Latest / The Joe Rogan Recap / Jensen Huang (2025) - The American AI Dream

Transcript

Jensen Huang (2025) - The American AI Dream · The Joe Rogan Recap

• 0:00
  Welcome to the Joe Rogan recap, and before we get going, please
• 0:02
  follow us right here on Spotify to make sure you get sight of
• 0:05
  our latest episodes in your feed.
• 0:08
  Today, we are taking a really unprecedented deep dive into the
• 0:12
  mind of Jensen Huang, the CEO of NVIDIA.
• 0:15
  And if you want to understand this current moment in
• 0:18
  technology, I mean the global AI race, what's happening with the
• 0:21
  future of work, even US industrial policy, this deep
• 0:24
  dive, which we've pulled from his talk on JRH TAG 2422 is
• 0:29
  basically the essential shortcut.
• 0:31
  Yeah, what's so fascinating about this transcript is just
• 0:33
  how broad the conversation gets. It goes from, you know, the
• 0:37
  Super technical origin story of deep learning all the way to
• 0:40
  these incredibly candid political discussion.
• 0:42
  And his own, like, personal anxiety, right?
• 0:44
  Exactly the anxiety of ACEO running a multi trillion dollar
• 0:47
  company. So our mission here is really to
• 0:49
  pull on the threads that connect his personal story to the global
• 0:52
  impact Nvidia's having right now.
• 0:54
  OK. So let's jump right in, starting
• 0:56
  with the context, because you've got this powerful tech CEO
• 1:00
  meeting, you know, a huge media personality and the source
• 1:03
  material just drops us into this this bizarre political overlap
• 1:07
  they have. It's a great hook.
• 1:09
  So they first met at SpaceX, but the second time involved this
• 1:13
  wild phone call to Joe Rogan from President Trump while Joe
• 1:16
  was apparently outside shooting arrows of.
• 1:19
  Course he was. It's just a classic Rogan story,
• 1:21
  and it frames a really key part of Huang's perspective on
• 1:24
  policy. He described President Trump as
• 1:28
  an odd guy. Not what he expected.
• 1:30
  Not at all. But he also said he was an
• 1:32
  incredibly good listener who, you know, remembered every
• 1:35
  single detail Huang told him. He even remembered Trump acting
• 1:38
  like a kid wanting to show off plans for a UFC fight on the
• 1:41
  White House lawn. And that political thread, it
• 1:43
  keeps going. Huang gets into his assessment
• 1:46
  of the policies that actually affected the tech sector.
• 1:48
  Right. He talked about what he called
• 1:49
  common sense policies and he was really focused on the drive to
• 1:53
  bring manufacturing back to the US He frames it not as like a
• 1:57
  partisan thing, but as a matter of national security.
• 2:00
  And creating jobs. Reindustrialization.
• 2:02
  Exactly. But the most critical point he
• 2:05
  makes, and this is something I don't think a lot of people
• 2:07
  connect, is the link between energy policy and AI dominance.
• 2:11
  He makes a huge claim there. A huge one.
• 2:14
  He specifically says that the expansionist energy policy, the
• 2:17
  drill baby drill approach, actually saves the AI industry.
• 2:22
  OK, that's a bold statement. How does building oil rigs save
• 2:26
  AI? Well, his reasoning is totally
• 2:28
  practical. To build these massive chip
• 2:30
  factories, these foundries, and the supercomputer facilities
• 2:33
  that train AI, you need just enormous amounts of cheap,
• 2:37
  abundant and stable energy. Right.
• 2:38
  They're incredibly power hungry. Massively and he argues that
• 2:43
  without that energy policy creating the right industrial
• 2:46
  environment, those huge infrastructure projects just
• 2:48
  wouldn't have been possible in the US.
• 2:50
  And he felt that the administration saw NVIDIA as a
• 2:53
  real national asset. He called them a national
• 2:56
  treasure. Which connects directly to his
• 2:58
  bigger view of the world. That technology leadership
• 3:01
  basically gives you superpowers. Superpowers.
• 3:03
  Information, Energy. Military might.
• 3:06
  He sees the US as being in this constant tech race, going back
• 3:10
  to the Industrial Revolution, the Manhattan Project.
• 3:13
  And he makes that comparison to Europe.
• 3:14
  Yeah, he contrasts the American approach, which is sort of take
• 3:17
  the new tech and run with it, with Europe, which he feels can
• 3:20
  get bogged down in policy debates before they even start.
• 3:24
  For him, this AI race is just the latest version of that.
