Lex Fridman Podcast
Lex Fridman Podcast

Mar 23, 2026 · 2h 32m

NVIDIA CEO Jensen Huang details the hardware engineering driving the AI revolution

#494 – Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution

As AI models demand unprecedented computational power, understanding NVIDIA's hardware strategy reveals where the limits of machine intelligence will be pushed next.

3 key takeaways
  1. 1NVIDIA relies on extreme co-design to build hardware and software in lockstep for maximum computing efficiency.
  2. 2The frontier of AI is shifting toward agentic scaling laws that require massive GPU clusters to support autonomous reasoning.
  3. 3Proprietary interconnect technologies like NVLink are critical to overcoming communication bottlenecks between modern GPUs.

Don't miss

Jensen Huang explains how the transition to agentic scaling laws will reshape the future of computational workloads.

The brief

NVIDIA CEO Jensen Huang sits down with Lex Fridman to trace the company's evolution from a specialized graphics accelerator pioneer into the multi-trillion-dollar engine driving the global artificial intelligence revolution.

Huang attributes the company's dominance to extreme co-design, a philosophy where hardware and software are built in lockstep. This unified approach allowed proprietary technologies like CUDA and NVLink to become industry standards.

The discussion details how the scaling laws of large language models are shifting toward agentic scaling, where autonomous AI systems require massive, highly interconnected GPU clusters to execute complex reasoning tasks.

Beyond raw computing power, Huang shares his unconventional leadership philosophy, emphasizing flat organizational structures and a first-principles approach to solving seemingly impossible engineering bottlenecks.

What was said on this episode

64 statements · 50 positive · 7 negative · 3 mixed · 4 neutral

  1. Jensen Huangon Extreme co-designPositive7:15

    Extreme co-design is necessary because modern AI problems exceed one computer’s capacity.

    “the reason why extreme code design is necessary is because the problem no longer fits inside one computer to be accelerated by one gpu”

    Listen at 7:15

  2. Jensen Huangon AI workloadsNegative7:17

    Large AI problems no longer fit within a single computer or GPU.

    “the problem no longer fits inside one computer to be accelerated by one gpu”

    Listen at 7:17

  3. Jensen Huangon Nvidia computing strategyMixed14:41

    Becoming more general-purpose creates tension with specialized computing optimization.

    “The better computing company we become, the worse we became as a specialist.”

    Listen at 14:41

  4. Jensen Huangon Computing architecture installed basePositive18:10

    An architecture’s installed base is its most important strategic asset.

    “the install base is in fact the single most important part of an architecture”

    Listen at 18:10

  5. Jensen Huangon CUDA on GeForce PCsPositive19:36

    Nvidia should distribute CUDA through every GeForce-equipped PC.

    “we ought to put CUDA on GeForce and put it into every single PC”

    Listen at 19:36

  6. Jensen Huangon Nvidia market capitalizationNegative20:52

    Nvidia’s market capitalization fell to approximately $1.5 billion after CUDA launched.

    “our market cap went down to like one and a half billion dollars”

    Listen at 20:52

  7. Jensen Huangon Nvidia CUDA investmentNegative22:17

    Nvidia’s CUDA investment increased GPU costs by 50%.

    “we increased our cost by 50%”

    Listen at 22:17

  8. Jensen Huangon AI training dataPositive29:59

    AI training data volumes will continue increasing, including synthetic data.

    “We're going to keep on scaling the amount of data that we have to train with.”

    Listen at 29:59

  9. Jensen Huangon AI model trainingNeutral31:08

    AI training is increasingly constrained by compute rather than data availability.

    “Training is no longer limited by data, is now limited by compute.”

    Listen at 31:08

  10. Jensen Huangon Test-time scalingNegative32:58

    Test-time scaling requires substantial computational resources.

    “test time scaling is intensely compute intensive”

    Listen at 32:58

  11. Jensen Huangon Agentic AIPositive33:46

    Agentic systems introduce a further AI scaling law through multiplying subagents.

