You're watching TBPN. Today is Wednesday, September 23rd, 2026. We are live from the TBPN Ultradome, the Temple of Technology, the fortress of finance, the capital of capital. Let me tell you about ramp.com. Time is money. Save both. Easy-to-use corporate cards, bill pay, accounting, and a whole lot more. All in one place.
What happened, John? The mallet broke.
No, you broke the mallet.
We broke the mallet. I mean, does that have anything to do with the number of days until Christmas? Are we at 90? Are we at a round number? So, we got to hit the gong. What are we?
I can't read that. Is it 90?
Okay. Well, yeah. If it's 92 days till Christmas, you got to celebrate. Of course.
That's right. Get ready.
TBPN road to Christmas continues. 92 more days. Get ready. Um, Nvidia CEO Jensen Huang went toe-to-toe with Ezra Klein on the New York Times podcast, The Ezra Klein Show. Uh, that's actually a crazy environment to go in as Jensen. I just think he's brave. Ezra is a real journalist.
It's like he's brave. He engaged with a real journalist.
No, it's like, you know, you have to imagine he was a bit traumatized after Dario.
Who isn't, you know, I don't think of as a journalist by any means, but he obviously has very strong beliefs and gets to the root question for a particular audience. Now, Dario, although he was going super hard on the open source question, on selling chips to China, on sort of AI safety and rollout and how the AI race should go, you know, Dario is not the person to push Jensen on like politics specifically, right? Like that's going to happen somewhere else. But yeah, it is a hot seat, but I think he did well. Two hours almost hanging out with Ezra Klein. Uh, there's some clips and I have some reactions and it's sort of interesting to dig into the mind of Jensen because he's simultaneously the biggest force in AI, the central bank of artificial intelligence according to The Economist. He's the backstop of all backstops. He's backstopping everything from, you know, the biggest data centers in the world lending investment-grade designation to the debt and the credit lines, uh, to like sort of acting as a soft landing for startups that get acquired at the application layer anywhere in the stack. Uh, he's been an acquirer that's sort of underwriting. Um, if you're a VC and you did a deal at a billion dollars, um, and you're kind of in trouble on it, there's a chance that you could get out because Jensen is willing to do a $10 billion, $15 billion acquisition. Whereas like in the prior era, Apple had the money, wouldn't do it. Google really didn't do that many deals that huge. Facebook would do one big one every 5, 10 years, but Jensen's been quick on it. And then he's also been investing directly in basically every lab and basically every AI project. So, uh, he's been an incredibly important force and yet he's now starting to stand alone in his sort of p(doom) = 0 take, which was, uh, very much the consensus in the technology community.
Would like a word, I think.
True, true. Yeah, I think it still is the consensus in the business community, in the finance community. Basically, anyone at the application layer is p(doom) = 0. Anyone deeper in the stack on the semiconductor side or the energy side, the buildout side, the neocloud side, they're all pretty much p(doom) = 0. But the labs, and Jensen's been playing there, he's at that level. He's as important a voice as Dario, Sam, Elon. And so, uh, to hear him not jump on the bandwagon when Elon, Sam, and Dario all agreed on pacing the frontier, he's saying, 'No, we don't really need to pace the frontier.' Um, it's interesting to hear him say this. There's a couple clips. Let's play one, and then I have sort of an allegory for where I think his position is. But let's play this.
So, what I've been hearing from the labs, what they've been saying publicly is that they are facing a hard problem.
Yeah, partially an engineering problem, partially an alignment problem, partially an operational excellence problem in Dario's framing.
And what they are worried about is that in competition with each other, in national competition with China, that they are being pushed to move too fast, that they all feel they're in a collective action dilemma. Now, I watch you on the All-In podcast stage. Donald Trump, President Trump gave you a call there.
Oh, no. This is not planned, but we know who it is.
And you and the president and the other members of the stage were very resistant to the idea any kind of regulation or collective action was needed.
And they're just playing right into the hands of a lot of people that don't want to see it happen. And that could be political people. It could also be China. And we're not going to let that happen. It's a hoax.
And you're right. You come out as pro-regulation.
It's illegal to slow down.
That's a form of regulation.
Like I think AI regulation is incredibly important. We need accelerationist regulation.
Really frustrate everyone. Uh, yeah, it's interesting. Didn't Jensen also do Juliana Kent on CBS just recently? I saw a clip from that. But very different clips going out. The one from Ezra Klein is talking about maybe we need to shut the labs down. A very, you know, in-the-weeds hot take about the current thing and the AI debate. I think the clip that I saw from CBS was talking about his leather jackets. Like I imagine that that interview will be more substantive when it gets clipped properly, but it's funny that that's the one that made it out initially. Uh, let's play the other clip from the Midas project. They said, uh, here's the clip of Jensen actually talking about slowing down. Why is there... release the product?
That's the simple answer. Um, if you're going to build a car, a self-driving car, and let's say it's a robo-taxi, and there's a really difficult condition.
And just as an engineer, we just have no idea how to solve this problem because these cars are not programmed. They're trained. And so we have no idea how to train these cars and we have no idea how to align them to the safety standards that are expected on the road. And so what's the answer? Don't ship it.
These products were unreleased.
These products were unreleased.
