#494 – Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution
Episode of Lex Fridman Podcast · March 23, 2026
Lex Fridman interviews NVIDIA CEO Jensen Huang on extreme co-design, CUDA, AI scaling laws, supply chains, power, China, TSMC, jobs, AGI and mortality.
Summary
Lex Fridman spoke with Jensen Huang, co-founder and CEO of NVIDIA, about how the company evolved from a graphics accelerator maker into a rack-scale and AI-factory platform company. Huang described extreme co-design across chips, networking, memory, power and software, his 60-person direct staff with no one-on-ones, and the costly decision to put CUDA on every GeForce GPU, which he said crushed margins and drove the market cap down to about $1.5 billion before building the install base he called NVIDIA's main moat.
Huang outlined four AI scaling laws (pre-training, post-training, test-time and agentic), argued that synthetic data removes data limits, and called OpenClaw the "iPhone of tokens." He discussed supply chain coordination with memory makers and TSMC, proposed using idle grid capacity through data centers that degrade gracefully, praised Elon Musk's approach to building Colossus, and explained his "speed of light" first-principles method. He attributed China's tech success to talent, provincial competition and open-source culture, and praised TSMC's orchestration, culture and trust, confirming Morris Chang once offered him the TSMC CEO role.
Huang predicted NVIDIA's growth is inevitable, saying $3 trillion in revenue is possible because computing has shifted from retrieval to generation. He defended DLSS 5 against "AI slop" criticism, named Doom and Virtua Fighter as landmark games, said AGI by Fridman's definition is arguably achievable now, and argued AI will increase the number of software engineers and coders, citing radiology. He closed on intelligence as a commodity versus humanity, his rejection of formal succession planning in favor of continuous knowledge transfer, and his optimism about human kindness and ending disease.
Chapters
- 5:47
Extreme co-design
Huang explains why distributed AI workloads require co-designing GPUs, CPUs, networking, memory, power and cooling.
- 9:52
How Huang runs NVIDIA
Huang describes his large flat staff, the path from accelerator to CUDA, and shaping belief systems before major announcements.
- 28:41
AI scaling laws
Huang outlines pre-training, post-training, test-time and agentic scaling and anticipating hardware needs, including OpenClaw.
- 43:42
Supply chain, memory and power
Huang discusses convincing suppliers to invest in HBM and LPDDR, rack-scale manufacturing, and using idle grid capacity.
- 58:45
Elon Musk, Colossus and speed-of-light thinking
Huang praises Musk's systems thinking and explains his first-principles 'speed of light' method.
- 1:07:38
China and open source
Huang explains China's tech success and NVIDIA's reasons for open-sourcing Nemotron models.
Linked moment · 1:15:49Skip to 1:15:49 using a listen link in the info box.
1:15:49TSMC and NVIDIA's moat
Huang praises TSMC's orchestration and trust, recounts Morris Chang's CEO offer, and names CUDA's install base as NVIDIA's moat.
- 1:26:43
Space data centers and a $10 trillion NVIDIA
Huang discusses computing in space and argues NVIDIA's growth is inevitable as computers become token factories.
- 1:40:39
Leadership under pressure
Huang describes decomposing problems, forgetting setbacks, and reasoning publicly to handle pressure.
- 1:54:27
Video games and DLSS 5
Huang addresses DLSS 5 criticism and names Doom and Virtua Fighter as the most influential games.
- 2:01:02
AGI and the future of jobs and programming
Huang argues AGI is arguably here by Fridman's definition, that coders will grow to a billion, and that workers should learn AI.
- 2:16:52
Consciousness, mortality and hope
Huang separates intelligence from humanity, rejects succession planning, and expresses hope for humanity's future.
Key points
What was said, by whom, with the exact words. Timestamps jump to the moment.
- Jensen Huang
Huang argues that install base, not elegance, defines a computing architecture, citing x86's survival.
“InstallBase is everything. InstallBase defines an architecture. Everything else is secondary”
- Jensen Huang
Huang says putting CUDA on GeForce consumed the company's gross profit and cut its market cap to about $1.5 billion.
“You know, our market cap went down to like $1.5 billion. And so we were down there for a while, and we clawed our way back slowly”
- Jensen Huang
Huang argues that because most training data will be synthetic, AI training is now limited by compute rather than data.
