The Thinking Machine

The Thinking Machine - Stephen Witt

Introduction:

“This is the story of how a niche vendor of video game hardware became the most valuable company in the world. It is the story of a stubborn entrepreneur who pushed his radical vision for computing for thirty years, in the process becoming one of the wealthiest men alive. It is the story of a revolution in silicon and the small group of renegade engineers who defied Wall Street to make it happen. And it is the story of the birth of an awesome and terrifying new category of artificial intelligence, whose long-term implications for the human species cannot be known. At the center of this story is a propulsive, mercurial, brilliant, and extraordinarily dedicated man. His name is Jensen Huang, and his thirty-two-year tenure is the longest of any technology CEO in the S&P 500.

Huang is a visionary inventor whose familiarity with the inner workings of electronic circuitry approaches a kind of intimacy. He reasons from first principles about what microchips can do today, then gambles with great conviction on what they will do tomorrow.

The success rate of parallel computing was zero percent before we came along,” Huang said, rattling off a list of forgotten start-ups. “Literally zero. Everyone who tried to make it into a business had failed.” Huang ignored this dismal record, pursuing his unconventional vision in open defiance of Wall Street for more than a decade. He looked for customers besides gamers, ones who needed a lot of computing power—weather forecasters, radiologists, deep-water oil prospectors, that sort of thing. During this time, Nvidia’s stock price floundered, and he had to fend off corporate raiders to retain his job.

With a near-monopoly on the hardware, Huang is arguably the most powerful person in AI. Certainly, he’s made more money from it than anyone else. In the strike-it-rich tradition, he most closely resembles California’s first millionaire, Samuel Brannan, the celebrated vendor of prospecting supplies who lived in San Francisco in 1849. Except rather than shovels, Huang sells $30,000 AI-training chips that contain one hundred billion transistors. The wait time to purchase his latest hardware is currently more than a year, and on the Chinese black market, his chips sell for double the price.

Huang doesn’t think like a businessman. He thinks like an engineer, breaking down difficult concepts into simple principles, then leveraging those principles to great effect. “I do everything I can not to go out of business,” he said at breakfast. “I do everything I can not to fail.

I found Huang to be an elusive subject, in some ways the most difficult I’ve ever reported on. He hates talking about himself and once responded to one of my questions by physically running away. Before this book was commissioned, I had written a magazine profile of Huang for The New Yorker. Huang told me he hadn’t read it, and had no intention of ever doing so. Informed that I was writing a biography of him, he responded, “I hope I die before it comes out.”

Still, Huang offered me access to a great number of people to report this book. I spoke with almost two hundred people, including his employees, his cofounders, his rivals, and several of his oldest friends. The beloved and even somewhat goofy family man who emerged from these interviews bore little resemblance to the unapologetically carnivorous executive who made Nvidia succeed, but it is these same attachments that spur Huang’s ambition: he spoke frankly with me of his insecurities, his fear of letting his employees down, his fear of bringing shame to the family name. Some executives speak of profit as “keeping score,” but not Huang; for him, the money is only temporary insurance against some future calamity. There was something a little touching about hearing a man worth a hundred billion dollars talk in this way.

Of course, Huang would work hard anyway. It is in his nature. If there is a theme to his life, it is amplification; he has executed on the same simple precepts of diligence, courage, and mastery of fundamentals again and again and again, to greater and greater effect. I was surprised to learn how much of the man he later became was present in the immigrant child arriving unaccompanied by his parents in the United States in 1973 to an environment so unconducive to flourishing that it seems a miracle he survived it. To understand Huang fully, we begin not at Denny’s restaurant, nor in the giant cathedrals of technology he later commissioned, but at this tiny rural school.

ONE: The Bridge

“As graduation approached, Huang got a job at Denny’s. The nationwide restaurant chain was known in that era for its burnt coffee, its reconstituted powdered eggs, its reheated sausage patties, and its round-the-clock operating hours. Huang loved the place. He began as a dishwasher and worked his way up to server. “I find that I think best when I’m under

adversity. When the world is just falling apart, I actually think my heart rate goes down,” he later said. “Maybe it’s Denny’s. As a waiter, you’ve got to deal with rush hour. Anyone who’s dealt with rush hour in a restaurant knows what I’m talking about.

By the time he graduated from high school, Huang had skipped a grade, was a nationally competitive athlete, and had a near-perfect GPA. Yet he opted out of the college-admissions scramble, choosing to enroll at nearby Oregon State University. There wasn’t much thought behind the decision, Huang told me, and no pressure from his parents to go anywhere else. His high school buddy Dean Verheiden was a legacy Oregon State student, and Huang chose to go as well. “I just followed my best friend,” he said.

TWO: Large-Scale Integration

“When LSI’s customers wanted new functions, most of the designers would simply respond, “There’s no way.” Huang would say, “Let me see what I can do.”

Huang would spend hours fiddling with the simulator, attempting to arrange the list of components to enable what the customer wanted. This was painstaking work, conducted without the assistance of graphical user interfaces or even color monitors. His focus was admirable, but Horstmann knew many engineers who could become similarly absorbed in technical problems; what set Huang apart was his ability to avoid dead ends. “Similar people, they get lost, right?” Horstmann said. “They just get lost in these deep, deep ratholes. He doesn’t. He has a great sense of seeing when a problem has reached a certain level of complexity, and he can’t easily make further progress, and he has to go in a different direction.”