Ah, so now it comes back to an engineering problem again. And so one is, one, you have to receive it. Second, you have to, you know, think about what's the, what you could have done, what's the solution for it. And then in the future you just, you know, improve your process so that you could avoid this from happening again. I am fairly certain, I am fairly certain they will say yes, they need, they know how to solve this problem. And if that's the case, then that's the problem, it's as simple as engineering. And now the alternative, the alternative is that if they say that, if they say the alternative, which is there is no way to contain our experiments, there's just no... when we test our AI models, uh, it will get out and it will damage the world. Then I think the answer is we have to shut the labs down.
Yeah. I mean, that is kind of what happened with self-driving cars, you know, like they do before they ship them, they do test them on roads, and they don't test them on open roads. They test them on closed courses before they move to open roads. They sort of have their own sandboxes, their own environments to test these in. Uh, I do wonder if there's a sort of a legislation or liability gap between the liability incurred by a self-driving car company who causes property damage from a car running into another car autonomously and an AI agent that hacks and defaces or causes some economic damage. I don't think there will be. I think the courts, if there was true economic harm, like one model accidentally took down a payment system for an e-commerce website, I think it would be pretty easy to sue that company and say, 'You caused me to lose this much revenue. You owe me.' And the courts would say, 'Sure.' And even if the lab argued, hey, we didn't tell it to take down your e-commerce system or your payment rails, uh, the judge and the courts would say, 'Doesn't matter.' Uh, even if you expressed a duty of care, like you still have to pay in this scenario. But it is possible that there's a gap there and that's where regulation could fit in.
Um, it's interesting. He spent the first like 20 minutes sort of steelmanning the jobs question and talking about jobs because I think that's throwing a lot of people off. Like we sort of moved past the job apocalypse, SaaS apocalypse narrative which was predicted from somewhat of the same community into actual apocalypse. And people are like, whoa, like you were wrong about the SaaS apocalypse. Like Salesforce is still doing fine, like Slack still exists. Um, and you were wrong about the job apocalypse. Like the unemployment rate is like 3% for American white-collar workers. You were predicting like 50% or 30%, uh, something like 10% overall. Uh, and even in the Philippines, I mean, I remember seeing, uh, I think it was Tristan Harris on Modern Wisdom, like he's now sounding the alarm bells about existential risk, but he was saying something like the Philippines would see like, you know, 90% of their economy go away, uh, because they're heavily dependent on call centers. Call centers are actually only like three and a half percent of the Philippines job market. Uh, they do in fact make things and have agriculture and all sorts of other economic endeavors going on in the country. Um, but even in the Philippines, like the unemployment rate is, uh, right now in the Philippines it's 4.9% in June of 2026. Now, it went up to 6% in August, but it still feels like the AI effect is pretty minimal, and most people were predicting that, uh, like the offshore call centers would be affected first. And that one seems like maybe we're there, but it's just taking so much longer that everyone feels very vindicated in saying like, hey, let's watch that play out first before we move on to the X-risk discourse potentially. Okay. I think that's a lot of what...
Well, and now with Muse floating the idea of having, uh, human-in-the-loop on personal agents, you can imagine those people would be former like...
Oh, they could just move over. Yeah. Yeah. Potentially.
Um, but yeah, I mean it's all the same, uh, you know, in the limit, in the exponential, add five orders of magnitude, maybe things look very different. Uh, but Jensen just rejects that. I mean, is he an AI-is-a-normal-technology guy? I saw Joe Weisenthal posting about this. He said that he had to differentiate between... Well, I got to pull it up because it's funny. Weisenthal. Um, Joe Weisenthal said he had to disaggregate. Where is it? He posts a lot so I got to dig it up. Um, normal technology... man, you post a lot.
He said, 'The splintering around AI discourse is really a sight to behold. I was in a conversation yesterday with some folks during which it became necessary to distinguish the people who see AI as normal technology from the AI-as-normal-technology people because there is a group of people that have rallied around a particular thesis, which is AI as normal technology, and that's different than people that are just casually into that idea because it's actually like a different ecosystem.' Anyway, um, it does feel like Jensen is AI-as-normal-technology. I mean, he's certainly seen plenty of technology revolutions come and go. And, uh, he's been in this industry for what, 30-odd years. Um, and I think on the jobs question, we were talking to Joe Weisenthal about this when he was in the studio. Uh, like where is the economic impact of the internet? Like why can't you see, you know, a kink in the graph of really any economic data when like the internet takes off? It's not like productivity went way up. There's nothing really to grab on to. You actually go back to 1970 if you want to see like the real trends shift. And there are other, you know, trade and globalization moves that have had bigger economic impacts than the internet, which is crazy to think because so much wealth was created, so many companies were created. Uh, and it did change the world, like the day-to-day experience, but didn't actually show up that much in the economic data. And that's what we're seeing now where jobs are sort of changing, tasks are changing, but, uh, we're not seeing dramatically different economic statistics. Like the 10-year at all-time highs or not all-time highs, but 19-year highs, you said, uh, over 5%. And yet everyone sort of a consensus agrees that that's because of the war. Are you rooting for people that are buying the bonds and getting higher yields now?
I don't know. Sometimes a big record-breaking number, it's just exciting to you. Just general, just real golden retriever mind.
You understand the implications and you...