“Training is no longer limited by data. It's now limited by compute. And the reason for that is most of the data is synthetic”
- Jensen Huang
Huang rejects the notion that inference is easy, arguing that inference is thinking and is compute-intensive.
“And that was always illogical to me because inference is thinking. And I think thinking is hard. Thinking is way harder than reading”
- Jensen Huang
Huang believes OpenClaw did for agentic AI what ChatGPT did 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”
- Jensen Huang
Huang claims NVIDIA scaled computing a million times in ten years versus about 100 times from Moore's Law, and that token costs fall an order of magnitude yearly.
“We progressed and scaled up computing by a million times in the last 10 years”
- Jensen Huang
Huang recounts convincing DRAM CEOs about three years earlier that HBM would become mainstream data center memory.
“I was able to convince several of the CEOs that even though at the time, HBM memory was used quite scarcely and barely by supercomputers”
- Jensen Huang
Huang proposes that data centers use the power grid's idle excess capacity by accepting graceful degradation during rare peak demand.
“And so 99% of the time, our power grid has excess power and they're just sitting idle”
- Jensen Huang
Huang attributes China's rapid innovation to talent, intense provincial competition and an open-source sharing culture, calling it the fastest innovating country.
“And so this is the fastest innovating country in the world today”
- Jensen Huang
Huang says NVIDIA has done business with TSMC for three decades without a contract.
“I don't know how many tens, hundreds of billions of dollars of business we've done through them, and we don't have a contract”
- Jensen Huang
Huang confirms Morris Chang offered him the TSMC CEO role and says he declined because of his responsibility to NVIDIA.
“And so I declined it, you know, not because it wasn't an incredible offer. It's an unbelievable offer, but I simply couldn't take it”
- Jensen Huang
Huang predicts that a $1,000-per-million-tokens price point for premium AI is coming soon.
“a thousand dollars per million tokens it's just around the corner it's not if it's only when”
- Jensen Huang
Huang says NVIDIA could become a $3 trillion revenue company in the near future, with no physical limits preventing it.
“is it possible for NVIDIA to be a $3 trillion revenues company in the near future? The answer is of course, yes”
- Jensen Huang
Huang says DLSS 5 is conditioned on the artist's geometry and textures and is a tool for artists, not post-processing AI slop.
“DLSS is integrated with the artist. And so it's about giving the artist the tool of AI, the tool of generative AI”
- Jensen Huang
Huang argues an AI agent could plausibly create a short-lived app that briefly reaches billion-dollar value, as many internet-era companies did.
“most of those websites were not anything more sophisticated than what OpenClaw could generate today”
- Jensen Huang
Huang cites radiology, where AI became superhuman yet radiologist numbers grew, as evidence that job purpose differs from tasks.
“Every radiology platform and package today is driven by AI. And yet, the number of radiologists grew”
- Jensen Huang
Huang predicts the number of software engineers at NVIDIA will grow rather than decline.
“The number of software engineers at NVIDIA is going to grow, not decline”
- Jensen Huang
Huang argues coding is now specification, expanding the number of coders from about 30 million to perhaps a billion.
“Describe a specification for a computer to go, telling the computer what to go build. How many people? I think we just went from 30 million to probably 1 billion”
- Jensen Huang
Huang recommends that every student and worker become expert in AI, saying he would hire the AI-expert candidate.
“Every college student should graduate and be an expert in AI. And everybody, if you're a carpenter, if you're an electrician, go use AI”
- Jensen Huang
Huang believes AI will recognize emotions like nervousness but that his chips will not feel them.
“I believe that AI will be able to recognize those and understand those. I don't think my chips will feel those”
- Jensen Huang
Huang argues intelligence is becoming a commodity and that humanity and character should be elevated instead.
“intelligence is a word that we've elevated to very high form over time. The word we should really elevate is humanity”
- Jensen Huang
Huang explains he rejects formal succession planning in favor of continuously passing on knowledge to his team.
“is to pass on knowledge, information, insight, skills, experience as often and continuously as you can”
- Jensen Huang
Huang predicts that understanding the biological machine is about five years away.
