Trump's Deal with NVIDIA, Genie 3 AI Paintings, Intern GPT-5 Test, AI Hedge Funds | Peter Harrell, Aaron Ginn, John Maslin

Watch. Monday, August 11, 2025. We are live from the TVPN Ultradome, the Temple of technology, the fortress of finance, the capital of capital. Do they do that in the Korean version of tvpn? Do they scream at the camera and they open up the. I need to watch this thing. If you're not familiar, there's a very. There's a very uncanny clone of this show's format.

Clone is at the timeline.

Yeah. Korean TVPN clone is at the time.

June park, which is. I believe I'm in favor of this.

I think it's cool.

Who is a technology brother. Okay, found NFM Live.

Yeah.

Over on LinkedIn. But we got to find their intro. We got to see.

Yeah, we have to dig into this.

I forgot that you're watching NFM Live in Korean would hit extremely hard.

It sounds like. It sounds incredible. But yeah, I mean, this is fun. Obviously, you know, we've played a lot with this format. We. We invented television. We invented news. And now people are taking our invention. We forgot to patent it, though. We should have patented news and television and shows and live stuff and business news. That would.

Technology news.

But we forgot to patent it. And so now everyone's copying us from over in Korea. But I mean, it makes sense. We don't really. I don't think we have a large audience in Korea and so have some fun over there. Hopefully it goes well. Good luck to you, NFM Live. I hope that you're massively successful. And I hope that, you know, you can start with, you know, the exact layout that we've done, and then you can kind of twist it and change it, make your own. And that's. That's how these things start. Tyler, what you got for.

Okay, so I just found it. They do not do any kind of interesting intro to start the show.

They just start the show.

No, no energy.

No energy. Okay.

Step it, guys gotta copy the important stuff.

Yeah, the important.

It's not about the overlay. It's about.

It's about the screaming. You losing your voice.

Losing my voice?

J. Kim, you gotta get. Bring up the energy.

Yep.

J. Kim, he's the host.

Jordy.

I clipped your intro, posted it, and.

Have never gotten more responses every time. That's the best intro we've ever done.

With just this one right now.

No, no, the one from last week.

We started the screaming.

I'm always trying to one up myself.

The audience really likes.

Well, it really. It really helps because we're always like just, you know, shuffling things around just one minute before we Go live. And then we really got to dial it in. And so the, the screaming really takes it up another notch. And you know what else takes it up another notch? Getting your business on ramp.com time is money save. Both easy to use. Corporate cards, bill payments, accounting, and a whole lot more all in one place. Switch your business to ramp.comghibli moment. Okay, sorry.

Wait. Important news.

Yeah. Please, please, please.

TVPN has hit substack.

Oh yes.

Tvpn.substack.com we have a daily newsletter now.

Yes.

It is loosely based around the content that we cover on the show.

Yep. So every day before the show goes live, we put together a run of show. So we, we, we select posts that we want to talk about. We give our little takes, we figure out our, our, our, the different things that we want to dig into at various levels of depth. We were, and we thought it might be fun to send those out to folks to both let you know, hey, maybe this is a show that you want to tune into. You want to hear us talk about these things in more depth. You can hop on the stream. You'll receive the email right when we go live. So you'll be able to click and tune in if you want. And also if you don't have a chance to listen to three hours of content that day, you can get a little bit of a summary in the, of the last day, of the last day and a little bit of preview of what's going to happen today. So some of the top stories and so some of the stuff we talk about on the show, we, we will have, you know, little summary teasers up there on our TVPN substack. So please go subscribe and give us any feedback. You can reply to any of the emails that we send you. Let us know what you think. If you have pushback on the takes, if you have extra information to add, if you think we missed something, let us know through that. And then if you like certain pieces of the format, we'll see some of the analytics. We're going to be tracking you. We're going to be seeing, we're going to be seeing what links you click on, the open rates, how far you scroll down. But if you particularly like something, just let us know.

Everybody says you guys make too much content. Can't keep up. Well, here's some more content.

I knew you were gonna say here's.

Some content to help you keep up with all the content.

Yeah.

So indeed.

And we couldn't spread this content all over the Internet without the help of Restream 1 livestream 30 plus destinations, multi stream and reach your audience wherever they are. You can sign up for free. Restream thank you for hosting our show. So we talked about it yesterday or we talked about it last week and we talked about it a bunch. Genie 3 hit the timeline was maybe the biggest news of last week in terms of artificial intelligence was crazy because it was GPT5, but GPT5 was more of this consumer update. We have some updates from Tyler. He's loving GPT5, interestingly, kind of contrarian. Takes as a lot of people were dunking on it. Tyler, what did you like about GPT5? Oh, and give me your evaluation.

Yeah, so I've been. Over the weekend, I've been using it more for coding, like various things. Definitely. I've been using the thinking mode like the whole time because I'm on the Pro subscription. So you can choose thinking. So it always thinks.

So there is a dropdown now.

Yeah, well, there always was a dropdown, at least for Pro, where you could do GPT5 or GPT5 thinking.

Interesting.

And now there's also. I don't know if this was always there, but you can choose to add the legacy models too. So I can also. I can still access the model.

I've been on the flagship model. I've just been yelling at it because.

I thought the model that was not available.

Okay, yeah, yeah, yeah, yeah.

So there's the model router, which I think is just GPT5 and that'll route sometimes to thinking.

Okay. But if I'm very stern with it, it usually thinks. It usually throws itself in thinking mode. Probably don't mess this up. Don't get it wrong. Then it'll be like, I'm thinking, I'm thinking, I'm thinking.

Yeah, but I just always use thinking because I'm on Pro.

So I'm just. Wait. So I have. I have three versions. I have GPT5, which is the flagship model. I have GPT5. Three thinking, which gets you more thorough answers. And I have GPT5 Pro Research Grade Intelligence. I want to be on Pro the whole time. So I think 200 bucks a month.

Pro was similar to O3 Pro, where it's going to be like a. It's going to take a long time.

It's going to take 10 minutes every time.

GPT5 thinking is similar to like O3 where it would just always think, but sometimes it's pretty quick.

Okay, well, I'm in. I'm on chatgpt.com. i can't find 40 and I'm wondering, where is my girlfriend? Where's my girlfriend?

I'm still using 4.

Oh, I actually can't see 4o. They apparently added 40 back.

So you can. Yeah, that might just be for. Plus that you can go back, but you can add legacy models.

This is for people who are really desperate and they replied to Sam Altman on X with a really funny post and he turned it back on just for them. So there's five people that have access to the old model, but everyone else is rugged on the new one. Okay, but give me your eval and give me your qualitative assessment of like. Of like, the model. Like. Like. Because it seems like you did. You did a test where GPT5 outperformed all the other models, but then simultaneously just seems like you've been having a good time. So break both of those down.

Yeah, so. So generally, like, just for coding tasks, I found it very useful. I wouldn't say it's like blowing my mind with how much better it Is than like O3 or 4.1 or something like that, but it is like, it's a good model. It's a good model, sir.

Yeah, like, I would agree with that. I'm trying to think of what I did this, I think this weekend.

Yeah.

A lot of times I was just like, what am I thinking of? I, you know, giving you some different stuff. How do I add something to my path? Top US Government contractors hedge fund research report. Like, it was doing a lot of this stuff. I was getting lunch with someone. They were recommending a piece of enterprise software. I just kind of described it. It nailed it very quickly. I don't think it took too long to think it was fine. I actually put in agent mode yesterday and I had it hunt around on the Internet for a Nintendo Switch 2. And it did find one in stock, which was, which was very useful because there are Nintendo Switch trackers that you can go and you can like and it scrapes all the websites. But this basically just did this on an ad hoc basis. It was good, it worked, it delivered. And then for other, like, kind of quick questions, it's always been. It's been quick and I feel like I should be. It's definitely going to speed me up because it knows to not. If I stick with the model router, it's going to respond quicker for the stuff that it knows.

So also, even if you're using GPT5 thinking, like on thinking mode, there's a button that says like give me a quick answer.

That's right.

So you can just tap it if it's taking too long. But yeah, also another thing was I forget exactly what I was writing, but it helped me with writing and I found it like better than I would have expected.

That's cool. Yeah. I feel like as part of the reasoning chain, the first response should just be the 4.0 response and then it should just surface that to you and it should just kind of say, here's what I know so far. And then I'm going to keep thinking, keep cranking on this to update this, go down the reasoning chain and then if you're good, I'd almost like to flip that UX around so that it's automatically going deeper and fact checking it and improving the response over time. Over 10 minutes, giving me, you know, a normal response within 10 seconds, then spending up to 10 minutes working on it. But if I'm satisfied with the 10 second result, I can just kind of click like I'm good. Yeah, that would be. That. That might be the next, like probably saving money too.

Yeah, that could be cool.

Anyway, and then give me your eval.

Yes.

So I thought of this thing. I think I looked it up. I think someone might have written a paper about this like a year ago. But it was like someone.

Quick question. Does the Korean TVPN have an internal. We need to figure that out.

You got to copy the intro and the intern.

You have to have an intern that you throw random questions at throughout.

We can look.

Okay. Anyway, back to your.

So basically my. The new benchmark I've kind of thought of, although maybe someone's done it before, I get to look more into it. But it's basically solving a Rubik's Cube. So yeah, solving an entire Rubik's Cube is like totally like way too hard right now. But if you basically just like start with a solved cube and then just do like one move, then usually they can get one move. If you do two moves they like, none of them can get it. Except GPT5.

Okay. Only GPT5.

Yeah. So you basically you can take like the state of the Rubik's Cube 2 move.

Champion of the world. Yeah.

And then if you put it in a string, you can format it nicely and then give it to it and then GPT5.

Thinking rotating the cube feels. It feels somewhat like ARC AGI in the same way of like there's this visual spatial test that then's represented in text tokens and, and an LLM should be able to Operate on those if it was able to transform that into true spatial thinking. But it's not quite there.

Yeah.

So I think something I'm going to try. I think I might, like, write a little blog post, but I want to try actually putting it in image format.

Right.

With, like, multimodal and stuff.

Are you jealous of our substack? You're like, oh, maybe I'll have a substack. Yeah. Oh, yeah, yeah. Maybe I'll just start one. No big deal. You can publish on ours.

Okay.

Yeah. But, yes, you should. You should flash that. So continue.