LSI’s most demanding customers were the computer-graphics designers, whose appetite for faster silicon knew no point of satiation. To serve them, Horstmann, with Huang’s encouragement, began signing contracts to deliver products that, internally, the two had no idea if LSI could actually make. Older engineers advised the two to be more cautious. Do you know what you’re doing? they’d say. If this fails, it may be the end of your career. “It was true, but that never troubled us,” Horstmann said. Almost everything Horstmann and Huang promised was eventually delivered.

Huang eschewed drama and led by example, driving himself hard, refraining from gossip, and carefully apportioning credit for good work. If a product was going to be late or if LSI couldn’t deliver on some promised function, Huang would immediately provide a detailed description of what had gone wrong, who was responsible, and what he was doing to fix it.

“When he said he was going to do something, there was a reasonable likelihood that he would actually do it, y’know?” Malachowsky said. Malachowsky struggled to think of other Silicon Valley product managers who fit that description.

If Huang had a flaw, it was that he embraced candor in the extreme, sometimes crossing into the territory of insult. The bluntness was part of his charm, of course, but it could leave people’s feelings hurt. He didn’t have much patience for people who disagreed with him, and he also seemed genuinely surprised that there were people working in his industry who didn’t want to spend fourteen hours a day fiddling with the circuit simulator. Of course, for quarrelsome workaholics like Priem and Malachowsky, these traits were only further evidence of Jensen’s managerial fitness.

THREE: New Venture

“Sun Microsystems had declined to pursue the consumer marketplace for PC video game hardware. So had Lori’s former employer Silicon Graphics, the industry leader in three-dimensional graphics. (Employees there were busy animating the CGI dinosaurs for Jurassic Park.) The failure of the major players to invest in PC gaming created a vacuum in the marketplace, which a brigade of start-up businesses was now scrambling to fill.

The concept was to take the hardware used to paint the wire-frame skeletons of model airplanes and dinosaurs and repurpose it to create controllable animated figures in three-dimensional games.

The product was known as a “graphics accelerator,” and at least thirty-five competitors were trying to build one. Huang worried there was no space for a thirty-sixth. The leading expert in computer graphics was Jon Peddie, who had written several textbooks on the topic. Huang had reached out to Peddie to get a sense of the market, and the two soon became friends, with Huang calling incessantly, asking questions late into the night. Peddie advised Huang that the space was too crowded and that many of the best engineers were already working for other start-ups. “I told him not to do it,” Peddie said. “That was the best advice he never took.

The office of the CEO was located on the first floor at a small circular table adjacent to the kitchen. By design or happenstance, Jensen had placed himself at the center of the natural flow of foot traffic—employees going to the refrigerator for drinks or snacks had to pass him. No matter how powerful he grew, he would seek to remain in the center of traffic for the rest of his career.

What he did have was the support of LSI’s founder, Wilf Corrigan. The following day, Huang and Priem traveled to Sequoia’s offices to pitch Don Valentine, the firm’s famously blunt founder. (Valentine’s favorite question for start-ups was “Who cares?”) The pitch went badly, with Huang fumbling over his presentation and Priem interrupting with irrelevant technical asides. After this uninspiring performance, Valentine took Huang aside. “Well, that wasn’t very good,” he said. “But Wilf Corrigan says I have to fund you, so you’re in business.

Coxe and Stevens agreed that it was Huang, specifically, rather than Nvidia’s proposal that attracted their attention. “The reason we backed these dudes is because they were world-class computer scientists,” Coxe said. “The average CEO will try to listen to the customer, but in computing, that’s a big mistake, because customers just don’t know what’s possible. They just don’t know what can be done!” Coxe observed that Intel and Microsoft had later struggled under more conventional management: “Jensen, from the beginning, was a world-class engineer who could see what was possible.

FOUR: Thirty Days

“When David Kirk arrived at Nvidia’s offices for the first time, in 1996, he could see at once that the company was doomed. Kirk was a graphics expert who consulted throughout the Valley, which was like being a connoisseur of failure. He had watched a great many start-ups falter, including his own, and Nvidia exhibited all of the symptoms of a company hurtling toward insolvency. The employees looked haggard and demoralized, the quirky product didn’t fit with the market, and the supposedly chummy founders were now deadlocked in a “technical discussion” that was obviously more than just a discussion and obviously about more than just technology.

Following the announcement, Huang reduced the head count from more than a hundred general staff to a skeleton roster of thirty-five engineers. Joining in the aftermath, Kirk walked into an eerie, half-abandoned office.

After a few weeks programming the emulator, Kirk realized he had a second, tacit role at Nvidia: curbing the technical ambitions of cofounder Curtis Priem. Kirk had invented the quadratic-mapping technique used in the NV1, but when he arrived at Nvidia, he advised the company to abandon it. “It was just an idea I had,” Kirk said. “I have lots of ideas.” But this only made Priem promote quadratic mapping more aggressively. Priem was a purist who dismissed technical compromises as spineless concessions to the money guys. “The way Curtis thinks is for the end point,” Malachowsky said. “But he doesn’t really have it in his makeup to, like, stay in business.”

Kirk soon realized that the abstruse question of whether or not to use quadratic mapping was a proxy for the more interesting question of who was actually in charge at Nvidia.

Spectators were important to Huang—when he dressed down an employee, he usually did so in public so that others could learn from the experience. (“Failure must be shared,” Huang said.) If a project was delayed, Huang would command the person responsible to stand up and explain to the audience, in detail, every single thing that had gone wrong.

Huang would then deliver a withering analysis of their performance. Such corporate struggle sessions were not for everyone. “You can kind of see right away who is going to last here, and who is not,” Diercks said. “If someone starts getting defensive, you just know that person won’t be long at Nvidia.”

