innovation is a necessity, not an option. Whiteboarding also requires active thinking and inevitably reveals how well (or not) any employee, including an executive, knows the material. (Page 1)
Employees must demonstrate their thought process in real time, in front of an audience; there’s no hiding behind neatly formatted slides or slick marketing videos. (Page 2)
He has run Nvidia for its entire three-decade history, the longest tenure of any current technology CEO. He has driven the company not only to survive, but to surpass all of its competitors in the unforgiving and volatile chip sector, and to surpass just about every other company on Earth, as well. (Page 3)
“Over the years, I realized what was happening, how people protect their turf and they protect their ideas. I created a much flatter organization,” Jensen said. His antidote to the backstabbing, to the gaming of metrics, and to political infighting is public accountability and, if needed, public embarrassment. “If we have leaders who are not fighting for other people to be successful and [who are] depriving opportunities to others, I’ll just say it out loud,” he said. “I’ve got no trouble calling people out. You do that once or twice, nobody’s going to go near that again.” (Page 6)
Although Jensen was happy to be back with his parents, he looked back on his time at Oneida Baptist as formative. “I don’t get scared often. I don’t worry about going places I haven’t gone before. I can tolerate a lot of discomfort.” (Page 13)
“I was very introverted. I was incredibly shy,” he said. “The one experience that pulled me out of my shell was waiting tables at Denny’s.” When Jensen was fifteen, his brother helped him get a job at a Denny’s in Portland. (Page 14)
Hoping to gain relevant work experience, Jensen applied repeatedly for an internship at a local technology company called Techtronic Industries but was rejected every time. (Page 16)
By day, he designed microchips; at night and on the weekends, he took courses at Stanford so that he could get his master’s in electrical engineering. On top of his work and his continuing studies, he and Lori had a son, Spencer, and daughter, Madison. Because he couldn’t take many classes at once, completing the master’s degree was a long, arduous process; he finally finished after eight years. (Page 16)
There, he was given a technical role working with customers. He was assigned to a start-up called Sun Microsystems, where he met two engineers, Curtis Priem and Chris Malachowsky, who were working on a secret project that promised to revolutionize how people used workstation computers—high-performance computers built to perform specialized technical or scientific tasks, such as three-dimensional modeling or industrial design. (Page 17)
Unfortunately, resilience matters in success,” he later said. “Greatness is not intelligence. Greatness comes from character.”17 And character, in his view, can only be the result of overcoming setbacks and adversity. To Jensen, the struggle to persevere in the face of bad, and often overwhelming, odds is simply what work is. (Page 18)
Priem an interview at Sun Microsystems. Sun was an early pioneer in high-end UNIX computer workstations, which it sold for thousands or even tens of thousands of dollars. It was founded by three Stanford graduate students—Scott McNealy, Andy Bechtolsheim, and Vinod Khosla—in 1982. (Page 23)
To make sure that Priem and Malachowsky got the chip they had drawn up, LSI assigned one of its rising stars to manage the Sun account—a relatively new hire named Jensen Huang. “This young kid had just joined them from AMD who had worked on microprocessors,” Malachowsky said. “Curtis knew what he wanted, I could design it, and Jensen helped us figure out how we were going to build it.” (Page 29)
Until the GX-enabled SPARCstation. For the first time, a realistic flight simulator became possible. Priem bought a workstation for his own personal use with his 60 percent employee discount, which shaved thousands of dollars off of the price. After he spent sixty hours a week at his day job, he would go home and get back to laboring on his new simulator program that would take full advantage of the new GX chips. Finally, he was able to realize his vision and complete the game, which he called Aviator. (Page 31)
Priem, in particular, was bothered by a culture where “it was easier to sabotage or get the other project killed than to come up with better technology.” He just wanted to make good graphics chips and had no interest in corporate infighting. (Page 33)
they had been so successful leading up to this time that they were more concerned with protecting success than driving for it. It was getting caught up in fear of failure. They stopped being very aggressive.” (Page 33)
“Chris knew every single struggle I went through, taking all the hits from Sun management,” Priem said. “He respected me taking all the arrows in the back. There were times I was so chastised by the VP of graphics that I’d be out with HR walking around the buildings in the park crying. It was just brutal.” (Page 34)
Eventually, Jensen decided that $50 million in revenue was possible. He was confident, as a gamer himself, that the gaming market was going to grow considerably. “We grew up in the video-game generation,” he said “The entertainment value of video games and computer games was very obvious to me.” (Page 38)
“We dropped the ‘I’ and went with NVidia to honor the NV1 chip we were developing,” said Priem, “and secretly hoped that someday Nvidia would be something that would be envied.” (Page 43)
Nevertheless, Corrigan promised to introduce Jensen to Don Valentine at Sequoia Capital. Valentine had invested in LSI Logic back in 1982, which earned him a handsome payout when the company went public a year later. He had hit it even bigger in other investments in tech companies such as Atari, Cisco, and Apple. By the early ’90s he was considered “the best venture capitalist in the world.” (Page 44)