    “the next scaling law is the agentic scaling law”

    Listen at 33:46

  12. AI intelligence will continue scaling primarily with compute.

    “intelligence is going to scale by one thing and it's compute”

    Listen at 34:39

  13. Jensen Huangon AI model architecturesNeutral35:13

    New AI model architectures emerge approximately every six months.

    “AI model architectures are being invented about once every six months”

    Listen at 35:13

  14. Jensen Huangon CUDAPositive36:40

    CUDA’s flexibility has allowed it to remain resilient while adapting to new algorithms.

    “CUDA has been so resilient on the one hand, and yet we continue to enhance it”

    Listen at 36:40

  15. Jensen Huangon NVL 72Positive37:05

    NVLink 72 can place models with four to ten trillion parameters in one computing domain.

    “We could now take an entire 4 trillion, 10 trillion parameter model and put it in one computing domain as if it's running on one gpu.”

    Listen at 37:05

  16. Jensen Huangon Vera Rubin rackPositive38:07

    Vera Rubin rack systems are designed to run agents that use tools.

    “this one is to run agents, and agents bang on tools”

    Listen at 38:07

  17. Jensen Huangon Agentic AIPositive40:37

    Agentic AI systems will use tools, access files, and conduct research.

    “it's going to use tools, that it's going to access files, it's going to be able to do research”

    Listen at 40:37

  18. Jensen Huangon OpenClawPositive41:55

    OpenClaw is a major milestone for agentic systems comparable to ChatGPT for generative AI.

    “I think OpenClaw did for agentic systems what ChatGPT did for generative systems. And I just think it's a very big deal.”

    Listen at 41:55

  19. Jensen Huangon Agentic system securityPositive43:15

    Agentic systems can be safer when sensitive access, code execution, and external communication are not all enabled together.

    “We could keep things safe if we gave you two out of those three capabilities at any time, but not all three.”

    Listen at 43:15

  20. Jensen Huangon Nvidia computing scalePositive44:25

    Nvidia claims computing scale increased one million-fold over the past decade.

    “We progressed and scaled up computing by a million times in the last 10 years.”

    Listen at 44:25

  21. Jensen Huangon AI token costPositive45:00

    AI token costs are declining by roughly an order of magnitude annually.

    “that token cost is coming down, it's coming down an order of magnitude every year”

    Listen at 45:00

  22. HBM memory will become mainstream in data centers.

    “this was going to be a mainstream memory for data centers in the future”

    Listen at 47:45

  23. Jensen Huangon Power grid capacityPositive54:10

    The power grid has excess capacity during approximately 99% of the time.

    “99% of the time our power grid has excess power”

    Listen at 54:10

  24. Data centers should receive less power during periods of peak societal demand.

    “the data centers would get less”

    Listen at 54:48

  25. AI data centers can reduce power use by accepting slower response times.

    “we could degrade our performance, reduce our power consumption and provide for a slightly longer latency response”

    Listen at 55:03

  26. Data centers should be engineered to reduce performance without losing data during power constraints.

    “we have to build data Centers that gracefully degrade”

    Listen at 57:11

  27. Jensen Huangon Elon MuskPositive59:50

    Elon Musk uses systems thinking to minimize processes while preserving necessary capabilities.

    “he has the ability to question everything to the point where everything is down to its minimal amount that's necessary”

    Listen at 59:50

  28. Jensen Huangon Nvidia engineering methodologyPositive1:02:56

    Nvidia evaluates engineering constraints against physical limits.

    “everything that we do is compared against the speed of light”

    Listen at 1:02:56

  29. Jensen Huangon First-principles engineeringMixed1:04:15

    Huang prefers first-principles redesign over incremental continuous improvement.