You don't even think about the implications potentially. You're just cheering for a bigger number. Yeah. It's like now it begins with a five. Great. I like it's a higher number.
Uh, but there's an interesting, uh, there's an interesting allegory, uh, around the effect that technology has on the labor market at least historically. Now things might change, but, uh, Stephen Covey highlighted this in First Things First, his book on time management. Uh, he said, 'I attended a seminar once where the instructor was lecturing on time. At one point, he said, "Okay, it's time for a quiz." He reached under the table and pulled out a wide-mouthed gallon jar. He sat on the table next to a platter with some fist-sized rocks. "How many of these rocks do you think I can get in the jar?" he asked. After everyone made their guesses, he said, "Okay, let's find out." He set one rock in the jar, then another, then another. Uh, I don't remember how many he got in, but he got the jar full. Then he asked, "Is that jar full?" Everyone looked at the rocks and said, "Yes." Then he said, "Ah." He reached under the table and pulled out a bucket of gravel. He dumped some gravel in and shook the jar, and the gravel went in all the little spaces left by the big rocks. Then he grinned and said once more, "Is the jar full?" By this time, people were in on him. They said, "Probably not. There's something else coming." "Good," he said, and he reached under the table and brought out a bucket of sand. He started dumping the sand in, and it went in all the little spaces left by the rocks and the gravel. Once more, he said, "Is the jar full?" "No," everyone roars. Uh, he said, "Good." And he grabbed a pitcher of water and began to pour, and the water went in between the rocks and the sand and filled up. Uh, and somebody said, "Well, there are gaps, and if you really work at it, you can always fit more into your life." He said, "No, that's not the point. The point is this. If you hadn't put these big rocks in first, would you have ever gotten any of them in?"' And so, the effect that Jensen's describing is that like at one point in human history, there were maybe only two jobs, hunting and gathering, something along those lines. Then we invent agriculture, then industrialization. At one point, everyone's farming. Now, very few people are farming, but we still have more food than ever.
There might have been a third job, singing. I could see Tyler while everyone else was hunting and gathering just kind of jester backing and just singing for all the hunters and all the gatherers, kind of like providing ambient entertainment. Yeah. But yeah, anyways,
But it does feel like that was the effect that the internet had. Like it didn't dramatically reshape the labor force, but like obviously lawyers use the internet to communicate. And now like you see the AI agents thing. It's like every little document will go through an AI pass. Every little interaction in the economy gets this like small effect, at least right now. And that's what Jensen's like living in. He's like, I have a real business to run in the real world today. So, he says he's worried about, or he says he thinks about the future, but really he's clearly very much living in the present.
It's interesting to think about how much, uh, like technology-driven efficiency gains just get effectively wasted.
So, I would imagine today it's faster to close a venture round post-term sheet than it was in like the '90s, right? You would think so with the internet and AI and all these different things, but I bet you that it's not as maybe as fast, as efficient as you might think if we can now generate docs on the fly and you can easily go back and forth on red lines. Yeah.
Right. Like maybe it went from six weeks to four weeks,
But theoretically it could have gone from six weeks down to five days, you know. Yeah.
Um, and I think you're just seeing that even with AI now, people can do their jobs faster.
And you've seen some management teams tell their employees, 'Hey, I know you're getting a lot more efficiency. That doesn't mean you should just do the same. I want you to do more work in the same amount of time. Don't do the same amount of work with less time, but you're just sort of like wasting a bunch of time.'
Yeah. Seems like no one's really saving time. Everyone's just doing more stuff.
Um, there is an interesting corollary of that, which is potentially these technologies, they don't necessarily change the growth curve, but they are responsible for the growth. Like if you don't have the internet speeding up commerce from a week to get an item to two days, that is actually the source of the 2% growth, and without the technology you have no growth. And so there is another side of an argument. Of course, there's like population growth and a whole bunch of other things that are affecting economic growth broadly. But there is a world where, yes, if you're compressing the timeline on everything, you're building the house faster, you're deciding to buy the house faster, you're exchanging everything faster as things move quicker. Even if you're not doing entirely like net new jobs, just the fact that you're doing them faster, you do more of them, and that's what actually creates the economic growth. I don't know. We'll figure it out. We'll get to the bottom of it tomorrow. Uh, let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange.
Uh, and yeah, it's always a good time to share the Financial Times black pill, white pill, uh, chart of which way things go. Either GDP goes to zero, infinity, or it stays the same. It's like that Mitch Hedberg joke. 'I used to be in a metal band. People either loved us, they hated us, or they thought we were just okay.' Stupid joke. Do you know Mitch Hedberg? Oh, he's a great like one-liner comedian. He's fantastic. RIP. Uh, anyway, um, let's move on to some other reactions. Uh, I'll tell you about Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. Uh, Claude Opus 5.5 drew every frame in this animation in JavaScript. They went all in on LLMs, on the big model, on the great model, and now it can do basically video generation, but do it in JavaScript, do it in Blender, do it in Python, do it in SVG. Uh, and you can just hill-climb on SVG, and people are sharing a bunch of cool demos, and good for launch. People are sick of benchmarks. People want to see visual stuff, entertain me, make a song. Uh, and uh, yeah.
Yeah, really, really cool style. Yeah. I'm already sort of mourning it just because this is probably going to be everywhere on the internet like by next week.