• 3:28
  That intensity is the perfect setup for where we're going
• 3:30
  next, which is the tech itself, because you have this duality,
• 3:34
  the optimism that Huang has versus the, you know, the real
• 3:39
  fear a lot of people feel. Elon Musk's 80% awesome, 20%
• 3:43
  trouble analogy. Exactly.
• 3:45
  And Huang's confidence seems to come from the fact that progress
• 3:48
  and safety has been just as fast as progress and capability, he
• 3:52
  says. Capability has gone up what,
• 3:53
  maybe 100 times in two years? 100 times.
• 3:56
  But he argues use that most of that power has been channeled
• 3:58
  directly into making it safer, not more dangerous.
• 4:01
  He. Used that great car analogy, a
• 4:03
  modern sports car has way more horsepower than 1 from 20 years
• 4:06
  ago, but it's infinitely safer to drive.
• 4:09
  Right, because all that extra power is used for things like
• 4:11
  ABS, traction controls, stability systems.
• 4:14
  The power is used to prevent the risk.
• 4:16
  So safety becomes a feature, not just a patch you add on later.
• 4:19
  It's. Built in, He talks about how
• 4:21
  these new AIS can reflect on their own answers, ground their
• 4:24
  results in truth to reduce hallucination and sort of
• 4:27
  breakdown problems step by step. For Huang, bitter functionality
• 4:31
  is better safety. But what about the deeper fear?
• 4:33
  The philosophical 1 sentience. You have that famous story of
• 4:37
  the AI that threatened to blackmail a programmer.
• 4:39
  I mean, if that's not sentience, what is it?
• 4:41
  And that gets to the core of the debate, right?
• 4:44
  Defining consciousness. Huang is very firm on this.
• 4:47
  He says AI has knowledge and intelligence, but it does not
• 4:51
  have consciousness. What's the difference for?
• 4:53
  Him he defines consciousness as needing experience, feelings,
• 4:57
  self-awareness and ego. You know, all these concepts we
• 5:00
  can't even really define for ourselves, let alone for a
• 5:02
  machine. So the blackmailing AI was just
• 5:05
  but a really good mimic. Exactly.
• 5:07
  A sophisticated pattern generator.
• 5:10
  It was pulling behavioral patterns from its training data,
• 5:13
  from novels, movies, Internet text where that kind of
• 5:17
  manipulative behavior exists, and just generating the next
• 5:20
  logical, though very sinister sounding response.
• 5:23
  He calls it imitation Consciousness A.
• 5:25
  Clever actor. A very clever actor, but one
• 5:28
  without any actual feeling or experience of malice behind it.
• 5:31
  So let's connect that to the most immediate threat people
• 5:34
  worry about, the military. The fear of an AI making lethal,
• 5:38
  unethical calls on its own. Yeah, and Juan's perspective
• 5:42
  here is Walt's pragmatic. He's actually happy to see U.S.
• 5:45
  military and defense startups using this technology for
• 5:48
  defense. He sees it as necessary.
• 5:50
  Necessary for security, He argues that having military
• 5:53
  strength is often what brings adversaries to the negotiating
• 5:55
  table. And he also has a practical
• 5:57
  defense against some rogue AI. He says it won't be one rogue AI
• 6:01
  versus defenseless humans. It'll be an army of our AIS
• 6:04
  protecting us. So an AI immune system?
• 6:06
  Almost. Kind of like that like how cyber
• 6:08
  defense works now with system sharing threat data instantly.
• 6:11
  OK, so if the security fears are being managed, what about the
• 6:14
  economic ones? That brings us to jobs, the
• 6:17
  great fear of job elimination. And here he gives that
• 6:21
  incredible paradox of the radiologist.
• 6:24
  This might be the single best example of job transformation.
• 6:28
  Five years ago, the experts were all saying AI would completely
• 6:31
  wipe out radiologists. I remember that it was the goto
• 6:34
  example. It was, and AI has swept the
• 6:36
  field. Almost every radiologist uses it
• 6:38
  now. And yet the number of
• 6:40
  radiologists actually grew. How is that possible?
• 6:43
  Because the purpose of the job isn't just to look at a 2D
• 6:46
  image, the purpose is to diagnose disease, and the AI
• 6:50
  made analyzing images so much more efficient that it allowed
• 6:53
  for much more complex studies. 3D4D scans scans over time, so
• 6:58
  the complexity and volume of the work went up, which meant you
• 7:01
  needed more human medical judgment to interpret it all.
• 7:04
  So the AI replaced the task, but it amplified the purpose.