    “I don't love the other methods, which is continuous improvement”

    Listen at 1:04:15

  30. Nvidia will likely produce approximately 200 Vera Rubin pods weekly.

    “we'll probably kind of crank out about 200 of these pods a week”

    Listen at 1:06:21

  31. Jensen Huangon ChinaPositive1:10:43

    China is currently the world’s fastest-innovating country.

    “this is the fastest innovating country in the world today”

    Listen at 1:10:43

  32. Jensen Huangon Open-source AIPositive1:13:49

    Open-source AI is necessary for many industries to participate in AI.

    “open source is fundamentally necessary for many industries to join the AI revolution”

    Listen at 1:13:49

  33. Jensen Huangon TSMC manufacturing systemPositive1:17:38

    TSMC operates manufacturing with high throughput, yield, cost efficiency, and customer service.

    “somehow they're running a factory with high throughput, high yield, really great costs, excellent customer service”

    Listen at 1:17:38

  34. Jensen Huangon Nvidia computing platformPositive1:21:21

    Nvidia’s most important competitive asset is its computing-platform installed base.

    “Our single most important property as a company is the install base of our computing platform.”

    Listen at 1:21:21

  35. Jensen Huangon CUDAPositive1:24:04

    Developers should target CUDA first because of its installed base and ecosystem.

    “If I were a developer today, I would target CUDA first.”

    Listen at 1:24:04

  36. Jensen Huangon Nvidia computing ecosystemPositive1:25:00

    Nvidia’s computing ecosystem spans essentially every global industry.

    “the ecosystem is so broad, it basically covers every single industry in the world”

    Listen at 1:25:00

  37. Jensen Huangon Planetary-scale computingPositive1:26:36

    Huang expects Nvidia’s computing model to expand toward planetary scale.

    “I'm hoping my next click is when I'm thinking about building computers. It's, you know, planetary scale.”

    Listen at 1:26:36

  38. Jensen Huangon Edge AIPositive1:28:08

    AI processing should increasingly occur at the edge near data sources.

    “AI ought to be done at the edge”

    Listen at 1:28:08

  39. Jensen Huangon Nvidia growthPositive1:30:46

    Nvidia’s growth is extremely likely and effectively inevitable.

    “I think that Nvidia's growth is extremely likely and in my mind inevitable.”

    Listen at 1:30:46

  40. Nvidia’s growth is extremely likely and effectively inevitable.

    “Nvidia's growth is extremely likely and in my mind inevitable”

    Listen at 1:30:48

  41. Jensen Huangon High-intelligence AI tokensPositive1:34:24

    Customers will soon pay approximately $1,000 per million tokens for high intelligence.

    “The idea that somebody's willing to pay $1,000 per million tokens, it's just around the corner.”

    Listen at 1:34:24

  42. AI-driven productivity will accelerate global GDP growth.

    “I am absolutely certain that the world's GDP is going to accelerate in growth.”

    Listen at 1:35:17

  43. Jensen Huangon Computing share of global GDPPositive1:35:25

    Computation’s share of global GDP will become approximately 100 times larger.

    “the percentage of that GDP that will be used for computation will be a hundred times more than the past”

    Listen at 1:35:25

  44. Jensen Huangon The Nvidia WayPositive1:36:03

    Nvidia could reach $3 trillion in revenue in the near future.

    “is it possible for Nvidia to be a, you know, $3 trillion revenues company in the near future? The answer is of course yes.”

    Listen at 1:36:03

  45. Jensen Huangon Nvidia GPUsPositive1:38:34

    Nvidia will eventually reach the future scale Jensen describes.

    “100%, we'll get there.”

    Listen at 1:38:34

  46. Agents are the fastest-growing application in history.

    “It is the fastest growing application in history.”

    Listen at 1:39:23

  47. Jensen Huangon OpenClawPositive1:39:29

    OpenClaw is the defining breakthrough application for AI tokens.

    “There's no question OpenClaw is the iPhone of tokens.”