Diercks believed there was a method to it. “He would never just yell at somebody,” he said. “He would wait for a meeting, with a bunch of people around, so he could make it an educational opportunity for everyone.” But Huang’s criticisms weren’t always constructive—sometimes they were just verbal abuse. One former employee recalled a time when he bungled a minor assignment.

The NV3 was mostly a copycat chip, but it had a couple of innovations. First, it could transport 128 bits at a time from memory to processing, double the industry standard. Second, it had Swiss Army multifunctionality: it could accelerate video games, it could resize a spreadsheet, and it could play a DVD. To emphasize this breadth of capabilities, the NV3 was rebranded as the Real-Time Interactive Video and Animation accelerator, or Riva 128.

The experience was liberating for Huang. Desperation, not inspiration, was the mother of victory. Huang encouraged his employees to preserve the mindset they’d adopted during the Riva crunch, asking them to constantly behave as if the company was teetering on the verge of bankruptcy even when it was making massive profits. For years to come, Jensen opened staff presentations with the words “Our company is thirty days from going out of business.” Even today at Nvidia, this sentence remains the corporate mantra.

FIVE: Going Parallel

“Kirk and Huang each combined a commanding breadth of technical expertise with a talent for low cunning. As Nvidia succeeded, many of the other graphics start-ups failed. Huang, sensing opportunity, created a master list of competitors on the whiteboard in his office. Then, in consultation with Kirk, he identified the two or three best engineers at each company and began strategizing on how to poach them.

Kirk recalled showing off the Riva 128 to a competitor at a trade show. When the engineer saw what it could do, he gave up on the spot. “I hired him within a few days—and that killed that company, right?” Kirk said. “Because, you know, I removed its brain.” Kirk, mild and professorial, had a predator’s instincts. “We had all the geniuses from all the other start-ups, and as we were successful in overtaking more and more of these little companies, the remaining companies had a harder and harder time staying alive.

But Huang’s knowledge of business books was encyclopedic. Dwight Diercks recalled Huang arguing with another executive about how much Nvidia’s products should cost. “The guy had an MBA, but he’d never read a book about pricing,” Diercks said. “Jensen had read probably ten or fifteen.” As the argument progressed, Huang halted the discussion and asked the MBA to name his three favorite books on pricing. The guy fumbled around for a bit, unable to name a single title. Huang listed out his three favorites, then told the executive he’d resume the discussion once he’d finished them.

Christensen’s insight was that it was easier to go up the escalator of profitability than down. Going down meant voluntarily shrinking profit margins by deliberately making inferior goods, which tended to upset investors and made executives feel like they were jogging in place. This led Christensen to his most enduring and most counterintuitive recommendation: “There are times when it is right not to listen to customers, right to invest in lower-performing products that result in lower margins, and right to pursue small, rather than substantial, markets.” It was a point that the buzzword discussion of “disruption” in the popular press

usually missed.

Nvidia didn’t have a mission statement (Huang didn’t believe in them), but Gibson’s observation might have served as one. The goal was immersion, total immersion, in digital worlds rendered with such pointillist detail that they made reality fall away.

SIX: Jellyfish

“Unencumbered by received wisdom, TD-Gammon discovered a new approach to backgammon. It determined that human players were risking too much to establish an advantage up front and that conservative openings were better. At the same time, it would often forgo a guaranteed win in the endgame in a greedy attempt to double its score, a strategy most human players considered reckless. In the middlegame, TD-Gammon made a variety of more subtle moves that human experts understood only after deep introspection.

But IBM failed to commercialize Tesauro’s project—why would a vendor of business servers sell commercial backgammon software to a few hundred customers? Why, indeed.

This adorable nook in the marketplace was filled in 1994 by the Norwegian researcher Fredrik Dahl. Dahl was an unusual man who enjoyed backgammon, chess, simulated tank battles, jiu-jitsu, and foraging in the woods for edible fungi. He worked for Norway’s defense establishment, where he simulated outcomes from a hypothetical Soviet invasion. His work drew inspiration from the 1983 movie WarGames, starring Matthew Broderick. In that movie, an AI attempts to start a nuclear war.

Dahl worked on this problem for many years. The obstacle was that he had almost no idea how his poker bot actually worked. The structure of its neural net was no easier to interpret than the nervous system of an invertebrate, and trying to tease out game-playing strategies by examining the individual weights of the grid was like trying to unravel consciousness

by looking at brain cells through a microscope. This was the criticism of neural nets and the thing that so biased the academic community against them. Once a neural net hit a training plateau—and they almost always did—there was rarely an obvious way to make it better. Classical programming was orderly and logical, but tinkering with neural nets required a different cast of mind. Dahl compared it to running a biology experiment: outcomes were unpredictable, and altering seemingly minor variables could have all manner of unanticipated results. Dahl tried everything he could think of to improve his no-limit poker bot. He fiddled with its evaluation function, he futzed with his computer’s memory, he replaced the activation trigger for the neurons, he even synthesized a simpler data universe for the bot to explore—but he never got it to play at an expert level.

SEVEN: Deathmatch

“Wall Street loved it. Nvidia shipped new cards on a six-month cycle, twice as fast as any other vendor. The company introduced a new product line for the back-to-school cycle each fall, then updated that product in the spring. Demand accelerated when flat-screen monitors arrived, and within a few years graphics accelerators were standard on most PCs. In early 1999, fewer than six years after its founding, Nvidia went public with a $600 million valuation. Sequoia, which had initially valued Nvidia at $6 million, tallied a hundred-bagger, subsidizing the losses from countless other speculative investments.