“Understanding the biological machine is right around the corner. It's not 10 years. It's five years probably”
People
In the conversation
- Lex FridmanAmerican podcaster and AI researcher; same host, second diarization label covering the interview segments
- Jensen HuangNvidia CEO who described GPUs as fungible, durable revenue-generating assets
Discussed
- Elon MuskEntrepreneur behind X, xAI and SpaceX, discussed in AI competition
- Morris ChangFounder of TSMC who offered Huang the TSMC CEO role
- Ilya SutskeverAI researcher who suggested pre-training was running out of data
- Peter SteinbergerCreator of OpenClaw, previously interviewed by Fridman
Topics
More from Lex Fridman Podcast
#500 – Khabib Nurmagomedov: Dagestan, MMA, UFC, Islam, Conor, Fedor & FootballAug 12, 2026
Khabib Nurmagomedov tells Lex Fridman about Dagestan's fighting culture, his father's coaching, the Conor McGregor fight, Islam, and football.
- #499 – Gary Gallagher: American Civil War, Slavery, Lincoln, Grant & Lee
Jul 28, 2026 · 3 h 53 min
Lex Fridman and historian Gary Gallagher discuss the American Civil War: slavery as its cause, Lincoln, Grant, Lee, key battles, the Lost Cause, and today's divisions.
- #496 – FFmpeg: The Incredible Technology Behind Video on the Internet
May 6, 2026 · 4 h 24 min
Jean-Baptiste Kempf and Kieran Kunhya explain FFmpeg, VLC, video codecs, handwritten assembly, open-source maintainer burnout and the Google security drama with Lex Fridman.
- #493 – Jeff Kaplan: World of Warcraft, Overwatch, Blizzard, and Future of Gaming
Mar 11, 2026
Former Blizzard designer Jeff Kaplan tells Lex Fridman about his path from failed writer and EverQuest player to World of Warcraft, Titan, Overwatch, and his new studio's game.
From the publisher
Jensen Huang is the co-founder and CEO of NVIDIA, the world’s most valuable company and the engine powering the AI computing revolution. Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep494-sc See below for timestamps, transcript, and to give feedback, submit questions, contact Lex, etc. Transcript: https://lexfridman.com/jensen-huang-transcript CONTACT LEX: Feedback – give feedback to Lex: https://lexfridman.com/survey AMA – submit questions, videos or call-in: https://lexfridman.com/ama Hiring – join our team: https://lexfridman.com/hiring Other – other ways to get in touch: https://lexfridman.com/contact EPISODE LINKS: NVIDIA: https://nvidia.com NVIDIA on X: https://x.com/nvidia NVIDIA AI on X: https://x.com/NVIDIAAI NVIDIA on YouTube: https://youtube.com/@nvidia NVIDIA on Instagram: https://www.instagram.com/nvidia/ NVIDIA on LinkedIn: https://www.linkedin.com/company/nvidia/ NVIDIA on Facebook: https://www.facebook.com/NVIDIA/ NVIDIA on GitHub: https://github.com/NVIDIA Nemotron: https://developer.nvidia.com/nemotron SPONSORS: To support this podcast, check out our sponsors & get discounts: Perplexity: AI-powered answer engine. Go to https://perplexity.ai/ Shopify: Sell stuff online. Go to https://shopify.com/lex LMNT: Zero-sugar electrolyte drink mix. Go to https://drinkLMNT.com/lex Fin: AI agent for customer service. Go to https://fin.ai/lex Quo: Phone system (calls, texts, contacts) for businesses. Go to https://quo.com/lex OUTLINE: (00:00) – Introduction (00:26) – Sponsors, Comments, and Reflections (06:34) – Extreme co-design and rack-scale engineering (09:20) – How Jensen runs NVIDIA (28:41) – AI scaling laws (43:41) – Biggest blockers to AI scaling laws (45:25) – Supply chain (47:20) – Memory (53:25) – Power (58:45) – Elon and Colossus (1:02:13) – Jensen’s approach to engineering and leadership (1:07:38) – China (1:15:51) – TSMC and Taiwan (1:21:06) – NVIDIA’s moat (1:26:43) – AI data centers in space (1:30:31) – Will…
Summary, chapters and key points were generated from the episode transcript on October 3, 2026. Quotes are checked word for word against the transcript.