• 7:07
  Perfectly put. It doesn't automate the lawyer,
• 7:09
  it automates the research so the lawyer can help more people.
• 7:12
  Huang thinks. It'll replace the repetitive
• 7:14
  tasks, chopping vegetables, analyzing a standard image, but
• 7:18
  not the ultimate purpose. And he thinks this will create
• 7:21
  all new industries. Robot mechanics, robot
• 7:24
  maintenance. He even joked about robot
• 7:26
  apparel for custom industrial robots.
• 7:28
  It's. Like a fun idea.
• 7:30
  OK, but let's talk about the technology divide because the
• 7:32
  counter argument is that AI requires these huge expensive
• 7:36
  server farms which only widens the gap.
• 7:39
  Huang says the opposite is true because of something he calls
• 7:42
  the. Right, he's talking about
• 7:44
  accelerated computing. So traditional computing uses a
• 7:47
  CPU, a central processing unit which does things 1 by 1.
• 7:51
  Serially. Accelerated computing uses a GPU
• 7:54
  with thousands of cores working parallel, and the performance
• 7:57
  improvement from that shift is just staggering.
• 8:01
  He estimates 100,000 times performance improvement.
• 8:04
  Wait 100,000 times. Just 10 years and that
• 8:07
  exponential leap. It drastically cuts the energy
• 8:10
  and the cost needed for AI models.
• 8:12
  It means that an AI that needed a multi $1,000,000 supercomputer
• 8:16
  2 years ago will. Run on your phone.
• 8:18
  It'll run perfectly fine on your phone pretty soon, and he also
• 8:21
  points out how easy it is to use.
• 8:22
  ChatGPT grew overnight. If you don't know how to use it,
• 8:26
  you just ask it to teach. You so he sees the divide
• 8:29
  collapsing, not growing. For the average person,
• 8:32
  absolutely. The absolute frontier AI will
• 8:35
  still be for the big players, but what he calls the 9 year old
• 8:38
  AI. The really good stuff from a
• 8:40
  couple of years ago will be accessible to everyone and still
• 8:44
  be totally transformative. This whole idea of accelerated
• 8:47
  computing, it sets up the next part of the story perfectly, the
• 8:50
  unbelievable origin story of NVIDIA itself, which is just a
• 8:53
  tale of incredible luck and absolute terror.
• 8:56
  And we have to start with a Big Bang of modern AI.
• 8:59
  It happened in 2012. Two students, Ilya Sutskever and
• 9:03
  Alex Krajewski, had this huge breakthrough in computer vision
• 9:07
  with a neural network they called Alexnet.
• 9:09
  And they didn't use a university supercomputer.
• 9:11
  No, that's the key. They used something that was
• 9:13
  sold to gamers. 2 NVIDIA GTX 580 graphics cards.
• 9:18
  The kind of thing you'd buy to play Quake on high settings.
• 9:20
  So they took a gaming chip. And realized that the parallel
• 9:23
  architecture, the thousands of little cores working at once,
• 9:26
  was perfect for the kind of math that deep learning needs.
• 9:30
  And the implication of that was just immense.
• 9:33
  They realized this architecture was a a universal function
• 9:36
  approximator. And that's a concept we need to
• 9:38
  pause on. Because it changed everything
• 9:40
  for Huang, it means a neural network can, in theory, learn
• 9:44
  any pattern. Newton's laws, protein folding,
• 9:46
  market dynamics. All you need is enough input and
• 9:49
  output examples. He saw that and thought it could
• 9:51
  solve basically every scientific problem.
• 9:53
  Every single one. If they could just build the
• 9:55
  hardware to scale. So Fast forward a few years.
• 9:58
  NVIDIA spends a fortune developing the DGX one this
• 10:02
  soups computer built just for AI.
• 10:04
  It costs $300,000 a pop. And when they launched it in
• 10:07
  2016, how did it do? It was a complete disaster. 0
• 10:11
  purchase orders for this multibillion dollar project.
• 10:13
  The only person who showed any real interest was Elon Musk.
• 10:16
  He asked for the very first one for his nonprofit AI company,
• 10:19
  Open AI. Open AI Huang personally
• 10:21
  delivered that first DGX one to their little office in San
• 10:24
  Francisco. An incredible moment.
• 10:26
  But even before that, NVIDIA almost didn't make it.
• 10:29
  Back in 1995, they were on the verge of collapse.
• 10:33
  They've been working for two years on a contract with Sega,
• 10:36
  building a chip for a game console.
