Today is March 5th, 2026, and my guest is author Stephen Witt. His latest book, and the subject of today's conversation, is The Thinking Machine: Jensen Huang, NVIDIA, and the World's Most Coveted Microchip. Stephen, welcome to EconTalk.
Oh, thank you so much for having me.
So, this is really an extraordinary book. It's a history, indirectly, of I'd say the last 30 years or so of the digital age. It's an incredible portrait of a visionary and his company. I think some of my listeners and viewers will have heard of NVIDIA but won't know much about it other than perhaps that it's the most valuable company in the world measured by market capitalization. Most of them, I don't think, will know who Jensen Huang is. And your book is a wonderful introduction to both Huang and NVIDIA. Let's start by going to the beginning. Jensen Huang, the founder and CEO. What is his beginning? He comes to the United States under unusual circumstances as a young, as a kid.
Yeah. Jensen was born in Taiwan in 1963. He moved to Thailand when he was about five years old. His father was an engineer who worked at a petroleum company. And then he came to the US when he was about 10. In 1973, there was a coup d'état in Thailand and it was violent. There were tanks in the street and his parents said, "Let's get the kids out of here." So they sent Jensen and his older brother ahead to the US. They were planning to move to the US, but Jensen showed up alone to live with his uncle. His uncle didn't know what to do with these kids, and so he looked for a boarding school that would take them. And he found the Oneida Baptist Institute in rural Kentucky in the United States, which I think he may have thought was a prestigious preparatory academy. But the boys show up there, sight unseen, two foreigners barely speaking English, and they realize they've arrived at a reform school for juvenile delinquents.
The grounds are littered with cigarette butts, and all the kids are basically criminals. And Jensen's first night there, he's put with a 17-year-old roommate who's recently been stabbed in a knife fight. So the kids are carrying switchblades. They're pretty poor backgrounds for the most part, mostly the children of tobacco farmers or coal miners, almost exclusively white. And at this time, actually, the Vietnam War is still going on, so a lot of racism against Asians. Jensen is called all sorts of racial slurs. It's a very difficult environment. But amazingly, Jensen thrives here. He does well. He's a good student. He's always been a good student, but he actually becomes one of the most popular students in the school and even a leader. And then sort of when his parents returned a couple years later and he was back to Portland, he's had this kind of unusual experience of being cast into basically a knife fight and surviving, right? And this kind of sticks with him. He would later say this is one of the most important things that ever happens to him. But still, I mean, despite surviving this kind of juvenile delinquent academy, he's still a nerd, you know, he's still front of the class, top grades, top test scores, ends up majoring in electrical engineering and gets a job in 1983 in Silicon Valley designing microchips, not software, but electrical engineering. And so really starts building computers from the transistor up, from the circuit board up, and has done so continuously ever since, ultimately revolutionizing what the computer can be.
And at one point, he's an undergrad at Oregon State. He ends up getting, I think, a master's at Stanford. Is that correct?
And at some point, he forms his own company. Is it 1993? Is that the right thing?
1993, the booth at Denny's Diner. If you're not familiar with the United States, Denny's is, you know, it's known more for the adequacy of its food.
Yeah. It's open all the time. It's open 24 hours. So, very popular.
So he forms this company in 1993 with two other folks, and what's the goal of the company? Why did they break off from where they had been before, and why did they break off? What was the vision that that company was going to fulfill? Because it's quite surprising.
