“The Nvidia Way” by Tae Kim: Review, Notes, and Analysis
Poor Ash’s Almanack > Book Reviews > Business & Strategy
Last Updated: October 2025
Overall Rating: ★★★★★☆☆ (5/7)
Per-Hour Learning Potential / Utility: ★★★★★☆☆ (5/7)
Readability: ★★★★★★☆ (6/7)
Challenge Level: 2/5
Blurb/Description: This book is the first to tell the full story of Nvidia, as well as its founder Jensen Huang. It details the company’s near-death experiences, its unique work culture, and the forward-thinking, strategic bets – like the development of CUDA – that positioned it to become the enabling technology for the age of artificial intelligence.
Summary: This book details the rise of Nvidia, arguing its success stems not from luck but from a unique corporate culture: “the Nvidia Way.” This culture, driven by co-founder and CEO Jensen Huang, combines unusual independence for each employee with the highest possible standards. It demands speed, accountability, and an intense focus on the future, driven by Huang’s belief that complacency is the greatest threat (The Innovator’s Dilemma is apparently one of Jensen’s favorite books). The book chronicles Nvidia’s history, from near-death experiences with failed chips to strategic masterstrokes. A central focus is the massive, costly bet on CUDA, a software layer that transformed GPUs into general-purpose parallel processors. This decision – more a decade and a half before GPT-3.5 took the world by storm – turned out to be key to the AI revolution. Despite being in the details, Jensen also takes a “CEO math” approach, which prioritizes big-picture thinking over false accuracy.
Highlights: The book offers a fantastic inside look at a unique and demanding corporate culture, grounded in specific, illustrative anecdotes. It details major failures, such as the first chips that “nearly killed Nvidia,” and Jensen Huang’s frank admission about the early days: “we sucked at a lot of things.” The explanations of key strategies are exceptionally clear, and the book provides in-depth reporting not found elsewhere.
Lowlights: The focus of the book is wildly misplaced; readers expecting interesting insights into Nvidia’s recent path to becoming the most valuable company in the world will be solely disappointed. While the book does not proceed exactly chronologically, the section starting in 2002 is on Page 133 (of 245), while 2013 – present is only covered from Page 187 on; in other words, the entire past decade gets less than 25% of the book, and the last two decades get less than half, while the 1990s get 50%. It’s not so much that the first section is too long, but rather that the rest of the book is comparatively too short. This is the primary reason I’ve rated it a 5 instead of a 6.
The author notes that he was a big fan of video games and the book seems to dwell for far too long on 1990s video game nostalgia, including career backgrounds of key individuals at other companies, things that happened with other graphics cards companies of the era, as well as too much detail on 1990s video games like “Quake” that might have been groundbreaking at the time, but are completely irrelevant today to a broad/general audience (I’ve played a lot of video games in my time, and found the discussion tedious).
There’s vanishingly little discussion of LLMs or transformer architecture and how NVDA ended up becoming essentially the sole-source provider of chips for AI, how and why others haven’t been able to make a dent in the last few years despite literally trillions of dollars of incentives to do so, and only brief discussion of recent strategic decisions such as the acquisition of Mellanox. Even the coverage of CUDA development and deep / machine-learning applications and the rise of GPGPU feels rushed given their importance. The book would have benefited from 50–100 more pages on the past decade or by trimming the dated 1990s focus in favor of serious analysis of Nvidia’s role in the AI revolution and why competitors like AMD and Intel fell behind.
Mental Model / ART Thinking Points:
- The entire “Nvidia Way” is an organizational design to combat complacency and the status quo bias. Jensen Huang is described as obsessed with avoiding the “Innovator’s Dilemma,” where, as the author puts it, “a company’s success often contains the seeds of the same company’s own failure.”
- Huang’s “CEO math” concept—using round numbers to think strategically—is a great example of prioritizing accuracy over “false” precision, a key part of the product vs. packaging mental model. He learned “false accuracy is pointless” if it slows you down.
- The company culture is built on a growth mindset forged through adversity. Huang’s advice to students is, “I wish ample doses of pain and suffering upon you,” believing that “Greatness comes from character,” which is built by overcoming setbacks. (Rob Henderson would strongly disagree with that, and I don’t think Jensen’s approach is for everybody.)
- Nvidia’s accountability system, which designates a “PIC” (Pilot in Command) for every project, is a powerful form of structural problem solving. It ensures there is “a name attached” to every deliverable, creating clear incentives.