Huang was now a centimillionaire, but his newfound wealth did not distract from his objective of crushing and absorbing the competition until only his firm remained. Dwight Diercks recalled no parties, no champagne, no sense of relief, not even congratulations from the boss. He shared with me an email he had saved from Huang. “The TNT2 team needs to do whatever it takes to get over the finish line,” Huang had written. The email continued in a tone of panicked desperation, with Huang complaining about missed deadlines and fretting over ascendant competitors.

Vivoli was a clever guy who viewed a limited budget as an opportunity. He had noticed that in making purchasing decisions, gamers relied on a half-dozen independent hardware reviewers. Vivoli reached out to the reviewers, informing them that the GeForce was the world’s first “graphics-processing unit,” or “GPU.” Vivoli’s team had, in fact, made this term up, but the reviewers began grouping products in the category. Soon, graphics accelerators were universally known as GPUs. “We invented the category so we could be the leader in it,” Vivoli said.

Nvidia’s frenzied six-month shipping cycle left the perfectionists at 3dfx at a disadvantage. At one point, one of 3dfx’s founders publicly speculated about declaring a truce between the two companies so that technical standards could be established before the next generation of products shipped. “That’s when I knew we had him,” Kirk said. “We were in a death struggle with 3dfx, and one of us had to die.

Garlick took the job and remained at Nvidia for the next seventeen years. “My theory is that Jensen is a good person at heart who had to be ruthless,” Garlick said. “As opposed to some other CEOs, who were ruthless at heart and trying to pretend to be good people.” Such were Huang’s charms that, out of the 120 employees he recruited, 106 joined the dark side.

Garlick was given access to Nvidia’s code base. He was appalled at what he saw. “Basically, it was cancer,” he said. “Y’know, cancer cells aren’t efficient. They just mutate, grow, and expand.” At 3dfx, Garlick had taken pride in the elegance of his programming, developing orderly systems with lucid commenting, allowing other programmers to easily maintain and improve his work. “In the time we spent making it clean, we went out of business,” he said. Nvidia’s approach was slapdash, with blocks of code written during some delirious midnight crunch serving as the foundation for critical systems. “What a shit show! The code was crap, the tool-chain was a mess, and the thing was, they didn’t give a shit!” Garlick said. “They didn’t give a shit about anything but the next tape-out.”

In this manner, Nvidia had accrued a great deal of “tech debt,” repeatedly taking shortcuts that led over time to less maintainable code and creating problems for programmers later on. But as Garlick acclimated to these shortcuts, he came to see the value of the Nvidia approach. “There was a bizarre brilliance to it all: just iterate, iterate, iterate, execute, execute, execute,” he said. “The way I see it now, tech debt is the battle scar of the

Survivor.

With the new hires, Nvidia had more than six hundred employees, up from just thirty-five four years earlier. The company relocated to a new headquarters down the road in Santa Clara, leasing a complex of curved, multistory, glass-and-steel buildings joined by skyways, festooned with sculpture, surrounded by parking, adjacent to the expressway, and spread across eleven acres of land. The new offices didn’t smell like takeout food. They didn’t smell like anything. Modernist respectability, with its boring and predictable implications, had arrived.

Kirk ran many of the hiring interviews. Years earlier, at his own start-up, he’d been forced to lay off a hundred engineers, an experience so painful that days later he himself had quit. Resolved never to repeat this experience,

Kirk determined that the best way to avoid layoffs was to be selective about whom he hired. The initial interview format at Nvidia consisted of several rounds of interviews, followed by a consensus hiring decision. But the technical staff, reluctant to make people squirm, stuck to standard interview bullshit: “Recall a time you overcame adversity,” “What’s your greatest

weakness?,” “Why are manhole covers round?”

Kirk, frustrated, felt that his staff were wasting time. He knew how Jensen would respond: by gathering the technical staff in a conference room and screaming at them. Like Diercks, Kirk believed that Huang’s outbursts were purposeful. “Yelling at people was part of this motivational strategy,” Kirk said. “You might think he’s just mad, but I think it was premeditated.

And it works! It annoys people, but it does work.” The audience, Kirk believed, was crucial: “He wants everybody to benefit. He would never just yell at some guy in the hall. When he’s torturing people, he’s forcing them to learn a lesson—and they certainly would never forget it.

Despite the success, Huang remained wary. In 1996, the leading graphics accelerator firm was S3 Graphics. By 1999, it was gone. In 1998, the leading firm was 3dfx. By 2000, it was gone, too. There was no guarantee the same wouldn’t happen to Nvidia. One of the business books stacked in Huang’s office, written by Intel CEO Andy Grove, was titled Only the Paranoid Survive.

Huang had managed to stay ahead of his competitors so far, but his asset-light “merchant” business was essentially just a collection of engineers sitting around a Silicon Valley office park. If those engineers weren’t constantly developing new, difficult-to-replicate technology,

manufacturers in Asia would start knocking off his chips, and Nvidia would cease to exist. “If we don’t reinvent computer graphics, if we don’t reinvent ourselves, and we don’t open the canvas for the things that we can do on this processor, we will be commoditized out of existence,” Huang later said. Not to gamble was the biggest risk of all.

EIGHT: The Compulsion Loop

“The accounting scandal occurred during one of the worst bear markets in history. Suffering under the simultaneous bursting of the dot-com bubble, the 9/11 attacks, and the Enron bankruptcy, the S&P 500 lost nearly half its value. Coincident with these misfortunes, Nvidia started squabbling with Microsoft. The dispute was attributed to pricing and intellectual-property issues, but Nvidia’s growing sense of entitlement played a role. Nvidia employees were unabashedly elitist. They considered themselves the best—and they were—but their pride could sometimes sound like narcissism.