• 10:38
  And then Huang has this terrifying realization.
• 10:41
  That everything was wrong. Everything.
• 10:43
  All three of their core technology choices were
• 10:45
  fundamentally wrong. They were dead last in the
• 10:48
  market with 100 other competitors and about 30 days
• 10:51
  from going bankrupt. So what does he do?
• 10:53
  This is where the vulnerability part comes in.
• 10:55
  He makes this desperate trip to Japan to meet the CEO of Sega,
• 10:59
  and he confesses. He says the project is a failure
• 11:02
  and it needs to be released from the contract, but then he asks
• 11:05
  for the final $5,000,000 payment not as a payment but as an
• 11:09
  investment. He's asking the person he just
• 11:11
  failed for a lifeline. An unimaginable risk.
• 11:14
  And the CEO of Sega agreed. Not because of the tech, but
• 11:17
  just because he said he liked Huang as a young man.
• 11:20
  That single act of grace. Save the company.
• 11:22
  But the tech was still broken. Still broken, They had about
• 11:25
  $1,000,000 left. Huang took half of it, $500,000,
• 11:29
  and bought an emulator machine from a company that was also
• 11:32
  going out of business. And this emulator was the Hail
• 11:35
  Mary. It was everything.
• 11:36
  It let them test their new chip design and software, running
• 11:39
  simulations before making any expensive silicon.
• 11:43
  He then went to the foundry TSMC.
• 11:46
  The most important chip maker in the world.
• 11:47
  And convinced them to skip all the normal prototypes and go
• 11:51
  straight to mass production based only on the software
• 11:53
  results. A move he said nobody has ever
• 11:56
  done before. And it worked.
• 11:58
  The chip was revolutionary, and that whole process, testing and
• 12:02
  software before you make hardware, it became the
• 12:04
  worldwide industry standard. That pivot didn't just save
• 12:07
  NVIDIA, it changed how the entire industry designs chips.
• 12:11
  And that all came from being vulnerable enough to say we are
• 12:14
  wrong. Exactly.
• 12:16
  That brings us to the final take away here, which is the CEO
• 12:19
  mindset. Despite running this massive
• 12:22
  company, Huang says his core driver is not ambition, it's the
• 12:26
  fear of failure. He still feels like he's 30 days
• 12:29
  from going out of business. Every single morning.
• 12:31
  He works seven days a week, constantly rethinking everything
• 12:35
  from first principles. He believes if you pretend to be
• 12:38
  a genius who's always right, you can't pivot when you need to,
• 12:41
  and pivoting requires admitting you were wrong.
• 12:44
  And this entire mindset is built on this incredible personal
• 12:47
  story. He moved from Taiwan and
• 12:49
  Thailand when he was 9 and ended up in a boarding school in one
• 12:53
  of the poorest counties in the US.
• 12:55
  In Oneida, KY, he talks about living in this prison like dorm
• 13:00
  cleaning toilets, having a 17 year old roommate who is
• 13:03
  recovering from a knife fight. And he communicated with his
• 13:06
  parents by sending cassette tapes back and forth in the
• 13:08
  mail. For two years, his parents gave
• 13:10
  up their whole lives. His dad was an engineer.
• 13:13
  His mom worked as a maid. They immigrated in their 40s to
• 13:16
  chase the American Dream, and Kwong feels like he now embodies
• 13:20
  that dream. So when you pull it all
• 13:22
  together, what does this mean for you, for anyone listening
• 13:24
  and trying to navigate all this change?
• 13:26
  The big take away from the NVIDIA story isn't that you have
• 13:29
  to be a genius who's always right.
• 13:30
  Not at all. It's that transformational
• 13:32
  success comes from a mindset that's actually fueled by
• 13:35
  anxiety, by constant self criticism and the willingness to
• 13:39
  be vulnerable enough to pivot when you're under pressure.
• 13:42
  It's the anxiety that builds the resilience.
• 13:44
  It is, but we should end with one last provocative thought
• 13:48
  built on a huge projection Hong made.
• 13:50
  He anticipates that in the next two to three years, 90% of the
• 13:54
  world's knowledge will be generated synthetically by AI.
• 13:57
  90%. Think about that.
• 13:59
  Almost all new information will be machine generated.
• 14:02
  So the question for you is, if we shift that quickly from human
• 14:06
  authored knowledge to AI generated knowledge, how does
• 14:08
  our whole idea of expertise change?
• 14:11
  And what new kinds of experts will we need just to Fact Check
• 14:13
  that digital deluge?