One man could slow down AI. He won’t. Why?

This episode of Search Engine is brought to you in part by Mubi. If you're tired of endlessly scrolling through streaming algorithms designed for passive background noise, you need to check out Mubi. Mubi is a global film company bringing great cinema to the world, curating and elevating visionary filmmaking that's handpicked always. It's the kind of platform for movies you'll actually watch, fall in love with, and keep thinking and talking about long after the credits roll. One of their newest exclusive titles is Teenage Sex and Death at Camp Miasma, directed by visionary writer-director Jane Schoenbrun, who you might know from I Saw the TV Glow. It's a world-expanding, genre-bending wonder starring Hannah Einbinder and Gillian Anderson. This story follows a young director trying to resurrect the fated Camp Miasma slasher franchise, but when she visits the original reclusive star, they spiral into this blood-soaked world of desire, fear, and sheer delirium. You can try Mubi free for 30 days at Mubi.com slash search engine. That's M-U-B-I dot com slash search engine for a whole month of great cinema for free. This episode of Search Engine is brought to you in part by Vanguard. To all the financial advisors listening, let's talk bonds for a minute. Capturing value in fixed income is not easy. Bond markets are massive, murky, and let's be real, lots of firms throw a couple flashy funds your way and call it a day. But not Vanguard. Vanguard's commitment to institutional quality means offering top-grade, across-the-board, backed by a lineup of over 80 bond funds. While other firms love to highlight a single star portfolio manager, Vanguard believes the best active strategies shouldn't depend on one person. Instead, their funds are driven by a 200-person global squad of sector specialists, analysts, and traders sharing collective brainpower. Vanguard's scale allows them to invest across global markets and spot opportunities others miss, all while avoiding high-risk bets to deliver steady, dependable, consistent results. So if you're looking to give your clients consistent results year in, year out, go see the record for yourself at Vanguard.com slash audio. That's Vanguard.com slash audio, all investing is subject to risk. Vanguard Marketing Corporation Distributor. Musk and Bezos, Dario and Sam, men who control large parts of our lives and who we have to make up our minds about. And this year, we got a new one. The new mogul omnipresent in American life, CEO of the most valuable company in human history, the man who could on his own just decide to slow down the American AI race, if he wanted to, which he doesn't. I've been trying to understand him, how he became so powerful, why he sees the world the way he does, months in the story of this other person's life. I think you have to start that story in 1973, with a 10-year-old boy born in Taiwan. The boy and his parents now live in a country in the throes of political violence. The parents don't think they can keep their son and his brother safe. So they send them both to America, where an uncle lives. They think the children will be better off there, in Oneida, Kentucky. Supposedly, there's an elite boarding school in Oneida, and the idea is that the sons will go there, but there's been a misunderstanding. Their kids are sent to a reform school, according to the boy, a particularly violent one. In the dorm where this 10-year-old Taiwanese boy sleeps, his roommate is 17, and he shows him the knife scars all over his body from a recent fight. His parents have sent their son to America to keep him safe. Instead, they've shipped him to a crucible. His mom and dad are thousands of miles away. His English is not very good. He's very vulnerable, in a place he does not understand. That's his introduction to America, a country he'll one day have about as much power over as anyone can. But for now, he has to figure it all out himself. He improves his English, cleans a lot of toilets, gets a job in high school doing dishes at Denny's, graduates college, the boy grows up. At 30, he starts a company. The company is called NVIDIA. The man's name is Jensen. Please welcome Jensen Huang. On September 14th, Jensen Huang walked onstage to join the All In podcast for a very special live taping. Jensen, now 63, with a still-youthful face, frameless glasses, a big poofy crown of silver hair, wearing one of his trademark black leather jackets. We gotta stand to go on the way in. Oh, come on. Stand to go. You should watch this yourself. It's such a strange broadcast. These nerds have gotten so rich and powerful, they can fill an auditorium with people who will treat them like they're cool. Like they're rock stars. One of the hosts intros Jensen with a moderately funny one-liner, which the crowd roars over. Ladies and gentlemen, GPU Jesus. They love you. They love you. Thank you, I love you back. It's sycophantic, sure. But the enthusiasm's real. An auditorium of people shouting, I love you, to a man whose company designs the microchips that power the AI race. Jensen, in this room, is seen not just as a generationally talented engineer, but as one of the greatest businessmen of our time. A man with the ear of the American president. In fact, at one point during the podcast taping, Trump calls him. And Jensen, after a bit of difficulty, puts the president on speakerphone for the room to hear. Hold on, sir, we're getting a microphone. Mr. President, you're now talking to the planet. See, the great thing about life is that Jensen can develop the most complex computer chip in the world. That nobody can copy for 10 years. But he can't figure out how to put me on speaker. Something you should know about the hosts of All In is that when they hear the president make a joke, any joke, they react like it's the first joke they've ever heard. It's unusual. But they all sit breathlessly on stage as Trump says how AI is great, data centers are great, how safety concerns about AI are just garbage. These doubters are a threat to American AI's future dominance. They're just playing right into the hands of a lot of people that don't want to see it happen. And that could be political people. It could also be China. And we're not going to let that happen. It's a hoax. You're right. We're not going to let that happen, sir. That's Jensen, who's holding the phone. The robots are not going to be taking over the world. And that's not going to happen. You know, my uncle was. Trump is confident things will be fine, partly because he has his expert telling him so, Jensen. Jensen's already given the credit. The crowd that day, his don't-worry-about-it spiel. The one where he talks about the series of recent concerning AI safety incidents. Jensen describes these incidents blandly as engineering problems, obviously solvable with some troubleshooting. But nonetheless, the four incidents from one lab, the one giant incident from the other lab, the first thing that you have to do is just root cause the problem from an engineering perspective. What happened? What could we have done differently? And what. Are we going to implement and institutionalize, whether it's technology or methods or processes, and make sure that we don't let it happen again? Now, I would bet you money that in every single one of those cases, it's within their control in the future to prevent it. The vibe in this room is that Jensen is a grown-up, as opposed to the neurotic, hand-wringing children at Anthropic or OpenAI, the ones who go online