Jensen's co-founders, Chris Malachowski and Curtis Priem, had been at Sun Microsystems, and Jensen had been at LSI Logic, and they knew each other actually because Sun was LSI's customer. So Jensen was essentially a sales guy, and Chris and Curtis were buying stuff. Now, that's a little demeaning to Jensen. Yeah, technically he worked in sales, but he still had this master's in electrical engineering. And in fact, Curtis and Chris were his most sophisticated and technically demanding customers. What they wanted to do was design a microchip that could work as a three-dimensional graphics controller for video games. Readers of a particular age will remember the Nintendo 64, which came out around this time and moved us from side-scrolling video games where we kind of moved across a static map to ones that were rendered in three dimensions with almost like a camera in real time, and we could move our point of view around to see what was going on. It was a radical upgrade. Basically, what we're doing is we're drawing points in space and then painting in the textures in them to make these kind of blocky polygonal figures and have them move around. So it's really actually a pretty tough math and physics problem. You wouldn't necessarily know that from what the game designers use to build with it, which is mostly, you know, gunfights and car chases and gore and zombies. But they roll out this 3D graphics chip. Initially, it's a flop. They almost go out of business, actually, the three of them. NVIDIA almost went out of business twice in the early days. But they managed to stabilize and find a niche market for this 3D graphics controller, especially for the video game Quake, which was a big hit in 1996 or 1997, kind of the first three-dimensional shooting game. The strange part about this, of course, is they're going to end up changing the world in rather extraordinary ways. But at the time, they were trying to convince their companies, before they founded their own company, that there was money to be made in video games, which was in 1993 or '92 when they were having that argument before they founded NVIDIA.
An absurd argument. They couldn't get anybody to fund them because the projected size of that market—and I've heard Jensen Huang talk about this in a talk he gave, I think at Stanford—the size of that market was estimated at zero, which is a small market, very small number, zero. And so no one really wanted to take a chance on it, but they believed in it. Why did they leave their comfortable jobs? And at this point, I think Jensen Huang had just become a husband and maybe a father. He had to be thinking about the future. Why did he feel that was a risk worth taking in 1993?
You have to be a little careful with Jensen, who has a tendency to reconstruct the past to fit the story that he wants to tell.
It was much larger than zero. There were 30 or 40 companies attempting to do exactly this thing. The market, they will tell you, Jensen will tell you, "Oh, there's no market." That was not true. It was obvious post-Doom, post-Mist that the PC gaming market was going to be big money. The size the market was going to be. The challenge there was that everyone saw that. I think there were 40-plus companies by 1994, 1995 trying to do the same thing. And in fact, NVIDIA was all the way at the back of the line. They were basically in last place. And as Jensen has described it, this was actually an opportunity for him, because when you're in last place, you can do anything you want. There's no real risk to it. You're going to go out of business anyways. So Jensen threw this—it's true—Jensen threw this huge Hail Mary to design his chips in a new way using a simulator rather than build a prototype. And this allowed them to skip about six months of work and actually arrive first to the market with what was basically, even by his own admission, an inferior knockoff product, but it came out faster, and that was enough to keep the company alive in those early days. I think the zero billion market came later. I think that Jensen did not identify that till later, but he wants a story.
Yeah, it's a nice story. It's a good story. This is one thing I noticed with Jensen, interviewing him a lot. You know, he's got great stories. He's got great anecdotes. The details of those anecdotes tend to shift around a bit over time. So as a journalist, you have to be a little careful with them. I always find that story interesting because, you know, the theme of it is, you know, trust—for some people anyway—and maybe he alludes to this—is, you know, trust your intuition. Don't listen to the so-called experts. And of course, that suffers from survivor bias. So he survived. He took a leap that worked out. Many, many other people, other potential founders or actual founders, took a leap for a market that nobody believed in and didn't make it. But anyway, I'll tell you what I think the lesson of that early video game era is.
The survivor—yes, there's survivor bias—but who gets to be the survivor? Jensen and his team would go into their whiteboard in their office and they would list out all of their competitors, and they would list out who the best engineers were working at each of their competitors, and then they would come up with strategic plans to poach that specific best engineer and get them to come work at NVIDIA. They called it brain extraction. And once they did, once they extracted that other company's brain, they would typically collapse very quickly because they no longer had their best person working for them. Jensen knows how to win in a knife fight, and the other guys didn't. Jensen had that ruthless killer instinct that you sometimes need in business, and he really made that the culture of NVIDIA. And the other guys really came from the gaming space. They were wearing flip-flops to the office. You know, it was fun for them. They didn't think like killers. Jensen thought like a killer. And of those 29, 30, 40 companies that were out there, by 2000, there was basically just one left: NVIDIA.