But, yeah, I was thinking of, like, trying different formats for the input. Maybe that'll help it out. Or doing, like, actual multimodal, like putting an image of the cube state.

Yep.

But I was, like, very surprised after two moves, I was kind of expecting, like, you know, you can look up all the algorithms you need online to finish the last. To finish the whole cube. So I was thinking, like, oh, yeah, let me just try the. Like. Like, I'll pass in with. With three moves to go. And, like, obviously can get that.

Yep.

And then it couldn't at all. And then I tried. So this morning I tried. That was with no tool use. So this morning I tried with GPT5 with, like, the agent.

Yep. So no tool use. It was still writing Python, Right?

Yeah, that's true. But it wasn't searching up, like, algorithms. But this. This morning I tried it with Agent. Agent, and it went to, like, a bunch of different. Yeah, Agent cube solvers.

Yeah, it did.

And it did not. It found them and it, like, said, like, oh, yeah, I'm using it. This is the best move. And then it, like, was totally wrong.

Okay. Interest. So that. That feels like the. That feels like a pretty good test and a pretty good eval for, like, the era of agents. Like, yeah, what I was describing this as, Tyler, I was saying, like, you have a random Rubik's Cube. If I gave it to you, and I was like, you need to solve this. And you didn't have all the algorithms memorized. Like, but you were like, it's my job to finish this cube. What would you do? You would Google how to solve Rubik's Cube. You might watch a YouTube video. You might, like, go to something, some cheat sheet that's like, type in all the colors on the Rubik's Cube and then it tells you exactly what to do step by step. And there are tons and tons of these, like, solvers. Like, like, you go out there, you Google how to solve a Rubik's Cube. You type in what you have on the colors and then it tells you move this one, rotate this one. The problem is, is that all those websites are completely built differently. Some of them are in Web gl, some of them are in, you know, like just old HTML. Some of them are the way the.

Model learns to solve a Rubik's Cube is by pretending to be a human and just Google searching.

That's how they will solve every problem because these tools exist. So it's like instead of like, instead of like rallying on door. On doordash, just go figure out how to use doordash. Go figure out like the answer to like booking the flight won't be like reconstituting an airline from scratch. It'll just be like Google it find like use Google flights. Right, like use, use whatever's out there. And so it was, it was impressive that it was able to perform at a higher level than the other models on the spatial reasoning required to finish the last two moves of the cube. But it was underwhelming on the agent side that it wasn't able to find and use a cube solver because it clearly hasn't been trained on that.

Yeah. But it is interesting that GPT5 thinking does like way outperform. It was like thing for like six minutes just to get two moves.

Wow.

Which like is really long. But it's like, it is like. I think it shows that there's like, you know, something. It's not just O3 packaged up as like in a router.

So did you have 03 or.

Yes, it did not and oh three didn't.

So. So there's something going on with the reasoning that is better. Like this is. It's weird for worse than 03.

Well, it's. We only, we never got actual 04. We only had 04 mini.

Okay, okay. So. So this could be like 04 under the hood. Like a better reasoning model RL on top of it. That's.

Yeah. Maybe smarter. Yeah.

And it's. And it's accessing that, you know, in a certain way, but it's still like nowhere near one shotting. Like just random hard problems.

Yeah. Yeah.

I had lunch with some friends on Saturday and one of the friends is a physics professor at a very prestigious college and he was able to get GPT5 Pro to one shot, a problem in about 25 minutes that he said his grad that would take his graduate students months to do on their own.

Historically, I can't even imagine the type of physics problem that Takes a month to do it.

I know I had him. I don't want to give away too much of the Alpha. He was like, his mind was like completely blown. This was on Saturday. He had just gotten access to the model and he was completely blown away.

It's such a weird era of the spiky intelligence because, like, pretty much any human should be able to solve the last two moves of a Rubik's Cube in like an hour of like tinkering with it and just trying stuff, moving back. Okay, what do you think?

An hour seems like a long time. Two moves.

It's two moves. Like, you can just look at it. Yeah, yeah. It doesn't seem that hard. Do we have.

Just look at it.

We have a Rubik's cube right over there. We have a normal Rubik's cube in the studio right somewhere. Let's, let's get that out there and see that. Toss that to me.

It's already solved the.

Yeah, yeah, yeah, There we go.

High stakes throw. That would have been deeply embarrassing for the firm if you dropped that.

So, yeah, it's like two moves off and then you have to look at it and then you have to, well.

Do a couple moves and then let's feed it to Tyler and he can work on it.

Oh, yeah. See if he can do it. Okay.

Anyway, let's see if GPT Pro can do it.

Oh, well. Oh, yeah. How many times did you test it?

Do you use Agent mode and tell them, like, figure this out, Figure this out.

Well, that's why I tried Agent Mode. Didn't do anything.

Didn't do anything.

Okay.

Oh, well, well, well. Let's tell you about figma.com. think bigger, build faster. FIGMA helps design and development teams build great things together, great products together. Get started for free. So despite all the GPT5 news, very exciting and just a bunch of like, progress that's happening. It feels maybe not as binary, not as, not as exponential as some people were expecting, but lots of cool things. Also, I wound up using Claude code for the first time and had interesting experience there. I had it do a deep research report. No, no, not at all. But I had to do a deep research report and instead of, instead of just spitting out like bullet points and tables, I had it just spit it out as HTML. And so it was able to, it was able to generate like, really, like, I'll show you, I'll show this to you. I don't know if you can see this, but, like, can you see this? Yeah, I got to take this off show, show everyone. This is not going to work, but basically it's able to, it's able to like show formulas, show different charts and graphs and just like use all the features of HTML, which is really cool. Now there were tons of like hallucinations in here and the actual quality of the research report wasn't that great. But we were talking to Doug at Semianlysis and he was saying that that cloud code is, has been really good for research reports. So I think I'm going to try and work that into my workflow a little bit more. And I feel like the next layup for Anthropic is just wrapping Claude code in a mobile app so that you get access to like a remote file, like directory that can actually write code. Deploy websites, do all this stuff but from the app and then, so then you can instantiate things. But the cool thing was that it felt like for the first time, won't.

That kill like $10 billion startups?

Maybe, maybe. But for the first time it felt like I was feeling this whole idea of like the generative ui, the on demand ui, because as like I was asking it to do this report on world models and it just generated like different sections of the, like different charts and graphs and put them all up there talking about how these work and it was able to just like create these little highlight, these little highlight sections for Odyssey, the company founded by Oliver Cameron. How much have they raised? Physical intelligence by Lockheed Groom and Carol Houseman who've been in the show. They've had a $400 million round. Jeff Bezos Thrive Lux the problem is like, I think I still need to fact check all this because I feel like the research wasn't as strong as deep research or what you get from GPT5. But it was very cool to see it. Just use all the HTML functionalities and buckets to create a really readable website, which of course was then immediately mobile optimized. Immediately you could read it this size, that size, and it just felt like the first moment of Generative UI on demand.

You need to experience these things firsthand before you can can really have that true aha moment.

Much like you need to experience Vanta firsthand. Automate compliance, manage risk, improve trust continuously. Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation, whether you're pursuing your first framework or managing a complex program. Okay, so back to the actual big AI news of last week. Yes, Genie 3 from DeepMind, Alexander Holinsky dropped A very fascinating series of viral videos on X. Saying something, something fun. We discovered you can use Genie 3 to step into and explore your favorite paintings. Here's a short visit into Edward Hopper's Nighthawks. And this video is crazy because, and I said this is a Studio Ghibli moment. As soon as this goes general access, like as soon as everyone has access to this, you're going to have the Studio Ghibli moment. Because everyone is familiar with that particular painting. Maybe not everyone, but most people. And it's just like you can go walk around it and feels like this real world. And I think that there's going to be so many examples of.

Think of the example that I gave where you just take three photos from, from like a specific moment, drop them and create a world, an entire world around it and then you can go relive this like three dimensional memory.

Exactly, exactly. And so it's this like it's not the. I think the AI generative AI generated stuff is by default like very mid, right? Like AI slop is pretty like mid. And like the AI generated photos, like if you just showed me a photo of a car or something that was AI generated, I'd be like, I would rather just see Max Verstappen driving. I'd rather see a more cinematic image of a Red Bull F1 car that's just taken with a really like high quality camera and tells a story and immediately builds my intrigue of like, oh, when was this? Was this during a race? Was he practicing? There's just so much more value when something's, even if, even if the fidelity is there, it's not quite there, but it's just less, less of like the story behind it. But when you take something that people know or is personal, like a family photo or a painting that they're a fan of and you, and you allow them the ability to customize and experience it. I think this will go way more viral. And so this, I think Genie 3 and I think world models generally are going to be the next like hot sector for AI. I think the GPT open source GPT OSS gptos is kind of cool. But you know, we've been there, we've seen that. I don't think that there's going to be that much new there. Claude 4.1 was kind of this minor spec bump and even GPT5 was, you know, somewhat incremental. Obviously very key for the business to get them, you know, seriously competing in the enterprise space and then also, you know, improve consumer usage. And just, you know, improve the experience of the ChatGPT app.

The meme potential of this, if you can generate a world and then just share the link to that world and other people can drop into it, it's just going to go incredibly viral.

And even just exporting videos of this is good content creation, even if you're just sharing the videos that you generated. But actually being able to share the link to is going to be really, really big. So I'm super excited about this. The big question from the show was, you know, what was going to be this the Ghibli moment for world models. And I think we know it now. It's not going to be, it's not going to be a Doom knockoff because you can just go play Doom. You can play the original Doom. You can also play the latest Doom which is like 4K60 frames a second graphics on Xbox and PS5. Like there are amazing ways to experience Doom if you're into Doom. And also people don't just play Doom to go walk around a cool designed like hell world. They want to actually have a gun with an ammo counter and a health bar and an armor and a boss and a story of some sort. Even if the story is like kind of blah, like they want to be able to progress and it turns it into this, this reward cycle of okay, I'm, you know, I'm having fun.

Viable rewards.

Exactly. And so I think exploring iconic images, paintings, memes, this is going to be a Ghibli moment. We got some recorded videos from the DeepMind researchers which we're showing you right now. And you can instantly see the virality. I can't wa for this to go into general availability because I think we're going to see some really crazy ideas come out of this that no one can really predict until you get it in the hands of, of just like millions and millions of people.