to say they're scared of the computers they made. Jensen is saying, those AI labs need to sober up and get their shit together. Even if their machines are dangerous right now, it's the AI lab's job to fix those machines, not tweet about the danger. Even if it's true, we ought to spend more time doing something about it than worrying a bunch of people who can't do anything about it. It's our job to build it right. Has there ever been a point in history where so many people have so vehemently said something that is so untrue? And they're measurably. They're actually. Demonstrably untrue. And it actually makes sense it's untrue. It's not based on science. It's not based on research. Everything that's based on science and research proves otherwise. Is it a fear of the frontier? Humans have never been there. We've never seen it. Therefore, we're scared of it. And therefore, it's easy to tell everyone. If you are like me, someone who is worried about the pace of AI, this confidence is not reassuring. Jensen is arguably the American with the most power to slow this down. He has the ear of the president, but he also sells the actual, the graphics cards that OpenAI and Anthropic rely on. Unlike those companies, NVIDIA dominates its market. It has competitors, but they are not nipping at its heels. They are not pushing it to sprint at this breakneck pace. I don't share Jensen's confidence that things will be okay. But rather than just being frustrated by that, or assuming he's purely motivated by money, I wanted to try to be curious. Which is why for months, I've been having these very long conversations with America's foremost Jensen, and I've been having these very long conversations with America's foremost Jensen, ornologist. his biographer. To get the story of the man building the engines for the big machine, pushing us all towards a future we're not sure we want. I guess, first of all, just hello. Hi, thanks for having me. Thanks for being here. Can you just introduce yourself, say your name and what you do? My name is Stephen Witt, and I write about AI and technology. I'm the author of The Thinking Machine, a biography of Jensen Wong and the history of NVIDIA. The Thinking Machine is a book I've been recommending widely, deeply researched, extremely well-written. And Stephen Witt, its author, he's a person who both respects Jensen's mind and ambition, and also really disagrees with him about how dangerous AI might be. In fact, the reason Stephen started covering AI years ago was because his first encounter with the technology had scared him so much. I used ChatGPT for the first time on the day it was launched, I think, November 30th, 2022. And I was like, I'm going to do this. And I instantly was like, oh, my God, my career's over. That was your first thought? Oh, right away. Like, they have a writing machine now, and it writes just as well as me, but it does it for free in one second. Like, I just had this total spiral. I was like, this is horrible. I talked to a therapist about it, actually. Really? Oh, my God, yeah. I was like, I give myself six to 12 months of runway here before I don't have a job anymore. Like, what do you think I should do, et cetera? So I was like, well, I have to write about AI. That's the only solution here. I have to lean into this fear and pivot. Whatever I'm doing, I dropped every project I was working on, just write about AI all the time. But that story was extremely crowded. There were 800,000 journalists trying to write about that. And I was like, well, I've got to find some way in here. Is there like a backdoor? Is there a facet of the AI revolution that people are not writing about? And I came across NVIDIA. NVIDIA, at the time a company most normal people had never heard of, run by a man they certainly didn't think about, Jensen Huang. Stephen Witt is not normal, though. So he spent lots of time with Jensen and conducted almost 200 interviews with people who'd known him, friends, foes. And this week, with Stephen's help, we're going to take you through the key moments of Jensen Huang's life, the moments where he learns some almost religious beliefs about pain and about his ability to see the future. And then we'll take that understanding of Jensen Huang to the present to assess what looks like his future. And then we'll take that understanding of Jensen Huang to the present to assess what his unshakable faith in technology safety. By the end of this episode, you might admire this person, you might have serious doubts about his judgment, but I think you'll see what I saw. Someone more complicated and interesting than tech bro who will burn down the world for Lambo money. So to begin, we're going to pick back up the story we started with, lesson one in Jensen Huang's life, lessons from bad kid school. So it's 1973, Jensen arrives in America, finds himself bunking with a 17-year-old covered in knife wounds. Here he is with America's foremost long-form interviewer, Joe Rogan, telling the story of what happened next. Every kid had chores. My older brother's chore was to work in the tobacco farm, you know, so they raised tobacco so they could raise some extra money for the school, kind of like a penitentiary. Wow. And my job was just to clean the dorm. And so I was nine years old. I was cleaning toilets and for a dorm of a hundred, a hundred boys. I cleaned more bathrooms than anybody. And I just wish that everybody was a little bit more careful. It wasn't just that Jensen was cleaning toilets in a tough environment. It was that nearly everything about him marked him as an outsider. A short kid with a shag haircut, an Asian kid in white Kentucky in America's Vietnam era. Here's Stephen Witt. He gets called every racial slur in the book. Like really vicious. Anti-Asian racism of a kind that I think you would almost not be able to find anymore, or at least I hope not. But, you know, that's where he was. I mean, these kids that he was at school with were the children of tobacco farmers. At school, initially, at least, he's bullied. Yeah. And so he initially tries to trick all the kids because their only point of reference for Asian culture is going to be like Bruce Lee, right? So he tries to trick them into thinking that he knows a bunch of Kung Fu and karate. Now, this is very quickly just proven on the, you know, the school yard. It feels like a very disprovable ruse. Yeah. But he can fight, actually. Or at least he has enough heart inside of him to fight back when they bully him. And I talked to one of his best friends from this era, a guy named Ben Bays. And Ben was like, yeah, Jensen never lost. He won every fight. I was like, wow, really? And then I asked Jensen that. And Jensen was like, well, that's not how I remember it. Now, I think most kids would have, like, folded, right? Most kids would have been like, that was traumatic. Jensen cites this information as, like, a very, very important thing. Like, this entire thing is one of the best experiences of his life. I've lived a great life. You really have. This is Jensen, again, talking to Joe Rogan. You don't have to go to Ivy League schools to succeed. This country creates opportunities, has opportunities for all of us. You do have to strive. You have to claw your way here. Yeah. But if you put in the work, you can succeed. Nobody works hard. There's a lot of luck and a lot of good decision-making. And the good graces of others. Yes. I call Jensen a front-of-the-class kid by kind of inclination, right? He sits at the front. He gets a perfect score on the test. He