Although Corrigan may have had doubts about Nvidia’s potential, he had none about Jensen himself. When he called Valentine after his conversation with the young, departing engineer, he didn’t pitch Jensen’s start-up idea; he pitched Jensen. “Hey Don,” he said, “we’ve got this kid who is going to leave LSI Logic. He wants to start his own company. He’s really smart. He’s really good. You guys should take a look at him.” (Page 44)
“Wilf says to give you money. Against my better judgment, based on what you just told me, I’m going to give you money. But if you lose my money, I will kill you,” Valentine told the Nvidia team. Nvidia secured $2 million of Series A funding from Sequoia Capital and Sutter Hill Ventures—$1 million apiece—at the end of the month. (Page 48)
they had succeeded on the strength of their reputation, not their business plan or their demo. It was a lesson Jensen would never forget. “Your reputation will precede you even if your business plan writing skills are inadequate,” he said. (Page 49)
While Priem was working on the design, Jensen focused on convincing Intel to support his new card. His contact at Intel was a young executive named Pat Gelsinger, who was responsible for managing revisions to the Peripheral Component Interconnect (PCI) expansion-slot standard for PCs that all forthcoming graphics cards would use. Jensen wanted PCI to add different types of throughput modes for the NV1 to take advantage of; Gelsinger was resistant. (Page 56)
“We thought we had built great technology and a great product,” Malachowsky said. “It turns out we only built great technology. It wasn’t a great product.” Sales were dismal, and most of the units sold during the holiday season were returned. By the spring of 1996, Diamond Multimedia had returned nearly all of the 250,000 chips it had ordered. (Page 59)
In the end, however, 3dfx executives opted not to make a move. Its executives believed that Nvidia’s bankruptcy was inevitable, and that it would be cheaper to wait until Nvidia collapsed so that they could pick up its talent and assets for a bargain. (Page 65)
Nvidia’s timelines were so tight that it was uncertain whether the chips would arrive in time for the conference—or whether they would be in good enough shape to showcase. The samples from the factory came only a few days before the event, and Nvidia engineers worked feverishly to troubleshoot any software bugs that remained. Their goal was to ensure the chips could run the Direct3D graphics benchmark that hardware manufacturers would use to evaluate their quality. Mere hours before the show started, the engineers managed to get the chips stable enough to work without crashing at unpredictable moments. (Page 71)
Within four months of the chip’s release, Nvidia had shipped more than a million units and captured a fifth of the PC graphics market. PC Magazine named the RIVA 128 an Editors’ Choice product, and PC Computing named it the Product of the Year for 1997.28 Large PC manufacturers, including Dell Computer, Gateway 2000, Micron Electronics, and NEC, all incorporated the chip into their computers for the holiday season. The torrid pace of sales allowed Nvidia to turn a $1.4 million profit in the fourth quarter of 1997—the company’s first profitable quarter since its founding four years earlier. (Page 73)
“We don’t do things like anybody else. If you come here and say, ‘This is how we did it before,’ we don’t care. We’re about doing things differently and better. When we were just twenty-five people, Jensen taught us to come here, take risks, do things outside the box, and make mistakes. I encourage you to do all three. But don’t make the same mistake twice, because we will fire you in a heartbeat.” (Page 77)
Jensen used this measure to gauge the performance of his employees. He would reprimand subordinates who set goals that referred to what the company had already done before or what the competitors were doing in that moment. As he saw it, he needed to prevent the kind of internal rot that he observed at other companies, where employees often manipulated their projects to provide steady and sustainable growth that would advance their individual careers, when in reality they were making only incremental improvements that actually hurt the company (Page 78)
They only cared about what would be possible with the maximum amount of effort and minimum amount of wasted time. Much of what the company learned on the RIVA 128 became standard in its future chip development. From that point on, Nvidia had software drivers ready at the beginning of chip production: the drivers would already have been tested across all the important applications and games and to ensure compatibility with prior Nvidia chips. This approach became a significant competitive advantage for Nvidia, whose rivals had to develop separate drivers for different chip-architecture generations. (Page 79)
Tester Henry Levin recalls that whenever he found himself working late, he was never the only one there. Even when he stayed to 10:00 p.m. or later, Nvidia’s graphics architects would still be at the whiteboard, passionately discussing chip optimization and rendering techniques. His contemporary, Director of Materials Ian Siu, has imprinted on his memory the image of colleagues spending the night at the office, even over the weekends, after bringing sleeping bags to work. Employees would also bring their kids to the office so that they could spend time with their families without leaving their workplace. (Page 80)
“I’m super tired. I need to get up. It’s hard,” she told herself. “But we need to kill Intel. Must kill Intel.” (Page 83)
“The first time we came in second place, Jensen sternly told me: Second place is the first loser,” Logan said “I never forgot it. I realized I’m working for a boss who believes we have to win at everything. It was a lot of pressure.” (Page 85)