And friend of the show, Logan Kilpatrick went to interview Demis on Genie 3 and Deepthink and a lot more. And the interview just went up like an hour ago. So. And we're working on getting the Genie team on the show hopefully this week.

I'm very excited for that. So the other interesting thing is like people have kind of stopped generating Ghibli's and images and ChatGPT is just like something that people use and people might generate a Ghibli for something if the meme calls for that. But I find myself generating more like turn this person into a bodybuilder, but also you have some funny idea and you go to images in ChatGPT and you say like I want to visualize this thing. What if we put a car in, in Central park and it was this color? Like AI images are clearly very useful for that.

You have continued to be a power user of make this person a Gigachad more so than, than the Ghibli's.

Yes, yes, yes, the Ghibli's. I haven't Ghibli'd in a long time, but I've certainly, I've certainly generated a lot. I mean I generate a lot of set design for like what the show should look like in the next version, a lot of ideas for, for marketing stunts and stuff like that. And so it's become this useful tool and then obviously it's baked its way into like the backgrounds of like true Hollywood production, all sorts of video production. It's just like generative imagery is just a tool both for designers to pull references and use a reference board or it's kind of like the first step in a plan to co create something actual.

It's how you speedrun to something that's mock up quality. Exactly. Not necessarily product or ready for release, but instead of, I mean I used to make. I think it's a way that you can, even if you're working with designers or people that are great at Photoshop, you can just generate a V1 that's pretty solid and say like I want to make this but make it like really good.

Yep. And so Genie 3 is close to mock up quality. Like if you, if you, if you wanted to, if you were pitching a video game or you were even pitching like a experiential world that you wanted to build or like you're building a store, a physical store for your business and you generate an image of what you think that one look what that should look like and then you want to let the client say hey, we're going to invest $5 million building out this, this, this Caffeine goo cafe in Nashville like our friend did. Like you could actually walk around it and really get an idea of what that is. Or you could totally do some of like the matterport things for. Okay, we took some photos of this house. We want you to be able to let you walk around like the tech is getting there. DeepMind is clearly scaling this. It's at 720p 24 frames a second right now, pretty close to optimal visual fidelity. They're going to get to 4K 60 frames a second soon, but there will be intractable problems that come up that I don't think you can just blindly throw scaling at, similar to what happened in LLMs. And so we're going to be running down the same path we did with language models. This is my take. So you'll use reinforcement learning to round off all the sharp corners or instead, like, kind of create spikes of spiky intelligence that create pockets of value within these generative worlds. So you can imagine if you're generating a video game, it's very important to have the main character look the same at the end of the game than at the beginning. So you might do some sort of, like, RL pass to make sure that character consistency is really important for your hero character. But really it doesn't matter if I, like, come back after 10 hours of wandering around like the video game, and like, a tree looks slightly different, like, it's fine. So. So there'll be areas where, where these companies and these, these foundation labs, like, go and apply a ton more RL to make sure that this particular feature works. But at the same time, it doesn't make sense to go and try and rebuild everything and bake everything into the model. So you're going to wind up with situations where just like ChatGPT got a web browser and a Python shell. You'll probably. These world models will need to be wrapped or interfaced with, like, a traditional database so that if you have an inventory and you want to track health points or you want to track ammo, that's not where you generate a world.

And then you make it fixed.

Yeah, yeah, yeah, yeah, yeah. Fix, fix like. Or fix like the underlying territory of the world, the underlying maps, so that certain elements, same elements, but then you're generating on top of that. Tyler, what's your take?

Yeah, I kind of disagree that, like, scaling won't fix this. If you look at the. I think if you look at Genie 2, it had this big problem with, like, consistency.

Right.

You would turn left, then you would turn right again and it would, like, look totally different. And I'm pretty sure in the paper they basically say that the consistency was just like an emergent property of scale.

Yep. So consistency over what time period, though?

I mean, over the minute. Long.

Yeah, I guess 100 hours. 100 hours. That's how long people like, like a.

Rolling, you know, 100 hours.

Like, that's how long people will play these things. And it's like they want to be able to go back to. If you're in Elder Scrolls, you want to be able to Go back to like the very first village you started in and see that the tree that you chop down is still. Look how far I've come. Exactly. Like, there are some, there are some crazy properties in, like in Elder Scrolls where it's like in the first village you accidentally killed the Blacksmith. And so 12 hours later you come back and you can't get the sword that you want. Why bother trying to bake all that into the model? Why not just use a database?

But I'm saying you don't need to think about baking it into the model. It's just going to be an emergent property as you scale.

Yes. But I do think you will wind up hitting plateaus of scaling. Just like, just like, you know, ChatGPT, like LLMs got pretty good at math where they could do two plus two and then they could do much more complex math. But at a certain point it was like, we're not. We're just going to give you a python shell so that you can just actually do the real math.

Yeah, that's fair, I guess.

Like, because even, even if, even if scale would solve like the really, really hard, intractable, huge, huge math problems. It's like, why burn the electricity to run that massive model to do some very, very basic calculation that would take like five flops on just a cpu.

Sure.

And so, like, you know, you don't need to. And I do think that there will be other, there will be other tools that it doesn't make sense to bake into the model. I mean, just think about like multiplayer elements. Like, how would that be baked in the model? Like, you'd want to sit on top of just like a normal social network. You want to sit on top, like of some sort of normal interface so that you can join the same game. Like, I don't want to recreate a new epic game store every time it launched.

Imagine like a new video game paradigm where the world is just constantly expanding and there's no real rules and characters or constantly emerging. But at the same time, imagine playing soccer and if you dribbled to the side of the pitch at one point, it just opened up and you could just keep dribbling. Eventually you'd get fairly. You'd be like, okay, I actually like having a rule set. I like having this sort of consistency in the game so that I have a framework in which I can play in. So in so many ways this will create entirely new formats, effectively. Like you're taking the concept of an open world video game and you're just opening it up infinitely. Yeah, but I think again, people will want that consistency of play and even a shared experience, right?

Yep. And I mean, so Gold Rock in the chat is saying $3,000 an hour to play at 64k 60fps. I mean that is a real consideration. The other, the other question is just, is just like, is there a way that you can pull forward adoption by like not being a deep learning purist? Even if, even if, even if deep learning is the solution over the long term while you're waiting to build the $1 trillion data center to actually hit the crazy, crazy scale. If you can get user adoption by just using a normal database to track your inventory in your health bar, you'll get more adoption and then you'll create this flywheel. It's the same thing with ChatGPT. Like you could, you could. We could have just been like OpenAI could have said, like, look, there's no reason to give it. To give an LLM access to a Python repl at a certain amount of scale. It'll be able to just reason about it and do, do the Python in its head. Right?

But, but you can pull, you can say like, actually you do want to. Like, if you use the world model as like an RL environment, then you can actually pull it forward that way by just like using it for other stuff. You know.

Maybe like, if you're using.

It, it's not a game, it's like a data set that you use for your. And then like, oh yeah, we can also release it as a game for like normal users. But really it's to train our robotics.

Really pilled on this robotics thing.

Yeah.

I think that because data is like.

The big issue, at least that's.

Yeah, yeah, yeah. Certainly for robotics. But I think that, I think that there's, I think there's a consumer product here.

I agree. But the way that you like fund it is through is using as maybe the opposite.

Maybe the robots learn from all the humans acting like robots in a game and then that, and then that generates a ton of, a ton of data.

That one hand watches the other a little bit.

Exactly, exactly. But my, but my real question is like DeepMind is obviously absolutely crushing on the research side. They're like really, really. It feels like they're really far ahead right now. This is clearly state of the art. The question is like, what does it take for Google to actually get this out in the world as a consumer product? Everyone says GPUs will melt if they launch this right now. Maybe that's true. Maybe they need to optimize it more. But the big question is like, is like they had a large language model around the same time as GPT 3.5 and they were really close in terms of their GPT 4 first Gemini launch at the similar scale. But ChatGPT launched earlier was more aggressive, broke through and became kind of the default. And I wonder if there's, if there's, if they should be more like risk on, in terms of actually productizing this and getting in people's hands. Like, clearly people are going to use this for bad things. They're going to put all sorts of references, reference images in there and be walking around them. We can imagine all the bad things that they'll do.

Well, you can imagine them being more aggressive around a product like this because it doesn't threaten their golden goose. Like, in hindsight it kind of makes sense that they weren't might threaten YouTube.

But.

Yeah, yeah, but you can also imagine where it's a catalyst for YouTube, where a bunch of there's like entirely new.

Vectors of you're streaming it, you're feeding it in, you're exporting it, that type of stuff.

And yeah, I can see this being like a net new category of content or just a tool that every creator can use to make better videos. Whereas.

No, I was saying it's basically the same thing with like they're to release VO3. It's like it's all the same safety constraints.

Right.

You can't put like some bad image into VO3 and then have it produce video.

Yes. So it's like most of it. So. Yeah, yeah. So I think it's even, even VO3, it feels like they're not being aggressive enough on the rollout of that.

That's probably also just like, it's just way too expensive to serve for Google.

The company that makes a billion dollars a day. Like, really, it can be that expensive? Like, I feel like, I feel like the ChatGPT, I mean, it says a.

Lot that you were on the 500amonth VO3 plan and you're still getting rate limit constantly.

Yeah, to the point where I've like sort of churned because I know that it's like such a rough experience to go in there and then like, okay, I need to plan it out. Oh, I can only do three queries and they'll come back the next day. It makes it really, really hard to actually do any real work.

The question is, how expensive is that for them to serve? It must be, it must be a lot.

It must Be it must be expensive a month plan. But. But what? Like what, what was the secret to ChatGPT beating Gemini? Like the Gemini app launched three months later. So moving first mover advantage clearly matters. Also Chat GPT, it seems like Sam Alton's never met a GPU he doesn't want to melt. He will, he will do anything he can just to scale, scale, scale, even if it's. I imagine the margins for the early ChatGPT launch were terrible and I imagine they were burning tons and tons of money in terms of inference costs and they were way, way upside down there. I mean ChatGPT was basically free for like months. Like, and it was like how like this clearly has, has operational costs, this has inference cost and they're just eating it. And so yeah, maybe being more, more aggressive on the financial side. And then I, and then I think.

