Get Your Weekly Digest

Receive the best podcast summaries and insights delivered to your inbox every week.

We respect your privacy. Unsubscribe anytime.

We Study Billionaires - The Investor’s Podcast Network

TECH002: Jensen Huang & NVIDIA w/ Seb Bunney - Review of The Thinking Machine by Stephen Witt

with Seb Bunney

Published September 24, 2025
Visit Podcast Website

About This Episode

The episode is a book-club style discussion of Stephen Witt's "The Thinking Machine," focusing on how NVIDIA evolved from a niche gaming graphics company into a central player in the AI revolution. Preston and Seb trace the technical and strategic milestones behind NVIDIA's rise-parallel processing, GPUs, CUDA, and neural networks-while examining Jensen Huang's leadership style, culture-building, and obsession with speed and iteration. They also touch on the implications and risks of AI, Huang's reluctance to address them directly, and preview their next book on OpenAI and Sam Altman.

Topics Covered

Artificial intelligence Machine learning Semiconductors Entrepreneurship Leadership and management Innovation strategy Product development Graphics processing units (GPUs) Robotics Reading and literature

Disclaimer: We provide independent summaries of podcasts and are not affiliated with or endorsed in any way by any podcast or creator. All podcast names and content are the property of their respective owners. The views and opinions expressed within the podcasts belong solely to the original hosts and guests and do not reflect the views or positions of Summapod.

Quick Takeaways

  • NVIDIA's breakthrough came from betting on parallel processing for graphics when most of the industry focused on sequential CPU computing.
  • CUDA, a software layer on top of NVIDIA GPUs, transformed the company from a gaming hardware vendor into a core AI and scientific computing platform.
  • Jensen Huang repeatedly pivoted NVIDIA away from crowded, low-margin markets toward "zero-to-one" opportunities where they could create new markets.
  • NVIDIA's culture emphasizes fast iteration, public feedback, and a relatively flat structure that gives even junior engineers direct visibility to leadership.
  • Simulation and AI now feed back into NVIDIA's own products, making graphics rendering and robotics training far more efficient.
  • Huang is intensely hardworking and publicly humble, yet can be sharply critical in internal settings, using public feedback as a way to teach in parallel.
  • Massive efficiency gains in GPU performance and energy use over just eight years are reshaping concerns about AI's computational and power demands.
  • Huang consistently deflects questions about the societal risks of AI, which stands out compared to his otherwise detailed and candid discussions about technology.
  • NVIDIA's "speed of light" principle pushes suppliers to reveal the absolute fastest possible delivery times regardless of cost, letting NVIDIA set its own tradeoffs.
  • Most consumers have little idea how central NVIDIA GPUs are to everyday technologies from gaming to photo search and text prediction, despite its multi-trillion-dollar valuation.

Podcast Notes

Introduction and episode framing

Host introduces Infinite Tech episode and guest

Preston frames the episode as a technology book review with his friend Seb, focused on "The Thinking Machine" by Stephen Witt[0:11]
They will discuss how NVIDIA evolved from gaming graphics to the center of the AI revolution and what Jensen Huang's leadership teaches about building markets and shaping tech's future[0:15]

Backstory on doing tech book discussions

Preston notes that when he and Stig started their original show they often read investing books and discussed lessons learned[1:51]
He and Seb now plan to do the same thing with technology books as part of Infinite Tech[1:33]
Seb mentions they recently spent a week in the mountains together and always return to talking about books, which led to this idea[1:30]

Overview of The Thinking Machine and NVIDIA's early history

Initial reactions to the book

Preston describes "The Thinking Machine" as an amazing book and says he really enjoyed it[1:58]
Seb says he thought he was familiar with the tech industry but had no idea how central NVIDIA is to modern AI and technology[2:38]
Seb found the book mind-blowing and eye-opening about NVIDIA's role in the world[2:38]

NVIDIA as the funnel for AI investment

Preston notes that people focus on hundreds of billions going into AI companies like OpenAI or xAI, but he hadn't initially thought to look one layer deeper[2:55]
He emphasizes that NVIDIA sits at the bottom of this funnel, harvesting revenue from AI companies via their chips[3:17]
He points out that energy companies powering those chips are also major beneficiaries of the AI boom[3:29]
This helps explain why NVIDIA's market cap has reached the trillions[3:35]