So he won like the battle royale of the 3D graphics controller market.
If you talk to people who experienced the other side of that, they were like, "God, he was just ruthless." I mean, he was just a shark. He destroyed our company without mercy, without pity. And, you know, that's not the story he's going to tell.
But that's what happened for sure.
Why did they call it NVIDIA?
Yeah. They wanted to make their competitors green with envy. They wanted people to envy them. And originally they called the company Envision, but this turned out to be the manufacturer of environmentally friendly toilet paper. So they went back to the drawing board. I think Chris, one of the co-founders, had a Latin dictionary, and Nvidia is the Latin word for envy. And so they called it Nvidia. That's where it comes from. And, you know, I would say they have done that. They have made their competitors green with envy.
And just to fast forward to the present, just to give people who don't follow this closely, they are the number one, as I mentioned, market cap company in the world. They're worth over four trillion, trillion with a T, trillion dollars.
Apple is second, the last I looked, about a week ago, and they're at 3-something, 3.6 and something like that. Well, not just that, but in fact, they recently hit the highest single concentration of any stock in the S&P 500 since Standard & Poor's started keeping track. So in real terms, inflation-adjusted, not only the most valuable company in the world, they're by some measures the most valuable company in history.
Yeah. So let's talk a little bit about how they got there. So they start off, and then we're going to talk about some of the personal issues and how the creativity of NVIDIA created—I think soon we will say—created the modern world, which is a frightening, weird thing to say, but I think it might be true. It's certainly the theme of your book. So they're in the gaming world, and somewhere along the line—and they're pretty good, but as you say, they have a lot of competitors. They do poach some engineers from them, but there's still lots of competition. But at some point, they realize that they can use their engineering ability to create chips that will facilitate other things besides car chases and killing monsters. And what's that transition like? And then bring us up to 2013, when they realize that there's this new thing coming called artificial intelligence, and that they might be able to contribute to it.
You know, we could divide the NVIDIA story into three phases. The first is the first, I would say, eight years or so, as they go from spitballing in a diner to actually joining the S&P 500 in 2001. And all of that is just the success and rise of their gaming product, but also the organic growth of the video game market, which was huge. Around 2001, they had started to notice that these graphics chips didn't work like normal computer chips. They were much more what is called arithmetically dense. So what that means, basically, it's parallel computing or accelerator computing. The microchip will pulse with each clock cycle. On a classic Intel CPU, about 3% of the microchip lights up with each cycle. So about 3% of the silicon is actually active with each pulse. For an NVIDIA chip, it was more like 30 or 40%. So they were doing a much greater volume of calculations per second, per tick. And you might say, why don't all chips work that way? Well, the answer is that the parallel computer approach is much harder to program, but when you do it right, it's much more dense and much faster. And what started to happen was that scientists noticed this: quantum physicists, people doing medical imaging, people with needs for very high-powered, high-performance computing. And they actually started to hack the video game circuits, the video game programming, just to get to these circuits. And Jensen saw this, and he's like, "Well, they shouldn't be hacking our video game stuff. I will build them a platform. I will build them software so that they can do science on these graphics processing units, on these GPUs." And that was a platform called CUDA. It was free. It wasn't open source, but it was free. It was an open platform just for downloading it. And you would do it, and then you could do science on GPUs. You could use them to do like medical imaging, quantum computing, all of this stuff. And so Jensen started to do all this outreach to scientists during phase two. Now they made a huge list of potential applications for parallel computing: weather forecasting, oil prospecting, all of these potential customers. And way down at the bottom of the list, in this tiny little use case that they barely considered, was something called computer vision.
