The Arms Dealer's Breakfast
On a Thursday morning in September 2023, at a Denny's restaurant on Berryessa Road in San Jose, California, Jensen Huang ordered seven items. A Super Bird sandwich. A chicken-fried steak. Pancakes, which he would roll around a sausage with his fingers. The journalist across the booth had just watched a video of a robot staring at its own hands in seeming recognition, then sorting colored blocks — a demonstration that had given him chills, the obsolescence of his species seemingly near. Huang, sixty years old, sarcastic and Teddy-bear-faced with wispy gray hair, dismissed the concern between bites. "I know how it works, so there's nothing there," he said. "It's no different than how microwaves work."
This was a peculiar thing for the most consequential arms dealer of the twenty-first century to say. Four months earlier, on May 25, 2023, when the Nasdaq opened, Nvidia's market capitalization had increased by approximately two hundred billion dollars in a single session — one of the largest single-day gains in stock-market history — after the revelation that OpenAI's ChatGPT had been trained on Nvidia supercomputers. By the close of trading, Nvidia was the sixth most valuable corporation on earth, worth more than Walmart and ExxonMobil combined. "There's a war going on out there in A.I., and Nvidia is the only arms dealer," one Wall Street analyst observed. And the arms dealer was sitting in a Denny's, eating a chicken-fried steak, telling a reporter that the technology undergirding his fortune was about as mysterious as a kitchen appliance.
The Denny's was not incidental. Huang had drafted Nvidia's founding paperwork at this very restaurant thirty years earlier, on his thirtieth birthday, over cheap coffee and Super Bird sandwiches. The CEO of Denny's was now giving him a commemorative plaque; a TV crew hovered nearby. But Huang's return to the booth where he'd bet his career carried none of the sentimental ceremony the occasion might have warranted. He told the waitress he'd once been a dishwasher at a Denny's in Oregon. "But I worked hard! Like, really hard. So I got to be a busboy." Then he tipped her a thousand dollars, stood up, and accepted his award.
Between the dishwashing and the plaque — between the nine-year-old Taiwanese immigrant who cleaned toilets at a Kentucky reform school and the man whose personal stake in Nvidia now exceeds forty billion dollars — lies one of the most improbable arcs in the history of American capitalism. It is the story of a patient monopolist who bet his company's future on artificial intelligence a full decade before the rest of Silicon Valley believed, who built the computing architecture that would make the A.I. revolution possible, and who did it all while opening staff meetings with the words: "Our company is thirty days from going out of business."
By the Numbers
Nvidia's Empire
$5T+Market capitalization (first company to surpass $5 trillion, 2025)
$155.5BRevenue, fiscal year 2025
$72.9BNet income, fiscal year 2025
~70%Gross profit margin on A.I. equipment
80%+Market share in GPUs for training and deploying A.I. models
32 yearsDuration of Jensen Huang's tenure as CEO (founded 1993)
36,000Employees worldwide
The Swinging Bridge
The story that Jensen Huang tells about himself — and he is a man who controls his narrative with the precision of a chip architect routing transistors — begins not in triumph but in displacement. He was born Jen-Hsun Huang on February 17, 1963, in Tainan, the historic capital of southern Taiwan, the second son of Huang Hsing-tai, a chemical engineer, and Lo Tsai-hsiu, a grade-school teacher. When Jensen was five, the family relocated to Thailand. But the Vietnam War was bleeding across borders, and Thailand itself was convulsing — "tanks were rolling down the streets," Huang has recalled, "grenades are going off. It's a full-on battle." His parents, who had formed a positive impression of the United States through his father's participation in a worker-training program at Carrier, the air-conditioning manufacturer, decided their sons' futures lay elsewhere.
In 1973, when Jensen was nine and his brother ten, the boys were sent as unaccompanied minors to the United States. They landed in Tacoma, Washington, to live with an uncle who promptly enrolled them at the Oneida Baptist Institute in rural Oneida, Kentucky — a school the uncle believed to be a prestigious boarding academy. It was, in fact, a religious reform school for troubled youth. Jensen's roommate, a seventeen-year-old, lifted his shirt on their first night together to display the numerous places where he'd been stabbed in fights. "Every student smoked," Huang has said, "and I think I was the only boy at the school without a pocketknife." His roommate was illiterate. The arrangement they struck — Jensen taught him to read; in exchange, the older boy taught Jensen to bench-press — carries the fairy-tale logic of a fable, except that Huang ended up doing a hundred pushups every night before bed, a habit that suggests not enchantment but survival.
Because Huang was too young for the academy's classes, he attended a nearby public school, where he befriended Ben Bays. Bays grew up with five siblings in an old house with no running water — one of Kentucky's tobacco-farming hollers, where most kids were poor and the vocabulary for describing a small Asian immigrant with long hair and heavily accented English had exactly one word. "The way you described Chinese people back then was 'Chinks,'" Huang told the New Yorker's Stephen Witt, with no apparent emotion. "We were called that every day."
To get to school, Huang crossed a swinging footbridge over a river — old planks, most of them missing, suspended by ropes high above the water. Local boys would grab the ropes when Jensen was crossing and try to shake him off. "Somehow it never seemed to affect him," Bays recalled. "He just shook it off." By the end of the school year, the boys who had tried to dislodge him were following him on adventures into the woods. Bays remembered how carefully Jensen stepped around the missing planks. "Actually, it looked like he was having fun."
Huang credits Oneida with building resilience. "Back then, there wasn't a counsellor to talk to," he said. "Back then, you just had to toughen up and move on." In 2019, he donated a building to the school, and spoke fondly of the footbridge — neglecting to mention the bullies who had tried to throw him off it. The omission is characteristic. Huang's relationship with suffering is not one of resentment but of utility. Pain, in his cosmology, is raw material. "People with very high expectations have very low resilience," he told Stanford's business school in 2024. "One of my great advantages is that I have very low expectations."