But for now, a good format. Uh, I would, you know, if you see maybe a coworker doom scrolling, send them a custom animation like this saying, 'Hey, stop doom scrolling.'
And do some work. Uh, you could maybe send this to the, uh, I think the president of Syria.
Oh, yeah. He was caught using Instagram Reels at the UN.
So, uh, naughty naughty as they say. Um, very, very cool.
Yeah. Mike Bird says, 'Total cultural victory. Man has yet to create an ideology that can compete with short-form video-based social media scrolling.'
Ahmed al-Sharaa spotted scrolling and sending Instagram Reels during UN General Assembly with the translator headphone on. He's scrolling. Hey, you know, you got to keep those group chats going. You got to keep the content flowing. You don't want to be falling off. He might be sending them to other world leaders. He might be recontextualizing what's happening at the UN with a funny reel. Like the person might be talking about some peace plan and he might be sharing some hilarious Instagram reel that makes fun of that to let his friends know that he's not buying it. Something like that.
He also could be, just his feed could be so dialed into just local political content. He could be mainlining public opinion and trying to understand what's important to voters heading into the next election. Yeah. You know.
Yeah. Yeah. Uh, there's one more Claude post I want to show, the GIF animated in Python and rendered in Blender. And at this point, you know, uh, I posted that AI video of you and people were asking like what was the workflow and I was like it's you just ask it to do exactly what you want it to do and you actually don't even need to know the word Blender, like you can just...
You can even misspell every word and it will still...
No, you just open up the voice mode and I just talk and I'm like, you can even do multiple things now. So, I was like, I have this children's story that I made up for my 5-year-old and I was like in one shot, I was like, take this story, turn it into a series of stories, then a series of children's books, go find a place that can print the children's books, illustrate it, turn it into short-form video, and also make a video game. And it did it all. It was just like spawning sub-agents to do it. And you can just do it all in one prompt. And I didn't need to be like, well, I want you to use JavaScript for this and Blender for that. You can just ask for what you want and basically get it from all the models right now. Uh, you have an interesting thesis. You're black-pilling on the application layer now. You think the models are getting so good that people are just going to use the AI tools. Does this change anything for you? Because I see this and like I still feel like these are great. There's probably a couple revisions. When I do the prompts, I still sort of filter and review. I'll usually get like 20 outputs, pick the best one. I actually talked to an AI video founder yesterday who's doing like AI movie production and whatnot. And it was absolutely printing using all the latest models. The business is doing fantastically. He's hiring four video editors like a week or a day or something, like he's hiring lots of people because there still is a lot of, it's not even prompt engineering. It's more like processing the output, curatorial work, understanding what is the right thing to fit together. And I think that mainly comes into picture when you're looking at something that's a bigger project. You want to go from a one-minute thing that might have some consistency, but when you go to two hours, the consistency becomes much more important. The style, the pacing, matching everything matters more. I don't know. Um, I think we might be gearing up for another application-layer versus model-layer debate. We saw this with the Harvey discourse earlier this, uh, I think we discussed that yesterday actually. People were sort of black-pilling because their margins went negative, but then all the models got cheaper since that released and so you think better and cheaper back up. Yeah. But of course, uh, if the models can do it at a base level, there is a world where every company just has, you know, a relationship with a foundation lab. So I don't know, we'll see. Anyway, people are feeling the AGI. Uh, the serious adult ML enjoyer at Anthropic. He said, 'I'm feeling the AGI.' I gotta follow this guy back. Uh, 'I'm feeling the AGI.' He made this cool video with Opus 5.5. Then Sholto comes in from the top rope. 'We don't have AGI. The job's not finished.' I love the debate. Even internally, people are saying they're maybe loosening up comms over there. I think it's cool. I like seeing different takes. Uh, of course, these companies are not monoliths. It's nice to actually get a glimpse into everyone's different perceptions. Everyone has different definitions for AGI, and I think it's cool to actually toy with them. Early in the show, I coined a definition of AGI, which was just purely economics. Just when AI revenues equal non-AI revenues, you have AGI. So like when the AI economy is as big as the human economy, then that's AGI. And of course, that's completely arbitrary. Who knows if that's a valuable metric. Certainly interesting, but I can definitively say we're not there because AI is contributing like a quarter percent to GDP and total lab revenues are in like the hundreds of billions while we're doing like tens of trillions in the global economy or in the US economy even. So, uh, interesting stuff, fun. Go play around with it. Let us know what you build. Uh, I like that they put the horse riding the astronaut on the moon. Uh, and I think there's probably some Blender under the hood, but it was cool that the model was able to do the post-processing and adding grain and texture that made it look very special. Like it popped out on the timeline, at least to me, because of that. Um, very cool. I also had an interesting test, like you really can hill-climb SVG. At one point, I just took an AI-generated image of a pelican riding a bicycle, which is of course because it's AI-generated, a GenAI image, like looks amazing. And then I just told Astra like, 'Turn this into an SVG pixel by pixel.' And I was able to send it to you. It was like a 40-megabyte SVG, but it looks exactly like a pelican riding a bicycle. And so it calls into question like what even matters, because you can use a generative AI tool to then create a Blender model to then create an After Effects file to then bake it down to a PNG or change it, like every format changes into every other format. Uh, use the best tool for the job. Uh, we're definitely in the regime of like how much did it cost to actually get that thing done as opposed to some like artificial syntax benchmark around like a line of code costs this much, like no one cares. No one cares what tools are being used under the hood. Uh, there's definitely a world where you go to a model, you ask for a thing, and if it needs to use Slack, it uses Slack, if it needs to use Photoshop, it uses Photoshop, if it needs to write its own thing in Python, it does that. It uses the right tool for the job and it intelligently chooses things just like a real person on your team would, which is exciting. Uh, what else?