    Listen at 1:39:29

  48. Jensen Huangon AGIPositive2:02:33

    AGI has already been achieved under the stated definition.

    “I think it's now, I think we've achieved AGI.”

    Listen at 2:02:33

  49. Jensen Huangon AI agents creating social applicationsPositive2:03:57

    AI agents may soon create viral digital influencers or social applications.

    “I wouldn't be surprised if some social thing happened or somebody created a digital influencer, super, super cute, or some social application that feeds your little tomagotchi or something like that, and it becomes out of the blue and instant success.”

    Listen at 2:03:57

  50. Jensen Huangon Computer VisionPositive2:05:32

    Computer vision reached superhuman performance around 2019 or 2020.

    “Computer vision was superhuman in 2019, maybe a little bit later. 2020.”

    Listen at 2:05:32

  51. Jensen Huangon Nvidia software engineersPositive2:07:04

    Nvidia’s software-engineer headcount will increase rather than decrease.

    “The number of software engineers at Nvidia is going to grow, not decline.”

    Listen at 2:07:04

  52. Jensen Huangon People able to code with AIPositive2:08:19

    AI may expand the number of people able to code from 30 million to 1 billion.

    “I think we just went from 30 million to probably 1 billion.”

    Listen at 2:08:19

  53. Jensen Huangon AI-skilled workersPositive2:13:36

    Employers should prefer candidates who are expert AI users.

    “I would hire the one who's expert in using AI.”

    Listen at 2:13:36

  54. Jensen Huangon AI use in work and educationPositive2:13:58

    Students and workers should become proficient AI users to improve their careers.

    “I would advise that every college student, every teacher should encourage their student to go use AI.”

    Listen at 2:13:58

  55. Jensen Huangon College students’ AI skillsPositive2:14:06

    Every college student should graduate with AI expertise.

    “every college student should graduate and be an expert in AI”

    Listen at 2:14:06

  56. Jensen Huangon Task-based jobsNegative2:14:58

    Jobs consisting primarily of automatable tasks are highly likely to be disrupted.

    “if your job is the task, then you're very highly going to be disrupted”

    Listen at 2:14:58

  57. Jensen Huangon AI chipsNegative2:17:39

    AI chips will not experience human anxiety, excitement, or other emotions.

    “I don't think my chips will feel those.”

    Listen at 2:17:39

  58. Jensen Huangon IntelligenceNeutral2:20:28

    Intelligence is a functional commodity distinct from humanity.

    “I actually think intelligence is a commodity”

    Listen at 2:20:28

  59. Jensen Huangon Human intelligence and humanityNeutral2:21:20

    Functional intelligence differs fundamentally from humanity and lived experience.

    “Intelligence is a functional thing. Humanity is not specified functionally.”

    Listen at 2:21:20

  60. Jensen Huangon IntelligenceMixed2:22:09

    Intelligence will become increasingly commoditized.

    “intelligence is going to be commoditized”

    Listen at 2:22:09

  61. Jensen Huangon DiseasePositive2:28:59

    Ending disease is a reasonable future expectation.

    “It's a reasonable thing to expect the end of disease.”

    Listen at 2:28:59

  62. Jensen Huangon DiseasePositive2:29:02

    Ending disease is a reasonable future expectation.

    “the end of disease”

    Listen at 2:29:02

  63. Jensen Huangon pollutionPositive2:29:05

    Pollution will be drastically reduced in the future.

    “It's a reasonable thing to expect that pollution will be drastically reduced.”

    Listen at 2:29:05

  64. Jensen Huangon Human biologyPositive2:30:16

    Human biology will be substantially understood within approximately five years.

    “Understanding the biological machine is right around the corner. It's not 10 years, it's five years probably.”

    Listen at 2:30:16

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.

Books & mentions

Lex Fridman Podcast

Explore more deep-dive conversations with the world's leading AI researchers and tech executives

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.