Every time he lost, Jensen would swipe his arm across the board, knocking over the pieces, and storm away. He would sometimes later insist on a rematch on the ping-pong table. Ribar graciously accepted, knowing Jensen was purposely shifting the competition onto more favorable territory. “He’s good at ping-pong,” recalled Ribar. “I’m okay, but he would just kill me to get his revenge. (Page 87)
When Nvidia was founded in 1993, Jensen struggled to find chip-manufacturing capacity. He had cold-called Taiwan Semiconductor Manufacturing Company (TSMC)—the best-regarded manufacturer in the world and the one Don Valentine at Sequoia had recommended that Nvidia partner with from the start—multiple times without success. (Page 88)
Companies that led one year, such as S3, Tseng Labs, and Matrox, were often displaced within one or two chip generations. “Mike, I don’t get it,” he said. “If you look at the PC graphics industry, why is it that one company can never hold a lead more than two years?” (Page 93)
Early on, Curtis Priem had invented a “virtualized objects” architecture that would be incorporated into all of Nvidia’s chips. It became an even bigger advantage for the company once Nvidia adopted the faster cadence of chip releases. Priem’s design had a software-based “resource manager,” essentially a miniature operating system that sat on top of the hardware itself. The resource manager allowed Nvidia’s engineers to emulate certain hardware features that normally needed to be physically printed onto chip circuits. This involved a performance cost but accelerated the pace of innovation, because Nvidia’s engineers could take more risks. (Page 95)
essentially a miniature operating system that sat on (Page 95)
Unlike Nvidia’s product road map, which created efficiencies by making multiple derivative versions of a single chip for a focused area of the market, 3dfx had an overly complicated lineup aimed at too many different customer segments and didn’t plan to reuse a common core-chip design. 3dfx then decided to expand into an entirely new part of the graphics industry. (Page 99)
Dwight Diercks said Montrym’s defection was “a watershed moment, because so many engineers revered John, and they all wanted to come work with John.” After Montrym joined Nvidia, every time the company posted a job opening for software developers or chip engineers, résumés and interview requests from Silicon Graphics employees would pour in. (Page 102)
Jensen was more defiant than enthusiastic. “We have had some setbacks, but I’m told I’m the hardest CEO to kill,” he told a Wall Street Journal reporter when asked for comment about the IPO. (Page 105)
Jensen and Diskin spent hours on their pitch material, sometimes working until midnight and then starting again at 8:00 a.m. the next day. Both men kept up the same intense work rate that had seen the company through its deepest crises and during the development of the original RIVA 128 and the race to get its derivative, the RIVA 128ZX, out to compete with Intel’s i740 chip. This time, they did not have the threat of bankruptcy hanging over their heads. Even so, they pushed themselves just as hard to salvage an opportunity to break into a lucrative new market. (Page 109)
It was the wrong thing to say to Jensen, who wanted to build a more communal culture. Nvidia as a whole was to be recognized for its achievements—not individuals. After Jensen came back from important business trips, Priem observed that he would always describe his own actions using the plural “we” instead of the singular “I.” Priem was initially skeptical, thinking, “What’s this ‘we’ stuff? I don’t know anything about negotiating contracts with fabs. But Jensen was right. We all did it together. We all share the credit.” When it came to chip designs, Priem could get possessive. He tended to talk about it as “his” work, “his” architecture. (Page 111)
He gave Priem a choice and an ultimatum: return to work full-time, transition into a part-time consulting role for Nvidia, or resign. Jensen even suggested a new mobile architecture project that Priem could work on, so that he could oversee one last job before retiring. Priem decided to leave Nvidia. “I was tired, beaten up, and demoralized. I needed to resign,” he recalled. “I always wish I could have stayed.” (Page 112)
But most of all, it had Jensen, who had learned how to manage the company as an extension of himself. Everyone at the company shared his singular focus on the mission. Everyone shared his work ethic. Everyone worked as fast as humanly possible in order to keep Nvidia one step ahead of the competition. And if anyone faltered or doubted, a sharp word from Jensen quickly brought that individual into line. Some investors did believe in Jensen’s future vision for Nvidia and his ability to keep the company trained on that vision. (Page 114)
In 1999, it launched the successor to the RIVA TNT2 series, which it called the GeForce 256. (Page 118)
Unlike the CPU, which sounded like the main piece of equipment essential to any computer, graphics cards were just one peripheral among many. A special designation for graphics chips that also drew an explicit comparison to the CPU would, for the first time, make them stand out as truly exceptional. (Page 119)
The world understood that CPUs were supposed to cost hundreds of dollars. Nvidia chips were sold at wholesale for less than $100 each, even though they were just as complex as, and had more transistors than, CPUs. Once the company started marketing all of its chips as GPUs, the pricing gap narrowed considerably. (Page 119)
Vivoli had another idea for the GPU’s launch: to actively intimidate rivals. An Nvidia marketer unfurled a banner advertising the GeForce 256 on a highway overpass that led straight to the headquarters of 3dfx (this was before its eventual bankruptcy). The banner announced that the new Nvidia GPU would change the world and crush the competition. The state police quickly removed the banner, which was being displayed illegally, and Nvidia received a formal reprimand. Still, it had served its purpose. “It was Art of War. We wanted to demoralize them,” Vivoli said. Nvidia was learning how to bend the world to its will. (Page 121)