- The book explores the massive tradeoff Nvidia made with its CUDA platform. It added a ton of cost and hurt profitability for years, but Huang remained committed.
You should buy a copy of The Nvidia Way if: you are interested in corporate culture, technology strategy, and a deep case study on how to avoid the “Innovator’s Dilemma.”
Length: ~245 pages (core text)
Reading Tips: Focus on the ‘why’ behind Nvidia’s strategies. The “whole cow” model, “CEO math,” and the “PIC” system are all solutions to specific problems (complacency, false accuracy, lack of accountability) that Jensen Huang identified.
Pairs Well With: The Innovator’s Dilemma (review + notes) by Clayton Christensen, which is cited as a major influence on Jensen Huang’s strategic thinking. The Frackers (FRK review + notes) by Gregory Zuckerman – another example of how innovation works in a very different industry. Finally, How We Got To Now (NOW book review + notes)
Reread Value: 4/5
More Detailed Notes + Analysis (SPOILERS BELOW):
IMPORTANT: the below commentary DOES NOT SUBSTITUTE for READING THE BOOK.
Page 1: The whiteboard is CEO Jensen Huang’s preferred medium for communication, earning him the nickname “Professor Jensen.” He uses it to collaboratively sketch out problems and ideas. The whiteboard is a central symbol of Nvidia’s culture, representing both possibility and ephemerality—the idea that “a successful idea, no matter how brilliant, must eventually be erased, and a new one must take its place.”
Page 2: The whiteboard is described as an accountability tool that “requires active thinking and inevitably reveals how well (or not) any employee, including an executive knows the material.” The book then connects Nvidia’s graphics processor architecture to its modern success, noting its design is well suited for AI workloads because of its ability to perform many calculations simultaneously, which is essential for large-language models.
Page 4: At a Caltech commencement speech, Jensen Huang reflects on Nvidia’s early failures. He resists nostalgia and looking back, admitting that in the early days, “we sucked at a lot of things.” He mentions the company’s first two chip designs, NV1 and NV2, which, per Kim, were “flops that nearly killed Nvidia.” His key takeaway from that period was: “We survived ourselves.”
Page 5: This page also touches on Nvidia’s multiple “near-death experiences.” Huang notes, “every single time, somebody left, and every single time, we picked ourselves up.” This forged a culture that “discourages looking back… in favor of focusing on the future.”
Page 6: The book defines “the Nvidia Way” – the company culture. They attribute their success not to luck or technical skill, but a “unique organizational and work culture.” This culture “combines unusual independence for each employee with the highest possible standards” and demands “an almost superhuman level of effort and mental resilience.” Jensen Huang’s core belief is that the company’s “worst enemy” is not competition, but “complacency that grips a successful company,” which he has structured Nvidia to eliminate. It’s an example of incentives and structural problem solving – I’m struck by how fiercely Jesnsen Huang has tried to resist the normal “big-company bureaucracy” bloat.
Page 11: Details Jensen Huang’s early life. He was born in Taiwan, and his family later moved to Thailand. Amidst political unrest, his parents sent him and his brother to live with an aunt and uncle in Tacoma, Washington. To learn English, his mother had the boys “each selecting ten words at random in the dictionary and memorizing them.” The page describes how Tacoma was “damp, dreary, and smelling of sulfur,” which reminds me of comments I’ve seen about the Seattle area in various other books such as books about Boeing, Starbucks, and Amazon.
Page 17: Covers Jensen’s early career and relentless drive. He completed his master’s degree over an arduous eight years, showing he can be “infinitely patient about certain things.” He left AMD for LSI Logic because LSI was pioneering new software tools for chip design. This move, which he knew “would mean taking a risk,” demonstrated his “forward-thinking nature.” At LSI, he met his future cofounders, Curtis Priem and Chris Malachowsky.
Page 18: This page explores Jensen’s core philosophy on “pain and suffering.” He believes that “Greatness is not intelligence; Greatness comes from character,” and that character can only be forged by “overcoming setbacks and adversity.” This is why his consistent advice to those seeking success is, “I wish ample doses of pain and suffering upon you.” This reflects a deep commitment to the growth mindset. Rob Henderson would strongly disagree with his wish, and I don’t think Jensen’s approach is for everybody (I strongly disagree as well).