Jurassic Park rendering anecdote

Preston highlights a line from the book that in the mid-1990s an NVIDIA-powered chip was used to render Jurassic Park[3:51]
The book states it took 10 months to render a three-second clip of Jurassic Park at that time[3:55]
Preston contrasts this with what GPUs can do today, calling the progress from the mid-90s to now mind-blowing[4:14]

Mind-blowing chip manufacturing precision

Seb reads a quote from the book describing the chip structures as "crystal canyons" sculpted with ultraviolet light at a precision that would impress a Renaissance master[4:46]
Engineers compare lithography precision to shooting a laser from the moon and hitting a quarter on a sidewalk in Arkansas[5:03]
Seb says the intricacy of chip manufacturing is absolutely mind-blowing[5:06]

NVIDIA's origin story and the leap to parallel processing

Jensen Huang's background and founding NVIDIA

Preston summarizes that NVIDIA started in the early 1990s, founded by Jensen Huang[5:36]
The book presents Huang as a total overachiever, very driven and intelligent, yet initially modest and not outwardly self-aggrandizing[5:48]
Huang was an electrical engineer fascinated by parallel processing rather than traditional serial processing dominated by Intel at the time[6:09]

From 2D games to 3D parallel processing

Seb stresses how big a leap parallel processing was compared to sequential CPU processing[7:15]
• Before GPUs, computers processed tasks sequentially, limiting ability to render complex game environments; games were mostly 2D like Mario or simple ping-pong style
Huang believed that processing graphics in parallel would allow much richer environments with fluid dynamics, shadows, and realism[7:15]
• Parallel GPUs transformed gaming by enabling more detail and engagement, then later spilled into non-gaming tasks as researchers started using GPUs for math-heavy simulations

Cutthroat GPU competition and survival

Preston explains that early GPU markets were extremely competitive, with 30-40 competitors in the 1990s[8:29]
Large CPU players like Intel didn't want to enter this niche gamer-driven market which seemed too small and specialized[8:33]
Despite this, NVIDIA kept competing in gaming GPUs and sold into a hardcore gamer customer base[8:40]

CUDA, general-purpose GPUs, and the rise of modern AI

The gamer with 30 GPUs and early GPGPU insight

Preston recounts a story from the book about a gamer who stitched together around 30 NVIDIA cards for a huge wall-sized gaming display[9:20]
The gamer realized the setup was doing an enormous number of computations per second just to render a game and wondered what else that compute power could be used for[10:24]
He contacted NVIDIA and Jensen Huang to ask if there were other uses beyond gaming, sparking NVIDIA's attention[10:15]

CUDA as a software platform on GPUs

Preston explains that NVIDIA created CUDA to make GPUs accessible for tasks beyond 3D graphics, representing a major turning point for the company[10:37]
Seb says researchers noticed that if GPUs could simulate realistic game physics, they could also crunch big scientific datasets[11:36]
• Researchers in physics, math, and data science started hacking GPUs for general computation, not just graphics
Seb says CUDA stands for compute unified device architecture and is a platform sitting on top of GPUs so developers can program GPUs using familiar languages like Python and C#[12:18]
CUDA opened GPUs to non-graphics workloads like computer vision, autonomous driving, speech recognition, and real-time translation[12:56]
Preston notes an interviewee in the book argues NVIDIA is as much a software company as a hardware company because CUDA created a powerful network effect[13:30]

CUDA as NVIDIA's moat and ecosystem

Seb uses a value-investor lens and asks what NVIDIA's moat is, concluding CUDA is central because everyone built free domain-specific packages on it[14:12]
• Machine learning practitioners, physics professors, and data scientists all created specialized CUDA-based packages for their industries
• Those packages created stickiness: once people got used to them, they kept buying NVIDIA chips

Neural nets, backpropagation, and GPU acceleration

Seb notes the book briefly touches on early AI history: 1940s "nervous nets" and later backpropagation, which let neural nets learn from mistakes[15:39]
Before GPUs, neural nets struggled because they were too complex and compute-intensive for sequential CPUs[16:01]
With GPUs and parallel processing, it became feasible to train large neural networks, changing AI's trajectory[16:27]