Huang’s anger, invisible to Romanosky, found increasing room to express itself at work. In early 2003, Nvidia shipped the infamous GeForce FX, prone to slow rendering speeds and known to gamers as “the dustbuster” for its faulty, overactive fan. The device was panned by reviewers and customers, including Huang’s thirteen-year-old son, Spencer. Jensen arrived home one evening to find a gaming magazine featuring a harsh review of the device waiting for him with a Post-it note attached. “Dad,” the note read, “I think you need to kick it up a notch.

Huang’s tirades inspired as much guilt as fear, and he often described, in detail, how in letting their customers down, Nvidia employees had let one another’s families down as well. (“I think I’m driven as much by guilt as anything else,” Huang told me.)

Nvidia conducted regular performance reviews of employees, and following the GeForce FX debacle, Clay feared that her next one would read RI: “Requires Improvement.” This, at Nvidia, was like being handed the Black Spot. For the GeForce FX, Clay had run four or five quality-control tests.

Many people at Nvidia told me that Huang’s anger enforced a kind of discipline within the company, in the manner of a military general or a pro football coach. “I’m not sure he yells more than any other Fortune 500 CEO,” one employee said. “Look, it’s not really his job to be your friend. It’s his job to push you beyond where you think you could ever go.

Of more than a hundred former and current Nvidia employees I spoke with for this book, almost all had a tender story about Huang to relate. One employee—the same one whom Huang had humiliated in front of dozens of people, asking for a full refund of his salary—told me that when he was later diagnosed with a serious medical issue, Huang offered to pay in full, out of pocket, for his treatment. When Ben Garlick decided to leave Nvidia for a start-up, he was startled to receive an impassioned, personal plea from Huang to stay.

In a series of chunk transactions between 2004 and 2006, Priem sold all his Nvidia shares. “That’s why the stock flatlined,” he said. “We basically sold into strength whenever it started going up.” Had he held those shares and done nothing but play cowboy for twenty years, Priem would today be worth more than $100 billion, making him one of the wealthiest people alive—but he told me he didn’t regret his decision. Doing so would have required him to have 99.9 percent of his net worth invested in the volatile stock of a risky tech company he no longer worked for, which didn’t seem like a good idea.

Channeling George Bailey, Priem asked me to consider where his vanished windfall profits had gone. “The shares went out there, but it’s not like they disappeared. It’s in pension plans. It’s in people’s houses. It’s sort of like I contributed $100 billion to our economy,” he said. “I’m on track to give away half a billion in my lifetime, and that has taken most of my time and effort. In the back of my mind, I’m trying to figure out what I would do with a $100 billion foundation, and it is not easy. I wouldn’t even know how to give that away.

NINE: Cuda

“To distinguish himself, Huang had to pursue a strategy that so defied conventional business logic that ATI wouldn’t follow. He had to build an exploratory product, like a $300 entry-level scientific supercomputer that not only didn’t have competitors but also didn’t even have obvious customers. The zero-billion-dollar market, by definition, was one that only he would participate in—one that only he would even see. Huang was going to build a baseball diamond in a cornfield and wait for the players to arrive.

Yet the architect kept his job. “Very rarely does Jensen make significant changes as a result of execution issues,” Halepete said. “He’s very conscious of having an even slightly chilling effect on people’s willingness to take risks and innovate. As a result, his level of forgiveness for even the largest screw-ups is extremely high.” Halepete surmised that the tirades were what Jensen did instead of showing you the door. “He will berate you, he will yell at you, he will insult you—whatever,” Halepete said. “He’s never going to fire you.

In pushing Shoquist, Huang was employing a scheduling technique he called the “speed of light.” He drilled this management concept into his employees with the fervor of religious doctrine—almost everyone at Nvidia I talked to referenced the “speed of light” at least once. “Speed of light” did not mean, as one might assume, to move quickly. Instead, Huang encouraged managers to identify the absolute fastest that something could conceivably be accomplished, given an unlimited budget, and assuming that every single thing went right. (For example, traveling from New York to London at the “speed of light” would involve perfect weather, zero traffic, and a supersonic plane.) Managers could then work backward from this unachievable constant to realistic but still impressive delivery times. “It sounds hard, but it really takes the pressure off of you,” Shoquist told me. “Once you understand the physical limits of what is possible, you understand the competition can’t go any faster either.

TEN: Resonance

“When Bill Dally wasn’t flying his plane, or applying for a patent, or reinventing the computer, he was riding his bicycle to the point of collapse, or rowing in Lake Tahoe, or competing in a downhill ski race, or sailing nonstop from Grenada to Antigua. Dally’s pace of invention made Kirk and Nickolls look lazy: he was the author of 250 technical papers and 4 textbooks, and he held 120 patents spanning an eclectic range of computing domains, ranging from complex circuit architectures to the chip that ran the power supply.

By 2012, the situation was becoming dire. Nvidia’s stock price had not appreciated in more than a decade, and although revenues and employment at the company had grown considerably, profits remained flat. Huang was bringing supercomputing to the masses, but the masses didn’t want it.