Page 31: Co-founder Curtis Priem had a long-running passion project: a flight simulator. His old Atari 800 computer “didn’t have enough graphical processing power” to run it. Only when he got a Sun SPARCstation with the new GX graphics chip (which he had worked on) did a “realistic flight simulator became possible.” This reminds me of some passages from How We Got To Now (NOW book review + notes) by Steven Johnson, particularly the line from Tae Kim when he notes “in fact, no card then on the market could bring to life the game that Priem had imagined.” Often, someone comes up with an idea for something but there is a bottleneck or constraint – Gregory Zuckerman’s excellent The Frackers (FRK review + notes) provides another great example by discussing how hydraulic fracturing, or “fracking” – a technique that had been around for quite a while – couldn’t reach its full potential until, in the late ‘90s and into the 2000s, oilfield technologists began to perfect the art of drilling long horizontal wells in thin layers of hydrocarbon-rich shale rather than traditional vertical wells that barely intersected with the oil-containing zones for a small portion of their length.
Page 36: As with the prior, Nvidia’s founding in 1993 was “perfectly timed” due to two major developments in 1992. The first was the industry’s adoption of the Peripheral Component Interconnect (PCI) bus, which transferred data at a much higher bandwidth. The second was Microsoft’s release of Windows 3.1, which “was intended to showcase the very latest in computer-graphics capabilities” such as “pixel-perfect [TrueType font] text… and high-quality video playback.”
Page 45: During an early pitch to venture capitalists, a partner at Kleiner Perkins insisted Nvidia should manufacture its own circuit boards. Co-founder Chris Malachowsky immediately rejected the idea, demonstrating a clear focus on their circle of competence. Malachowsky asked, “Why would we compete on a resistor? … We’ll stick to what were good at.” Kim notes that “Nvidia had to focus its resources on the single best opportunity rather than spread themselves thin chasing all possible ones.” Opportunity costs.
Page 58: Details the failure of Nvidia’s first chip, the NV1. A key miscalculation was memory: PC memory prices had “plummeted from $50 per megabyte to $5 per megabyte,” which meant the NV1’s stinginess with integrated onboard memory “was no longer much of a competitive advantage.” In other words, memory was no longer a bottleneck, so Nvidia’s approach – which shortly prior might have been very valuable – was no longer adaptive.
Page 60: The team learned critical lessons from the NV1 failure. They had “overdesigned the product” by including features like audio, when the market simply wanted “the fastest performance for the graphics.” Customers told them, “We really like your technology, so if you guys get rid of the audio, come back and see us.” Jensen summarized the lesson: “We learned it was better to do fewer things well than to do too many things…. Nobody goes to the store to buy a Swiss Army knife. It’s something you get for Christmas.”
Page 61: Jensen reflects that “we were just bad at our jobs. The first five years of our company – we had really talented people working super, super hard, but building a company is a new skill.”
Page 66: Following the NV1 failure, Nvidia was in a dire financial position. With only $3 million left in the bank, the company had just nine months of cash remaining. To survive, they had to pivot. Kim notes that “The NV1 had been designed to do what Nvidia engineers wanted rather than what the market wanted. The proprietary standards… showcased technical acumen but ended up alienating manufacturers.”
Page 77: Jensen implemented a culture of working at the “Speed of Light.” This philosophy demands that projects be “constrained only by the laws of physics,” not by internal politics, bureaucracy, or financial concerns. Nvidia’s unique culture:
“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… we take risks, do things outside the box, and make mistakes… but don’t make the same mistake twice.” – Michael Hara, Nvidia’s former director of marketing, on Nvidia’s resistance to status quo bias
This high-pressure environment was paired with a policy of “firing fast” if a new hire wasn’t working out.
Page 78: The “Speed of Light” concept was applied to the RIVA 128 chip. While most chips took two years to develop, Nvidia had only nine months. The Speed of Light – “the theoretical limit of what you could do – that’s the only thing we were allowed to measure against,” according to former executive Robert Csongor. This reminds me a lot of Brad Stone’s The Everything Store – Bezos was famous for wanting things to be done faster than was possible.
Jensen identified the bottleneck by asking engineer Dwight Diercks, “What’s the main limiting factor?” Diercks identified the software drivers as the “primary obstacle,” allowing the team to focus their efforts.
Page 83: Intel entered the graphics market with its i740 chip, a direct threat to Nvidia’s RIVA 128. Jensen responded with an impassioned speech at an all-company meeting, declaring, “Intel is out to get us and put us out of business… Our job is to go kill them before they put us out of business. We need to go kill intel. Must kill Intel.” This created an intense fight against a company 860 times its size.