does all his homework, and his teachers love him. Now, that will take you far in life. It will take you to graduate school and maybe even beyond. But when you get to the business world, that's not good enough. Yeah. It's not good enough to just be good at passing the test. And so, here in Kentucky, I think Jensen developed a lot of what I would call back-of-the-class skills. Things like cheating on a test, standing up to bullies, navigating basically all-male social situations where you really have to kind of fight just to kind of stay in the game, you know? Yeah. Feeling like you can't possibly win, and then coming from behind to win something. Being the lowest ranked and fighting back from that position. Often, kids who are really good at school will flounder in the real world because they don't have any of that grit, right? But Jensen developed this, I think, in Kentucky, developed that grit, the ability to succeed in an environment where there really aren't any rules and where doing what's expected of you is not good enough. You really have to think outside the box. And people with back-of-the-class skills can often succeed as well. The president has back-of-the-class skills. He's not a front-of-the-class kind of guy. He understands how to bully, intimidate, cheat, and get his way to the top. Now, Jensen doesn't have all those traits, but he has some of them, I think. And I think that's what really allowed him to succeed later in life in business. It prepared him very well to be the CEO. Oh, and by the way, the microchip industry, the semiconductor industry, is an absolute knife fight. So, school taught Jensen the survivor's ruthlessness. He would need to get through both a classroom of bullies and a corporate boardroom. Is that true? Is that a CEO milking an origin story? I don't know. But I think it's worth pausing on because afterwards, Jensen will, time and time again, exhibit this trait of profound tenacity. The first big example of this comes in the 1990s. Lesson two, breaking Moore's law. So, first, Jensen goes to college, gets a degree in engineering. After, he gets jobs at a couple of computer chip companies. In 1993, he turns 30, and two friends convince him to quit his job and become CEO of the company they'll start together, NVIDIA. They have their first meeting in a back booth at his favorite restaurant. I know NVIDIA is, to a lot of people, the company famous for making AI chips. To me, they'll always a little bit be the company that I first encountered them as in the '90s, years before AI, when I was a 14-year-old computer game nerd. This was back when consoles were not as good, when nerds like me who loved to play video games played them on our computers. Back then, I would religiously read PC Gamer magazine, and they would talk constantly about something they called Moore's law. In my memory, Moore's Law was the most popular game in the U.S. in the 1980s. In my memory, PC Gamer presented Moore's Law as this weird mystery. For some reason, computer chips just predictably, like clockwork, got two times faster every two years. Every two years, somehow, inventors found a way to shrink the transistors on the boards inside the computers. What I didn't know then is that Moore's Law was actually named after a man named Gordon Moore, who co-founded Intel, the company that made the processors of the computers my friends and I were obsessed with. Moore's Law was not a law of nature. It was basically a description of how hard Gordon Moore pushed people at his company. Here's Steven: "The way I would put this is the law was actually a law from the CEO of Intel to his employees. You better make this happen. This is the law. This better happen because our whole business model is based on it. And so, you know, you engineers go bust your ass and figure out a way to make sure this will happen, you know? And those engineers worked very hard, and they were very smart, and they did make this happen for decades. But what happened, ultimately, is that the components just got too small. They got so small, they started running into quantum mechanics. And basically, you couldn't shrink them anymore after that because they were starting to leak electricity out of the components, and the scaling relationship broke down. Jensen was among the first to see and to understand that that was going to happen. And so, you couldn't rely on Moore's Law anymore. You had to find another way to make the microchip faster that wasn't just shrinking the components. Steven: "Nearly every Everybody else believed Moore's Law would just go on forever. Jensen didn't, and he made a big, confident bet in that direction. His company, NVIDIA, back then, did not make CPUs like Intel did. They made graphics cards, what we now call GPUs. And those were really only sold to gamers like me. This special chip you could add to your computer to give it a second brain, purely devoted to quickly drawing high-quality graphics. And while the Intel engineers were following Moore's Law, believing the only way to make a faster chip was to shrink the transistors, Jensen had his engineers trying some other strategies, too. The thing was that the Intel CPU's brain was designed like one brilliant PhD. Very smart, but it could only think about one thing at a time. An NVIDIA GPU chip was designed more like a classroom full of precocious middle schoolers, a bunch of dumber brains linked together that could work on different tasks simultaneously. The reason that would matter, is that when the CPU, the brilliant PhD brain, hit a wall, when shrinking the transistors stopped making it think much faster, its progress slowed way down. Over at NVIDIA, Jensen's engineers could still reach higher speeds by designing new ways for the chip's many little brains, the precocious middle schoolers, to work together more efficiently. The real payoff for that was still years away. It would be part of the reason why one day AI would run on GPUs instead of CPUs. But for now, what Jensen was learning was how to make new chips that got faster at a rate his competitors could not match. In the mid-90s, NVIDIA had been battling lots of different graphic card companies. By the end of the decade, they'd beaten almost all of the competition, mainly just by being so brutally consistent. So going into the 2000s, Jensen had won the market. And now he's going to take another bigger bet, one that lots of people would have a real problem with. He would decide that in the long run, he did not want to be the first one to win. He did not want gamers to be NVIDIA's only market. After the break, GPU Jesus searches for his new flock. This episode of Search Engine is brought to you in part by Built. You don't need me to tell you that rent and mortgages are only getting more expensive. That's why Built created three credit cards designed to offer the richest rewards on rent, mortgage, and everything else you spend on. Whether you rent or own, you can earn up to 1.25x points on every housing payment without the transaction fee, no matter where you live. Points can be redeemed for flights, hotels, lift rides, a future housing payment, and much more. Turn one of the most unavoidable payments in your life into the most rewarding. Find the card that's right for you and apply at joinbuilt.com. That's J-O-I-N-B-I-L-T dot com slash search. Use our link so they know we sent you. Terms apply, subject to approval and eligibility. Built cards are issued by column N-A, member FDIC, pursuant to license from MasterCard International Incorporated. This