Early graphics pipelines involved fixed-function stages each with a handful of hardwired operations. Nvidia and its competitor graphics-card makers each defined how its chips would handle all four stages in the pipeline; third-party developers could not change how the chips rendered anything, meaning that they could only create visual effects and artistic styles from a menu of options set by the chip designers Because every programmer had to use the same handful of fixed-function operations, every game on the market looked similar—none could stand out through visuals alone. David Kirk, Nvidia’s chief scientist, wanted to change all this by inventing a true GPU. His idea was to introduce a new technology called programmable shaders. These would open up the graphics pipeline to third-party developers, giving them the ability to write their own rendering functions and exert more control over how they presented their games visually. The shaders would allow developers to make visuals in real time that rivaled the best computer-generated graphics in movies. (Page 121)
In February 2001, Nvidia released the GeForce 3, whose programmable shader technology and support for third-party development of its core graphics functions made it the first true GPU. Kirk’s analysis was proved to be correct. The GeForce 3 was a blockbuster success. By the third fiscal quarter of 2001, Nvidia’s quarterly revenue reached $370 million—an 87 percent year-over-year increase. (Page 122)
Vivoli and Diskin went to meet with Jobs at Apple’s headquarters. During the first part of the demo, the Nvidia team showed off Luxo Jr. using similar shots and angles as the original. It was suitably impressive. Jobs said that it “looks good.” Then they ran the demo again, but this time Vivoli started clicking around the demo, which changed the position or the angle of the camera. The camera movements showed that, unlike a static video, Nvidia’s chips could render the entire scene in real time. The user could change and watch the scene from any angle with realistic lighting and shadow effects. Now Jobs was blown away. It was impressive enough that Nvidia’s GPU could render animations in real time, and with comparable visual fidelity, that had taken Pixar’s supercomputers weeks to generate. But on top of that, it offered real-time interactivity. Jobs decided that the Power Mac G4 computer would offer the GeForce 3 as a premium option. (Page 125)
Nvidia would go from no presence in Apple laptops to a nearly 85 percent share of Apple’s entire computer lineup in a matter of years. Diskin got the chance to prove himself, and Nvidia’s chips, thanks not only to his demo but also to his quick thinking and his willingness to challenge one of the most intimidating figures in the tech industry. (Page 126)
During the NV30’s development, it became clear that the chip would not win many of the benchmarks for the games that consumers then cared most about. For the first time since the NV1, Nvidia was about to release a card that was not at the very top of the market in terms of performance. ATI, in contrast, had agreed to sign the contract with Microsoft so that it could optimize the R300 with Direct3D 9 from the start. The chip and the new card that housed it, the Radeon 9700 PRO, worked perfectly and was fully compliant with Microsoft’s latest release of the API. (Page 129)
Harris learned there was a chip team within Nvidia working on a secret project code-named the NV50. Most chip designs were only one or two generations removed from the current architecture. The NV50 was Nvidia’s most forward-looking chip under development: it would not be released for several years. It would have its own dedicated compute mode, so that its GPU would be easier to access for nongraphics applications. Instead of Cg, it would utilize extensions to the C programming language, a widely used general-purpose language. And it would enable parallel compute threads with access to addressable memory—in essence, allowing the GPU to perform all the functions of a secondary CPU that might be needed in scientific, technical, or industrial computing. (Page 139)
G80 for use in its GeForce line of graphics cards, alongside CUDA in November 2006. It would be the company’s first GPU chip with a computing function. The G80 boasted 128 CUDA cores, which are extra hardware circuits used to support CUDA functionality. The GPU was able to run up to thousands of computing threads concurrently across those cores by using a hardware multithreading feature. By comparison, Intel’s main Core 2 CPU at the time only had up to four computing cores. (Page 141)
“CUDA added a ton of cost into our chips,” Jensen acknowledged “We had very few customers for CUDA but we made every chip CUDA compatible. You can go back in history and look at our gross margins. It started out poor and it got worse.”8 Nevertheless, he believed so strongly in CUDA’s market potential that he remained committed to the course he had chosen, even as his investors demanded a strategic course correction. (Page 142)
After the morning presentations, it was clear that the analyst group was skeptical about CUDA and saw it mainly in terms of the negative impact it was having on Nvidia’s profit margins. (Page 143)
Nvidia saw a prime opportunity to both help universities and drive adoption of its GPUs. After making several ad hoc donations, Kirk formalized a program with Caltech, the University of Utah, Stanford, the University of North Carolina at Chapel Hill, Brown, and Cornell. Nvidia would provide graphics cards and financial donations to the schools, and in exchange the schools would use Nvidia hardware in graphics programming classes. “It was not entirely selfless,” Kirk said. “We wanted them to use our hardware instead of AMD’s hardware for their teaching.” (Page 144)