Homework as Courtship
After a couple of years, Huang's parents secured entry to the United States, settling in the suburbs of Portland, Oregon, and the brothers reunited with them. Jensen excelled at Aloha High School with a ferocity that reads, in retrospect, as the energy of a boy who'd survived a reform school and was determined to outrun every disadvantage of his biography. He was a nationally ranked table-tennis player. He joined the math, computer, and science clubs. He skipped two grades and graduated at sixteen. "I did not have a girlfriend," he said.
At Oregon State University, he majored in electrical engineering. His lab partner in his introductory classes was Lori Mills — earnest, nerdy, with curly brown hair, one of approximately three women among two hundred and fifty electrical-engineering students.
Competition among the male undergraduates for Mills's attention was fierce, and Huang felt himself at a disadvantage. "I was the youngest kid in the class," he said. "I looked like I was about twelve."
His strategy was characteristically systematic. Every weekend, he called Mills and pestered her to do homework with him. "I tried to impress her — not with my looks, of course, but with my strong capability to complete homework," he said. Six months of homework sessions passed before he worked up the courage to ask her on a date. She accepted. They married after graduation. Lori, herself a microchip designer, initially earned more than Jensen. ("She actually made more than me," he said, with the half-proud, half-sheepish tone of a man who keeps score.) She would eventually leave the workforce to raise their two children — Spencer, who became a speakeasy proprietor, and Madison, who went into hospitality, before both, after years of what Huang calls "paternal browbeating," joined Nvidia.
The Latin Word for Envy
Following Oregon State, Huang landed at Advanced Micro Devices in Sunnyvale, California, then moved to LSI Logic Corporation, where he designed software tools for chip architects and rose to become director of a company division — all while attending Stanford for a master's in electrical engineering by night. He was, by all accounts, an accomplished product manager with a reputation for speed and technical mastery that exceeded what his titles suggested. He had told his future co-founders, Chris Malachowsky and Curtis Priem, that he aimed to run something by the age of thirty.
Malachowsky and Priem were veteran microchip designers at Sun Microsystems, the fabled Silicon Valley workstation manufacturer. Malachowsky was a hardware architect with deep roots in chip fabrication; Priem, a graphics specialist who dreamed of making competitors "green with envy." When they decided to leave Sun and start a company, they recruited Huang — younger, but already clearly the one who should run things. "He was a fast learner," Malachowsky said, with the understatement of a man who has watched his protégé become one of the wealthiest people alive.
On April 5, 1993 — Huang's thirtieth birthday — the three men sat in a booth at the Denny's on Berryessa Road. They had $40,000 in capital. They wanted to design graphics chips. They initially called the company NVision, until they discovered the name belonged to a manufacturer of toilet paper. Huang suggested Nvidia, riffing on invidia, the Latin word for "envy." They selected the Denny's because it was quieter than home and had cheap coffee, but Huang had a deeper affinity for the chain. He'd worked at one in Oregon during the nineteen-eighties — first as a dishwasher, then as a busboy. "I find that I think best when I'm under adversity," he said. "My heart rate actually goes down. Anyone who's dealt with rush hour in a restaurant knows what I'm talking about."
I do everything I can not to go out of business. I do everything I can not to fail.
The founding myth of Nvidia is a Denny's booth and three electrical engineers. But the company nearly died before it could grow. Huang and his co-founders bet on quadrilateral-based graphics primitives rather than the industry-standard triangles. It was a principled technical choice and a near-fatal commercial error: soon after Nvidia shipped its first product, Microsoft announced that its graphics software would support only triangles. Short on money, Huang pivoted to the conventional approach, laid off more than half of the company's hundred employees, and bet Nvidia's remaining funds on a production run of untested microchips he wasn't sure would work. "It was fifty-fifty," he said, "but we were going out of business anyway."
The product, RIVA 128, hit stores in 1997 with Nvidia holding enough cash for one month of payroll. The gamble paid off: a million units sold in four months. But the near-death experience embedded itself in the company's psyche like a constitutional amendment. For years afterward, Huang opened staff presentations with the words "Our company is thirty days from going out of business." The phrase remains the unofficial corporate motto.
A Fleet of Motorcycles
To understand what Nvidia actually makes — and why it matters — requires a brief excursion into the architecture of computing. In standard computer architecture, a microchip called a central processing unit does most of the work. Coders write programs; the C.P.U. processes one solution at a time. For decades, the dominant C.P.U. manufacturer was Intel, and Intel's approach — serial processing, one task after another — defined the industry's physics. Nvidia embraced an alternative. Its graphics-processing unit breaks complex mathematical tasks into small calculations and processes them simultaneously, a method known as parallel computing. A C.P.U. functions like a delivery truck, dropping off one package at a time. A G.P.U. is a fleet of motorcycles spreading across a city.
In 1999, shortly after going public, Nvidia introduced a graphics card called GeForce. Dan Vivoli, the company's head of marketing, called it a "graphics-processing unit" — coining the term. "We invented the category so we could be the leader in it," Vivoli said, articulating a principle Huang would deploy repeatedly over three decades: if the market doesn't exist, create it.
GeForce's popularity was driven by the Quake video-game series, whose grenade-launching, monster-rendering gameplay demanded exactly the kind of parallel-computing horsepower that Nvidia's cards provided. PC gamers, perpetually seeking an edge in deathmatch mode, bought new GeForce cards with every upgrade cycle. The business was lucrative, but Huang saw something beyond gaming. In 2000, Ian Buck, a graduate student at Stanford studying computer graphics — intense, balding, radiating intelligence, the kind of computer-science hot-rodder who would spend two decades pushing chips to their limits — chained thirty-two GeForce cards together to play Quake on eight projectors. "It was the first gaming rig in 8K resolution, and it took up an entire wall," Buck said. "It was beautiful."
Then Buck asked the question that would change the trajectory of computing: What if these graphics cards could be used for something other than launching grenades at friends? With a DARPA grant, he hacked the cards' primitive programming tools to access the parallel-computing circuits beneath, effectively repurposing the GeForce into a low-budget supercomputer. Soon, Buck was working for Huang.
