to go. And then those same people will tell you a story about the time they saw him throw a water bottle at somebody's head, or the time that he screamed at them, made cutting, withering insults about their performance and their capabilities. So he has both of those things inside of him. He's a very complex person.

Yes, very much so.

I think he has in fact evolved into it. Now, it's hard to say. I must say I don't know for sure, but I did talk to one of his closest and oldest friends, who was friends with him really in his early 20s when they worked together, and he told me that the yelling part of it didn't really start to happen until Jensen became the CEO. You know, Nvidia's early history was they struggled a lot. They nearly went bankrupt twice, if not more. And there was a time where Jensen had to lay off about 70% of the company. Their first product was a dud. It was returned to stores. It just didn't perform the way they wanted to, and they had to kind of reboot the whole company from scratch and pursue a much less humble vision than the kind of big one that they'd originally pursued for computer graphics. And I think somewhere in there, Jensen realized to motivate his employees he had to act as if the company was constantly teetering on the verge of bankruptcy. Even today at Nvidia, the motto is — this is the largest company in the world — he will tell his employees, 'This company is 30 days from going out of business.' Now, there were times where that was literally true. There were times in Nvidia's history where they didn't have enough money to pay the rent and barely make payroll. Now it's not true. But still, conceptually, for Jensen, it is true. Remember what I said at the beginning, that the reason I wrote the book about Nvidia is I was just the same reason I do anything: I was afraid of the consequences of what would happen if I didn't do it. I have this in common with Jensen, actually. This is what motivates him. He's totally driven by neurosis and fear, right? He's not like an optimist in any way. He's constantly thinking of various catastrophic outcomes for Nvidia and how he might avoid those from occurring. It's really different in that way, I think, than Jobs. I've never met anyone quite so tightly wound. I think it's hard to be Jensen. I wouldn't trade places with him.

Well, there's that. And if you stand around Jensen, you get to add three zeros to your net worth.

Yeah.

If you put up with it, you made money. I mean, but the thing is, actually, there was a long period of time where Nvidia's stock price was in the toilet. I mean, it was one of the dogs of the S&P 500. If you bought the stock in 2001 on the day it entered the S&P 500, as many index funds and mutual funds were more or less required to do, you did not see a positive return on your investment until about 2012, 2013. So there was 11 years of just darkness, pain. People thought he was eccentric and weird, pursuing these weird visions for supercomputing with a graphics processor. What are you doing? Didn't you go to Camp Hampton Nintendo at one point?

That was before they went public. I mean, that was very early. You know, what was happening was you had your classic '90s, early '90s, late '80s side-scroller video games, you know, like Mario Land, I guess it was. And then we went to Super Mario World or Galaxy, and Mario became a 3D character, explore a 3D world. And to make that happen, you need a graphics controller to kind of create the images for you. And that's what Jensen was making, except he was making it for the PC. And so all these classic PC games like Quake and stuff were doing this. And just technically, I'll be fast about this part because I know it's boring. But to do that, basically you have to render the perspective. You have to render a three-dimensional perspective on a two-dimensional screen, right? And the mathematics you do that with is something called a matrix multiplication. And you have to make that frame render something like 30 to 60 frames per second. The more frames you can get, the smoother it looks. So they built basically a machine that did tons and tons and tons of matrix multiplications. And then by complete coincidence, that turned out to be exactly the thing that they needed for AI.

I say by complete coincidence, not completely. It's a little more complicated than that, but they got lucky. Let me tell you that much.