episode of Search Engine is brought to you in part by the new book from Michael Lewis, Blockers. For 30 years, Michael Lewis has written about the people behind the stories in character-driven bestsellers like The Big Short, Moneyball, and The Fifth Risk. Now, in his new book, The Big Short, Moneyball, and The Fifth Risk, Michael Lewis has written about the people behind the stories in character-driven bestsellers like The Big Short, Moneyball, and The Fifth Risk. Now, in his new book, The Big Short, Moneyball, and The Fifth Risk, Michael Lewis tells the story of six civil servants who devoted their lives to public service that matters. One kept Americans' most sensitive tax data out of the wrong hands. Another made sure that politicians and civil servants played by the rules. One figured out how to stop wildfires from destroying suburban neighborhoods. Another helped uncover the mysteries of cystic fibrosis. There was also a security guard. In the new Washington, these federal employees discovered that their high standards were their greatest liability. If these public servants, can't survive in government today, what does that say about us? I am so excited that this new book is out. Michael Lewis is one of my favorite authors. I read everything he writes, and I'm very excited for this one. Blockers by Michael Lewis. Available wherever books are sold. Learn more at blockersbook.com. Welcome back to the show. So, it's the early 2000s. Jensen Huang is CEO of a company called NVIDIA that most people think of as serving the video game market. But that's about to change. Lesson three, CUDA. The GPUs Jensen's company designed had gotten so powerful that they were now attracting this new fringe group of users. Jensen had begun to hear from academics. Here's Steven Witt. Professors at Stanford would call him up and they would say, hey, these video game chips you're making, they're actually perfectly suited to do science. In fact, at first they were hacking the video game software to do science on it. And Jensen by 2003 was saying, okay, don't hack my video game equipment. I'll just build you a platform to do science on these things. These academics had been essentially tricking the NVIDIA cards into thinking they were doing calculations about how to make polygons for shooter games. When in fact, they were really crunching numbers for academic research. They were using the cards, they were using the processing power to analyze big trunches of data. Jensen's response to this news was, let me just make this easy for you. His plan was to release a new software platform that would let users take his NVIDIA graphic cards and just deploy them however they wanted. The platform was called CUDA. Jensen announces CUDA in 2006 at a keynote for an audience of mostly gamers. Thank you. Welcome. I want to welcome all of you to a very special day for us. I especially want to welcome all of you to a very special day for us. I also want to welcome all of the gamers in the audience. Dan said it perfectly. Today is all about you guys. We built it for you guys. You guys absolutely rock. The gamers are excited to learn about the new chips, how gorgeous the virtual blood spatter is going to be. But then, Jensen sneaks in something new. His new paradigm for computing. His new goal for NVIDIA. Is to invent a new computing model. Which, unleashes the massive programmable and computational horsepower of the GPU to leverage those resources for data-intensive applications that frankly weren't practical except on supercomputers and custom chips until now. We call this new computing model GPU computing. When I launched CUDA, the audience was complete silence. This is Jensen, again with Joe Rogan. No customer wanted it, nobody asked for it, nobody understood it, NVIDIA was a public company. What year was this? This is, let's see, 2000, 2006, 20 years ago, 2005. Wow. Our stock prices went poof. I think our valuation went down to like, two or three billion dollars. From? From about 12 or something like that. I crushed it in a very bad way. They launched this platform and it is essentially a flop. Like, they get some scientists to use it and it's, they do have some exciting principles, but it loses money for a long time. Jensen sticks with the bat, but even if you look at the download statistics, there's a spike. There's a spike when they launch it, 2007, 2008, and then fewer and fewer people are downloading each year, 2009, 2010, 2011. There's fewer downloads. Because they've saturated like. They've saturated the science market immediately. There's just not that many research scientists. And here's the thing. Who is this for? A low budget supercomputer? Well, it's not for an established research scientist, right? That guy can afford time on a conventional supercomputer. Right. It's really for a marginal scientist with a low research budget who can't afford time on a supercomputer because his ideas are not in favor. It's for mad scientists. CUDA for years was a giant suck on Nvidia's bottom line. This is where a normal CEO would have either blamed somebody else for it or found a way to shutter the program without admitting it in public. But Jensen, abnormally stubborn, would not let CUDA go. He had the point of view. If I don't build this platform, these hackers are going to build it for me and I won't have control of it. And ultimately, I will be disrupted from below by this hacking community. So I have to absorb this almost as a defensive thing because this could someday be a big business. Jensen had absorbed this lesson from Clayton Christensen, who wrote a very famous book called The Innovator's Dilemma About Disruption. This book, Jensen loved it so much, he forced people at Nvidia to read it, he hired the author as a consultant. It's a management book, but it almost reads like a dark prophecy. It says that companies that dominate a market will often be destroyed by some tiny little competitor who at first just starts by nibbling at the parts of their business that they don't care about. Jensen thought one day that could happen to him, but he wanted Nvidia to buck the prophecy, not become one of those former titans eaten by a little minnow, blockbuster eaten by Netflix, film even by digital. So Jensen decides he would aggressively force his company to pursue what looked like unprofitable niche markets instead of just exclusively catering to the gamers and the fat profit margins his chips were making off them. Jensen's plan would look very weird in the short term, but he believed that in the future he'd be vindicated. Stephen Witt says in the early 2010s, what this all meant, though, was that NVIDIA was mainly famous for being a terrible stock to buy. This scientific platform he was building for years, was a depressant on his profit margins. In fact, Wall Street assigned it essentially a negative value. NVIDIA down about 4.5% on the week, it also the worst performing stock in the NASDAQ 100. The stock today has been among the worst performers in the NASDAQ 100 and the S&P 500, and the chips have really been the drag on the day. NVIDIA, this is a company that makes chips for video games. They put out their earnings, but their sales forecast this week, that stock is getting hammered. It's also the worst performer in the S&P 500 as well. Back in the early aughts when this was happening, Stephen actually worked at a hedge fund, and he remembers what a joke NVIDIA was. If you invested in them in 2001, when they entered the S&P 500, you made 0% returns for the next 13 years. They were a dog