Eventually, he found himself pitching Richard Blahut, the head of the Electrical and Computer Engineering Department at the University of Illinois at Urbana-Champaign. Blahut told Kirk it was a really good idea but said that if Kirk was serious about it, he should teach the class himself. (Page 145)
In 2007, Kirk flew from Colorado to Illinois every other week to give his lectures. At the end of the semester, the students carried out CUDA research programming projects and published their work. Other researchers around the country began to request lectures and teaching materials from Kirk and Hwu, so they recorded their classes and made their videos and notes freely available online. The following year, Nvidia named the University of Illinois at Urbana-Champaign the first CUDA Center of Excellence and provided the school with more than $1 million as well as thirty-two Quadro Plex Model IV systems, each with sixty-four GPUs—the most advanced machines Nvidia made. “David Kirk and Wen-mei Hwu were the evangelists,” said Bill Dally, Kirk’s eventual successor as Nvidia’s chief scientist. They “taught teacher’s courses around the country to basically spread the religion of GPU computing, and it really took off.” (Page 146)
But because Kirk’s was the first course of its kind, there was no common syllabus or set of standards, no textbook to use. So Kirk and Hwu wrote one. Their first edition of Programming Massively Parallel Processors, which was published in 2010, sold tens of thousands of copies, was translated into several languages, and was eventually used by hundreds of schools. It was a major inflection point in attracting attention, and talent, to CUDA. (Page 146)
Second, he started conducting annual two-day technology summits where Nvidia employees could interact with and learn from scientists themselves. Dozens of researchers from the life-sciences industry—chemical engineers, biologists, pharmacologists, as well as the software developers who supported their work—arrived in Santa Clara from across the United States but also from Europe, Japan, and Mexico. On the first day, Nvidia’s engineers would tell them about future improvements to CUDA, including software and hardware advances. The scientists and developers would then give their feedback. (Page 147)
The scientists and researchers appreciated Nvidia’s transparency and willingness to listen, in turn. “They saw us as resources,” said Ross Walker, a biochemistry professor at the University of California, San Diego. “We could go tell them, ‘we need this feature,’ and they would change the design of the chip or add it to CUDA. There was no way on Earth Intel would have ever done anything like that.” (Page 147)
Afterward, he went to Jensen. “Thank you for coming to the meeting. I really appreciated it,” Moore said. Jensen, however, saw the meeting differently. “It ended well, but Derik, let me tell you your failure here.” The remark shocked Moore. “It scared the crap out of me,” he recalled. “The failure here was you did not tell us what the company was going to ask in advance,” Jensen said. “Nobody likes surprises. Don’t ever let that happen again.” (Page 156)
Nvidia invested heavily in deep learning from the outset, dedicating substantial resources to creating CUDA-enabled frameworks and tools. This proactive approach paid off when artificial intelligence exploded in the early 2020s, because Nvidia was already the preferred choice of AI developers everywhere. Developers want to build AI applications as quickly as possible with minimal technical risk, and Nvidia’s platform is far more likely to have fewer technical problems, because the user community has already, over more than a decade, fixed bugs or figured out optimizations. Other AI chip vendors never really had a chance. (Page 158)
Nvidia made a general-purpose GPU that represented the first major leap forward in computational acceleration since the invention of the CPU. The GPU’s programmable layer, CUDA, was not only easy to use but also opened up a wide range of functions across scientific, technical, and industrial sectors. As more people learned CUDA, the demand for GPUs increased. (Page 158)
But in the new, larger Nvidia, he found that it was difficult to reach all the employees on a consistent basis. Jensen decided to offer Nvidia employees more direct criticism in larger meetings, so that more people could learn from a single mistake. “I do it right there. I give you feedback in front of everybody,” he said. “Feedback is learning. For what reason are you the only person who should learn this? You created the conditions because of some mistake you made or silliness that you brought upon yourself. We should all learn from that opportunity.” Jensen displayed his trademark directness and impatience in all settings. He would often chew people out for fifteen minutes straight, regardless of the venue. (Page 161)
Do your job. Don’t be too proud of the past. Focus on the future. (Page 162)
As a result, he rarely held one-on-one meetings with his direct reports, at least when it came to such open-ended topics. Instead, he focused on providing them collectively with information from across the organization, as well as with his own strategic guidance. This would ensure that every part of the business was aligned and allow Jensen to manage more executives in a manner that actually added value. (Page 163)
When a typical CEO would have eight or nine people in a room for big executive meetings, Jensen would have a packed house. “Everyone could hear what he was telling the executive staff,” Keane said. “It kept everybody in sync.” (Page 165)
“Strategy is not words. Strategy is action,” he said. “We don’t do a periodic planning system. The reason for that is because the world is a living, breathing thing. We just plan continuously. There’s no five-year plan.” (Page 166)