From Dave over on X. He flagged this first.
Let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish.
We are one week out from the Roadster reveal.
And I'm... Did you listen to the end of the show when you bounced?
We went deep on the Roadster. We got a scoop. We got a scoop from our guest who was next to a Tesla employee on a plane and was like pulling facts about him just to go leak it on TBPN. Yeah. Crazy. No, no, I mean I don't know how real this was because the guy could have been messing with him. I don't know. It's all rumor mill stuff, but the guy was basically like, 'Yeah, I was next to a Tesla employee and he was saying that the jets are not to make it fly. It's actually for downforce. It's going to suck it to the ground and it's going to go 0 to 60 in 1 second.' Very cool prediction. Very different from what we were saying. So, who had the lowest?
Yeah. The only thing is it doesn't track...
Who had the lowest number. So, you're closer to win if that's what happened.
Well, it doesn't track with what Elon was saying on Rogan, which is that it can't fly.
And that's what I was getting at. It was like if you can have a fan or a jet that sucks you to the ground for one purpose, you can probably reverse those and boost up at least for a little bit for a little jump. But we'll see. We're counting it down. Well, in other EV news,
The road to Christmas, the road to Roadster, but we're still keeping an eye on it.
Saudi Arabia is aiming to disrupt the auto industry with these radical EVs.
They have a brand called Ceer. XeroJ.
It's coming as a sedan and an SUV.
It looks like sort of a futuristic like Lamborghini option.
Yeah, there's a little bit of Huracan wedge in there. Little Cybertruck in there.
Uh, is that... this is Saudi Arabia's first true production car brand. So making these, there's such a hard trade-off between the wedge looks great but it's a terrible use of space. Like the optimal use of space is like the new Waymo vehicle which is basically just a box or like a bus.
You just care about utility now.
I'm just saying it's a trade-off. Like the more wedgie you get, the less utilitarian the vehicle's space is. Uh, of course you need to have...
This one looks like somebody broke up with the new generation of Prius and now they've leveled up again. That was the original like read on the Prius is like, 'Who hurt you?' You know, cuz the new Prius is looking pretty stanced, you know? It's a level up.
Is it the Prime? Prius Prime looks pretty good. Yeah, they've done a good job.
Anyways, uh, 850 horsepower tri-motor electric powertrain. Uh,
It does look insane. I wonder where they'll hit in terms of price. That'll be interesting to see. Um, I mean it's cool. Saudi Arabia obviously known for, you know, pretty outrageous car culture in many places.
What if they could make this the performance of the Lucid at only 60% of the cost?
Yeah. Still hundreds of thousands of dollars. Uh, hopefully not that. Um, but uh, it'll be interesting.
Yeah, I have no idea where they would price this because it could be sort of like a national treasure, a point of pride. It could come in at a very high price. At the same time, uh, EV is a tough sell when you're in the six figures. Uh, should we watch the trailer? The official trailer for... you can see everything. We watched the preview clip. Now we get the full trailer. And I want you to pay attention to the quotes because the reviewers were floored by this movie. I've never seen quotes like this on a movie.
Uh, people really, really enjoyed this.
Yeah. Seriously. While they pull that up, let me tell you about public.com. Investing for those who take it seriously. They got stocks, options, bonds, crypto, treasuries, and more with great customer service. I...
feel you did nothing wrong.
So where did things get lost? Like how are you where you are right now?
I wish I knew, you know? I don't know. I don't feel like I understand that. So, I know you guys invited me to stay here. Does any part of you worry about me being here 24/7?
Well, what if I see... 24/7.
He's actually, I think, living with Elizabeth Holmes.
And it's crazy because only Nathan Fielder would even think to throw out an idea of, hey, would you mind if I moved into your house before you go to prison and documented the entire process? Because you would think that there would be no circumstances where that would be a good idea.
I think it's just pure upside. There's very little downside to doing this. I mean, you're going to prison. Like worst case, you look worse, but you're in prison either way. Like, yeah, the only way you can go is up and actually endear some people to you, which is the attempt. We'll see how it lands. Let's keep watching.
I think this is the moment in time.
No other film I've ever experienced... this story before I'm gone.
I fell in love in that first conversation before or after that she revealed she was the founder of Theranos.
I mean, weren't you worried she was sort of scamming you?
100%. But what did she have to gain from me?
Well, isn't your family rich?
Yeah. I think that there's like a lot of different levels of wealth.
You've never... What is this for?
Why are they doing a 3D scan?
That's just something Nathan would do.
Do you mind if we scan your body?
I saw that I was like, 'This has nothing to do with Theranos or going to prison.'
He... master... can prepare you.
I may think about this movie for the rest of my life. That is a crazy quote.