The Zero-Billion-Dollar Market
Since 2004, under Buck's oversight, Nvidia developed CUDA — Compute Unified Device Architecture — a software platform that would allow any programmer to harness the parallel-computing power of Nvidia's chips for general-purpose tasks. Huang's vision was audacious: enable CUDA to work on every GeForce card, effectively democratizing supercomputing. "We were democratizing supercomputing," he said, deploying the verb with the conviction of a man who does not use words loosely.
Wall Street reacted with dismay. Huang was spending billions on a new chip architecture targeting an obscure corner of academic and scientific computing — a market that, at the time, was "certainly less than the billions they were pouring in," as Ben Gilbert of the Acquired podcast noted. Huang argued that the mere existence of CUDA would enlarge the supercomputing sector. It was a classic zero-billion-dollar market play — a term Huang coined to describe products that have no competitors because they don't yet have obvious customers. "One of the things you can definitely guarantee is where there are no customers, there are also no competitors," he told Caltech graduates in 2024.
The market's skepticism was vindicated, at least initially. By late 2008, Nvidia's stock price had declined by seventy percent. Downloads of CUDA peaked in 2009, then fell for three consecutive years. Board members worried that the depressed stock price would attract corporate raiders. "We did everything we could to protect the company against an activist shareholder who might come in and try to break it up," Jim Gaither, a longtime board member, said. Dawn Hudson, a former N.F.L. marketing executive who joined the board in 2013, described a "distinctly flat, stagnant company."
During this wilderness period, Huang kept faith. He has cited a visit to the office of Ting-Wai Chiu, a professor of physics at National Taiwan University, as sustaining his conviction. Chiu was attempting to simulate the evolution of matter following the Big Bang. He had constructed a homemade supercomputer in a laboratory adjacent to his office — GeForce boxes littering the floor, the machine cooled by oscillating desk fans. "Jensen is a visionary," Chiu said. "He made my life's work possible." Chiu was the model customer. The problem was that there weren't many like him.
Nvidia marketed CUDA to stock traders, oil prospectors, molecular biologists. At one point, the company signed a deal with General Mills to simulate the thermal physics of cooking frozen pizza. One application that Nvidia spent almost no time thinking about was artificial intelligence. There didn't seem to be much of a market.
Prophets in the Wilderness
At the beginning of the twenty-tens, A.I. was a neglected discipline. Progress in basic tasks — image recognition, speech recognition — had been halting. Within this unpopular field, an even less popular subfield attempted to solve problems using "neural networks," computing structures inspired by the human brain. Many computer scientists considered neural networks discredited. "I was discouraged by my advisers from working on neural nets," Bryan Catanzaro, who would become Nvidia's lead deep-learning researcher, said, "because, at the time, they were considered to be outdated, and they didn't work."
Catanzaro described the researchers who persisted as "prophets in the wilderness." Chief among them was Geoffrey Hinton, a professor at the University of Toronto — a British-born cognitive scientist who had spent decades arguing that layered neural networks could learn to represent the world, largely to polite academic indifference. In 2009, Hinton's research group used Nvidia's CUDA platform to train a neural network to recognize human speech. He was surprised by the results, which he presented at a conference. Then he reached out to Nvidia. "I sent an e-mail saying, 'Look, I just told a thousand machine-learning researchers they should go and buy Nvidia cards. Can you send me a free one?'" Hinton said. "They said no."
Despite the snub, Hinton encouraged his students to use CUDA. One of them was Alex Krizhevsky — Ukrainian-born, possibly the finest programmer Hinton had ever encountered. In 2012, Krizhevsky and his research partner, Ilya Sutskever, bought two GeForce cards from Amazon on a tight budget. Krizhevsky began training a visual-recognition neural network on Nvidia's parallel-computing platform, feeding it millions of images in a single week. "He had the two G.P.U. boards whirring in his bedroom," Hinton recalled. "Actually, it was his parents who paid for the quite considerable electricity costs."
What Krizhevsky built can now be mentioned alongside the Wright Flyer and the Edison bulb. AlexNet, the neural network he trained in his parents' house, entered the annual ImageNet visual-recognition contest in 2012. Neural networks were unpopular enough that Krizhevsky was the only contestant to use the technique. AlexNet scored so well that the organizers initially wondered if he had somehow cheated. "That was a kind of Big Bang moment," Hinton said. "That was the paradigm shift."
AlexNet's nine-page architecture paper has been cited more than a hundred thousand times, making it one of the most important documents in the history of computer science. Krizhevsky pioneered several key programming techniques, but the central finding was stark: a specialized G.P.U. could train neural networks up to a hundred times faster than a general-purpose C.P.U. "To do machine learning without CUDA would have just been too much trouble," Hinton said. Within two years, every entrant in the ImageNet competition was using a neural network. By the mid-twenty-tens, neural networks trained on G.P.U.s were identifying images with ninety-six-percent accuracy, surpassing humans.
Huang's ten-year crusade to democratize supercomputing — the billions poured into an obscure corner of scientific computing, the frozen-pizza simulations, the seventy-percent stock decline, the board members sweating activist shareholders — had succeeded. Not because frozen pizza needed simulating. Because a Ukrainian programmer in his parents' bedroom needed two graphics cards and a software platform, and both happened to exist.
G.P.U.s showed up and it felt like a miracle.
The Friday E-mail
What happened next was less a strategic pivot than a detonation. Huang concluded that neural networks would revolutionize society and that he could use CUDA to corner the market on the necessary hardware. He announced — again — that he was betting the company. "He sent out an e-mail on Friday evening saying everything is going to deep learning, and that we were no longer a graphics company," Greg Estes, a vice-president at Nvidia, recalled. "By Monday morning, we were an A.I. company. Literally, it was that fast."