So by 2008 or 2009, this program is up and running. They've got, you know, a few, let's say, 100,000 downloads or so per year. But it's not really a success, and it's extremely resource-intensive. And this is the zero billion market. This is Jensen inventing this platform from scratch, basically, and losing money to do so. Now, you might ask, who is this for, right? Really? Well, it's not actually for mainstream research scientists, because those guys are well-funded. Those guys can afford time on a classical supercomputer. Who needs a jerry-rigged, homebrew supercomputer built out of graphics cards? Well, it's a marginalized scientist. It's a scientist without a lot of research money. It's a mad scientist, basically. And at the very fringes of science at this time, in the computer vision world, the very smallest, most kind of meager customer were these guys doing this form of AI. And it was an unpopular form of AI that did something called a neural net, which simulated the firing of the neurons in the human brain. I have the 2011 artificial intelligence textbook by Russell and Norvig on my shelf. It's about 1,100 pages long. Of those 1,100 pages, about 16 total are devoted to neural nets. And that was the state of the art in 2011. This was a dead technology. Nobody believed in neural nets. And these guys were fringe mad scientists working on the absolute limits of computer science, but Jensen had built them this tool. And around 2010, 2011, and 2012, the mad scientists get a hold of two GPUs, two NVIDIA graphics cards, total cost about $1,000, and they use CUDA, this platform, to jerry-rig a supercomputer. They start simulating the neurons of the human brain on this kind of tiny homebrew supercomputer, and they usher in a scientific revolution.
Because as it turns out, the thing that the neural nets were missing was just firepower. They were just missing computing power. And when you unite these two technologies, you have an extraordinary breakthrough. That was known as AlexNet, where suddenly computers, which had struggled to label images, this computer vision program, suddenly computers can see and they can identify images correctly with an unprecedented level of accuracy on basically the cheapest commodity—no, not commodity, but cheap hardware that you can just buy at Best Buy. You can have a revolution in computer science. You don't need a supercomputer. And this inaugurates the third phase of NVIDIA, which is the AI phase. Now, up until this time, NVIDIA had actually been struggling. If you look at the period from 2001 to say 2012 or 2013, NVIDIA stock goes nowhere. And in fact, Jensen was not well regarded. There were activist investors taking positions in NVIDIA demanding change. It was a stagnant company. They were having to reform the board, and the reason was this supercomputing effort, this platform. Wall Street did not believe in it. It was the zero billion dollar market. Jensen was spending a billion dollars plus in research money a year to pursue what looked like these marginal weirdos, weird customers, weird ideas, right? I can't express to you how fringe these AI guys were. They were not popular. Even in the AI community, they weren't popular. And AI itself was viewed as a career graveyard at that time. I mean, you went into AI because you wanted to be a research academic. You didn't start a company. The amount of total venture capital investment in 2010 into AI was closer to zero than any other meaningful number. But they had this breakthrough, and it started to build the modern world, inaugurating phase three of NVIDIA, which basically was, you know, a rocket ship to planet money. You know, it was just a galaxy made of money, as it turns out.
Yeah. I guess we'll get into that. Yeah. I mean, one of the—I'm going to give you the incorrect interpretation. I'll let you correct it before I get very far. In your book, I looked at NVIDIA's market cap some number of years ago, something before 2020, it was maybe around 2020, and I may have the numbers wrong, but it doesn't matter. It was worth at that time—remember, this is in the middle of this phase you're talking about—it was worth something like $300 billion. And I think they were the 15th most valuable company in the world, which is no small feat. And so you could describe NVIDIA's success as the following. And this is not—I want you to refute this, but this was my impression as I started to read more about NVIDIA as a newcomer and just starting your book. I thought, well, it's a company that starts with this graphics program that's used for gaming, and they get lucky because it turns out there's a big demand for this that was unexpected in things outside of games, and they profited from that. And so when AI came along, which was lucky for them, the demands on computing were so intense, they had the best chip. So they fell into this extraordinary frenzy of VC investment that we're in the middle of right now. And it may—some people think it won't last very long, and some people think it's just getting started. And that's beyond the scope of this conversation. But obviously NVIDIA's market capitalization, the value of the company, reflects this incredible surge in artificial intelligence work. And after I read your book, I realized that's not the right way to look at it. And so for starters, they weren't particularly profitable in advance of this revolution, but it wasn't just luck. Obviously, there's some luck involved, but they helped create it. Explain.