Transformers and the "Attention Is All You Need" paper

Seb explains that transformers, introduced in 2017, transformed AI by shifting from narrow specialist models to more generalist models[16:41]
• Instead of encoding explicit meanings of words, transformers learn context by predicting what comes next in sequences of words
• Seb gives an example: given "green, ribbit, lily pad, amphibian," an AI can predict "frog" without knowing the semantic meaning of frog, just statistical co-occurrence
Preston mentions the Google paper "Attention Is All You Need" (by Vaswani et al.) as the key milestone enabling transformers, focusing on contextual associations between letters, words, and paragraphs[18:39]
They connect this milestone with the later parts of the book that show NVIDIA at the center of deep learning and AI[19:01]

Themes: zero-to-one strategy, iteration, and vision

Zero-to-one markets and avoiding red oceans

Seb highlights a major theme: Huang focuses on creating markets (zero-to-one) rather than fighting for share in existing "red ocean" markets[20:03]
He references Peter Thiel's "Zero to One" and the idea of vertical progress (creating something entirely new) versus horizontal progress (incremental replication and scaling)[20:43]
Seb notes that throughout the book Huang repeatedly chooses to reshape industries-like GPUs and parallel processing-rather than compete head-on in established ones[20:10]
NVIDIA considered entering phone chip markets but Huang recognized there was no edge in an already crowded space, so he wrote off sunk costs and moved on[21:15]

Early near-death experiences: NV1, NV2, NV3

Preston recalls that NVIDIA's early NV1 chip sold reasonably well but was killed overnight when Microsoft released a new graphics protocol[21:16]
The follow-on NV2 chip flopped due to timing and intense competition[21:16]
The company was close to failing and pinned its survival on the NV3 chip, which they could not prototype physically due to lack of funds[21:57]
• They designed NV3 entirely via simulation and sent it to the foundry without having tested it on real silicon, a "Hail Mary" move
NV3 worked, keeping the company alive and enabling the next stage of growth[23:15]
Preston finds it ironic that simulation saved NVIDIA early on, and today NVIDIA builds hardware used to run massive simulations and AI models of reality[22:55]

Operating as if the company might die tomorrow

Preston says Huang views NVIDIA culturally as if it could die tomorrow, a mindset rooted in those early near-death experiences[22:55]

Simulation, robotics, and digital training grounds

Cosmos: hyper-realistic training environments

Seb mentions that training robots in the physical world is slow and risky because robots can fall and be damaged[23:30]
He describes "Cosmos" as a hyper-realistic digital environment that obeys physics, fluid dynamics, gravity, cause and effect, and object permanence[23:30]
The goal is to train robots extensively in this digital world so they are highly proficient before entering the real world[22:35]
• Example: a warehouse robot can simulate every possible path to pick packages in a split second in Cosmos instead of learning slowly in a real warehouse
Seb says Cosmos is free to use, reflecting NVIDIA's desire to support advances in science, robotics, and AI[24:46]

Iterate, execute, and messy code

Seb notes that NVIDIA's focus is on rapid iteration rather than perfect, clean code[24:58]
He cites a quote from the book where someone reviews NVIDIA's code and calls it "like cancer" and poorly written, yet acknowledges there was brilliance because it worked and shipped fast[24:56]
NVIDIA shortened chip development cycles from one- or two-year cadences down to six months, emphasizing speed to market[25:58]

Lidar, compression, and efficiency gains in AI and graphics

Lidar and precise 3D mapping for AI

Preston explains lidar as an emitter-sensor system using light at frequencies humans can't see to map physical environments in 3D with millimeter precision[32:10]
Lidar can scan a room, collect returns, and produce a highly accurate depth map for AI and robotics training[32:11]
He notes how converging lidar, AI, and GPUs is "beyond exciting" for training humanoid robots and other systems[33:08]

AI-assisted rendering and ImageNet

Seb describes the ImageNet competition, where teams build models to categorize images, such as enabling Google Photos to find all pictures with cats[33:43]
He notes that the AlexNet team used NVIDIA GPUs and neural nets, winning the 2012 ImageNet contest with much lower error rates and blowing away the competition[34:22]
Seb says NVIDIA later applied AI back onto their own GPUs: for a 4K screen with 8 million pixels, they only render 500,000 with the GPU and let AI generate the rest[34:24]
• By focusing compute on fewer pixels, they can increase detail there and rely on AI to fill in the full image, making rendering far more efficient and realistic