Well I still think for Google, if they were to come out and launch a ChatGPT like product that didn't have for example like a knowledge cutoff that ChatGPT did at that time already. So there's a post here from an account that does a weekly newsletter of quotes from earnings calls and they aggregated a bunch of quotes on AI driven changes in Google search, reducing referral traffic and impacting users user engagement. So these are just different CEOs. In the last week, the one here, the portion of our traffic that comes from Google search has declined from 52 to 28%. Another one from the Groupon CEO. Google is significantly changing the behavior of search result pages. What we see is that we have declining traffic. Another one on Google. When Google provides the AI response, there's a much lower click rate on it because it's typically a user getting the answer in that response and they don't need to go further. NerdWallet CEO, the Google search is still pretty challenged. And another one, we can't deny it, it is a risk on page views. And so again all these AI features are reducing people's need to just navigate the web and drive. Some of which is ultimately, I'm sure paid traffic.

Yeah, and maybe the world model stuff, video game stuff is it's like not as threatening to their core business. I mean it clearly is not. And so maybe Google will move faster on this and own the consumer product. But I feel like in the next 12 to 18 months we are going to see some sort of explosion in consumer AI driven around world models, something built on top of this, whether it's a game or meme engine or some sort of product. And it's interesting because there are two or three serious, serious companies building in this space. We talked to Descartes who's building in this space. Fei Fei Li has World Labs, I believe that's building this space. And it feels like there's this dance from like, okay, you've done enough research. It's time to productize and you need to completely turn the ship ship of your company towards how can we get this out the door in a way that's sticky and viral but also doesn't completely put us out of business because of the inference costs. But we'll keep digging into it. We'll keep having people on the show to talk about it.

Team over at DeepMind is posting. I just put a post in the chat.

While you're pulling that up, let me tell you about graphite.dev code review for the age of AI. Graphite helps teams on GitHub ship faster, higher quality software faster. You can get started for free. Let's pull up this post from DeepMind from Vedant Misra.

He is counter posting.

Oh, I saw this. This is so good.

Sam Altman, if we can pull this up, this is the Korean TVPN is going to run into some challenges very quickly on pulling up posts.

There we go.

Here we go.

Oh wait. This is a different post than one.

Multiple posts.

If you time multiple.

Yes.

Okay. So in this one, Google is the rebel. Are the rebels. He is. What he's saying, I don't even know at this point.

It's hard to tell who's the Death Star.

Okay, I finally landed on it. On what? Sam Altman, the Steel man of the Sam Altman Death Star post. The Death Star was the was the overwhelming weight of numbered, numbered language model releases. And he was, he was feeling the crushing weight of GPT5 needing to be an order of magnitude bigger and an order of magnitude better and a massive qualitative step up and a binary change in the overall experience. And he blew that up. He blew up the expectations that numbered releases would be correlated directly to specific model capabilities. That's the closest I could get on the steel man of the of the Death Star post.

The vague Alternatively, alternatively, it was provocative.

It got the people going.

Got the people going.

It did. It did drive a lot of, a lot of speculation, a lot of frustration.

Well, over the weekend you posted there's 100 million in computer hardware locked in the replies. It's your job to get it out. So if you were on X over the weekend, Donald boat at Laserboat 999 was on a tear. If you had a viral post over the weekend, he was going to be in your replies saying, I agree, the only way we can drive culture, science and industry forward is if you overnight me a motherboard compatible with an AMD Ryzen. Blah, blah, blah, blah, blah. And I believe he's made some tremendous progress in terms of putting together this computer.

I was talking to him a lot over the weekend, tracking the story. He said he was shy, didn't want to come on the show. But I. A lot of people are questioning, like, is this some scam? Martin Scowley was wondering if it's a scam. I think he's just having fun and he's a good poster. And he is just. He's kind of begging and kind of shamelessly asking for stuff, but having fun. And it's funny. And it's basically like the tipping version of the ex creator payout. Like he's just getting extra stuff. So he has received almost everything. It's also hilarious, the type of accounts that he responds to because it's everything from like Elon and Sam Altman to Dasha from Red Scare. And like, he's all over the place. He's not, he's not purely just going after like the richest people in tax. Yes. He has a specific taste in poster. And I mean, when he started this project, he already had 100,000 followers. Was prolific poster. But he is apparently going to have some sort of MIT physicist put the PC together. He says he doesn't know how to code and he's not using it to like build software. He's more just interested in playing Battlefield 6, which had a beta this weekend that apparently did very well and apparently blew out the Call of Duty beta. I guess. So there's this big debate over did Microsoft overpay for Activision because they acquired Activision for something like $70 billion. They got call of Duty and you would expect the Call of Duty would continue to defeat Battlefield, which is an EA property.

Blizzard boss Mike Ybarra says Battlefield six to boot. Stomp Call of Duty Black Ops seven.

Yep. Rough, rough. So, but you know, it's. It's fun. You know, he's 21, he's having fun on the timeline, getting an incredible PC that will be fully Battlefield 6 ready. I was in a similar boat when the first Battlefield game came out. Battlefield 1942. It's probably like 2003 or something. I had to beg my mom for a cutting edge Nvidia graphics card because I couldn't run it on my current PC I had to build. I had to reply said mom, you, me, Amalfi Coast.

I was so Apple pilled from a young age. I had to. I refed hundreds of soccer games to buy A maxed out MacBook Pro, which is still like terrible.

Yeah, it's not a gaming gaming rig.

But I would use that to play.

Yeah, it was always PC Master Race. And the last PC I built, the last bastion of the Windows ecosystem in my life is sitting right over there. I built that not for gaming.

I never even noticed that.

I built that PC for rendering computer graphics using. Oh my God. Tyler's picking it up. He's going to show off the thing. Donald Boat, if you want an older PC with a. What does it have? It has two Nvidia 1080 TIS in there. It's not bad. It has Ryzen Threadripper. I think it has 64 threads actually. It might even have. It's probably not as fast, but it's pretty solid CPU in there. That's a Rig 2. It's a Rig. I built that from scratch and I was rendering with scraps in a cave. With scraps in a cave. I was rendering computer graphics on Cinema4D. Did a little Houdini on there. You let it cook. It was a fun time. And then I also played some virtual reality on there. Had a good VR rig set up every once in a while. It was fun.

This is like again, this is like Hunter Pydon reminiscing on smoking crack.

Yeah, never get John's crack. I mean dollar Boat was like totally getting me back into it. I was like, maybe I should.

Maybe I should build a new rig.

Maybe I should build a new rig and play Battlefield 6. This seems like really good in the two minutes I have for free time this week. Absolutely not. Really quickly. Polymarket has which company has the best AI model at the end of 2025 and Google is still on a tear. The there was a major switch after GPT5 dropped. Google sitting at 75% for August 30th and way up there for September 30th and. And even by the end of the year, Google is slated at 54%. So DeepMind is in. Is in control of the Rebel Alliance. Maybe they are the Death Star because.

They all eyes on.

They cannot be definitely Gemini 3. Although of course this is benchmark based. So anyway, Donald Boat, he's having fun. People are starting to say, oh, maybe it's AstroTurf. There's thousands of likes on every post. They think it may be fake. I Think it's organic. I think the kid is just out there hustling, having some fun.

It's hilarious.

It is hilarious. And it's bringing me a lot of joy. And I think that like the. No one is getting over their skis.

It won't work again.

No.

The next person that tries to do this.

Yep, yep.

It's gonna be like Donald Boat knockoff.

Yep. It's. It's a one time. Funny, funny like meme. That. That, that does very well. And I don't think anyone here. There's no promise of financial reward. And most people are like Sam Altman can afford a graphics card. Will Miniatis can afford an AMD Ryzen 7, 8 core desktop processor or, or a motherboard, you know, And I don't think, I think everyone was kind of chipping in what was reasonable to that. Where even if it's a scam, if it takes you for 200 bucks, it's not that big of a deal. It's. It's a lot of fun.

It's not a scam in my view. It's entertainment.

I think it's entertainment and I think I did. I was trying to push him. Like, hey, you're gonna have some doubters. You gotta go live. You gotta stream Battlefield 6. When you get this thing set up, you don't need a face cam. Although he was asking for a blue Yeti microphone and that was suspicious because he was like, I'm really shy. I'm like, if you're shy, what do you want the microphone for? What's going on here? But overall, I think he's gonna build the PC. I hope he streams Battlefield. I hope he's good. Could be a lot of fun. Anyway, let me tell you about Julius. What analysis do you want to run? Chat with your data and get expert level insights. In seconds, you can ask Julius to analyze your data. It's loved by over 2 million users and trusted by individuals at Princeton, BCG, Zapier and TVPN.

We, you guys already know we love.

The CEO of Rahul Sonwalker.

Rahul.

He's been on the show three or four times. He's been in person on the Friday show.

That's right. But absolute legend. Go try out Julius. You're gonna love it. Whether you're working in marketing, finance, product, you can use it to understand your business better.

I was talking to a VC over the weekend who was lamenting missing Julius because he understood who Raoul was. But Raoul was working on a previous company and Julius was a pivot out of that apparently. And he was like, I should have just done it on the basis of who Raul was and knew that he would figure something out. But we didn't like the particular. That particular business, that particular market.

And great founders are just one pivot away.

Exactly. And so, yeah, when you find someone, you got to let them cook and let them figure it out.

Yeah. Well, you know who else is having fun cooking on the timeline?

He's monitoring the situation professionally now.

Former OpenAI researcher Leopold Oschen Brenner. His fund Situational awareness is up 47% in the first six months of the year.

Love to see it.

You may have first become aware of the podcast he did with Dwarkesh and the essay that he wrote alongside it. Situational Awareness is probably my favorite name for any type of fund in a long time. It just says so much. It's like, hey, we're aware of the situation, we're monitoring it and we're deploying. We're trading against the situation.

I like it. It's definitely up there. It's in the conversation. But the bar is extremely high. You're going up against Cerberus, you're going up against Ares, the God of war. The ceiling for quality of fund name is so high when you're in the public markets. Wouldn't say this about VC funds because they all, they all have very similar Blue Owl. Blue Owl is a great name for a fund, a private credit fund, but I mean, Citadel, It's a very strong name. You know, Bridgewater, Blackstone, Blackrock.

I love it because it's unique.

It is unique. It is unique.

It aligns with this very viral thesis. It says. It says exactly what you're getting into.