Page 85: Jensen channels his inner tiger mom: “Second place is the first loser.”
Page 94: Jensen became obsessed with the problem of market leadership, noting “why is it that in the PC graphics industry, one company can never hold a lead for more than two years?” He identified a core mismatch: PC makers refreshed products twice a year, but a “new chip design took eighteen months.” Graphics technology advanced so fast that new chips were “functionally obsolete” by the time they were released. Jensen realized speed to market – i.e. shortening the design cycle – could help “keep Nvidia ahead of the competition… forever.”
Page 95: Nvidia’s solution to the 18-month design problem was to release a new chip derivative every six months. This faster cadence was made possible by co-founder Curtis Priem’s “resource manager” architecture, which was essentially a “miniature operating system” on the chip itself. This “secret sauce” allowed engineers to emulate new hardware features in software, accelerating innovation and reducing risk. If a feature wasn’t ready in hardware, they could “emulate it in software.”
Page 115: This page introduces the central concept of The Innovator’s Dilemma. The theory posits that “a company’s success often contains the seeds of the same company’s own failure.” Successful incumbents focus on their “current, profit streams” and ignore low-end disruptors, who then “iterate and innovate faster” and eventually take over the market.
Pages 116-117: Nvidia’s “whole cow” strategy was a direct response to the Innovator’s Dilemma. The company realized that it was throwing away parts that didn’t meet quality standards for high-end applications – but it cost merely a few dollars to package these chips into lower-end graphics cards that could serve other applications. Not only did this allow them to fend off lower-end competition, but it also massively improved their margins and yields. This strategy, called “ship the whole cow,” ended up nearly putting competitor S3 Graphics out of business.
Page 119: The now-ubiquitous term “GPU” (Graphics Processing Unit) was a deliberate marketing invention for the GeForce 256. The team wanted to signal that this chip was special. They wanted it to be clear that the chip was very poewrful. This branding created an “explicit comparison to the CPU,” helping to “make them stand out as truly exceptional” and justifying a premium price.
Page 131: The failure of the NV30 (GeForce 5800 Ultra) chip was a low point. The chip’s fan was notoriously loud, earning it a “leafblower” nickname. Jensen was furious, shouting at an all-company meeting that the NV30 was a “piece of shit … How could you not see the leaf blower issue?” Others noted that their competitor, ATI, missed a huge opportunity. If Jensen had been running ATI, “he would have put Nvidia out of business” with aggressive pricing. Instead, ATI kept its pricing high (equivalent to the NV30), allowing Nvidia to limp along.
Page 137: This page details the origins of GPGPU (General-Purpose computing on GPUs). In 2002, a researcher named Mark Harris wanted to simulate clouds and “hacked” an Nvidia GeForce 3 to do it. He and other researchers realized they could use the chip’s “programmable shader” functions, originally designed for pixels, to perform “matrix multiplication” for scientific computing.
Page 138: Using GPUs for non-graphics purposes required a “very specific skill set.” Researchers had to “‘translate’ their real-world problems into functions” using niche graphics programming languages. Mark Harris coined the term “general-purpose computing on GPUs,” or “GPGPU,” and created a website, GPGPU.org, which became a hub for this budding community. This is a sore point for me personally because I owned Nvidia in the early 2010s when it was perceived as a company levered to a supposedly-dying category (PC graphics) – everyone thought mobile games would take over. I was well aware of GPGPU at the time, and yet…
Page 139: Nvidia realized GPGPU was too difficult to work with and while the hardware was powerful, the software / ability to work with it was the bottleneck to mass adoption. Jensen appeared to be a master at design-based thinking. To open the market, they began work on the Compute Unified Device Architecture, or CUDA. CUDA was revolutionary because it “utilize[d] extensions to the C programming language,” a widely used language. This made it “possible for not only graphics programming specialists but also scientists and engineers to leverage the GPU’s power” – specifically, the “parallel compute threads with access to addressable memory… allowing the GPU to perform all the functions of a secondary CPU that might be needed in scientific, technical, or industrial computing.”
Page 142: The first CUDA-enabled chip, the G80, was an enormous financial bet. It cost $475 million to develop – a third of Nvidia’s total R&D budget over four years. The massive R&D spending caused Nvidia’s gross margin to fall from 45.6% in 2008 to 35.4% in 2010, just as the GFC destroyed both consumer demand for high-end graphics cards and corporate demand for GPU-powered workstations. Jensen acknowledged, “CUDA added a ton of cost… We had very few customers,” but he remained committed because “we were convinced that accelerated computing would solve problems that normal computers couldn’t. We had to make that sacrifice. I had a deep belief in its potential.”