“Nothing stays. Nothing festers. You answer and move on it,” former head of human resources John McSorley said Jensen would often respond to e-mails within minutes of receiving them and wanted a response from an employee within twenty-four hours at most. The responses had to be thoughtful and backed by hard data. Those that fell short of his high standards would get a typically sarcastic response: “Oh, is that right?” (Page 169)
At the whiteboard, Jensen will sketch out how to organize a particular market, how to accelerate growth of a particular product, and the software or hardware technical stacks involved in a particular case. His whiteboarding creates a specific kind of meeting, one dedicated to solving problems, not reviewing things that have already been done. (Page 170)
But there is more to it than that suggests. Whiteboarding forces people to be both rigorous and transparent. It requires them to start from scratch every time they step up to the board, and therefore to lay out their thinking as thoroughly and clearly as possible. It becomes immediately apparent when someone hasn’t thought something through or bases their logic on faulty assumptions, unlike with a slide deck, where you can hide incomplete thoughts in pretty formatting and misleading text. At the whiteboard, there is no place to hide. (Page 172)
Icahn observed that competent executives often get sidelined in favor of more likeable but less capable ones because of behavioral incentives inside companies. The personalities who ascend the corporate ranks resemble college fraternity presidents. They become friendly with the board of directors and are not threatening to the current CEO. They’re not prodigies, but they’re affable, always available for a drink when you are feeling down. As Icahn put it, these figures (they are mostly men) are “not the smartest, not the brightest, not the best, but likeable and sort of reliable.” (Page 175)
Jobs was right. Under Ballmer, Microsoft missed the shift to mobile computing and also made a series of terrible acquisitions, including aQuantive and Nokia. Microsoft’s stock price fell more than 30 percent during Ballmer’s fourteen years as CEO. Apple had previously faced its own challenges under a chief executive with more of a business than technical background. Jobs was famously ousted by Apple’s board of directors in 1985, who replaced him with John Sculley, a former marketing specialist at PepsiCo. Sculley met with some initial success, including with his strategy of selling incrementally better computers at higher and higher prices. He then made several misguided technology-product decisions, such as introducing the Newton personal digital assistant and selecting PowerPC processors for the Mac in the early 1990s. The stagnation in technical innovation brought Apple to the brink of bankruptcy later in the decade. (Page 176)
Intel offers another example. Bob Swan joined the chip maker as its chief financial officer in 2016 and rose to the position of CEO two years later. Swan had a primarily financial background; he had previously held CFO roles at eBay and Electronic Data Systems, the company founded by former IBM salesman H. Ross Perot. Under Swan’s leadership, Intel suffered from repeated delays in moving to more advanced chip-manufacturing technologies and its next generations of processors, falling behind its main CPU competitor, Advanced Micro Devices. Worse, it seemed Swan was mainly focused on executing a substantial multibillion-dollar stock buyback program and issuing billions in dividends to lift the company’s stock price, which siphoned money away from R&D investments. Intel floundered so badly that it lost significant market share across its businesses and yielded pole position in the technology of CPUs to AMD, which was then led by Lisa Su, who, in contrast to Swan, had a strong engineering pedigree. (Page 176)
Under Swan, Intel made a string of poor product decisions. On the AI front, it shut down Nervana Systems, even though the start-up had a promising product that was nearly ready. Instead, the company restarted its AI efforts with Habana, effectively negating the prior several years of development time. Nvidia’s head of GPU engineering, Jonah Alben, commented on Intel’s AI plans after the company acquired Habana. “Intel’s AI strategy is like throwing darts. They don’t know what to do but feel like they need to buy something, so they are buying everything,” he said In 2021, Swan resigned as Intel’s CEO and was replaced by Pat Gelsinger, who came with an impressive background in engineering. One of his first decisions was to halt the stock buybacks. (Page 177)
Krewell knew Jensen wasn’t scheduled to speak and asked him what he was doing at the conference. Jensen replied, “I’m here to learn.”7 Nvidia’s CEO had not assigned someone to attend and take notes on his behalf. He had shown up himself so he could absorb the recent developments in artificial intelligence. He wanted to be deeply involved in the space, attending sessions and talking with presenters, students, and professors. Later on, he began hiring many of the people he met at the conference. (Page 178)
Most senior leaders at Nvidia agree that Jensen relies on his people to exercise their good judgment on interpreting his directions. He doesn’t want to control every decision; in fact, being overly prescriptive can stifle the very independence and bias toward action that he seeks to cultivate. Rather, he wants to make sure that they have done their diligence and considered all possible effects of their decisions. (Page 180)
Nvidia’s extreme work culture stems from the chief executive himself, who lives and breathes his job and looks down on anyone who isn’t as committed. “I don’t actually know anybody who is incredibly successful who just approaches business like, ‘This is just business. This is what I do from 8 to 5, and I’m going home, and at 5:01, I’m shutting it down,’ ” Jensen has said “I’ve never known anybody who is incredibly successful like that. You have to allow yourself to be obsessed with your work.” (Page 181)