Yep. And I mean, and it was, it was incredibly accurate. Like when he, when he posted the original situational awareness essay, it was something that was consensus in San Francisco and contrarian everywhere else. And so what I loved about this was that there were so many people that were seeing what was going on in San Francisco in AI and they wound up starting seed funds. And he was like, no, the real alpha is in the public markets in these like long tail trillions of dollars of market cap. Exactly.

Going to fluctuate based on AI adoption.

Exactly, exactly. So. So his portfolio is. Is very interesting. It's all deeper in the supply chain. Notably Nvidia is missing. Maybe that was too consensus, but he's been long a bunch of things. Now there was a little bit of confusion because in the 13F, which was.

Only up 32% year to date. John, it's barely in the conversation. Trying to put up 47% on a billion plus.

You got it. Yeah, you got to go. Would have been a laggard in the fund. Wow. It's crazy. But yeah. So the 13F came out, which is the fund filing showing the position of the hedge fund, showing what they're in. There was a little bit of confusion because intel was ranked at the top, but that was because it was a call. And when you buy a call, the total value that you, of the, of the stock that you have the right to purchase is listed. And so if you have the right to purchase a billion dollars in intel stock, that will be listed as a billion dollars, but in fact you will not have actually invested a billion dollars in that call. And depending on where, how far out of the money that is, that can be very, very cheap.

Yeah. And people were questioning like Hedge, did this fund already blow up? Because intel is obviously.

No, no. They might have had to write down those calls or get out of the position. There's a million things that could do. But there's, there's a bunch of interesting things in here. And so front page of the Wall Street Journal for Leopold. You love to see it. At age 23, he emerged last year as a precocious artificial intelligence influencer, which I think sells him way, way short. But he did publish a widely read manifesto and then he also did a Dwarkesh Patel episode which I think a lot of people saw clips from.

If you're an AI researcher now and you do a podcast and it goes a little viral, congrats, you're now an influencer. Influencer.

But the, the influence of that, of that episode is crazy. So he, so he did Dwarkash. Most people do dwarakash for like an hour, two hours. His episode is over four hours long. It's so, so long. The PDF is 170 pages. It's a multi part essay series. It's a fantastic, it's a fantastic episode. It's a fantastic PDF. It's a billion dollar PDF, as they say, the PDF that spawned $1 billion in capital. The 23 year old with no professional investing experience quickly raised more money for a hedge fund than most pedigreed portfolio managers when they strike out on their own. As valuations of Nvidia, OpenAI and other artificial intelligence companies continue to soar, so do investments in hedge funds. Hoping to ride the AI wave. Ashen Brenner San Francisco based firm Situational AW now manages more than 1.5 billion people familiar with the Matter said he has described the firm as a brain trust on AI and he certainly seems to be in the thick of it, cozying up to Dwarkesh Patel. He's got Petra, he's got Nat Friedman in. He's got a, he's got a really stacked, stacked roster that. Of people that know what's going on at the cutting edge of AI. Where the plateaus will be, where the investments will be made, the distribution of, where different, where different data center investments will be made, where the different bottlenecks will be. That's really, really key to the strategy.

Yeah, it shouldn't, it shouldn't be a surprise that he was able to accumulate this much capital so quickly because you know, again, if you're, if you're Daniel Gross or Nat Friedman and you want to allocate to a manager that deeply understands AI, you're, you're probably better off finding a, you know. Yeah, incredibly intelligent 23 year old that was at a lab that can then deploy capital for you against all these different trends.

It's basically like 1.5 billion. What is that? Like one researcher's lifetime earnings. So he just needs one.

Just one.

Just one.

Duncan is a core LP in the fund as well.

Still got it, baby. Course he is, Duncan.

Of course he does.

Absolute picker.

Cover of Colossus.

Cover of Colossus. His strategy involves betting on global stocks that.

Let's give it up for global stocks.

That stand to benefit from the development of AI technologies such as semiconductor infrastructure and power companies, along with investments in a few startups, including Anthropic. Oh, he's in Anthropic.

There we go.

He told investors he plans to offset those with smaller short bets on industries that could get left behind. Situal situational awareness gained 47% after fees in the first half of the year. Some of the people said in the same period, The S&P 500 gave in about 6% including dividends, while an index of hedge funds compiled by research firm Pivotal path rose about 7%. And so he is crushing the competition. We'll get into some of the pushback here, but I am very excited for this. Aschenbrenner, a native of Germany, briefly worked as a researcher at OpenAI before being pushed to out. It was very interesting. He called OpenAI's security measures egregiously insufficient. And then he shared a document with external researchers about how OpenAI was doing security and that violated company policy. So he was pushed out. I don't know how big of a deal that really is. I don't know, like you should be able to talk to other people. Obviously you shouldn't like violate company policy. Kind of weird to like talk trash about your own company's security things. And then also we're in this weird phase where like, okay, like you got a habeas corpus when you talk about the security stuff. Because like if you're, if you were saying, if you were running around the building being like, we're all going to get turned into paperclips, we're all going to get turned to paperclips and then it's three years later and like nothing's happened. It's kind of like, come on man, like what were you doing? But if he was sounding the alarm about geopolitics, which is something he did in situational awareness or you know, AI psychosis, like there are a lot of things that are good to research in safety. And so his overall take was he, I think he benchmarked the entire industry and found that you know, like less than 1% of AI headcount was going towards like safety and the alignment problem. And so he identified that as a problem to kind of go after. But now, you know, he's, he's more focused on the public markets. So. Aschenbrenner, native of Germany, worked at OpenAI. He named situational awareness after his 165 page essay he wrote about the promise and risks of artificial superintelligence. He recruited Carl Schulman and another, another AI intellectual. He used to work at Peter Thiel's Macro hedge fund as director of research. Great pickup. The firm's backers include Patrick and John Collison, the billionaire brothers who founded payments company Stripe, as well as Daniel Gross and Nat Friedman, whom Mark Zuckerberg recently recruited to help run Meta platforms AI efforts. Graham Duncan, a well known investor who organizes the story Sohn conference, is an advisor.

It's really a who's who it is.

We're quote, we are going to, we're going to have way more situational awareness than any of the people who manage money in New York.

Wait for, wait for this line. He says we're definitely going to do great on investing.

I love it. Confidence, baby. In another sign of the demand for Auschwitgen Brenner services, many investors agreed to lock up their money with him for years. Other recent launches include an AI focused hedge fund from Value Aligned Research Advisors, a Princeton, New Jersey based investment firm founded by former quants Ben Hoskin and David Field.

Ever really a former quant.

Once a.

Quant always a quant.

That fund has a billion and VAR also manages about 2 billion in other AI focused investment strategies. VAR's investors have included the philanthropic foundation of co founder of Facebook Dustin Moskovitz, according to regulatory findings. Veteran hedge fund firms are entering the fray too. Last year Steve Cohen tapped one of his portfolio managers at Point72 Asset Management Eric Sanchez to start an AI focused hedge fund that Cohen planned to stake with $150 million of his own money. Assets of the fund called Turion after AI thought after AI theorist Alan Turing, who's on the madness list, baby, we'll see if he can hold on in the next revision which is coming soon. Turian a lot of people is up about 11% this year and gained 7% last month. So a little bit of a slower start for Turion compared to situational awareness that seems to be absolutely swinging for the fences. It is no surprise that thematic funds are springing up to capitalize on the AI frenzy. In years past, hedge funds specialized in the transition to clean energy and investing in with an environmental, social and corporate governance lens proliferated in response to client demand. Identifying a winning theme isn't the same as trading it well. Investors taste can be fickle. Many prominent ESG hedge funds have either shrunk or gone out of business.

But was I don't know, it's hard to say that smart money was saying that ESG was going to be such a winning theme.

Yeah, it was odd because it wasn't some underlying technological shift. It was more just like values based investing. It was values based investing, but it was also predicated on more demand because everyone there it was like a game of musical chairs a little bit. It was like well I think that Harvard is going to want an ESG strategy so I'm going to build one. And Harvard says well I'm going to want one because Calpers is going to want one.

And there was a time that I think consumers specifically were demanding sustainability from brands. It was really like a feature to a product, but I think that is changed some degree for sure. The market swoon. That followed the January release of an advanced low cost language model from Chinese company Deep Seek showed the fragility of the valuations of AI winners. Though the market has roared back since.

I don't go in long Jevons Paradox Come on, come on, come on.

AI focused investors argue the long term trend of development and adoption are inevitable, even if there are bumps along the way. With so many publicly traded companies that operate in the AI adjacent economy today Stock picking funds often pile into the same positions as one another and more generalist hedge funds. Vistra, a power producer that supplies the juice to AI data centers, was a top three U.S. position of both situational awareness and VAR investors as of March 31, according to each of their most recent securities filings. Other hedge fund managers are debuting funds to make investments in privately held AI companies and startups. Gavin Baker's Atreides Management.

Come on, that's also in the top 10 best names ever. Atreides Atreides is fantastic. It's from Dune.

Paul Atreides Atreides Management teamed up with Valor Equity Partners to launch a venture capital fund earlier this year that has raised millions from investors. Millions is such a funny way to describe a fund.

I'm sorry.

I would hope it's in the millions. Otherwise, what are you doing? 10 dozens of dollars from Oman's sovereign wealth fund. Each firm separately invested. Xai at least one portfolio manager is planning an AI hedge fund as a comeback vehicle. Sean Ma wound down his Hong Kong based firm Snow Lake Capital. Sounds cool. I mean, anytime you add smash capital to the end of always an improvement. A word or a couple words, it always elevates everything involved.

Yes.

After it agreed to pay out 2.8 million to settle securities and Exchange Commission charges last year that the firm participated in stock offerings of companies that it had also bet against, Ma took over an investment firm called M37 Management in Menlo park earlier this year. He's currently fundraising for a hedge fund focused on AI software and hardware. So lot of action.

Did you know that Meta is calling their AI lab TBD Lab or something like this? You see this?

Since when?

I don't know. I thought it was msl. But the journal saying Meta's TBD Lab tackles new version of llama. I think this is some sort of like sub lab focused on open source or llama. Specifically at the forefront of the push to build computer a computer mind smarter than any humans is a team dubbed TBD Lab, which houses many of the researchers the company has lured from rival labs, in some cases with pay packages of tens or hundreds of millions of dollars. TBD Lab, as in to be determined, is spearheading work for that on the newest version of llama, the company's large language model, according to people from familiar with the matter. I don't know. Just interesting. That feels like some weird leak that like the journal's like maybe misinterpreting. Anyway, back to Leopold. Interesting, interesting strategy. I think a lot of people are kind of so young Macro has an interesting take here. Leopold Aschenbrenner's fund has outperformed basically every mainstream hedge fund year to date and he's running a billion dollars of capital. The Gulf billionaires and pension funds are watching this. Capital management will soon become an activity exclusively done by chronically online zoomers. And there's certainly a lot of value in having.