Feels like there's going to be some twist or something. Very interesting. But I'm...
No, I think the whole thing is going to make the viewer feel like they're in a really bad acid trip.
The pauses, the silence, the way the lighting, the sort of sense of impending doom, right? Because she's about to go to prison.
Yeah. The kids, you know, being present in these scenes. Like the whole thing is just like incredibly dark and weird. I don't know. I feel like it has to break expectations in some way because the average viewer is going to sit down and believe that she is guilty and that she lied and is like a sociopath because like that's sort of her brand, right? Is that like she never admitted any guilt and was guilty in the Alex Gibney retelling of the story and obviously the court findings. And so for it to be like that, I feel like there has to be some twist or something. I don't know. I would be shocked if it's just like, yep, like she defrauded investors and didn't really build a good blood testing device and now she's in jail. Like the end. Like would that be...
I don't think it would get those quotes.
I don't think it would get those quotes. So I feel like there's going to be something crazy that happens, some like complete twist, like maybe she's innocent or something, or maybe she's like, I don't know, guilty of something else. There's...
The question is how does the... who's the guy that she would call... Balwani?
Oh yeah. Uh, R... I actually don't know. Um, yeah, like because during the court case there was a question about his responsibility, um, should he bear more of the responsibility. And so, uh, but he doesn't seem to be in the trailer, so I don't know. Um, because one weird like twist would be if you came away being like that guy's more responsible than I thought, you know. But anyway, let me tell you about Figma agents. Meet the canvas. Your AI agents can now create and modify your Figma files with design system context. Uh, we have the president and CEO of Qualcomm, Cristiano, in the waiting room with us in the TBPN Ultradome. Cristiano, how are you doing?
Very good. How about yourself?
We're doing fantastically. Where are you calling in from?
Maui, Hawaii. It's our Snapdragon Summit. This is day two.
How long have you been doing Hawaii? Why Hawaii?
Uh, it's probably more than 10 years. I think it's, uh, we have a lot of, you know, global press that come here from all over the world, a lot of partners. Nobody complains about coming to Hawaii.
Yeah, I can imagine. Yeah, it's a good place to be. Uh, what are the highlights? What are the messages that you're trying to drive home today and with this conference?
Look, at the Snapdragon Summit, we always announce our latest, uh, you know, Snapdragon flagship processor for smartphones, but this one is special because, uh, for the past couple of years, we have been seeing, we have a clarity of vision. We have seen how smartphones...
Yes. How does your business actually change in an era of folks wanting to do more on-device inference and AI on something like an Android phone, one of your big partners? How do you think you need to adapt? Is there something you need to change, or has this been years in the works? You think you're ready?
No, years in the work. Look, we were kind of a long time ago. It feels like a long time ago when we were kind of showing, you know, multi-billion parameter models running on devices. Look, there's this whole conversation about AI running on the cloud and on the edge, and people often ask the question, why the edge, why the cloud? And I think that's probably the wrong question.
It's almost, I like to go back and say, okay, everybody get their phone out of their pockets, get your 200, 300 apps that you have, and let's go have a conversation about what part of your app runs on the device or runs on the cloud. And if you and I know, even though we build incredible processors, if you put on airplane mode, you don't use your phone. So at the end of the day, AI is no different than that. It's going to be running on the device and on the cloud. It's going to be all transparent to you.
Yeah. How are you viewing what the next couple years will look like in automotive? It feels like the demands for AI on device at the edge are even more important potentially, but maybe it's equally important. How are you seeing the automotive business develop?
Well, automotive has been ahead of that. Like we've been very fortunate. Automotive is a new business for Qualcomm, and we now serve all the car companies in the world. And what has been interesting about that is AI on the edge actually is being very big in automotive because of assisted driving and autonomous driving. That's kind of AI running on a processor in the car. But what's happening with the car right now is when you're behind the wheel and you don't have any legacy of OSs and apps, an agentic experience is very natural when you're behind the wheel. We also see now the fusion of the systems in the car that was designed for navigation, like the cameras for example, being also used for see what I see and have agent experiences. Like, you know, people are in their car and they say, well, the agent, this restaurant on the right, what's the Yelp review and do they have availability for lunch right now? Those are kind of some of the experiences we're seeing, and that's accelerating for us.
Got it. How are you processing the narrative around like the CPU crunch? We saw the agentic era boom, and you know, there were the whole websites that were going down because so much code was getting pushed. It feels like agents, like the idea that everything would be done at the model layer on a GPU, is not the way things are playing out. CPUs are incredibly important, especially in the data center. I'm wondering if are we going to experience like another CPU crunch as we enter the era of like personal agents, because we saw the effect that the enterprise agents and software coding agents had on the CPU demand. What does the next couple months even look like for CPU demand in the data center?
Okay, let me separate that conversation into, I think, one topic is there's so much more demand for compute than availability right now, and across the board. So I think what we've seen this right now, and by the way, as a semiconductor company, I'll tell you the supply chain, it's operating at 100% capacity. Everything is short because there's so much more demand than the availability of compute. Now, the second part is the way we see it, and we're just entering the data center, but we kind of see that also on devices at the edge. The data center is going to have to evolve, and it's already on its way to do that, into a heterogeneous compute. You have different engines to do different things. CPUs are going to be very important for orchestrators and agents, and CPU demand will continue to rise. And that's also true on the other side of data center, phones. As those agents get deployed on devices, they have a lot of CPU demand, but you're going to have different engines for different things. Like for example, for inference, we see there's an engine now for prefill, engine for decode, decode attention, and so forth.