AI as compression analogy

Preston compares AI-driven rendering to compressing a large WAV audio file into a smaller MP3 that sounds the same to human ears[36:32]
He says AI is effectively compressing data, finding ways to represent complex outputs more efficiently across file types, including images[36:26]

DGX-1 and massive efficiency improvements

Seb discusses the DGX-1 system described in the book, a top-of-the-line GPU box in 2016 used to train AI neural nets and sold for about $250,000[37:26]
He notes the first DGX-1 went to OpenAI, and Elon Musk personally received it at their office[38:02]
Seb references a 2024 conversation where Huang shows a mini version one-tenth the size, with six times the processing power and one ten-thousandth the energy use compared to DGX-1[38:48]
He argues this 10,000× energy efficiency gain in eight years is mind-blowing and reframes concerns about AI's energy demands[38:05]

Jensen Huang's leadership, culture, and structure at NVIDIA

Public feedback and "torturing to greatness"

Preston contrasts Huang's very humble public persona with book descriptions of him sometimes dressing down employees harshly in public[28:24]
Seb notes his background in somatic therapy and says Huang appears to use public feedback deliberately to turn one person's mistake into a collective learning moment[28:24]
Seb cites a phrase from the book that Huang "tortures to greatness," rarely firing people but pushing them hard to improve[29:19]
Preston suggests Huang's public corrections parallel the idea of parallel processing: many people learn simultaneously from one incident[28:39]

Flat organizational structure and weekly emails

Seb explains that NVIDIA maintains a relatively flat structure; junior engineers can sit in on executive meetings and share opinions[39:14]
Employees send Huang weekly emails listing their top five current projects or interests, and he randomly samples and reads many of them[39:37]
Seb says this practice reinforces visibility, approachability, and cross-pollination of ideas while avoiding the "telephone game" of hierarchical filtering[39:58]

Loyalty, wealth creation, and employee sentiment

Preston notes that many NVIDIA employees love working for Huang and see him as a celebrity inside the company[37:59]
He points out that employees who hold stock have repeatedly seen extra zeros added to their net worth every few years as the company's valuation surged[38:02]
Preston wonders how much of employees' affection stems from financial gains versus respect for Huang's leadership style and admits the book left him unsure[38:29]

Reciprocal relationship between NVIDIA and AI community

Seb emphasizes Huang's point that NVIDIA and AI researchers have a reciprocal relationship: GPUs enabled neural nets, and AI advances then improved NVIDIA products[34:46]
He reiterates that NVIDIA is not solely dictating the future; they provide tools that others use to invent new AI techniques, which NVIDIA then leverages back into hardware and software[33:47]

Speed of light principle, pricing physics, and strategic pivots

Speed of light principle in supply chains

Preston recounts a story where Huang asks an executive for the absolute fastest possible delivery time for certain components, regardless of cost[40:18]
Vendors initially quoted slower timelines, having filtered out extremely fast but very expensive options[40:26]
Huang publicly rejects the initial answer and demands the true "speed of light" figure-the theoretical fastest achievable time plus cost, like a physics limit[40:18]
Knowing this hard boundary lets NVIDIA decide when to pay for extreme speed, such as if a customer urgently needs massive numbers of GPUs[41:37]

Pivoting away from existing markets

Seb notes NVIDIA dropped its early quadratic rendering approach when it caused Windows crashes and almost bankrupted the company, pivoting quickly after a critical cash injection[24:33]
He also cites NVIDIA's decision to abandon the burgeoning mobile phone chip market when they realized it was not a true zero-to-one opportunity[24:16]

AI risks, Jensen's reluctance, and closing

Huang's avoidance of discussing AI risks

Seb highlights that near the end of the book the author asks Huang about AI risks, and Huang responds by downplaying concerns and focusing on historical examples like agriculture and electricity lowering marginal costs[41:55]
Huang insists NVIDIA is a "serious company" doing "serious work" and rejects comparisons to science fiction scenarios[41:57]
Seb and Preston note that NVIDIA executives were extremely disciplined and well-prepared, but consistently disinclined to discuss potential future implications of their technology[42:31]
Seb interprets Huang's intense reactions to AI-risk questions as driven by fear or discomfort, since his response seems disproportionate to the stimulus[42:59]