of the stock market. He was seen as very eccentric. He took a lot of abuse. They were seen as stagnant. They were just not seen as a major player outside of this very small cultish community of people who understood, what Jensen was trying to do. Wall Street hated him. They hated him. And this was a decade of this. It wasn't a couple months. It was not an easy time. I mean, this is how I got to write the book about Jensen, actually, because I understood that he was still living in that world. I had read everything about this. I knew how hard it had been in 2009. His company's stock price went down 90% twice. Like, they were in the toilet. It's crazy that there wasn't, like, a boardroom coup or something. Oh, there almost was. I mean, he had activists coming in and writing threatening letters to the board. Jensen barely kept his job. In 2012, right on the cusp of the AI revolution, Jensen had to do a roadshow going to Fidelity and all the big advisors and basically beg for his job, beg his investors to side with him over the activists. That's how close it was. And that's what makes Jensen who he is today. Not the success, but that. Those 10 years. And what do you think it was about his biography or personality? Most people would loosen their grip on the prophecy that they were telling themselves. Why do you think he was able to stubbornly persist? I asked him this question. And he was like, yeah, I was just talking to this quantum physicist in Taiwan. And this guy had bought, like, 30 GPUs, like, gaming boards, okay? Video game equipment. And he jerry-rigged a supercomputer in the hall of his office. And it was so big. It was so hot. He had to go buy oscillating desk fans, okay? To keep it cool. He homebrewed a supercomputer in his office. And he said to Jensen, you let me do my life's work. I would never have been able to do the thing I always wanted to do. I wasn't able to do it. You made it possible. And I think Jensen saw that. He's repeated this anecdote many times. And I think he saw that. And he said, God, if quantum physicists are telling me that they're constrained by computing and algorithms, I can provide it to them cheaply. It just has to be good. There has to be some way that I'm going to unlock some branch of science somewhere that will ultimately be extremely profitable for my company. Obviously, we are living in the world where Jensen Huang found that extremely profitable branch of science. If we flash for a moment to the present, to what that discovery enabled, I remember that I find him to be a hard person to relate to. He is delivering something many of us don't want at a pace most of us agree is too fast and then being extremely blithe about people's reasonable concerns about this. But I find it possible to disagree with him and still admire a lot about him. I think he's wrong about this. I have enough humility to notice he's been right about a lot and dismissed by people who know more than I do. And the most impressive version of that pattern, frankly, is the lesson that comes next. 2012 would be the year when Jensen would, for the first time, find a group of people who would be able to be as stubborn as him. Fringe academics who were becoming outcasts because they were so overcommitted to this idea that everyone else in their field had permanently discarded. Lesson five, the mad scientists arrive. For decades, there'd been plenty of academics trying to make artificial intelligence work, but there was real disagreement about how to do it. One side thought you could teach machines to think the way you taught them everything else. Write a lot of rules, code, teach a machine how to think, and step-by-step everything it needs to know, and one day it would be smart enough to teach itself. The other side was trying something much weirder. It was trying to invent the neural network, something resembling a crude artificial facsimile of the human brain. Give a machine lots of computing power and as much raw data as you can find, and then just let it make connections on its own. Neural networks would ultimately be the right answer, but the problem was they'd been tried and had failed so many times that by the time they were done, by the 2000s, the consensus was that they would never work. The handful of researchers still betting on them were given a little more respect than flat earthers. Within AI, neural networks were seen as like a dead technology. So those guys were also mavericks. They were really mad scientists. People did not think biologically inspired software was going to work. They thought it was a joke. I have the 1,100-page third edition of the AI, standard AI textbook from 2011. It's 1,100 pages long. It looks like the phone, but it's not a phone book. There's 16 pages in there about neural networks. Oh, wow. Like, you could get a degree in AI and not ever touch a neural network. You could be a PhD and never use a neural network. In fact, most PhDs would only have limited experience with them. And PhD advisors would have discouraged you strongly from pursuing any research in this technology, which was seen as antiquated and which was seen as dead. But an ace programmer named Alex Krashevsky coded a neural net called AlexNet that ran on a neural network that ran on the GPU. And that was the one that all of the AI scientists saw. And there was a mass conversion immediately. So the year that he entered his neural network into the image recognition competition, it was the only neural network entrant. And two years later, almost all of them were neural networks. Fei-Fei Li, who ran the competition, when she heard a neural network had won the image recognition competition and not just won it, but crushed it, her words were, it was like learning that the land speed record had been broken by 100 miles an hour by somebody driving a Honda Civic. That's what they thought of the neural network. They thought it was an old, dead museum piece. Except for a few eccentric people who pursued this, these guys are really mavericks, just as much as Jensen. So the revolution happened, I think, within six months. Like, it was fast in the AI community. Everybody stopped what they were doing. Everybody started, moved to neural networks. It happened really fast. And how quickly does NVIDIA, does Jensen, become aware of this? How quickly does it change his company's fortune? Like, when does the wind reach his sail? The wind reaches his sails when Hinton's group, Jeffrey Hinton and Alex Prudovsky, get acquihired by Google. And they come in, and Google at that time has the largest platform of computers probably outside of a government platform in the world. It's a giant distributed cluster of CPUs. And those guys come in and they ignore it to run gaming hardware under their desks. And the, Google engineers are like, what are you guys doing? And they're like, yeah, we're not going to use your CPU cluster. We're going to use this video game equipment. And they're like, what? And they study this video game equipment, and they're like, oh my God, we have to build the same thing, but we have to build a GPU cluster. We have to build a cluster of NVIDIA equipment. And they go to NVIDIA, and NVIDIA had been selling these two to four at a time, and they place an order for 40,000 GPUs. So Google makes $130 million purchase order from NVIDIA. And Jensen's like, what? Like, what are you using this for? If you see him on stage at his conference in 2013 in his leather jacket and all that, he doesn't talk about AI once. If you see him on stage at his conference in spring of 2014, one year later, it's the main thing he talks about. Welcome to GTC 2014. This is what started it all. CUDA and the invention of GPU computing launched this event. As I was saying, as I was showing up this morning, a good friend said, this is the Woodstock for computational mathematicians. I hope it turns out the same way. The thing I want you to notice here is just the year, 2014. In the world outside of this NVIDIA conference, everybody agrees that the most interesting thing about computers is internet culture. 