Around the same time, Huang approached Catanzaro with a thought experiment. "He told me to imagine he'd marched all eight thousand of Nvidia's employees into the parking lot," Catanzaro said. "Then he told me I was free to select anyone from the parking lot to join my team." It was the gesture of a CEO willing to subordinate every existing business line — billions of dollars in gaming revenue — to a technology whose commercial applications were, at that moment, almost entirely theoretical. "I didn't want him to fall into the same trap that the A.I. industry has had in the past," Catanzaro said. "But, ten years plus down the road, he was right."
The bet metastasized. In 2016, Nvidia delivered its first dedicated A.I. supercomputer, the DGX-1, to a small research group at OpenAI. Huang personally carried the computer to OpenAI's offices.
Elon Musk, then chairman, opened the package with a box cutter. The following year, researchers at Google introduced the transformer architecture for neural-net training. The year after that, researchers at OpenAI used Google's framework to build the first "generative pre-trained transformer" — G.P.T. — trained on Nvidia supercomputers, absorbing an enormous corpus of text and learning to make humanlike connections. In late 2022, after several versions, ChatGPT was released to the public, and the entire world noticed what Huang had been building for a decade.
Marc Andreessen, of the venture firm Andreessen Horowitz, had seen it coming. "We've been investing in a lot of startups applying deep learning to many areas," he said in 2016, "and every single one effectively comes in building on Nvidia's platform." What Andreessen understood — what Wall Street took another six years to internalize — was that CUDA had done for A.I. what the App Store had done for mobile: created a platform so comprehensive that the ecosystem couldn't leave.
Haiku and Ransom Notes
Nvidia's headquarters in Santa Clara consists of two enormous buildings, each in the shape of a triangle with its corners trimmed — a shape replicated in miniature throughout the interiors, from the couches and carpets to the splash guards in the urinals. Employees call them "spaceships." The buildings are cavernous, filled with light, and largely empty post-Covid. Underneath the north-campus bar, in windowless laboratories, pallid young quality-control technicians wearing earplugs push microchips to the brink of failure amid a constant whine of high-pitched fans trying to cool overheating silicon. It is these chips that have made the A.I. revolution possible.
Huang governs this empire through a management style that defies every Silicon Valley convention. He has approximately sixty direct reports. He holds no regular one-on-one meetings. There are no fixed divisions or hierarchy. Instead, employees submit a weekly list of the five most important things they are working on, and Huang surveys these e-mails late into the night. He communicates by writing hundreds of e-mails per day, often only a few words long. One executive compared the e-mails to haiku. Another compared them to ransom notes.
Rene Haas — who worked at Nvidia in the early 2010s before becoming CEO of the British chip designer Arm, and who considers Huang both a former boss and personal mentor — identified the logic beneath the apparent chaos. "It's a very unique culture," Haas told the Financial Times. "The benefit of that is transparency and speed. And I think that is one of the things that Nvidia is really, really good at. They move very, very fast, they're very, very purposeful." Huang organizes the company around projects rather than traditional hierarchies, allowing him to reach any layer of management and extract answers directly.
Wandering through Nvidia's campus, Huang often stops at the desks of junior employees and quizzes them on their work. A visit from Huang can transform a cubicle into an interrogation chamber. "Typically, in Silicon Valley, you can get away with fudging it," the industry analyst Hans Mosesmann said. "You can't do that with Jensen. He will kind of lose his temper." Huang himself acknowledges the mismatch between internal and external processing. "It's really about what's going on in my brain versus what's coming out of my mouth," he said. "When the mismatch is great, then it comes out as anger." One employee offered a more visceral comparison: "Interacting with him is kind of like sticking your finger in the electric socket."
Yet Nvidia has remarkably high employee retention. Jeff Fisher, who runs the consumer division and was one of the company's earliest hires, is now extremely wealthy but continues to work. "Many of us are financial volunteers at this point," Fisher said, "but we believe in the mission." Catanzaro left for another company, then returned. "Jensen is not an easy person to get along with all of the time," he said. "I've been afraid of Jensen sometimes, but I also know that he loves me."
Jensen is not an easy person to get along with all of the time. I've been afraid of Jensen sometimes, but I also know that he loves me.
The Architecture of Failure
Perhaps Huang's most radical management belief is that "failure must be shared." In the early two-thousands, Nvidia shipped a faulty graphics card with a loud, overactive fan. Instead of firing the product managers responsible, Huang arranged a meeting in which they presented, to a few hundred colleagues, every decision that had led to the fiasco. Nvidia also distributed to the press a satirical video, starring the product managers, in which the defective card was repurposed as a leaf blower.
Presenting one's failures to an audience has become a beloved ritual at Nvidia — beloved by those who survive it. "You can kind of see right away who is going to last here and who is not," said Dwight Diercks, Nvidia's head of software. "If someone starts getting defensive, I know they're not going to make it." The practice functions as both accountability mechanism and cultural selection pressure. It ensures that institutional knowledge about failure circulates rather than being buried, and it identifies the people capable of the radical intellectual honesty that Huang demands.
This culture of transparency extends to Huang's own reasoning. At Stanford, he described his philosophy of sharing his thought process: "If you send me something and you want my input on it, and I can be of service to you, and in my review of it, share with you how I reasoned through it, I've made a contribution to you. I've made it possible to see how I reason through something." It is a leader's version of showing your work — not dictating conclusions but modeling cognition, so that the organization learns not just what to think but how.
The Cousin Question
Nvidia's fiercest rival is Advanced Micro Devices, and the rivalry carries a familial tang. Since 2014, A.M.D. has been run by
Lisa Su — another gifted engineer who immigrated to the United States from Taiwan at a young age, who attended MIT, who rebuilt a struggling chip company into a formidable competitor. In the years since Su became C.E.O., A.M.D.'s stock price has risen thirtyfold, making her second only to Huang as the most successful semiconductor chief executive of this era. Su is also Huang's first cousin once removed.