Smith, forty-two, was youthful and energetic, with curly hair and a boyish face. He liked to question operational decisions in excruciating detail: he once managed to replace the entire twelve-person board of Darden Restaurants while holding less than 6 percent of the company’s stock on the basis of a 294-slide plan to turn around the struggling Olive Garden chain. Starboard’s Olive Garden slideshow became a legendary document among equity analysts, particularly slide 104, which criticized the restaurant’s breadstick strategy. (Historically, Olive Garden waiters would bring one breadstick for every guest, plus one for the table; they would then refill the breadstick container as needed. But over time the quality of service deteriorated, and servers just started dumping a bunch of breadsticks on the table, reducing the amount of food that customers ordered.) Slide 163 noted Olive Garden had also stopped salting the pasta water in a misguided effort to extend the life of the cookware. “How can management of the world’s largest Italian restaurant chain think it is OK to serve poorly prepared pasta?” Starboard asked.

Disruptive technologies, Christensen had observed, often grew out of hobbyist communities. They were developed using “bootlegged resources” in which “off-the-shelf components” were redeployed for something other than their intended purpose. They started out wonky but rapidly improved along attributes of performance that established players ignored.

But even once you had absorbed this lesson, it wasn’t easy to implement. Pursuing niche markets cost profits, making investors question your sanity. This, too, Christensen had foretold: “One of the reasons managers at established firms find it difficult to serve emerging markets is that their investors and customers tell them not to.”

That was the real secret of The Innovator’s Dilemma, which readers often missed. It was not a book about how to succeed; it was a book about how not to fail. Christensen’s book wasn’t a how-to for start-ups but a counterinsurgency manual for senior managers at stagnating firms. Thirteen years in, Huang felt that Nvidia was at risk of becoming such a firm, and it was as much paranoia as optimism that led him to pursue the mad-science market.

Huang’s marginally profitable customers were scientists. They were scientists engaged in research, and in serving them, it was just possible he might enable one to change the world. Lateral technology transfers of this type had happened before. In the early 1600s, Dutch craftsmen working in the spectacles business realized they could rearrange their eyeglass lenses to view distant objects. (One story credits the discovery to two children trying to observe a weather vane.) The lenscrafters flooded the Dutch patent office with designs for telescopes, and within a year, Galileo was pointing one toward the heavens, becoming the first human to describe the phases of Venus, the moons of Jupiter, and the rings of Saturn. Made from modified eyeglass lenses, Galileo’s telescope had less magnifying power than a pair of modern bird-watching binoculars, but it forever changed our understanding of the

universe and our place within it. By shipping low-budget supercomputers to the mad scientists, Huang hoped to enable a similar revolution.

Huang did not have a concrete vision of what the future of technology would look like. Some technologists did; for example, Elon Musk began with a vision of himself standing on the surface of Mars, then worked backward to build the technology he would need to get himself there. Huang went in the opposite direction; he started with the capabilities of the circuits sitting in front of him, then projected forward as far as logic would allow. Only there, at the frontier of reason, would he allow himself to take a single step forward into the nebulous realm of vibes.

“Catanzaro was convinced the solution was to redesign the microchip anew. He cofounded the UC-Berkeley Parallel Computing Lab in the mid-2000s, along with several colleagues. There, Catanzaro made a list of existing parallel applications. The business problem, he could see, was that even for the supposedly hungriest customers, the demand for computing power was capped: once you sold an oil prospector a supercomputer, you saturated demand for years. What you needed, Catanzaro figured, was an application that was so hungry for computation that it could never be satisfied. You needed another application like 3D graphics that demanded more computer power once its initial needs were fulfilled. Eventually, Catanzaro deduced what had to be parallel computing’s killer app. “The answer to that was AI,” Catanzaro said. “I came to AI from the bottom up. I came from a circuits perspective. I felt it was just inevitable that AI was the most important computational workload.

Dally published many of his discoveries in academic journals for public consumption and with no financial reward. Often, he coauthored papers with engineers at AMD and Intel. Dally’s openness surprised a lot of people and sometimes led to pursed lips inside Nvidia, but Dally was playing the long game: he figured it was better to advertise what he was doing to other leading scientists so that they would come to work alongside him. “We’ll get the best academics to join the company because they’ll see our publications,” he would say. “The quality will speak for itself.

He could talk extemporaneously for more than two hours at a time, and in these presentations he would often revisit the same themes: the importance of the “speed of light” scheduling concept, the pursuit of the fabled “zero-billion-dollar market,” and above all, the ever-present danger of creeping bureaucracy.

THIRTEEN: Superintelligence

“It took a sharper eye to spot the differences. There was the vision question, with Musk moving backward from fantasy and Huang moving forward from reality. There was also the topic of loyalty. Musk did not value it; he often fired people arbitrarily and without warning, in one case canning the entire Starlink engineering team almost at random on a Sunday

afternoon. Huang almost never fired anyone, and when he did, it was only after multiple cautions and the offer of a performance-improvement plan. It took truly egregious behavior to get kicked out of Nvidia, and many employees worked there for decades, including boomerang hires like Catanzaro and Aarts. Even when operating economics forced Huang to shutter a division, he reassigned employees to other useful tasks. In 2019 Curtis Priem returned to Nvidia’s offices for the first time in sixteen years to join Huang and Malachowsky for a reunion of the company’s founders. “I was astounded at how many people were still there,” he said. “Jeff Fisher, his kids were working for Nvidia.

FOURTEEN: The Good Year

“To outsiders, Nintendo resembled an uncrackable safe. Even by Japanese standards, the company was insular. It was headquartered in conservative Kyoto, with most of its decisions made by iconic game designer Shigeru Miyamoto and the small clique of executives who surrounded him. Miyamoto had been in his early thirties when he’d produced Super Mario Brothers and The Legend of Zelda; now in his sixties, gaming’s Walt Disney had lost none of his passion.