Online zoomers.

Yeah. Let's give it up for profound. Get your brand mentioned on ChatGPT. Reach millions of consumers who are using AI to discover new products and brands. You can get a demo.

So funny. I'm looking at the Comments on, on WSJ.com giving a 23 year old one and a half billion to invest is pretty solid proof of a market top.

How old was Ken Griffin when he started Citadel again?

Wasn't he like 18?

Oh yeah, he was at Harvard. He was an undergrad. He hadn't even graduated yet. That's right. Interesting.

Just say you never studied the greats.

Just say you've never studied the greats. Yeah, I mean like, I don't know, like of course there is a risk that there's, that there's some bubble and they're risk.

Of course we're in a bubble. That doesn't mean you shouldn't try to make money on it.

Exactly. But, but, but there is a risk that you get caught over your skis the wrong moment. There's a pullback and the fund draws down. Really crazy. But again, Citadel drew down like 70% in the housing crisis.

All the greats have suffered a massive drawdown.

It happens.

Look at Bezos.

I feel like Leopold is, this is what he's meant to do and, and I'm rooting for him. I'm excited for it. I think it's, I think it's very, very cool. Anyway, this will be, this will be, will this be good news for Leopold? He's not in Nvidia. You know, in some ways you could say he is net short Nvidia because he is long. Everything but Nvidia at this point. Is this good news? Is he cheering? Is he sad? The news is that breaking from the Financial Times, Nvidia has agreed to share 15% of revenues from H20 chip sales in China with Uncle Sam. According to export control experts, no US company has ever agreed to pay a portion of their revenues to, to obtain export licenses. We were talking about this this morning at the gym.

Sam eats first and I was like.

I was like, certainly like this has to happen like all the time. I, I, I chat you and it's like, actually it's unconstitutional. You can't do this at all. And so I'm sure some folks worried about Trump's overreach and abuse of the federal system will be up in arms over this particular so we're going to.

Have Peter Harold on the show in.

20 minutes and then we're going to talk to Aaron Giddon about it too.

January 2021-2022 Harold served at the US White House as a Senior Director for International Excellence Economics, jointly appointed to the National Security Council and the National Economic Council. He CO led President Biden's EO 14017 supply chain resilience agenda, worked on the global digital 5G and telecommunications strategy, spearheaded negotiations with the EU on data data privacy, and served as a White House representative on a number of related matters. So excited to get his takes, he shared a bunch on LinkedIn yesterday evening or this morning. So we'll kind of wait to get into that.

The wording in this Journal article is so wild because it's not saying like the IRS or something, it's saying the Trump administration will receive 15% of the sales as part of a deal to approve exports of Nvidia's H20 chip. So it's like where exactly is the money going? Like, I guess it goes into the treasury, the Commerce Department. But it's just funny to phrase it as like Trump gets to control, but hopefully it goes down towards paying down the debt. Like I guess that's the bone.

GPT says 15%. That's all off the top. You don't got to worry about nothing else. No problems, no supply chain disruptions.

So anyways, exports of the H20 and Mi 308 were halted in April. So this actually applies to AMD as well, which also has a similar like nerfed GPU for China.

And Trump had a quote back in April.

They yeah, they, they, they, they halted the trade of these H20s. Jensen Huang has gone on a charm offensive since then, says the Wall Street Journal, talking to officials in both countries about the need to do business with each other. We follow rules the US Government sets for our participation in worldwide markets, Nvidia said in a statement on Sunday. While we haven't shipped H20 to China for months, we hope export control rules will let America compete in China and worldwide. This is an interesting one. I mean there are good arguments on both sides. You know, if you're super, super intelligence pilled, if you're AGI pilled, you probably don't want that in the hands of a rival country. But if this is more like Microsoft Excel, why not go sell some seats over there? Why not go sell some licenses? I mean, we sell airplanes over there. We sell 747s.

We don't sell F35s.

Yeah, but is this an F35 or is it.

That's a big question.

Or is it souped up xl? You know, and that's the debate. And I think a lot of people are shifting over into the camp of like, this is knowledge retrieval. This is like pretty generic hardware. This is not, this is not as, as a crit, as critical of infrastructure.

Trump saw the GP GPT5 launch. Okay, we'll give him some chips.

I mean, why were we saying that he didn't care about the deficit because he was super AGI pilled? Why is he now focused on the deficit and not worried about competition with China? Because he's less AGI pilled now. So he says, you know what? Like, I don't think God in a box is coming down the pipe anytime soon. And so let him have him. Whatever, it's fine.

Trump on Monday left open the possibility that Nevette Nvidia could export its Blackwell chips for a higher price. The Trump administration had closed the door on the export of that technology to China, even after reversing course on the H20. On Monday, Trump said that he'd consider allowing it. The Blackwell is super duper advanced. I wouldn't make a deal with that, although it's possible, Trump said, make a deal somewhat enhanced in a negative way. Blackwell, in other words, take 30 to 50% off of it. But that's the latest and greatest in the world. Nobody has it. They won't have it for five years.

Okay, Wildcard. Nvidia's got to track this.

And he said again, Trump said Jensen will return to the White House in the future to discuss selling an unenhanced version of Blackwell. I think he's coming to see me again about that. But that will be an unenhanced version of the big one. You know, we will sometimes sell fighter jets to a country and we'll give them 20% less than we have.

Okay, see? See, we do sell fighter jets to come to countries. Let's go. Raghav says it should go into the sovereign wealth fund, which he can eventually help pay down the debt. Yeah, well, that makes sense.

A lot of people have been venmoing the government.

The. So Nvidia is going to need to track this. It's very similar to sales Tax they need to get on numeral sales tax on autopilot, spend less than five minutes per month on sales tax compliance. Maybe this is a line expansion for numeral Track your export. Your export bill. In other chip news, Tom Cotton is diving back into the Lip Bhutan news at Intel. The CEO, of course, Liputin, had to visit the White House and has been singled out by Trump as unacceptable. And Trump called for his resignation. We kind of debated that on the Friday show, but Tom Cotton, the I believe, Republican senator, he said the new CEO of Intel reportedly has deep ties to the Chinese Communists. U.S. companies who receive government grants should be responsible stewards of taxpayer dollars and adhere to strict security regulations. The board of intel owes Congress an explanation. And so Lipoutin is set to visit the White House today. He probably is there currently or was there earlier today after President Trump called for his removal last week over ties to Chinese businesses. This is. Lubouton is not. He was born in Singapore and he's an American citizen, but he had a venture fund that did a bunch of deals in China. And so that's kind of, I think, the main focus. And when you look at the web of like, all the deals that he's done and all the company, it's like a lot of companies. And so it looks like this crazy spider web. It looks very like red, stringy. So Thanh is expected to have a wide range. Yeah, we gotta get the whiteboard out. Break it down. Get Lipp Bhutan on the show.

Tyler, could you whiteboard out Lip Bhutan's portfolio?

You're going to have to. You're going to have to look it up. There is a. There is a crazy chart over here somewhere. But Thanh hopes to win the win Trump's approval by showing his commitment to the country and pledging the importance of keeping Intel's manufacturing capabilities as a national security issue. Trump has aggressively pushed US companies to make changes raging ranging from eating the cost of tariffs to doing business with more conservatives and other political allies. So the, you know, Trump said the CEO of Intel is highly conflicted and must resign immediately. There's no other solution to this problem. The odd thing is that it's like, okay, so, so like intel with, like, they're really lagging. Semiconductor technology is now critical to America, but Nvidia H20s aren't and we can let those go to China. So there does seem to be some, like, incongruence here. Right. Have you sorted that out at all.

The conversations we had last week? It feels like intel is still incredibly strategic. And important. And we need to take it very, very seriously. We need to make sure that they can like the United States. It's the foundry business in particular. If you didn't hear our interview with Fabricated Knowledge Doug over at Semianalysis on Friday, go listen to it. But yeah, I think universally people agree that intel is important. It's in a bad situation. But so the question then becomes, is Lipp Bhutan the right guy to steward it? And ultimately it's the intel board that should be determining that. Right? Yeah, Doug did say he had issues with the. The board wasn't exactly excited about their leadership over the last decade and a half.

Yeah.

But again, this, this kind of thing typically feels like.

Feels like if you were trying to keep semiconductors out of the hands of China, you would be anti age 20 sales and anti lip Bhutan in. In intel potentially. And if you had more laissez faire attitude, you would say, yeah, lip Baton is fine and H20 sales are fine. But it's weird to be like age 20 sales are fine but Lip Bhutan is not fine. That's where there's like some incongruity. But I think this will all be resolved. It's probably just one of these Trump like gambits where art of the deal, he's going to ask for intel to, you know, donate to some sort of thing or, you know, pay something.

I think they should donate to themselves.

Donate to the shareholders with a buyback maybe. I don't know.

Just figure it out.

Donate the. Donate the fab business to tsmc.

No, I think, I mean, I think that's the whole point. It's like we can, TSMC can be producing chips in the United States, but we still shouldn't let a company like intel just fail at a point where it should be on the path to thriving. Mike Moritz was in the Journal over the weekend.