Huang says he didn't know Su growing up and met her only after she was named C.E.O. "She's terrific," he said. "We're not very competitive." This is the kind of statement that, inside Nvidia, would provoke a knowing silence: employees can recite the relative market share of Nvidia's and A.M.D.'s graphics cards from memory. Their personalities are a study in contrast. Su is reserved, stoic, possessed of what Mosesmann calls "a great poker face." Huang is temperamental and expressive. "Jensen does not," Mosesmann added, "although he'd still find a way to beat you." The gross profit margin on Nvidia's equipment approaches seventy percent — a ratio that attracts competition the way chum attracts sharks. Google, Tesla, Amazon, and numerous startups including Cerebras, which makes a "mega-chip" the size of a dinner plate, are all developing A.I.-training hardware. "They're just extorting their customers, and nobody will say it out loud," Cerebras's C.E.O., Andrew Feldman, said of Nvidia. Huang's counter was characteristically reframing: "The more you buy, the more you save."
Speaking Universes into Existence
When asked in September 2023 if he was taking any gambles that resembled the ones he'd made twenty years ago, Huang responded immediately with a single word: "Omniverse."
Inspired by a V.R.-architecture experiment in which he'd strapped his building's architect, Hao Ko, into a virtual-reality headset connected to a rack of G.P.U.s to simulate the flow of light across Nvidia's headquarters, the Omniverse is Nvidia's attempt to simulate the physical world at extraordinary levels of fine-grained detail. Since 2018, Nvidia's graphics cards have featured "ray-tracing," which replicates the way light bounces off objects to create photorealistic effects. Inside Nvidia's executive meeting center, a product-demo specialist showed a three-dimensional rendering of a gleaming Japanese ramen shop — light reflecting off the metal counter, steam rising from bubbling broth — with nothing to indicate it wasn't real. Then he demonstrated "Diane," a hyper-realistic digital avatar that speaks five languages. The imperfections were the most affecting: blackheads on her nose, trace hairs on her upper lip. The only clue that Diane wasn't human was an uncanny shimmer in the whites of her eyes. "We're working on that," the specialist said.
Huang's vision is to unify Nvidia's computer-graphics research with its generative-A.I. research, creating image-generation systems so sophisticated they can render three-dimensional, inhabitable worlds populated with realistic people — while language-processing A.I.s interpret voice commands instantly. "The programming language of the future will be 'human,'" Huang has said. Users will speak universes into existence. Digital twins of our world will train robots and self-driving cars. Combined with V.R. technology, the Omniverse could allow users to inhabit bespoke realities.
It is the kind of vision that makes even hardened technologists dizzy. But inside Nvidia's spaceships, the executives building this Manhattan Project of computer science responded to existential questions with the equanimity of appliance salesmen. When a journalist wondered aloud if an A.I. might someday kill someone, Catanzaro said: "Eh, electricity kills people every year." When he wondered if it might eliminate art, Diercks replied: "It will make art better! It will make you much better at your job." In May 2023, hundreds of industry leaders had endorsed a statement equating the risk of runaway A.I. with that of nuclear war. Huang didn't sign it.
"Horses have limited career options," he said, when asked about economists who compared A.I.'s potential displacement of humans to the Industrial Revolution's displacement of horses. "For example, horses can't type."
The Leather Jacket and the Moss Garden
At sixty-two, Jensen Huang is compact, polished, and known among colleagues for a quick temper and visionary leadership — traits that coexist without apparent tension. He enters a room looking, as TIME described him in 2025 when naming him among the Architects of A.I. for its Person of the Year cover, as though he might erupt or collapse. Then someone puts on music and he dons the trademark black leather jacket and appears to transform — not just the uniform but the body language and optimism of a man who genuinely believes he is building the most impactful technology the world has ever known.
His relationship with public performance is characteristically contradictory. "I hate public speaking," he told a journalist at an event before several hundred architects, then went onstage and performed with relaxed confidence for an hour. "I'm not a great speaker, really, because I'm quite introverted," he told the New Yorker. ("He's a great entertainer," his friend Ben Bays countered.) "I only have one superpower — homework," Huang said. ("He can master any subject over a weekend," Dwight Diercks responded.) "I don't really think I've done anything special here. It's mostly my team." ("He's irreplaceable," Jim Gaither said.) The contradictions are not affectations. They are the syntax of a man who learned, on a swinging bridge in Kentucky, that survival requires both absolute self-reliance and the performance of nonchalance.
He works seven days a week. He is either working or thinking about work every waking moment, he told Stripe CEO Patrick Collison. He reads the weekly reports of thousands of employees. He writes hundreds of e-mails a day. His wardrobe — the black leather jacket, black jeans, black shoes — has been featured in the
New York Times Style section and is widely imitated by subordinates. (When architect Hao Ko showed up in an identical outfit, Huang spent six minutes roasting his pants for having too many pockets. "
Simplify, man!")
He never reads science fiction. He dislikes speculation. He reasons from first principles about what microchips can do today, then gambles with great conviction on what they will do tomorrow. "I'm never satisfied," he said. "No matter what it is, I only see imperfections."
At Caltech's 2024 commencement, Huang closed with a story from Japan. Watching a gardener painstakingly tend to Kyoto's famous moss garden, he realized that when a person is truly dedicated to their craft — when they prioritize their life's work — they always have plenty of time. "Prioritize your life," he told the graduates, "and you will have plenty of time to do the important things."
In 2025, Nvidia became the world's first five-trillion-dollar company after Huang announced plans to build supercomputers for the U.S. government and forecasted an additional five hundred billion dollars in orders for Nvidia's A.I. chips. President Trump, who had become a regular late-night phone partner, told him: "You're taking over the world, Jensen." Memes depicted Nvidia as Atlas, holding the stock market on its shoulders. When asked by TIME if there was anything he was envious of — this man who had named his company for envy — Huang said no. He tallied what he was grateful for: his happy marriage, his adult children, his two dogs, who had both received clean ultrasounds that day.
The deepest revelation about Jensen Huang may be the simplest: he is not, and has never been, motivated by the destination. He is motivated by the work. He told a crowd at TiEcon, when asked what still drives him, that he doesn't have anything else to do besides serve as CEO of Nvidia. He told Caltech's graduates to find their GPU, find their CUDA, find their generative AI — to believe in something unconventional and unexplored and dedicate themselves to making it happen. He has been doing the same thing for thirty-two years, from the Denny's booth to the five-trillion-dollar company, and the thing that makes him unusual among the century's great capitalists is not vision or cunning or even the leather jacket. It is endurance.