He lacks sympathy for anyone who works less than he does, and he does not believe that he has missed out on anything in life by giving himself so completely to Nvidia. When 60 Minutes interviewed Jensen in 2024 and asked about employees who said working for him was demanding, that he was a perfectionist and not easy to work for, he simply agreed. “It should be like that. If you want to do extraordinary things, it shouldn’t be easy.” (Page 181)
In 1981, IBM introduced the IBM PC, revolutionizing the world of computing. The computer manufacturer made two critical choices for the PC that would define the industry. The first was choosing an Intel 8088 chip as the PC’s processor. The second was deciding on MSDOS, from a small software start-up called Microsoft, as the PC’s operating system. (Page 182)
Intel, by contrast, missed a pair of generational opportunities: the arrival of smartphone processors and the rise of AI software. In 2006, Steve Jobs asked Intel CEO Paul Otellini whether the chip maker would be willing to supply processors for the upcoming iPhone. In a fateful decision that would prevent Intel from participating in the future of the smartphone chip market, Otellini declined. “There was a chip that they were interested in, that they [Apple] wanted to pay a certain price for and not a nickel more, and that price was below our forecasted cost. I couldn’t see it,” he said in a 2013 interview with The Atlantic. “The world would have been a lot different if we’d done it.” (Page 183)
Also in 2006, Intel sold its XScale unit, which was developing power-efficient ARM-based processors for mobile devices, to Marvell Technology for $600 million. This left the company without important expertise just before the smartphone market came to be dominated by such processors. (Page 184)
Intel made a series of missteps in its core business. It was slow to purchase and introduce new chip-manufacturing equipment from the Netherlands-based company ASML, which uses the advanced chip-manufacturing technology called extreme ultraviolet (EUV) lithography, and it underinvested in production techniques that are based on EUV lithography. As a result, Intel fell behind TSMC in its ability to produce more advanced chips at high volume. In 2020, when Intel announced another round of delays in the transition to seven-nanometer manufacturing, many customers abandoned it for competitors such as Advanced Micro Devices, which designs semiconductors and pays TSMC to make them. And in the same year, Apple started to replace Intel as a Mac processor supplier with its internally designed chips that are based on the ARM-chip architecture that powers the iPhone and which are now used across its entire Mac lineup. (Page 184)
Nvidia also made smart acquisitions, including the high-speed networking leader Mellanox, to fill out the company’s data-center-computing product offering. Nvidia took these decisions in the face of demands from Wall Street to reduce costs and increase profits—exactly the kind of strategy that Intel adopted when it declined to pursue ARM architecture and GPUs. It was an instance of the innovator’s dilemma: Intel, as the incumbent, failed to capitalize on new technology, allowing the more agile Nvidia to undercut its entire business model. (Page 185)
So far, every major computing era has seen technology favor the big players who can develop a market-leading platform—a “winner take most” dynamic. WinTel’s dominance in PCs is a model for Nvidia’s leadership in AI hardware and software. In an August 2023 report, Jefferies analyst Mark Lipacis estimated that WinTel generated an incredible 80 percent of the operating profit of the PC industry era With the rise of the internet, Google captured 90 percent of the search market And Apple has been able to generate nearly 80 percent of the profits of the smartphone industry era. (Page 185)
For the first fifty years of computing history, the most important chip inside the computer was the central processing unit, or CPU. The CPU is a generalist, capable of performing a wide variety of tasks. It moves from task to task with great speed and can dedicate significant processing power to each operation. Nevertheless, it can handle only a few operations simultaneously because of its limited number of cores, which process only a few computation threads at once. The GPU, in contrast, is optimized for volume over complexity. It contains hundreds or thousands of tiny processing cores, enabling it to break down tasks into numerous simpler operations executed in parallel. (Page 190)
In the third contest, which occurred in 2012, University of Toronto professor Gary Hinton and two of his students, Ilya Sutskever and Alex Krizhevsky, put forward an entry they called AlexNet. Unlike the rest of the field, which had started developing algorithms and models before optimizing them for use on ImageNet, the AlexNet team took the opposite approach. They used Nvidia GPUs to support a small-scale deep-learning neural network that was fed ImageNet content and which then “learned” how to build relationships between images and their associated tags. The team did not set out to write the best computer-vision algorithm possible; in fact, they did not write a single line of computer-vision code themselves. Instead, they wrote the best deep-learning model they could—and trusted it to figure out the computer-vision problem on its own. (Page 196)
Jensen announced the change in strategic focus in a company all-hands meeting. “We need to consider this work as our highest priority,” he said He explained that Nvidia had to get the right people working on AI. If they were currently assigned to something else, they would change focus and work on AI, because it was going to be more important than anything else they could possibly be doing. (Page 198)