He wasn't in the Journal. He was in the pink sheets in the, the. In the Financial Times opinion section. Intel must not bow to Trump's attack on its CEO. The latest character that US President Donald Trump has attempted to assassinate is Libouton, recently appointed chief executive of Intel. Firing from his perch on Truth Social, the president wrote, the CEO of Intel is highly conflicted and must resign immediately. For context here, Mike Moritz is Sequoia, not founder, but longtime gp. The Financial Times describes him as a Silicon Valley investor. Quite do the job. It's technically true, but Sequoia, I believe, invested in intel and has a long relationship with the company. And Mike Moritz is, I believe, anti Trump but Then there were a few other folks at Sequoia who were pro Trump in this last cycle. So there's a little bit of a divide there. But I think Mike Moritz is out of Sequoia at this point or has maybe moved on to a different piece of the organization. It's all very complicated. I can't keep it perfectly in order. But says while presidents have sometimes expressed their outrage about the behavior of the corporate sector, Trump's assault has no modern precedent. The source of the President's ire and that of Tom Cotton, Republican chair of the Senate Intelligence Committee is is Tan's business dealings with China as both venture capitalist and in his prior role as chief executive of California based Cadence Design Systems, a maker of electronic design systems. The President and Senator pointed to a $140 million payment agreed by Cadence last month as part of a plea deal to settle claims of illegal exports to China between 2015 and 2021. They have also lambasted Tan's venture investments in a number of Chinese semiconductor companies. Companies. In addition, they are seeking to question whether with ton. With Tan at the helm, intel is jeopardizing the 8 billion dollar grant it was given under the Chips and Science act that was passed under. Passed by the Biden administration. The attack on Tan comes at a time when the President's policy towards China and the Chinese tech sector seems to change by the hour. Of course, it's much easier to smear a man with accusations than examine his record and character. Tan, 65, is a longtime American citizen and much admired figure in Silicon Valley who is known for operating with a soft spoken, authoritative dignity. I've known him as an acquaintance for almost 40 years.

The whiteboard is going up.

Oh, there we go. Oh, he's connected to smic, too.

Yeah, so he was. His. His firm was called Walden International.

That's right.

And they were. Yeah, they were early in smic. Seed investor.

No way.

I'm looking for more smic.

For reference, is the. Are they the Nvidia of China or are they the ts? I think they're the TSMC of China. And then SME is the ASML of China.

Yeah.

And then Semiconductor Manufacturing International Corporation.

So they actually do the fab. They are a fab.

And it is partially state owned.

Yes.

Like many companies.

Interesting. Okay.

And then I don't know if it's AMEC or amec. That's like a tool etching tool company for semiconductors.

Yeah. So more. So more of an ASML type competitor.

Yeah, yeah.

Actually make the lithography machines that then are used in the fab to make the chips that are designed by Huawei or any other client. Interesting. I didn't realize that. Mike Moritz and Lipp bhutan go back 40 years. Wow. They met when Tan was 25 years old. What a legend.

Did you know Intel's first external investor was the meme account Arthur. Arthur Rock?

Oh, yes, it was Arthur Rock. That's right.

Of course, not the current account Arthur Rock, but the original Arthur Rock. The legendary Arthur Rock.

Yeah.

Who? His firm was Arthur. Just Arthur. Arthur Rock and Company.

Yes. And you know what the Rock stands for, right? Rockefeller. It's Rockefeller Metal.

Wow.

Yeah.

I did not know that.

Yeah, the Rockefeller money got passed down.

This goes way deeper.

This goes deeper than we thought.

A tinfoil hat.

The Rockefellers are involved. Born in Malaysia and educated in Singapore, Tan studied at mit, where he was awarded a master's degree in nuclear engineering. After moving to San Francisco, he joined the Walden Group, one of California's early venture firms. Instead of plowing the Silicon Valley furrow, he elected to head to Taiwan. Later, he made investments in China, the majority in companies with deep technology roots. His deep understanding of both Taiwan and China in is an asset many US Chief executives would like to possess. When he later became CEO of Cadence, which had fallen on hard times, the wise acres assumed he would fail given his lack of operating experience. Instead, he transformed the business. In his 12 years at the helm of the company's revenue tripled to almost 3 billion. And the stock price appreciated by over 3200%. Today, cadence is worth close to $100 billion. Wow. That is a phenomenal performance.

Earlier again, Lip Bhutan made a seed investment in SMIC in 2001.

Wow.

Exited it fully in 2021.

Yeah. And I mean, you think about the. The vibe around doing business in China in 2001 was like, absolutely, absolutely. These are. These are our boys. We make shoes and Nikes there. We make iPhones there. Like, it was, like, completely reasonable. There was no great power competition discussion whatsoever. All that has cropped up in the last decade. And so earlier this year, Lip Bhutan became the chief executive of Intel, a job which in Mike Morris's book amounts to public service. It is no secret that over the past 20 years, intel has ceded its mantle as the world's most important semiconductor company to companies like Nvidia, Broadcom, tsmc, and Samsung. Six different leaders have failed to return intel to the glory days that it enjoyed under the late Andy Grove. It failed to capitalize on the rise of the mobile phone industry and was outgunned by Nvidia as the supplier of silicon engines for the AI revolution. Its manufacturing prowess, a rarity among US chip companies, has been consigned to the shade by tsmc. Oh, Mike Moritz, what a great writer. Today intel is also ran lacking much in the way of financial firepower. The last thing tan needs is to be distracted by a vindictive political sideshow. Now the intel board must decide whether to march to the beat of so many other corporate leaders and capitulate to the president's artless bullying, or to set an example for other companies and display some backbone. Early signs of defiance are encouraging. Let's hope that Intel's overseers abide by the refrain that old Tammy that that old Tanny Tammy Wynette song and continue to stand by their man. What a hilarious reference. Do you know that song Stand by My man by Tammy Wynette? I do not. Famously quoted by Hillary Clinton during the Monica Lewinsky scandal where she said that she was not going to leave Bill Clinton over the scandal, but she was also not doing what Tammy Wynette said. She was not merely standing by her man, but she wound up she did stand by her man in some ways. And they did not get a divorce. But fascinating quote, fascinating article. I think like he can probably riz this up. I think tan can go to the White House, do some deal. Art of the deal both ways. Figure out some way one hand washes.

The other and be fine for tan. Yeah, obviously. Get a cool leather jacket.

Cool leather jacket. Dust off those golf clubs for sure. Bring a Big Mac for sure. Pick up some McDonald's on the way.

Smart.

And throw in a Vacheron historique.

Maybe bring a bar of gold or, or a lot, even even potentially numerous bars.

Yeah, definitely, definitely bring the biggest chip that intel has ever produced. Maybe just a whole wafer. Because he's probably seen Blackwell. You've seen bout Blackwell is pretty big, right? You gotta bring something even bigger and just be like, we're making the biggest and best chips in America. We're American made. I'm America all the way. We're gonna build a fab at Mar a Lago. We're gonna build it in Key west. And I think he gets it done. Anyway. Should we move on to linear? Linear is a purpose built tool for planning and building products. Meet the system for modern software development. Streamline issues, projects and product roadmaps.

Go to link when we get on the show, we will be giving him a live demo of Linear.

We should. We should.

Linear will be the key to getting intel back on track. Let's make it happen.

Okay. Tyler cowen, he's loving GPT5 and he has a post he says somewhat unrelated but somewhat related, of course, on the topic of AI, he says the quote, AI is causing a slowdown for new hires hypothesis has pretty much been abandoned. The trend seems to predate GPT4, and the trend seems to apply to tech jobs that have no serious AI competition. Interesting. I was digging into the hiring numbers across the world because if, if, if AI is causing a slowdown in new grad hires, I would have to assume that it's also causing a slowdown in India because there's so much business process, BPO business process, outsourcing and outsourced IT work, outsourced software development work that happens in India. And I dug up the unemployment data in India, split it out by rural and urban, and it doesn't seem like there's a huge spike. Maybe that's coming, but we're certainly not there yet.

I think Tyler's post was in response. There was a New York Times article Yesterday said goodbye $165,000 tech job student coders seek work at Chipotle as companies like Amazon and Microsoft layoff workers and embrace AI coding tools, Computer science graduates say they're struggling to land tech jobs. And there's a story of Manasseh Mishra, remember seeing tech executives on social media urging students to study computer programming. The rhetoric was if you just learn to code, work hard, and get a computer science degree, you can get six figures for your starting salary. Is now 21, recalls hearing as she grew up in San Ramon, California. Those golden industry promises help spur Ms. Mistra to code her first website in elementary school, take advanced computing in high school, and major in computer science in college. But after a year of hunting for tech jobs and internships, Ms. Mishra graduated from Purdue University in May without an offer. I just graduated with a computer science degree and the only company that has called me for an interview is Chipotle. I guess my first question is, was this for coding? Coding creating software for Chipotle? Because having a burrito tracker on par with the Domino's Pizza tracker could be pretty cool.

It could be very good.

And again, I think one thing that could happen is hiring managers and executives telling themselves we should be able to do more with AI. We don't need to hire for these people. Even if AI is not at the point where it truly replaces a person, it's just sort of forcing people into this mindset of like do we actually need to hire this person?

Yep.

And again, we'd called out before. Microsoft has laid off more people this year than they have in the past few years combined. And so again, I think.

But at the same time, don't know how big the hiring glut was from Zirp and Covid. Like that could still be working its way through because companies really, really staffed up and they staffed up remotely and they staffed up young people.

Yeah, we've heard a bunch of stories this year of companies that had people.

Had eight different jobs, remote engineers, that.

They were kind of like, you can move back to the office or you.

Can move on at the same time. Raghav Meta in the chat says it's really bad in India. I have it on very good authority. I believe you. It's. It's just not showing up in the data yet. I'm interested to track it more deeply and actually understand what's going on at some of the big. At the big IT outsourcing firms in India. Are they actually scaling down? Do we have hard data on it? And then. Yeah. What is driving it? Is it actually AI or is it something else that's going on?

But regardless, at the same time, Salesforce is saying AI is now doing everything at Salesforce and yet head count is.

We heard the same thing from Klarna and it was unclear if that was just like a right sizing of the business for where they were in terms of scale. But I don't know. It's interesting. AI is everywhere at the same time. There's a lot of these tools are just extra levers and the dynamic is that, yes, it helps you write more code, but there was always way more code to write than your business was able to write. And so you're really just unlocking new capabilities, capability and just moving faster. Totally. Well, if you need an AI native CRM, get on. Adio Customer relationship Magic. Adio is the AI native CRM that builds scales and grows your company to the next level.

You can start for free. Well, on that note, we have our next Peter. Play that in the restream waiting room. Let's.

There we go. Welcome to the stream. Peter, how are you doing? You look fantastic.

Great. I was told I should wear something, you know, not just a black suit for you guys. So there we go.

Fantastic.

Looking good.

How are you doing? I'm great.

I mean, you know, the east coast, it's been a little cooler than it was a couple of weeks ago. It's the 90s are coming back, but not Yet. So yeah, no complaints. And you know, I am for better actually really for worse, not an Nvidia shareholder so I don't have to directly deal with their 15 paid to the government.

Did, did the stock pop today? What actually happened with.