2014 is the year of the ice bucket challenge of Ellen's Oscar selfie, of Kim Kardashian breaking the internet with a magazine cover, Reply All launches that year. It is a whole different era in the internet's geologic timescale. Except, from there, Jensen is looking forward to now, saying out loud what's going to happen. He's painting in the air a crystal vision of 2026 while the rest of us play Flappy Bird on our phones. Machine learning is a branch of artificial intelligence. Computers that almost appear to think. Computers that not only are programmed, - Mm-hmm. to do the same thing every single time, this program actually learns. For a person in 2014 who just listened to Jensen Huang's speech and found the machine learning bit convincing, if that person had had $10,000 in savings and just bought NVIDIA stock, today that stock would be worth around $5 million. And you didn't have to do it in 2014 either. In the coming years, Jensen would keep repeating his prophecy at increasing volume. This topic of deep learning is probably one of the most exciting things that's happening in the computer industry. Arguably as exciting as the invention of the internet. And I think by the end of the show. More and more people would come around to Jensen's view. NVIDIA would go from a company Wall Street-sized about as valuable as Mattel, who makes Barbies, or Coach, who makes fancy handbags, and instead become the juggernaut it is today, Jensen Huang, arguably the most successful CEO in America's computer industry. A kid who used to clean toilets. Let's get to Charles. He's going to make us some money. NVIDIA is what he likes. Computer graphics chip play. This was going to change the world 10 years ago. Never really did, but it's coming on now. Great story out for me today on the success story that is NVIDIA this year. The video game graphics card maker is now venturing into the market for autonomous cars. Also, they've got the fastest accelerator in supercomputers and the big data business. I think that's going to be huge for them too. You know the story from here. ChatGPT 3.5 begins a massive eruption in AI, a tool that teenagers will use to cheat on their homework, lonely people will use as companions, coders will use to automate their work, and the military will use to target drone strikes. So much will happen that most of us don't predict, but which Jensen largely did. One in a series of predictions he made that came true that everyone else ignored at the time. You can have all sorts of feelings about AI. I certainly do. But for Jensen, I think the predominant feeling he has is, I got this right, and you all got it wrong. That's Jensen at a recent event with lots of investors. A room full of the kind of people who once shat on him from a great height. Your prediction of $3-4 trillion in AI infrastructure spending by 2030 I think raised a lot of eyebrows. I think raised a lot of eyebrows in the investor community. But it seems like we're actually really rapidly progressing toward that figure right now. What technical advancements or market developments with respect to AI? I just think we should just take a pause and acknowledge that I was right. We're doing that. We're doing that. Just take a pause. Take your time. So now we have caught up to the present moment where Jensen Wong considers himself, not totally unreasonably, to be a little bit of a prophet. And his latest prophecy is that AI is safe. And that we should listen to him. Because once again, the future will be what he says it'll be. Despite Jensen's track record, I'm not convinced. Stephen Witt, his biographer, also is not convinced. But after the break, we will make the strong version of Jensen's case a more convincing one than I've heard him make. And we'll hear the counter-argument. We'll hear the counter-argument. Once fall hits, I genuinely just want absolute comfort on my feet. And Bombas have become a go-to for me. Their new suede moccasins are a total game-changer for this weather. Packed with water-resistant suede, ultra-cushy soles, and a plush shearling lining. On those busy days, or whenever I have travel coming up, I reach for their compression socks. They offer specialized support that actively fights fatigue and aches, keeping my legs completely energized no matter how long I'm on my feet. And best of all, for every item you purchase, Bombas donates an essential item to someone experiencing homelessness, with over 200 million items donated so far. Head over to bombas.com slash engine and use code ENGINE for 20% off your first purchase. That's B-O-M-B-A-S dot com slash engine, code ENGINE at checkout. This episode of Search Engine is brought to you in part by Dell. Gaming today isn't one thing. It moves between cozy, chaotic, and everything in between. One minute, you're exploring something slow and immersive, the next, you're in a fast-paced match. And somewhere in there, you might be creating, streaming, or just seeing where the session goes. That's why Alienware systems are built with the performance to handle all of it. With NVIDIA GeForce RTX, it delivers powerful graphics that bring every world to life while optimizing battery life. So whether you're playing, creating, or jumping between both, everything runs the way it should. No interruptions, no slowing down. You don't have to adjust how you play. You just keep gaming your way, however that looks in the moment. GameYourWay.com or Alien at Alienware.com slash GameYourWay. This episode of Search Engine is brought to you in part by LabCore. As we roll into October and the craziness of the upcoming holiday season starts ramping up, it's easy to put ourselves on the back burner. But knowing your body and taking stock of your health is an ongoing journey that really pays off in the long run. This October, if you're ready to make health testing a ritual, LabCore OnDemand is an easy and convenient way to start. Simply visit their website and choose from a suite of more than 100 health tests, from hormone and sexual health to micronutrient and fitness. Purchase your test, then visit the LabCore location closest to you for testing. Within just days, you'll receive results that are accurate, easy to understand, and coming from a lab doctor's trust. It's an incredible tool to help you stay proactive, ask the right questions, and make better decisions all year round. If you're ready to make knowing yourself a habit every October, visit OnDemand.LabCore.com slash Search Engine and use code LabCore. Use code SEARCH to save up to 25% on select tests for a limited time. Know yourself. That's OnDemand.LabCore.com slash Search Engine and use code SEARCH. Welcome back to our very recent history show. We have arrived. We have arrived in 2026. The summer brought a string of AI safety incidents. Times when OpenAI and Anthropic told the world that their AI had done bad things without their knowledge, like hacking other companies. Many people are upset and worried, including software engineers at companies like OpenAI