How are you thinking about the changing landscape of consumer devices? I mean, Meta Connect is today. I'm sure that they're going to announce some new consumer devices. But we talk to founders all day long that are building just like really cool small robotics projects, wearables, devices. There's rings and wristbands and ankle bands and anything you can wear, put on your body. It's happening. And I'm wondering if you're seeing that actually start to move up your to-do list as like you got to engage with that smaller community even though it might be niche. And are you actually trying to engage with like the smaller device community on the consumer side at a conference like this?
Look, and I'm going to say this in all humility, but the majority of those new classes of devices are actually used in AI. And there is now a multitude of those products now, and there's big companies and small companies. And the reason is because when you think about agents, and agents are not bound by OSs or applications, and you have multimodal, like think about glasses for example. I'm actually a big believer that glasses is going to see an inflection point. The glass is prime real estate, close to your eyes, to your mouth, to your ears, your head turns, your camera sees it.
And then those things like see what I see, read what I read, hear what I hear are going to come up. But we've seen all sorts of form factors. We've seen earbuds with cameras, jewelry, pins, buttons, and this is kind of the new personal AI category, and I think that would be a very big category.
How are you grappling with the open-source ecosystem versus closed-source tooling to enable the next generation of devices to integrate? I mean, obviously, you've already won a huge portion of them over, but you want to keep that forever, right?
Oh, thank you for asking this question. I think it's maybe a great opportunity for me to make a plug about what we're doing with Modular. So, I don't know if you heard about what we're doing with Modular.
Look, we made an acquisition of a company, and that's a great team. It's the Modular team. I think the founder is Chris Lattner. I think he's probably a legend within the computer science world. He was the inventor of the Apple Swift programming language. He was the inventor of LLVM.
And he built a stack which is like CUDA, but is designed to work on any hardware. Doesn't matter, you know, CPU, accelerators, and will run on Nvidia, on AMD, on Qualcomm, or on any hardware whatsoever. So we bought that company, and we're making that open source. That's what we're doing because we actually believe that the industry will benefit from an open-source stack that scales from the data center across different hardware in the edge, and it doesn't matter. I'm going to celebrate that stack in each and every one of my competitors because we probably need an open stack to drive innovation in AI, otherwise it's just one company doing most of the innovation.
Thank you, that's a very helpful explanation. We were just earlier in the show talking about Jensen Huang sitting down with Ezra Klein, engaging with some of the very deep questions about AI jobs and existential risk. Is this something, is this conversation that we're seeing bubble up in the public sphere actually making its way to the C-suite of your customers that you engage with, or are you sort of in a mode of put one foot in front of the other and let those conversations take place in other platforms? Have you been engaging with all these debates around slowdowns and open-source AI versus closed-source AI? It feels like every week there's a new big hot topic and meaty almost sci-fi scenario to engage with. But what has your approach been as the CEO of an important company in the space?
Well, I wish we had at least like half an hour to have this conversation. This is a very big topic. It's a very big topic. But look, in one minute that we have, I'm going to try to maybe give you an answer.
Solve the whole problem. One minute.
In one minute. [laughter] I think there's a lot of different conversations, and look, it kind of changes. For example, in the United States, you see a lot of conversation about safety of the models. Look, I think people talk about safety in general, but there are very specific things. I think cybersecurity is actually a big one, and I think it's really important. It's important to have products that are done responsibly. I think nobody's going to argue against that. Hey, do I want you to build a product that is going to go crazy? No, no, I don't want you to do that. I think that's kind of a logical thing. And cybersecurity is actually a serious issue. There's a big surface area. But if you go to places like China, the conversation is very different. It's about putting AI in every car, in every phone, every PC, every industrial, and it's kind of very different.
Yeah. Yeah, it is. Yeah. We were talking about that, how there's just such a wide gap in the discourse between the impact of putting AI in all these little places and then you have like the bigger questions. But we'll get to that next time. We'd love to have you back on the show. We can spend a full hour solving...
30 minutes we can get to the bottom.
Yeah. Yeah. Then we'll solve it. It will be solved. But congratulations for calling in from your... Thanks so much for taking a couple minutes today to come chat with us. Have a great rest of your day. We'll talk to you soon.
Cheers. You too. Great talking to you guys. Thank you.
Goodbye. Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it. Coming back on the show, we got Talia Goldberg, partner at Bessemer Venture Partners. We very much enjoyed her last appearance.
What's going on? How you doing? Hey, great to see you guys.
You had a hot take last time that went very viral and proved true. Everyone thought, everyone was like, 'No, it's about to collapse.' And you were like, 'No, I think it's okay if there's some, you know, margin compression in the short term. Things will iron out, models will get cheaper.' And everyone was like, 'She's not taking finance seriously.' And here you are vindicated.
Vindicated. So, congratulations. Victory lap package.
Your margin is my opportunity.
There we go. There we go. How are you processing the current moment? What's exciting to you? Is the lab trade over? Is there more opportunity above the fold at the application layer, below the fold at the semiconductor neocloud buildout level? Like what's exciting to you personally?