Back in Kentucky, on the swinging bridge above the river, the planks were rotting and the gaps were wide and the local boys were shaking the ropes, and a nine-year-old immigrant with no pocketknife stepped carefully around the missing boards. Bays, watching from the bank, noticed something he didn't expect. Jensen wasn't just surviving the crossing. He was studying the bridge.
Jensen Huang has led Nvidia for more than three decades — from a Denny's booth to the world's most valuable company. The following principles are distilled from his decisions, his management practices, and the culture he has built. They are not platitudes. They are the operating logic of a man who has bet his company's existence at least three times and won each gamble by margins that looked, in the moment, like recklessness.
Table of Contents
- 1.Build for the zero-billion-dollar market.
- 2.Treat near-death as an operating principle.
- 3.Reason from first principles, then bet with total conviction.
- 4.Invent the category so you can lead it.
- 5.Make failure a public institution.
- 6.Replace hierarchy with information velocity.
- 7.Pursue the speed of light.
- 8.Build the full stack.
- 9.Use suffering as raw material.
- 10.Show your reasoning, not your conclusions.
- 11.Abandon commodity work relentlessly.
- 12.Outlast everyone.
Principle 1
Build for the zero-billion-dollar market.
Huang coined this phrase to describe CUDA — a product that, when launched in 2006, had no competitors because it had no obvious customers. Wall Street punished Nvidia for the investment; the stock fell seventy percent. But the absence of customers also meant the absence of competitors, which meant that when demand eventually materialized — when AlexNet demonstrated that G.P.U.s could train neural networks a hundred times faster than C.P.U.s — Nvidia owned the entire ecosystem.
This is not the "build it and they will come" naïveté it superficially resembles. Huang's zero-billion-dollar markets are chosen on the basis of deep technical reasoning about where computing architectures are heading. The bet on CUDA was informed by Huang's intuition that
Moore's Law was decelerating and that parallel computing would become essential. The bet on A.I. in 2013 was informed by AlexNet's results and Hinton's research. The current bet on Omniverse is informed by the convergence of ray-tracing and generative A.I. Each zero-billion-dollar market is, in Huang's framing, the logical destination of trends already visible to anyone with sufficient technical depth.
At Caltech, he articulated the inverse logic: "With no more markets to turn to, we decided to build something where we are sure there are no customers. Because one of the things you can definitely guarantee is where there are no customers, there are also no competitors."
Tactic: Identify the technology that will be essential in five to ten years but has no commercial application today — then build the platform that makes it accessible, and own the ecosystem before demand arrives.
Principle 2
Treat near-death as an operating principle.
The RIVA 128 episode — betting the company's remaining funds on untested chips when Nvidia had one month of payroll left — could have been a traumatic memory. Instead, Huang institutionalized it. "Our company is thirty days from going out of business" became the unofficial corporate motto, repeated at the start of staff meetings for years.
This is not performative paranoia. It is a deliberately constructed organizational psychology. Nvidia's chip business runs on brutally short product cycles where a company is only as good as its last card. Huang's urgency is calibrated to that reality: in a perpetual upgrade cycle, complacency is indistinguishable from extinction. The near-death framing also inoculates against the bloat that typically accompanies success — the committee-driven decision-making, the political turf wars, the loss of speed that large companies almost inevitably suffer.
Tactic: Embed your company's closest brush with failure into its institutional memory — not as a cautionary tale but as an operating tempo. The urgency of genuine survival is irreplaceable; if you've lost it, reconstruct it deliberately.
Principle 3
Reason from first principles, then bet with total conviction.
Huang does not read science fiction. He dislikes speculation. He reasons from what microchips can do today, then extrapolates with an engineer's precision to what they will do tomorrow. "You can learn how something can be done and then go back to first principles and ask yourself, 'Given the conditions today, given my motivation, given the instruments, the tools, given how things have changed, how would I redo this? How would I reinvent this whole thing?'" he told Stanford.
The critical move is what happens after the reasoning: total commitment. When Huang concluded in 2013 that neural networks would revolutionize society, he sent the Friday e-mail declaring Nvidia an A.I. company. When he concluded in the mid-nineties that triangles would beat quadrilaterals, he laid off half the company and bet the rest on a production run. The first-principles reasoning provides the warrant; the conviction provides the velocity.
This combination — analytical rigor followed by existential commitment — is rare. Most founders are either careful analysts who hedge their bets or wild gamblers who trust their gut. Huang is both, sequentially.
Tactic: Separate the reasoning phase from the commitment phase. During reasoning, be relentlessly analytical and open to being wrong. Once you've reached a conclusion, commit as if the company's survival depends on it — because in Huang's experience, it usually does.
Principle 4
Invent the category so you can lead it.
When Nvidia introduced the GeForce card in 1999, Dan Vivoli didn't just name a product — he named a category. "Graphics-processing unit" didn't exist as a term before Nvidia coined it. "We invented the category so we could be the leader in it," Vivoli said. The same logic governed CUDA (creating the "general-purpose GPU computing" category) and the DGX line (creating the "A.I. supercomputer" category).
Category creation is Huang's alternative to competing on market share. "We never talk about market share in our company," he has said, "because the concept of market share says that there are a whole bunch of other people who are doing the same thing. And if they are doing the same thing, then why are we doing it? Why am I squandering the lives of these incredibly talented people to go do something that's already been done?"
N
Nvidia's Category-Creation Playbook
Three decades of defining new markets before competing in them.
| Year | Category Created | Conventional Wisdom |
|---|
| 1999 | GPU (GeForce 256) | Graphics acceleration is a niche feature |
| 2006 | General-purpose GPU computing (CUDA) | Supercomputing belongs to national labs |
| 2016 | A.I. supercomputer (DGX-1) | A.I. is an academic curiosity |
| 2020s | Industrial metaverse (Omniverse) | The metaverse is a gaming concept |
Tactic: Don't fight for market share in existing categories. Define a new category where you are the default leader, then build the ecosystem — developer tools, software stack, community — that makes the category synonymous with your product.