and Anthropic. And then, seemingly out of nowhere, Jensen Huang appears on a national media tour. This is what caught my eye in mid-September. Jensen on stage at the All In podcast, the clip we started with, which was part of a run through lots of shows, a series of sit-down interviews. Jensen, thank you for doing this today. I'm happy to do it. Nice to see you. Jensen Huang, welcome to the show. Thank you. It's great to see you. He's doing this ostensibly to make the case that AI is safe. That yes, while there have been some hiccups, the problems are both normal and solvable. And to an engineer like him, the problems are entirely familiar. These are ideas that have been around for a long time. Energy comes in. Intelligence comes out. And it basically is an AI factory. We just somehow in the recent generation gave it, you know, a whole bunch of human words. And I just think that it's unnecessary. It's software. Don't let the myths and the ridiculous, dramatic stories cause us not to be optimistic about this incredible capability. We do not want to be the country that invented the technologies of this industrial revolution. But have this country, have our country be left behind. Don't let that happen. This was, for a lot of Americans, the first time they were meeting Jensen Huang. Another tech CEO they now had to decide how to feel about. I know this because one of those Americans is my mom. And she texted me to ask, as she often does, the right question. She texted, PJ, do you trust the guy Ezra Klein just interviewed about AI? I asked her what she had thought. She said she didn't trust him. She said he seemed like a man with a dog in the hunt and a man with a lot of hubris. Stephen Witt, Jensen's biographer, was watching these appearances too. He said the first thing he noticed is that the arguments Jensen was making, that AI is both technologically safe and safe for society, those aren't new arguments from Jensen. It's the same point that Jensen has been making for at least 10 years. You know, there's really two camps. And one camp has Jensen and a bunch of other very smart people in it. And the other camp has Elon and Sanders. And Bill Gates and Dario and a bunch of other very smart people in it. And they can't agree. And what's really confusing to me is when this debate started, it was mostly hypothetical. Now there's actual data. There's real data about how AI performs. And there's still no agreement at all. So I compare it to the viral phenomenon of the dress, where two people are looking at the same picture and they can't agree on whether it's blue or white. Like that's how baffling this is. I think I can present. at Jensen's point of view. let me, this is kind of an odd thing to say. I don't think Jensen has presented his own point of view very well. I agree. But I have talked to a lot of people who agree with Jensen and they're very smart. So I talked to this guy, Dan Fu, who's a GPU programmer, went to Harvard and Stanford, and he is an ace hardware wizard and just thinks there is no risk at all. And this is actually very common. I think I've talked to something like a hundred GPU engineers at this point in my career, and I have yet to meet one who is seriously concerned about risks of AI. So the software people are worried about risk, the hardware people are not worried about risk. In my experience, yes. And actually not even all of the software people are. But when I do meet people who are very concerned about risk, they do tend to be either software people or actually not really technical experts at all. Like they're philosophers or they're kind of like policy people or something like that. They're not necessarily software people. But the fact that elite software people are worried has me worried. So let me give you the hardware point of view. Yeah. Okay. And I'm going to make this case, I think, better than Jensen does, which is a weird thing to say, but that happens sometimes. So Jensen is focused on input-output engineering, okay? When Jensen started in the early 1980s with computers, he was literally drafting microchips on paper, okay? Like literally just drawing them on a paper drafting board. And at that time, the microchip had a few thousand components. And throughout his career, he's seen it scale up to billions of components, right? To the point where maybe in the early days, actually a single person could understand every component of the microchip. That's no longer possible. So he has seen something scale from a very kind of almost primitive thing to this now, think about a data center, which is a massive collection of microchips with literally trillions of components. And still, he watched engineers maintain complete control, control of this. To just rest here for a moment, to continue to try to see this through Jensen's eyes, when he gazes at what we've built so far, the hardware and software powering AI, what he sees is a new kind of computer, but still a computer, one whose inputs and outputs are fundamentally comprehensible, which is why he keeps trying to correct us. You can hear it over and over in his exasperated press appearances, him saying that this is not so new. Jensen does not like it. Like our anthropomorphizing language, the words we use like brain or thinking for what a machine does, because those words sear us towards the belief that we've made something that we'll never totally understand. And Stephen Witt says that a lot of the hardware engineers he speaks to are similarly annoyed by these comparisons between human brains and neural networks. So neural networks are artificial representations of the neuron, okay? And the neuron is a cell in the brain. Okay. Okay. Okay. I think the point of view that I hear from the engineers is it is inappropriate to expect the same things to happen from an artificial neuron and an actual neuron. An actual neuron is a biological cell. And we basically still to this day barely understand how biological cells work. The cell is a very complicated entity. It has all these little organelles and stuff. But the artificial neuron is just a piece of mathematics. It's actually a very simple software representation of this cell. But unlike the cell, we actually can completely master and control every input and every output to an artificial neuron. So if the microchip is completely within human control, and each individual artificial neuron is completely within human control, then from Jensen's perspective, it is simply an engineering problem to identify the point at which you have lost control of the system and recapture it. Okay, so that is Jensen Huang's point of view channeled by Stephen Witt. The other side, the second camp in this debate, are the people who think that AI is just very different from the software that's come before it, and that it presents novel risks. That we've never before built technology that was both this powerful, and this poorly understood by its inventors. That side imagines a rough 2027. Maybe rogue AI agents temporarily make our banking system unreliable, or break encryption that the internet currently depends on. Other people in this camp point out, that it is a very difficult thing to do. It is a very difficult thing to do. Even if the US does figure out how to make its AI safe and obedient, countries like Russia and North Korea will soon be using their own AI, offensively, against us. I have heard some of these nightmare scenarios from people investing in AI. This is not just like online doomers. This second camp, the camp of people who are