Page 150: Introduces academic researcher Ross Walker, who was “disillusioned” with the bureaucratic process of allocating supercomputer time. Proposal funding rates were in the “low single digits,” and the system favored already-famous scientists. Walker described it as a catch-22: “The way to get supercomputing time was to already have resources.”
Page 152: Ross Walker discovered CUDA as a solution. He had first tried AMD’s competing cards but found their software “immature and were not easily programmable.” In contrast, CUDA was a “much easier environment to work with.” In 2009, he released a GPU-enabled version of his AMBER simulation software that “ran up to fifty times faster” than the CPU version. CUDA “made it possible for scientists to carry out important experiments on affordable hardware.” Specifically: “you could buy a $100 CPU and four $500 GeForce cards and have a workstation that was as powerful as a full rack of servers. It was a game changer.”
Page 157: Nvidia’s sales strategy was built on value, not price. The company “never discounted its chips,” not even to match competitors. The sales leader’s message was: “We don’t sell on cost… We believe that we bring exceptional value to the customer.” The CUDA platform itself, with “more than 5 million developers” and thousands of applications, creates an “incredible network” that reinforces this value. This is a clear example of a network effect.
Page 161: Jensen is known for giving harsh feedback publicly, based on the belief that “I give you feedback in front of everybody… we should all learn from that opportunity.” The book gives an example of him “publicly and repeatedly” humiliating Mike Rayfield, the project manager for the Tegra 3 chip, at an all-hands meeting. This “at-times harsh approach” was part of a high-pressure culture where people either proved themselves or left.
Page 166: This page details Nvidia’s “mission is the boss” philosophy. This concept is designed to make employees focus on “the work, not the hierarchy.” To enforce this, every project has a “PIC” (Pilot in Command) who reports directly to Jensen. This system creates “far more accountability” because “Everything has to have a name attached to it… you have to know who’s the PIC.” PICs were granted Jensen’s authority. Intriguingly, “after Jensen organized Nvidia’s employees… they were treated as a general pool of talent and not divided by business units or divisions. This allowed the people with the right skills to be assigned to projects on an ad hoc basis. It also helped mitigate some of the ever-present job security that plagues corporate America.” Another unique cultural advantage.
Page 170: An anecdote illustrates Jensen’s demand for intellectual honesty. When an employee (Berger) gave a vague answer, Jensen “unleashed a tirade.” The employee later recalled the key lesson: “The one thing with Jensen is don’t bullshit him… The appropriate answer is: I don’t know, Jensen, but I’ll find out.” This forces employees to take ownership of information and is different from other corporate cultures.
Page 179: This page explains “CEO math.” Jensen learned from his professor, Donald Amort, to always use round numbers and avoid getting bogged down in decimal points. He “rejected such precision if it slowed him down too much,” based on the lesson that “false accuracy is pointless.” This allows him to do “big-picture strategic thinking without getting bogged down” in spreadsheet details. An example of accuracy vs. precision.
Page 185: The book contrasts Nvidia’s success with Intel’s failure. Intel canceled its Larrabee GPU project and “failed to capitalize on new technology,” a classic instance of the “innovator’s dilemma.” While Intel focused on protecting its existing profits, Nvidia “invested so much in CUDA that it became a foundational ecosystem for AI developers,” allowing it to win the new computing era.
Pages 190 – 192: Nvidia recruited Bill Dally, a “master of parallel computing,” from Stanford. The page contrasts the CPU and GPU. The CPU is a “generalist” that “can only process only a few operations simultaneously” because of its “limited number of processing cores.” The GPU, by contrast, is a specialist built for parallel tasks. Over time, the lines between CPUs and GPUs has blurred, and GPUs have grown from graphics into more general-purpose chips.
Page 193: YouTube was a surprise enabler of deep learning, as one of the key things needed was a large dataset.
Page 194: Using Nvidia’s CUDA language, a team wrote new optimizations that “consolidate the work once performed by two thousand CPUs across a mere twelve Nvidia GPUs.” This demonstration was “the spark that ignited the AI revolution.”
Page 221: An interesting discussion of Nvidia’s $11B revenue result compared to estimates of $7.2 billion, thanks to the takeoff of AI.