They built the modern world. I mostly reject the narrative that they got lucky. I mean, yeah, they didn't identify AI as their big customers when they were building CUDA. But they were very deliberately trying to unlock new branches of science. I mean, that's why you do this, right? Maybe you can't predict in advance exactly which new branch of science that you're going to unlock, but they were certainly trying to unlock some kind of scientific revolution with the understanding that when that happens, now you can build a whole platform around that and you can build a whole ecosystem. You know, it wasn't a charity. From the start, they were engineering what they called vendor lock into the parallel computing platform, into the CUDA system. Once you learned how to do science on a GPU, you were basically locked into this relatively expensive—actually quite expensive—hardware upgrade ladder basically forever. And that had long been true of the gamers. You know, NVIDIA had always been good at going into the video game companies and even like having a guy kind of embedded in the video game company, helping the game developers optimize their game for NVIDIA hardware so that when it came out, all the gamers could put, you know, "Powered by an NVIDIA chip. That's the best platform for this particular game." And all the gamers would go out and spend $1,000 on NVIDIA hardware. And then they did the same for scientists. As the AI revolution started, all of the science grew, the entire ecosystem grew around this particular chip, right down to the guts of the machine. They struggled a bit, actually, getting scientists to use CUDA sometimes, because they would already have their own programs, and the scientists would be like, "What do you mean I'm going to refactor a million lines of code to do weather forecasting on a GPU? I'm not going to do that. It's going to take forever. It's going to take years. I don't want to do that." But with AI, there was nothing to refactor. There was nothing to rewrite. It was being built from scratch around this platform, right? It was good, actually, that it was kind of a backwater, because that meant they could just rebuild everything and build it for the first time from scratch. And so that turned out to be enormously profitable. Now, along the way, two things happened that really turbocharged NVIDIA.
One is, and this came as a shock to a lot of people, AI is a brute force problem. It's basically linear. The more computing power you throw at the computer, the smarter it gets. And the demand for AI is functionally unlimited. Why would you not want something more intelligent? You're never satisfied. You always want a more intelligent system. So the smarter the computer gets, kind of counterintuitively, actually increases demand as new potential applications are unlocked. The second thing that really helped NVIDIA was in 2017 at Google, they introduced a new deep learning or AI kind of architecture, a new blueprint for AI called a transformer, which was basically a funnel that took massive, massive amounts of data and distilled intelligence from it. The best-known transformer model is the generalized pre-trained transformer, or GPT, ChatGPT—that's where that comes from. Now, this is great for NVIDIA that this works, because it really turns AI into like heavy industry. Basically, giant barns called data centers, giant warehouses that have to be full of NVIDIA equipment running 24/7 around the clock to distill intelligence, to distill insights from massive, massive, massive amounts of data. It's almost like an oil refinery or something. It's like this big heavy industry project. And for NVIDIA, this is like the best thing that can ever happen, because it hundred- or even thousand-x's the demand for their microchips. And so this is the point at which NVIDIA, which had actually been doing well already based on the computer vision results, starts to really rocket from being, you know, a 5-10 billion market capitalization to 500 billion and then ultimately 5 trillion. Jensen calls these data centers AI factories, where data goes in and intelligence comes out.
But why is it that in these server farms, this ugly, anonymous, unbranded thing in a relatively deserted part of, say, America, why is it, quote, full of NVIDIA equipment? What's the alternative if NVIDIA disappeared today? What would replace it, if anything? And why is it not as good?
Yeah. So why can't we just use an AMD chip?
Well, we can. A what chip? An AMD?