Principle 5
Make failure a public institution.
The defective graphics card with the overactive fan. The satirical video repurposing it as a leaf blower. The product managers presenting, to hundreds of colleagues, every decision that led to the fiasco. This is Huang's most distinctive management innovation: failure as institutional theater.
The practice serves three functions simultaneously. First, it distributes knowledge — the specific decisions that caused the failure become organizational learning rather than private shame. Second, it functions as a selection mechanism: people who get defensive don't survive at Nvidia, and the public-failure ritual identifies them quickly. Third, it normalizes the admission of error at every level, which is essential in a company that makes existential bets and needs rapid course correction when those bets go wrong.
Huang frames this as intellectual honesty. In a company where the CEO routinely bets the entire enterprise on unproven technologies, the ability to say "this didn't work, here's why, here's what we learned" is not a nice-to-have cultural value — it is a survival mechanism.
Tactic: Create a formal ritual for presenting failures publicly — not as punishment but as institutional learning. The people who can do this honestly are the ones you want; the ones who can't will self-select out.
Principle 6
Replace hierarchy with information velocity.
Huang has roughly sixty direct reports. He holds no regular one-on-one meetings. There are no fixed organizational divisions. Instead, the company is organized around projects, and every employee submits a weekly list of five priorities. The result is what Haas described as "transparency and speed" — a flat information architecture that allows Huang to reach any layer of the organization and extract ground truth.
The logic is deliberate. "Prior models of leadership were drawn from the battlefield," Huang has said, "where only the general makes strategic decisions, while the foot soldiers fight on the ground." In Huang's model, information flows at maximum velocity from any point in the organization to any other, with the CEO serving less as a hierarchical apex and more as a pattern-recognition engine operating across the full surface area of the company.
This structure demands an almost inhuman tolerance for cognitive load — Huang reads thousands of employee updates weekly, writes hundreds of e-mails daily, and personally quizzes junior engineers on their work. It also demands a particular kind of employee: one who thrives in radical transparency, can operate without the cover of organizational layers, and doesn't need the psychological comfort of a clearly defined reporting structure.
Tactic: Organize around projects, not divisions. Create mechanisms that allow leadership to access ground-level information without intermediaries. Accept that this requires extraordinary personal bandwidth from leadership — and extraordinary self-direction from employees.
Principle 7
Pursue the speed of light.
When scheduling, Huang asks employees to consider "the speed of light." This is not simply an instruction to move quickly. It means: determine the absolute fastest a task could conceivably be accomplished — the physical limit, the theoretical minimum — then work backward toward an achievable goal.
The V.R.-architecture story illustrates this. When Nvidia was designing its headquarters, architect Hao Ko's V.R. headset originally took five hours to render design changes. At Huang's insistence, the engineering team got the speed down to ten seconds. Ko understood the logic in retrospect: "If the headset took five hours, I'd probably settle on whatever shade of green looked adequate. If it took ten seconds, I'd take the time to pick the best shade of green there was." The speed of light is not about efficiency for its own sake. It is about how velocity changes the quality of decisions by enabling more iterations within the same time frame.
Tactic: For every critical process, identify the theoretical minimum time for completion. Make that the target. The point is not just speed — it's that faster iteration cycles produce fundamentally better outcomes because they enable more experimentation per unit of time.
Principle 8
Build the full stack.
Nvidia designs chips, builds hardware systems, writes the software platform (CUDA), develops the programming tools, maintains the developer ecosystem, and now offers cloud computing services. This vertical integration — from silicon to software to systems — is the structural source of Nvidia's moat.
The full-stack approach means that customers who adopt CUDA become embedded in an ecosystem that is extraordinarily difficult to leave. The twenty-five million CUDA downloads, the forty thousand companies and fifteen thousand startups building on Nvidia's platform, the decades of accumulated software libraries — all of this creates switching costs that no individual competitor can replicate by building a better chip. Competing with Nvidia requires not just designing superior silicon but replicating an entire computing paradigm.
Ben Thompson of Stratechery has described this as Nvidia's deepest strategic insight: the company doesn't just sell hardware. It sells a computing platform, and the compound effect of its full-stack approach produced, over the past decade, what Huang calls "a million-x speed-up" in A.I. computing.
Tactic: Don't compete on a single layer of the stack. Build vertically — from infrastructure to tools to ecosystem — so that customers adopt not just a product but a platform, and the switching costs compound over time.
Principle 9
Use suffering as raw material.
The Oneida Baptist Institute. The swinging bridge. The bullies. The reform school. The missing planks. Huang does not narrate these experiences as trauma; he narrates them as manufacturing inputs. "Greatness is not intelligence," he told Stanford students. "Greatness comes from character. And character isn't formed out of smart people, it's formed out of people who suffered."
This is not motivational poster bromide. It is the operating philosophy of a man who was called a racial slur every day as a nine-year-old, who was nearly shaken off a bridge by local boys, and who responded by leading those same boys into the woods by the end of the school year. The experience taught him that resilience is not a personality trait but a practiced skill — one that can be cultivated through exposure to adversity and the deliberate refusal to internalize it.
He applies this framework to his company: the thirty-days-from-bankruptcy motto, the public failure rituals, the relentless product cycles. Nvidia's culture is designed to produce institutional suffering — not gratuitous pain, but the productive discomfort of operating at the edge of capability under conditions of genuine uncertainty. "If you want to do extraordinary things, it shouldn't be easy," he told 60 Minutes. "It should be like that."
Tactic: Don't insulate your organization from difficulty. Design systems that create productive discomfort — short deadlines, public accountability, existential framing — while providing the psychological safety to learn from failure rather than being destroyed by it.
Principle 10
Show your reasoning, not your conclusions.