I mean, yes and yes. I'll say a few things I've been thinking about lately. One is physical AI, and one is this new concept that a portfolio company of mine coined called token market fit.
And I'll talk about, like, I love this idea of token market fit. I wish I had invented it, but they did. And the idea of token market fit is, simply put, like, what are the areas and the categories where we've found the ability to productively use, the average end user, like $10,000 a month of tokens? And if you actually look at where a lot of the spend has gone, it's really like there are basically only three categories I can think of right now that have like real token market fit. One is in coding, like, you know, enormous spend in coding. The second is in video and in media, and in your line of work where you can very productively spend, you know, huge amounts of money creating great content. And then maybe the third is in high-frequency trading. But if you look at, like, on average, you know, legal, customer support, sales, like, we actually haven't really hit like product market fit in a real way in a lot of those areas. We will, and I think the bottlenecks are moving away from things like code and video.
Is it possible that you can have product market fit without massive costs? Because like... I would argue that you have like token market fit in something like legal, but it's just not, you're not going to see massive spend in the way due to the nature of the work. Right. So it's like possible that you can have... I'm just, this is maybe a question, is like, can you have token market fit with just like relatively modest per-person spend at a company?
Is this sort of the argument that like we need a lot more software, but maybe we don't need that many more legal briefs or something, like the market size is small? Is that...
Yeah, there's in some ways almost like infinite software, like a product is never finished, but in legal work, like you do a deal and then it's sort of like done and there's not, you know, I don't know, my question, yeah, my question is like, I believe there's going to be some categories where like AI is incredible, it transforms like the task or the job, but you just don't end up with that much, it's just very efficient.
And you don't end up with like, you know, exceptional spend. Right. You're saying like $10,000 a month per company in a category.
It's possible. I mean, I think like I guess, yeah, there's a difference between product market fit and token market fit. But legal is like a really interesting example. Bessemer's been a longtime investor in Lora, which is like a great example of this. And I think they're still very early actually in this transition towards what's possible and what spend even for their customers is possible. So I think legal is like yet to be unlocked in a lot of ways. And in a lot of use cases, Lora is still co-pilot. It's not really autopilot. It's not like truly doing the work of lawyers, and you don't see like large law firms like, you know, massively changing their team compositions yet. I think that is still to come. And that's on the come, and as we can figure out how to productively leverage models to give more work to agents, like we will be spending more on agents and on models in fields like legal too.
Yeah. Yeah. I mean, we saw that transition maybe last year, earlier this year with like the software engineer who says like, I don't read the code anymore, and that can sometimes be a little bit risky, but in a lot of places... the lawyer who's like, I don't even read the contract. That would be token market fit. Like if I heard a, you know, a star lawyer say, 'Yeah, I don't even read the filing anymore, the document,' I'd be like, 'Okay, yeah, superintelligence is here. It's working.' Like I don't...
Yeah. Well, the difference is like software, you can ship a feature and it can be broken 1% of the time. Where in legal, if you have a 1%, you know, major error rate in contracts, like it's a lot of money on the line.
Yeah. Maybe you have to be a little bit higher up. Yeah.
Yeah. Look, there are these categories that, like, I guess the point is like they're still to come and like they're still growing. And so like what are those next areas with token market fit from an investment perspective are interesting because while you're right there will be some that are maybe a lot more efficient, a lot of the dollars are in the token flow.
Yeah. How important do you think it is to be the entry point to a certain token flow, to be an aggregator in the strategy parlance? I was just thinking about FAL and Higgsfield. They, you know, access to incredible models. There's a few other companies in the category, but a lot of people, the workflow is like go to their preferred AI agent, then send an API key over, and they've already put some credits in an account, and then they're using another tool to sort of access that inference. And there's a world where that billing relationship lives within Claude Code or lives within Codex, and they're taking sort of an app store cut. Is that a nightmare? Is that a death knell for the inference provider, or could that actually be the future way of these companies working together in a positive way that actually grows the market and the distribution so much that it offsets whatever rent the model labs and the front doors are taking?
Yeah. Yeah, look, I think it's probably not like black and white in every single category, and so it's somewhat different. Like even in, you know, in the case of FAL as an example, like they are the front door, and so, you know, that you can even access a model like, you know, in the past, models like Nano Banana and Google's models even through FAL, so you don't have to go to Google. So like they have their cake and eat it too. But we do see this playing out, and I think like Instinct and Muse and what's happening with Meta or Amazon and Shopify are other really interesting analogies of this aggregation-disaggregation theory. And at a high level, like I am a believer like there will be places for both, and like both exist. Amazon exists and Shopify exists, and like the tension between the two is real.
Personal agent predictions. How are you thinking about the category?
Um, well, we're very small seed investors in Instinct, so we're, you know, very bullish on their... I don't know if I should be offended by that sound or...
No, no, no. It was a small check, but now it's...
But because you went in the seed, it's big now.
I wish we were bigger. I wish we were bigger investors. But it is such a magical experience. I'm obsessed with these products. Like other than when ChatGPT launched, like this is the second 'oh my god' moment that I've had. And like my parents are having this moment and my sister's having this moment, and they don't even use technology. And so it's kind of crazy.