FIFTEEN: The Transformer

“Words themselves meant nothing: in isolation, they were arbitrary collections of sounds. The only way to capture their meaning was to draw links between them and other words in the text. So if you had a knowledge graph linking the words “hop,” “green,” “tongue,” “flies,” and “amphibian,” then you knew enough to guess that the word in the center was “frog.” Not only that, but the graph should look the same in German, French, Swahili, or Vietnamese. The word wasn’t the letters “f,” “r,” “o,” and “g”—those letters were just placeholders. The word, in a cognitive sense, was that unique map of links to the rest of the vocabulary.

Uszkoreit, seeing the analogy with the brain, wanted to do the opposite, piping massive amounts of text, words, and computing firepower through a simple yet elegant funnel. Uszkoreit outlined his thinking in 2023: “If you are given a piece of hardware that has the very key strength of doing lots and lots of simple computations in parallel, as opposed to complicated, structured computations sequentially, then really that’s the statistical property you want to exploit.

This elegant architecture, designed to do the simplest thing conceivable—just take one step at a time—was like a skeleton key for AI. In 2017 the team published its results in the Neural Information Processing Systems journal, which had published the original AlexNet results. The paper needed a name, so Jones, channeling the Beatles, suggested “Attention Is All You Need.” This was an off-the-cuff joke that he didn’t think the team would actually use. Later, he would meet people with the sentence tattooed on their arms.

SIXTEEN: Hyperscale

“Nvidia had grown too large for Huang to really understand everything that was going on, but it was not his style to delegate. To maintain resonance, he needed to keep communication open with the frontline employees. Sometime around 2020, Huang asked everyone at the company to submit a weekly list of the five most important things they were working on. Every Friday from that day forward, he received twenty thousand emails. Brevity was encouraged; Huang would randomly sample from this pool of correspondence late into the night. In turn, he communicated to his staff by writing hundreds of emails per day, often only a few words long. (One executive compared the emails to haiku. Another compared them to ransom notes.) His responsiveness was superhuman. “You’d email him at 2 a.m. and receive a reply at 2:05 a.m.,” Dally said. “Then you’d email him again at 6 a.m. and receive a reply at 6:05 a.m.

SEVENTEEN: Money

“Like many firms, Nvidia allowed employees to purchase stock at a discount to market prices. What set Nvidia’s program apart was that employees were allowed to purchase stock at a discount to the lowest price at any point in the last two years. These purchases were capped at a certain dollar amount, but as the stock went vertical, the program basically turned into free money, and those who maxed out their contributions each year made the trade of a lifetime. With the windfall extending deep into middle management, some newer employees expressed concerns that the nouveau-riche veterans were entering a state of “semiretirement.” Executives disputed this characterization. Jeff Fisher, who ran the company’s gaming side, had been among the first thirty employees. “Many of us are financial volunteers at this point,” he said, “but we believe in the mission.”

The lure of developing this revolutionary technology offered purpose beyond what money could buy. This was especially true of the old guard, who’d spent years explaining to baffled peers why they were working for a gaming company and who constantly had to correct the pronunciation of the firm’s name. AI had not been a consideration for these veterans, and they were as surprised to be working on it as anybody. “There was no way me, or anybody else, could have dreamed at the time that this stuff that science fiction writers might come up with has become a reality,” said Jay Puri, Nvidia’s head of sales, who started work at the company in 2005. The value of Puri’s shares exceeded $700 million by 2024, but he felt that the interesting work at Nvidia was only beginning. “Maybe I’m biased, but I think it really is the most important technology company of our time,” he said.

EIGHTEEN: Spaceships

“Catanzaro was uncorked now—I sensed that he didn’t often get to share this perspective at his job. “It doesn’t need to inhabit this biosphere. In fact, it doesn’t need to be on the Earth, either, because the thing about artificial intelligence is that it travels at the speed of light. Humans, you know, we actually have to lug bodies around. Artificial intelligence can move along a radio signal as long as there’s an antenna on the other side.” Free of the limitations of biology, Catanzaro explained, AI would rapidly spread throughout the solar system and beyond. “Humans are naturally confrontational—like, we’re territorial animals, and it’s built into our limbic system to defend our turf,” he said. “AI, if it’s truly intelligent, the things that it’s interested in are so much bigger than the little thin crust of Earth that the humans live on. I don’t think that it’s going to be interested in taking that from us. Rather, I feel like AI is going to want to take care of us.

NINETEEN: Power

“These GPUs used a lot of juice. A standard Google search required about a third of a watt-hour’s worth of electricity. With generative AI enabled, the same Google search required ten times that, which was enough to power a light bulb for about twenty minutes. Ask GPT to write you a five-thousand-word term paper, and you used enough energy to run a microwave for an hour. Industrial demand was greater; executives were excited about the prospect of replacing human labor entirely.

Firms that attempted to replicate GPT with in-house data often produced shambolic “knowledge engines” that were little better than toys. These AIs supplemented the standard large-language-model training set with emails, mission statements, patent applications, legal memoranda, and other exciting selections from the internal corporate syllabus. As the buzz percolated through middle management, executives at marketing, media, and health-care firms launched ambitious initiatives, sometimes openly telling staff that many employees would be laid off once the neural nets were working. But much of what was produced was vaporware: late, expensive, and barely functional. Many users felt that AI technology simply

wasn’t ready.