Most leaders transmit decisions. Huang transmits cognition. "If you send me something and you want my input on it, and in my review of it, share with you how I reasoned through it, I've made a contribution to you," he told Stanford. "I've made it possible to see how I reason through something. And by reasoning, as you know, how someone reasons through something empowers you."
This is pedagogical leadership. By modeling his thought process rather than issuing directives, Huang builds an organization that can make Huang-quality decisions without Huang being present. It explains how he can manage sixty direct reports without one-on-one meetings: the company has internalized his reasoning framework, not just his conclusions.
The approach also explains Nvidia's ability to pivot rapidly. When the Friday e-mail declared the company an A.I. enterprise, the organization could execute the pivot over a weekend because employees already understood how Huang thought about technology transitions. They didn't need step-by-step instructions; they needed to apply the reasoning framework to new conditions.
Tactic: When reviewing work, don't just correct conclusions — narrate your reasoning process aloud. Teach your organization how you think, not what you think. This scales leadership beyond the bottleneck of any individual decision-maker.
Principle 11
Abandon commodity work relentlessly.
Nvidia has walked away from multiple profitable businesses that became commoditized. The logic is brutal but consistent: "Why am I squandering the lives of these incredibly talented people to go do something that's already been done?" Huang asks. He views commodity work as an existential threat — not to margins, but to the talent density that makes Nvidia's next-generation bets possible. The best engineers want to work on problems no one has solved. If the company is competing on price in an existing market, those engineers leave.
This principle is the inverse of the zero-billion-dollar market thesis. You enter markets where there are no customers and no competitors; you exit markets once they become crowded with both. The discipline of abandonment is at least as important as the discipline of investment — and considerably rarer.
Tactic: Regularly audit your product portfolio for commoditization. Walk away from businesses that others can do, even if they're still profitable, to preserve your ability to attract the talent that can build what no one else can.
Principle 12
Outlast everyone.
Jensen Huang has been CEO of Nvidia for thirty-two years. He is the longest-serving tech CEO in Silicon Valley. He founded the company at thirty and is now sixty-two, and the unofficial motto remains "thirty days from going out of business." The longevity is not incidental to the strategy. The zero-billion-dollar market thesis requires patience measured in decades — CUDA took nearly a decade to find its killer application. The full-stack approach requires compound investment over time horizons that would destroy a CEO who was managing to quarterly earnings. The cultural practices — the failure rituals, the flat hierarchy, the reasoning-first leadership — require decades of reinforcement to become genuinely embedded.
⏱
The Patience of a Patient Monopolist
Nvidia's major bets and their time-to-payoff.
1993Nvidia founded at Denny's with $40,000 in capital.
1997RIVA 128 saves the company; 1 million units in four months.
1999GeForce and the IPO; "GPU" coined as a category.
2006CUDA launched; stock drops 70% over the next two years.
2012AlexNet proves GPUs can train neural networks 100x faster than CPUs.
2013Huang sends the "Friday e-mail" pivoting Nvidia to A.I.
2016First DGX-1 delivered to OpenAI by Huang personally.
2023
Huang has never entertained the possibility of stepping down. When asked what still drives him, he said he doesn't have anything else to do. The answer is both self-deprecating and revelatory: Nvidia's strategy is inseparable from Huang's continued presence. No successor could replicate the technical intuition, the institutional memory, the cultural authority, the willingness to bet the company, the six-hundred-e-mails-a-day cognitive bandwidth. Nvidia's deepest competitive advantage may be that its founder simply refuses to leave.
Tactic: Build for the long term by staying for the long term. The compound returns on culture, institutional knowledge, and ecosystem investment accrue disproportionately to organizations whose leadership endures across multiple technology cycles. Patience is a structural advantage, not a personality trait.
In his words
I find that I think best when I'm under adversity. My heart rate actually goes down. Anyone who's dealt with rush hour in a restaurant knows what I'm talking about.
People with very high expectations have very low resilience — and unfortunately, resilience matters in success. One of my great advantages is that I have very low expectations.
I hope you believe in something. Something unconventional, something unexplored. But let it be informed, and let it be reasoned, and dedicate yourself to making that happen. You may find your GPU. You may find your CUDA. You may find your generative AI. You may find your NVIDIA.
Every industry needs it, every company uses it, and every nation needs to build it. This is the single most impactful technology of our time.
I'm never satisfied. No matter what it is, I only see imperfections.
Maxims
-
Thirty days from bankruptcy. Urgency is not a response to crisis; it is an operating tempo. The company that behaves as though it has one month of runway will outperform the company that behaves as though it has decades.
-
The zero-billion-dollar market is the safest market. Where there are no customers, there are no competitors. Build the platform first; the demand follows the infrastructure.
-
Horses can't type. Technological displacement threatens the inflexible, not the human. The correct response to A.I. is not fear but adaptation — and the refusal to be a horse.
-
Deep learning is not an algorithm; it is a method. The A.I. revolution is not about a single breakthrough. It is about a new way of developing software, which means a new way of thinking about computers.
-
The more you buy, the more you save. When your product creates more value than it costs, price objections are a sign that customers haven't yet understood the economics. Reframe the discussion around total value, not unit cost.
-
Failure must be shared. Private failure becomes organizational ignorance. Public failure — presented honestly, without defensiveness — becomes institutional knowledge that compounds over time.
-
The speed of light is a planning tool. Determine the theoretical minimum time for a task, then work backward. Faster iteration changes not just efficiency but the quality of decisions, because it enables more experiments per unit of time.
-
Don't squander talented lives on commodity work. The best people want to solve problems no one has solved. If the market is crowded, the talent will leave before the margins do.
-
Greatness comes from character, and character is formed from suffering. Intelligence is common. The willingness to endure discomfort, absorb failure, and keep building is rare — and it is the only reliable predictor of extraordinary outcomes.
-
Prioritize your life, and you will have plenty of time. The Japanese moss gardener tends each blade with devotion. Urgency and patience are not opposites; they are the same discipline applied at different time scales.