Speculation on personal psychology and background

Seb notes the book mentions Huang immigrated to the U.S. at age 10, always feeling like an immigrant and the younger of two brothers[43:35]
Seb draws a parallel to research in "The Talent Code" about youngest siblings often overachieving, wondering if Huang has a strong need to prove himself[44:18]
Preston agrees that Huang's intensity around AI-risk questions is one of the most interesting and puzzling aspects of the book[45:17]

NVIDIA's scale and public invisibility

Seb says many friends with gaming consoles and computers don't even know what NVIDIA is, despite using its technology[33:04]
Preston contrasts the ubiquity of iPhones, which people see and hold, with NVIDIA GPUs that operate invisibly inside data centers and devices[33:54]
He notes that NVIDIA's market cap is roughly a trillion dollars higher than Apple's at the time of recording, underscoring how much value is tied up in largely unseen chips[32:48]

Next book selection: "Empire of AI"

Preston announces the next book they will cover is "Empire of AI," about OpenAI and Sam Altman[1:09:04]
He openly states a bias that he is not a fan of Sam Altman as a person based on what he has read, though he recognizes OpenAI's achievements as mind-blowing[1:08:57]
Seb says he uses ChatGPT and sees OpenAI's release as profoundly changing the world, but notes Altman's public behavior raises questions he hopes the book will explore[1:09:34]

Closing remarks and Seb's own book

Preston praises Seb as thoughtful and says he loves having these kinds of conversations, which mirror their real-life discussions[1:10:12]
He mentions Seb's own book, "The Hidden Cost of Money," describing it as a Bitcoin book and recommends listeners read it[1:10:30]
Seb says he feels lucky and grateful to be on the show, having listened to Preston for over a decade and enjoying bringing the show back to book discussions[1:10:50]

Lessons Learned

Actionable insights and wisdom you can apply to your business, career, and personal life.

1

Building enduring companies often requires pursuing "zero-to-one" opportunities-creating entirely new markets or use cases-rather than fighting for scraps in crowded, existing markets.

Reflection Questions:

  • • Where am I currently competing in a red-ocean market instead of looking for a problem no one else is addressing?
  • • How could you reframe your product or skills to create a new category rather than be compared directly to incumbents?
  • • What is one niche or emerging problem you could start exploring this month that might become your own zero-to-one opportunity?
2

Relentless iteration and fast execution can beat elegant but slow approaches; shipping working solutions on short cycles compounds learning and market advantage.

Reflection Questions:

  • • In what areas of your work are you over-optimizing for perfection instead of speed and feedback?
  • • How might shortening your development or decision cycles change the trajectory of your current project?
  • • What is one process you can deliberately make "good enough" this week so you can ship, learn, and iterate faster?
3

Creating platforms and ecosystems-like software layers that others can build on-can become a powerful moat that locks in users and attracts compounding innovation.

Reflection Questions:

  • • What parts of your work or business could be turned into a reusable platform or toolkit for others?
  • • How would your strategy change if your goal were to enable thousands of other people to create value on top of what you build?
  • • What is one API, template, or shared resource you could create in the next quarter that would make you or your organization more of a platform than a standalone product?
4

Flat information flows and direct access between leadership and front-line contributors make it easier to pivot quickly and spot transformative ideas early.

Reflection Questions:

  • • Where are the "telephone games" in your organization or life that distort information before it reaches decision-makers?
  • • How could you create more direct channels for people closest to the problems to share their insights with you?
  • • What is one new routine-like a weekly summary email or open meeting-you could start to improve the quality and speed of information you receive?
5

Technological progress brings both benefits and risks, and leaders who build powerful tools have a responsibility to engage honestly with questions about long-term consequences.

Reflection Questions:

  • • What are the second- and third-order effects of the technologies or systems you are contributing to today?
  • • How might openly discussing potential downsides of your work change the safeguards or design choices you make?
  • • What concrete step could you take this month to better understand or mitigate the risks associated with the tools or processes you help advance?

Episode Summary - Notes by Phoenix

TECH002: Jensen Huang & NVIDIA w/ Seb Bunney - Review of The Thinking Machine by Stephen Witt
0:00 0:00