TWENTY: The Most Important Stock on Earth

“Management professors theorized that a chief executive should ideally have between eight and twelve direct reports. Huang now had fifty-five. He had no right-hand man or woman, no majordomo, no second-in-command. Huang also had no designated successor, and as Nvidia grew, its C-suite actually shrank, meaning that there was no scapegoat for mistakes. Board members spoke of his irreplaceability; it was not an exaggeration to suggest that Huang had personally saved the American economy from recession. The US stock market, over the course of Nvidia’s rise, had pulled away from markets in Europe and Asia.

This was a little surprising, for while working at Nvidia was stimulating, it was never exactly fun; the corporate culture that Huang fostered was closer to Microsoft than Google, closer to IBM than Apple. But years earlier, Chiu, the Taiwanese physicist, had told Huang that he’d allowed him to do his “life’s work.” The phrase had stuck with Huang, and now he wanted to offer that same opportunity to his staff. “We want NVIDIA to be a place where people can build their careers over their lifetime,” the company wrote in its annual report. “Our employees tend to come and stay.”

The appeal lay in what Nvidia allowed you to achieve. It was not a secret that Huang pushed people hard. Thus, he attracted determined workaholics seeking to establish legacies as inventors. In the same way that a bestselling author didn’t stop writing, even many wealthy Nvidia engineers kept showing up to work each day to attack difficult technical problems. Those engineers collectively held more than fifteen thousand patents, but there was always something left to build.

Two fringe strains of computer science, starved of investment, hated—no, detested—by industry and researchers alike had somehow unified to form a thriving, sprawling entity now careering toward sentience. “I just thought, there is no way that Nvidia is this lucky,” Aarts said. “There’s no way that deep learning just fits this perfectly because Nvidia has never put any effort into it!”

Huang called it “luck, founded by vision.”

For Dally, it was Huang’s tireless work ethic that made Nvidia succeed. Even Dally, who left no spare second in his day, could not quite believe the superhuman efforts of his boss. “The rest of us are just here to reduce the bandwidth demands on Jensen,” Dally said. “I mean, when does he sleep?” Diercks agreed: “His hobbies are work, email, and work.”

Plenty of people worked long hours, though. Jens Horstmann attributed Huang’s success to his adaptability. “I’ve often asked myself, how is it that we started in the same cubicle, you know, with a similar IQ, both working equally hard,” Horstmann said. “How is it that this person not only built this amazing company, but also a network around him of people that—that would just die for him if needed?” Huang, Horstmann believed, had changed himself many times. He recalled Huang at LSI, pushing the simulation software to its outer limits. “Now, he’s still doing the same thing, but what he’s engineering is himself. He was not born as a great CEO; he was not destined to be one. He transformed himself into one, just by abstracting! Just by problem-solving the inputs and outputs of what a good CEO should be.

TWENTY-ONE: Jensen

“Jensen contradicted himself frequently, sometimes offering opposing viewpoints within the same interview. He wasn’t playing devil’s advocate, exactly—he just liked to attack ideas from both sides. “He’s not trying to be a politician,” Horstmann said. “He’s not trying to stay on message. He’s trying to process real-time input, and he’s willing to entertain a contradictory thought for a while.” What might appear to be a definitive pronouncement was often just Jensen thinking out loud. Only once he started to repeat himself was it time to pay attention. When an idea really struck Jensen, it slowly built up steam over a period of days or even weeks. It cycled into his vocabulary and was repeated at every meeting. Concepts like the “zero-billion-dollar market” or the “speed of light” hadn’t come to Jensen in a flash; they’d arrived as polished nuggets of wisdom after spending months being tossed in the rock tumbler of his mind. Having arrived, they were then drilled so thoroughly into his employees that his staff sometimes sounded like characters from The Manchurian Candidate, repeating Jensen’s catchphrases verbatim with a glassy look in their eyes. Even employees who hadn’t worked at Nvidia for years could still recite the catechism from memory.

TWENTY-TWO: The Fear

“When I shared Jensen’s objections with Bengio, he grew agitated. “Of course, there is no data!” he said. “Humanity hasn’t disappeared yet! Are we going to wait until we have repeated the death of humanity multiple times to decide that, oh, now we have data?!” He made a good point. All the data in the world would not have predicted the breakthrough of AlexNet or the success of the transformer architecture. Twice in ten years, AI had experienced unforeseeable and permanent upgrades to its capabilities. Bengio did not think that the current models posed an immediate threat to human life—but what about the next breakthrough? No one could say what it might bring or when it might happen.

Still, if Bengio, Hinton, and Sutskever had been sidelined by capital, the points they made remained valid. They had seen better than anyone the potential of what AI technology could be, and they had the academic credentials to prove it. If they were worried now, I felt it was worth listening to. “Right now there are ninety-nine very smart people trying to make AI better and one very smart person trying to figure out how to stop it taking over,” Hinton said.

TWENTY-THREE: The Thinking Machine

“Looking back, it became clear to me that Jensen had wanted to lose his temper; he’d made a conscious decision to thrash me. Once the performance had started, his fury was genuine, but it was all in service of a larger point he wanted to make. It wasn’t just that Jensen didn’t read science fiction—it was that he actually hated science fiction. He was a

serious man.

The reason that Jensen had succeeded in fields where others had failed—parallel computing, AI, the Omniverse—was precisely because he didn’t tolerate airy speculation about the future. He examined technologies coldly, from first principles, swayed neither by optimism nor fear but only by a cold and patient sense of business logic that he alone could push to the outer limits of corporate foresight. Beyond that he did not look or care to imagine. The potential for human extinction was not a question of corporate strategy and thus, to him, was as foolish as drawing a dragon on the unexplored portion of the map.