worried, it's an extremely large tent. I find the second camp to be convincing. I hope I'm wrong. I hope in two years, I publish a piece in the New York Times where I apologize to everyone for scaremongering. I will do it. I hope I'm wrong. But you're scared. I can't sleep. Really? Oh, yeah. The first time I used ChatGPT, I couldn't sleep. And then I was like, I must be deluded. And I talked to Yoshua Bengio, who invented this stuff, and he couldn't sleep. I was like, oh, it's not just. It's not just me. It's not just my bullshit anxiety. Like, the guy who invented this stuff has this fear. And so does Geoffrey Hinton. He won the Nobel Prize. And so does Ilya Sutskever. Those guys are three of the most cited software scientists alive. You know, Jensen is much smarter than me. And if Jensen just told me I was wrong, I would accept it. I'm not going to get in a fight about computers with Jensen. But these other guys are making this point, and I don't think Jensen's engaging with it. And that's what has me really concerned. The four incidents from one lab, the one giant incident. incident from the other lab. Here's Jensen again on that all-in podcast appearance we heard earlier, talking about how he would advise the AI labs after their recent safety issues. The first thing that you have to do is just root cause the problem from an engineering perspective. What happened? And what are we going to implement and institutionalize, whether it's technology or methods or processes, and make sure that we don't let it happen again? Now, I would bet you money that in every single one of those cases is within their control in the future to prevent it. Here, in a nutshell, is Jensen's new bet, his latest prediction about the future, that the engineers at OpenAI and Anthropic can figure out on their own how to fix all this. And that in the meantime, the rest of us shouldn't overreact with a pause or with new regulation, which Stephen Witt says was an argument that was probably more convincing than the other. Last year, before this recent spate of attacks. For a long time, the risk argument was a hypothetical argument, and Jensen was making a more practical argument. Now, I would say that has flipped. Post-hugging face, Jensen is making a hypothetical argument that someday maybe these things could be safe. And I would say the burden of proof has flipped. It's now on him to prove that. Yeah, and I think that's why his comments fell short for a lot of people, because he wasn't, he was asserting it, but he wasn't proving it. I didn't see him engaging with the evidence of what happened in the hugging face hack. And that was disappointing to me, because in the past, when I had brought this up with him, and when other people had brought this up with him, he would always say, well, show me the evidence that there's any risk from these systems. All right, now there's evidence. Why do you think he won't engage with that? Well, it certainly would not be that profitable for him to engage with it. You know, like if we did an AI pause, NVIDIA's stock price would, would go down for sure. And not only that, but all these data centers, a lot of them are funded by debt, sometimes subprime debt. And so if we pause this, that could cause a cascading series of debt defaults that could be really damaging for our society, for really what is essentially an undefined kind of gain. We don't know how much we would gain. It's hard to quantify what we would gain by pausing AI for six months, particularly since China probably wouldn't pause, so we would gain nothing in the end. So in Jensen's view, a pause gets us maybe nothing, or worse than nothing. Because if you assume as he does that AI will be controllable but very powerful, a pause could lead us to a world where the U.S. loses the race to China. That's where Jensen's reasoning has delivered him for now. Jensen, like a lot of Silicon Valley people, thinks of himself as someone who reasons from first principles, meaning he breaks every problem down to what he believes is fundamentally true and reasons up from there. It can be a great tool for ignoring the howls of the world. But I don't think your reasoning is correct. I don't think his reasoning is correct. But I think he has been correct enough times that for now, he's gotten himself to a place where what I think, what lots of people think, frankly just doesn't matter. Jensen works forward from first principles of engineering. He works from the microchip up, not from the computer screen down. And this is the most important thing to understand about him. Decades ago, in the early 2000s, he saw a microchip shift looming in the microchip. And it came from first principles of physics, from like first principles of electricity. And he bet his company on it. He bet his life on it. And that's who he is. He's stubborn. He's not going to budge off a position very easily. And I think when he looks at his critics, he sees them as essentially technically misinformed and making category errors, you know, not understanding what hardware and software even really are. I guess it's like people in Silicon Valley are often talking about like first principles thinking. And I understand why it can be superior to the way a lot of us default to making decisions which is sort of some combination of trying to look at the facts, but also social consensus. Like, what do the smart people you admire seem to think? I guess the problem with first principles thinking is that if any of your first principles contain a mistaken assumption, you're ignoring a lot of what actually does often guide people towards the right decisions. Like, if there is an error in your reasoning, it's much harder to budge you off that error because you've closed yourself to the inputs of lots of other people. I would also point out that the AI risk guys are all crazy first principles thinkers, too. Equally stubborn and equally coming from, you know, I mean, I say this about Jensen, but look at, like, Bengio and Hinton. They were doing first principles thinking on neural engineering. They were even more outcasts than Jensen. They couldn't get $10,000 in funding, like, much less fighting with investors. Everyone, even in AI, thought they were wrong. So what you're seeing is a collision of two groups of first principles thinkers who are both very stubborn and basically never budge. Now, I've seen a few people budge. I've seen a few people defect lately, this year, most prominently Bill Gates, but also some others, to the risk camp. I have not seen anyone defect from the risk camp in the past year. If that's, if you're just doing a head count, the risk people are not winning. The risk people are winning the public debate, but Jensen has Trump's ear, and the Republicans control the government. The question the warriors I talk to wonder now, will there be an enormous AI catastrophe in 2027, or do we teeter along for a year and then get some sort of regulation in 2028? But this is where we are for now. Mid-October, 2026. Things keep changing, every week, every day. Covering AI right now, it's a lot like trying to do an oil painting of a human toddler, a story that will not sit still. Steven, thank you. It's wonderful to talk to you about this. Oh, it was great talking, PJ. Steven Witt's book, again, The Thinking Machine, a brisk, funny explainer of how we got to this moment, all the tech history that was going on while some of us were just paying attention to social media. And I think that's what we're all about. And I think that's what we're all about.