Right. So there's kind of what is their mainstream business right now, which is, you know, I mean, graphics is like less than 10% of what they do now. Now what they sell is selling, you know, microchips to data centers. I mean, that's where this boom is happening, this massive basically new industrial revolution. And, you know, they had some competition for a little while, but basically they're the only game in town. I think it's like above 80% market share, if not 90%, doing this. And they release chips on an extremely fast cadence, like every six to 12 months. So there's this Nvidia upgrade ladder that everyone's more or less committed to in perpetuity. And the hyperscalers pay it because what do they care, honestly? That's not what they're worried about. You know, like OpenAI is not worried about saving money on microchip costs, right? That's not what they're thinking about right now. They're worried that they're going to lose to Anthropic, and switching off Nvidia's stack would mean refactoring all this code, and you would be standing still for six months. And they just cannot afford to do that. They just cannot afford to do — they can't stand still for one second, much less try and save money on not giving it to Nvidia by refactoring a bunch of like low-level computer code. They're not going to solve the Riemann zeta hypothesis doing that, right? Which is where they're at, right? So they're racing towards artificial superintelligence. It makes no sense for them to attempt to save, you know, some money on microchips. Like, who cares? They have the money they need if they need it. You know, they can turn on the faucet anytime they want. So that puts Jensen in a very privileged position. He's also a very diligent and excellent partner. I mean, he embeds his guys with all these guys to make sure that they always get what they need. This has always been what Jensen was excellent at. Going back to his 20s, even before he founded Nvidia, he really went to the most demanding customers on the planet and said, 'What would you do if I gave you a million times more computing power?' and then built systems for them. And so for a long time, that was actually the 3D graphics guys. Those are the most demanding customers on the planet. Now it's AI. What the future will be, you know, it's probably more AI at this point. I don't think anything else is on the horizon. So that's the core business. And then there's all these kind of businesses around that. Jensen has, I don't know, 500 initiatives or something. It's not that many, but it might be a hundred. And all of those are like, you know, let a thousand flowers bloom kind of philosophy. The big one that's succeeding right now, I would say, is robotics. Jensen has been investing in robotics for over 10 years. In fact, when I asked him in 2023, like, what's your next CUDA? Like, what's your next big platform play that might lose money for years but will ultimately add five trillion dollars to your market cap? He immediately answered by referencing robotics. So way back then even, and have been investing in it forever. I just wrote a piece about humanoid robots for The New Yorker, and I got to see 10 or 11 different models of robot from different manufacturers, all kind of different form factors, all kind of different abilities, capabilities, design. Every single robot without exception had an Nvidia microchip as its brain. So it's another dominant platform that he has, right? And what's happening here is, all right, we have these what are called training chips that we — this is what we sell to the data centers, but the robot has to think quickly, right? It has to respond in real time. It doesn't have time to go communicate with the data center and come back with an answer. And so they build what's called an edge chip, an edge inference chip that goes right in the brain and does most of the AI processing right there. Now I'm pointing at my head. Most of them actually have it in their chest, but some of them do actually put it in the head. The robot — and these chips are really powerful, okay? Like your brain will consume about 20% of the energy that your body produces. An Nvidia chip will drain up to 60% of the robot's battery. So what looks like it's mostly moving around, you know, motors and stuff, that's not what's drawing the brain — the battery down. It's this really powerful brain. And as far as I can tell, they have virtual platform monopoly on this already. Could be worth — could be worth trillions of dollars. You know, and Jensen has 40 other things like that.

Jensen's vision is that everything that moves will be autonomous. That's what he says. So whether it's a humanoid or whether it's a self-driving car, self-driving vacuum, toilet cleaning robot, he envisions a world in the near future where you will have either one humanoid gadget in your house doing all your chores or a fleet of smaller purpose-driven gadgets doing all of your chores. And I think he is agnostic about actually which of those two worlds occur. He should benefit in either one. But that's what's happening. And I see a lot of investment coming there. You know, I should say the robots right now have not had their ChatGPT moment. They will not blow you away with their capabilities. In fact, they're sort of clownlike and they fail a lot. But, you know, having covered AI for some time now, I truly think that it's only a matter of time. And I'd actually be surprised if we don't have a significant breakthrough in the next 12 to 24 months in robotics. I have yet to see the hill that AI can't climb.

So what's missing there is we have these vast repositories of data for text, right, and for mathematics and for other things, but what we don't have is a big free library of joint action trajectories, right? So people are attempting to build it, right, either by reverse engineering it from like YouTube videos or putting like factory workers in motion capture suits or, you know, first-person view cameras and whatnot, so they can build these data libraries up and up. You know, the company I talked to, 1X, for my article, they're deploying robots in the home. And at least part of the play is just that they're not going to work that well at first, but they're going to use teleoperators to control the robot, and then that teleoperations group is going to sit directly next to the AI group at 1X headquarters working in tandem to take teleoperator data and turn it into like more actionable AI data. It might take a minute, you know, it might take a minute. Data is, you know, the thing about the AI that we have now, to the extent it has any limitations at all, it doesn't learn very quickly. Like, for example, if you wanted to teach a robot to shuffle a deck of cards, given current technology, you'd probably have to train it on like, I don't know, like a million hours of card shuffling. I learned to shuffle cards, I think it took me an hour, might have taken me two hours, you know. So I learn much, much, much faster. I have much more physical intelligence in terms of my ability to learn complex physical tasks than the robot does. On the flip side, the robot can clone itself in a simulator and do a million runs at once. So who cares if it takes a long time to learn, you know? Like, it has advantages that I don't have. This was a point Geoffrey Hinton made to me. Like, yeah, you know, actually, it's true that AI doesn't learn in quite the same way we do, but that doesn't mean it's inferior. Maybe it's smarter.

Yeah.