“The entire team—the GPU group, Jensen, and myself—agreed to incorporate significantly more support” for AI despite how late in the development process they were, said Dally. That “support” included the development of an entirely new type of tiny processor, called the Tensor Core, which was integrated into Volta. (Page 199)
Kirk had championed the new division precisely because he saw that the most complex problems in computer graphics would require sustained research over time, even if commercialization could take a lot more time. Within a few weeks of starting, Luebke had three new coworkers. At their first team lunch with Kirk, they asked where they might start. Kirk was noncommittal: he told them it was up to them to figure out what their job would be. He did offer some basic guidelines, at least. They should work on something important to the company. They should create significant impact with their projects. And they should focus on innovations that would not occur in the regular course of Nvidia’s business—inventions that would not be possible without dedicated, long-term work, of the kind that the rest of the company was not set up to do. (Page 210)
While it would roll out new chips and boards on a very fast schedule, it would now, with Nvidia Research and other groups, pursue “moonshots” at the same time. “When we got to the next-generation Ampere, we had enough momentum for ray tracing and DLSS to make that product a home run,” Jeff Fisher said. It was a further, institutionalized form of protection against the kind of stagnation that Clayton Christensen warned about in The Innovator’s Dilemma: the inevitable desire to focus on the company’s core, profit-generating business at the expense of investing in more exploratory innovations that might not be commercially viable for years. According to Jon Peddie Research, Nvidia’s share of the discrete or add-in board GPU market has stayed at roughly 80 percent over the past decade, as of this writing. Even though AMD offers better price-to-performance based on traditional metrics, gamers keep choosing Nvidia for its ability to innovate. Both ray tracing and DLSS have become must-have features that developers have incorporated into hundreds of games. And the features perform the best on Nvidia graphics cards, making it difficult for AMD to compete effectively. (Page 216)
The only thing we’re optimizing for is this: Is it incredibly cool, and are people going to like it?” (Page 218)
Its first-quarter data-center business for fiscal 2024 rose by 427 percent from the prior year, to $22.6 billion—driven primarily by artificial-intelligence chip demand. Unlike software, which is easy to scale at essentially no incremental cost, Nvidia is producing and shipping complex high-end AI products and systems, some of which contain up to 35,000 parts. There was no precedent for this level of hardware growth at a technology company the size of Nvidia. (Page 222)
Jensen is always trying to figure out the next thing and what Nvidia can do to prepare itself to take advantage of it. In early 2023, he was asked by a student to predict what will follow, and build upon, AI. “There’s no question,” he said. “Digital biology is going to be it.”12 Though biology is one of the most complex systems, Jensen explained that for the first time in history, it could be digitally engineered. With AI models, scientists can now start to model the structure of biological systems in greater depth than ever before. (Page 227)
The report revealed that CIOs plan to increase their AI compute hardware spending by more than 40 percent annually over the next three years, going from 5 percent of total IT budgets to more than 14 percent in 2027. One-third of CIOs also said they will defund other IT projects in order to support the new AI investments. (Page 231)
A former senior executive at a large software company said that he was always struck by how you could talk to multiple Nvidia employees and they would never contradict one another. The message from the top was consistent, and Nvidia staff learned it and made it their own. He drew a contrast with almost every other company he’d ever worked with, whose representatives sometimes argued with each other in front of external clients. (Page 233)
Jensen can also reach down into the organization at any time and award stock directly, without waiting for an annual compensation review. This allows him to ensure that people who are doing great work feel appreciated in the moment. It is yet another sign of his interest in every aspect, and level, of the company. (Page 238)
“That kind of characterizes Jensen because he always wants to know what’s going on out there,” Jeff Fisher said. “He just wants to know what’s going on out in the world, so he can make better decisions.” (Page 242)
“LUA means pay attention because you’re talking about something important and you need to do it properly,” Catanzaro said. “He does not like it when people put up an abstraction or sort of puffery to deflect an answer to a question. Everyone who works for Jensen has heard LUA.” (Page 242)
Curtis Priem might have decided to make the NV1 chip more like everything else on the market, and it might have succeeded. But that would have deprived Nvidia of the chance to learn from its failure and respond with the RIVA 128—the chip that saved the company. “Nvidia would have been a failure if NV1 had not failed,” Priem said. (Page 244)
We are sometimes told, by various self-help experts and gurus, that we can make more money while working less. Jensen is the antithesis of that notion. There are no shortcuts. The best way to be successful is to take the more difficult route. And the best teacher of all is adversity—something he has become well acquainted with. It is why he still keeps going at a pace that would see most other people, at any age, burn out. (Page 245)
“Criticism is a gift”: Providing direct feedback leads to continuous improvement. (Page 247)