It's. It's a normal day, it's up like 1%.

Okay. Nothing crazy, not a huge move.

Before we dive into all the news, would love to just give us the highlights from the bio. You've done quite a lot, but I think it'd be helpful context before we get into discussion.

Yeah, no, look, it's great to be with you guys. Thanks for having me on. I currently am a non resident fellow at the Carnegie Endowment for International Peace which is a think tank in dc. I also am an attorney and have a trade law practice sort of focused on trade and sanctions and export controls kinds of issues. I was in the Biden administration at the White House where I dealt with the global economics file, including a bunch of the export control issues and been in private practice prior to that was the State Department for a while and when I was much younger was a reporter covering campaigns and elections like many, many years ago.

That's super cool.

It's kind of been around.

Well, breakdown. So before we get into the Nvidia AMD stuff and the news from yesterday, take us through kind of your experience of this year so far because there's been a lot of different stories.

Go back further. Chris Miller writes chip War. Everyone wakes up. Is that what happened? Like when, when did D.C. wake up to. Maybe we need to think more critically about the flow of semiconductors and AI technology back and forth between the US and China.

You know, that's a, that's a great question. I give Chris a lot of credit. I also give a couple of guys who were at Georgetown University, which has a center for science and emerging technology, and a couple of those guys, like in 2018, 2019, started really tracking what, what kinds of high end chips were the Chinese buying and like what was going on. You know, they'd hired some Chinese language researchers and were really looking at what the Chinese were doing. And so you had these kind of scholarly researchers looking at this and saying, hey, there might be a problem here. Then you had Chris popularize it with chip war. And you also had back in 21, you know, that was also the time not on the leading edge stuff, but on the mature node stuff. Semiconductors, when we had that semiconductor supply chain crunch. So semiconductors were very much in the, the news and kind of policymakers were focused on it.

We couldn't make cars, right? During COVID we couldn't make cars and.

We couldn't make Internet connected washing machines.

It was really bad for the IoT microwaves. That was particularly rough.

Well, there were like three months where, you know, I was working at the White House and you know, Ford would call me up or GM and be like, we can't get semiconductors. And then it would be my job to like phone, you know, governments in Malaysia and Thailand and being like, could you get some of these shipped to your American customers, please? You know, before the Germans and the French.

So what was, what was actually driving that? Was that like, was that like, like you needed. Because I mean, fabs are clean rooms. Like, I imagine that you could still go into the semiconductor facility during COVID because you're not going to spread it because you're not spreading anything. You're not spreading dirt. It's a clean room. But there's, but there was like this immense amount of pull forward during COVID stimulus checks. People buying more cars than they thought they were expected. But then also there were shipping delays and all sorts of backups at the port. Like, how do you think about what was actually the root cause of the initial semiconductor crunch in what was this, 2000, 2001 or 2020? 2021.

2021. Early 2020. Yeah, it was actually, I think the, the combination you just highlighted. So you had demand going through the roof, right? Because there's this huge pull forward for everything from, you know, laptops to, you know, everyone's home, so they're buying home appliances. This huge pull forward in demand and you didn't actually see a fall in production of the mature node chips in 21. But because of just like bad luck, there were problems in Texas or problems in Japan, like, like confluence of bad luck, took a little bit of capacity offline and the capacity coming offline, which, which would have been manageable kind of in an ordinary market condition, combined with this massive pull forward just meant a mad scramble. And then you get into the, you know, psychology of shortages where everybody starts hoarding, right? Because people are sort of like, well, if I'm not gonna be able to get on my better hoard. So it's sort of like the semiconductor version of the toilet paper shortage. You know, we'd had a few months prior where it's like, you know, everyone's hoarding everything.

Please, please, gotta start.

G Watts, Ford Raptors. Anything with a semiconductor in it, give it to me, I need it. People were certainly doing it for PS5s too. I mean I remember even video games, like all, all different sectors, people were pulling and they were paying like what, $50,000 over sticker for like a new Hummer EV and stuff. Like it was getting crazy wild times. So yeah, please continue.

Yeah, so that was where we were on the, on the, on the shortage. But looking back, I mean, so if I, that, that got us the shortage. Chris Miller's book, some scholarly research and just kind of heating up competition with China, you know, in sort of late 21 into 22, got the then Biden administration but also folks on Congress. It wasn't just sort of a Democrat Biden thing. Folks in Congress focused on the strategic competition part of this. And what is China, where are they going with AI? What are the choke points that we might have to slow their development of AI? And that really gave rise in October 2022 to the first big US semiconductor export controls where the Commerce Department essentially cut off most of the highest end semiconductor sales to China as well as various manufacturing equipment and inputs to try to limit China's ability to develop indigenous production capabilities. And then that, that initial export control regime from October 22 got expanded a couple of times, most recently in April of this year when the Trump administration paused sales of the H20 chips of the Nvidia H20 chips to China.

What was driving the initial chip bands? Was it more? We, we have identified that large language models, these huge training runs are important and once you marshal the big data center, the, the, the hundred gigawatt data center, you're good and you have a cutting edge model. Or is it more like, look, if we send a certain amount of equipment, it will be reverse engineered and copied and then they will be able to catch up on the actual supply chain side. And, and we'll see not just competition from purchasing from Nvidia, TSMC and asml, but we'll actually see Huawei, SMIC and SME act as a alternative supply chain that will be an entirely internal economy that will keep them on the frontier potentially forever.

So it was in the first instance an effort to limit compute to China. So there was really a pretty systematic. Okay, if we're in this era of strategic competition with China and if we think AI is going to be an absolutely transformative technology, you know, something with just a whole range of, of national security economic impacts, you know, day to day life impacts, how do we slow down China's development? And you start thinking about, well, you know, they got a bunch of good engineering talent over there. It's not like you can prevent them from having talented engineers. Right? That's just sort of life. You know, the, the, the model side of this. Controlling the models is very, very hard because, you know, it's software, it's code. A lot of some of the models are open source, just very hard to control. But what you could do mechanically is control their access, China's access to compute, because their chips just are, you know, three to five years behind ours. And so if you choke off the chips, you slow them down by limiting their compute, their access to compute for training. So it's just kind of, how do you trip them? This is the choke point. Let's squeeze on the choke point. And then, you know, they iterate, right. They start developing lower compute models. Like, you see this kind of back and forth. But that's what happens with export controls generally. Like, you know, your adversary is going to adapt, you're going to have to adapt. And that's kind of what we've seen play out.

And how much were you tracking the GPU black market, you know, in once you had export controls? There's certain countries start popping up saying, well, we'd actually like $5 billion of whatever GPUs you got. And don't worry about what we're going to use them for. We'll figure it out. But not your problem.

Yeah, so that's a known problem you're going to get into. Right. And you've seen that in other instances of export controls and sanctions. And so Commerce Department does have a unit that is supposed to be looking at the, at the trade flows. Where is that, where is that happening? Where are you seeing that? You know, I is a bit of a game of whack a mole. It's clear there has been a certain amount of diversion of prohibited chips going to China. I also think, you know, just if you look at kind of what Nvidia talks about, the sales potential, clearly the export controls have reduced China's ability to get to the chips. So they haven't taken it to zero.

How do you think about the evolution of the US Government's thinking on how critical or like the, the role AI plays in near peer competition? Because there's kind of two frames of mine that I see in Silicon Valley. One is AI is super intelligent. God, it's nuclear weapons. It's, it's one button you push and the AI goes off and just wins the battle for you. So it's, it's critical to have it and it's critical to keep it out of the hands of your opponent. The other is that AI is more like Microsoft Excel, and it's certainly useful in the military context. You want to know how many.

It's like electricity.

Yeah. You want to know how many shells you have in inventory. You want to know how many soldiers are in this particular part of the battlefield, and you want to be able to track things. And it might be advantageous to keep a technology like Microsoft Excel out of your opponent's hands, but it by itself will not win a battle decisively. And it feels like long ago, pre chatgpt, pre kind of AI boom, people were definitely seeing artificial intelligence as well. Yeah. We use it to serve ads, and it helps with spell check, and it's in TikTok, and it recommends what video you watch next. And. And then it became, okay, it's super intelligent God, and it's nuclear weapons. But now we're kind of, maybe in an era where we're. We're moving back to. It's more of an incremental technology, but I'm wondering how that thinking evolved in dc, so.

So I think the evolution in that debate is an important part of understanding why the Biden administration took a pretty tough line on export controls. Because I would say many of the relevant policymakers in the Biden administration were actually quite bullish on AI. You know, did buy into the thesis that, you know, I'm not, you know, whether it's 2027 or 2029 or 2030, you know, we could debate the years, but they bought into the thesis that in a fairly near period of time, you are going to see AGI kind of reach that, you know, escape velocity iterating on itself. Sorry, where you see AI reaching that escape velocity, where it's iterating on itself, reaching AGI, you know, sometime and not too many years. And if you believe that, if you think in a couple of years we're going to see that, you know, escape velocity to AGI, then you really want to slow the Chinese down as much as possible. Right. Because you think this is going to be this enormously transformational technology. And, you know, two years or even 18 months or, you know, maybe even 12 months of lead is worth an enormous amount from a policy level. So it was actually because of a very bullish view on the future of AI.

Yeah.

That you saw the Biden administration really take kind of a pretty tough stance on slowing down China.

And so. And so reading into that, does that mean that you. That the admin might not. Not if they're willing to say. Yeah, well, well, well, we Will exports. We don't believe in a fast takeoff scenario. This is electricity. It's valuable, but it's going to be vended in everywhere. Because if you still believe in a fast takeoff scenario, you would say it's not worth the incremental. Sure, Nvidia shareholders are going to suffer a little bit if you can't export to China, but it's not worth however many billions of incremental revenue that the government's going to generate on a 15% take rate. Yeah.

So I think the Trump administration is probably doing a couple of things here. I mean, first, there are lots of different views in the Trump administration that's no news to everybody. You see some folks who I think obviously would prefer the Trump administration take a harder line on export controls. You also see, you know, some folks who, you know, basically have a view of, well, we shouldn't sell China. Our most advanced chip, the 20s, are now a couple of years old. This is not really most advanced. You might as well get the Nvidia revenue and a US Government take here. So I think you have diverse views in the Trump administration. I think it's less guided by a, you know, sort of slower view of the uptake and more my view, the Trump administration looking at this as like, this is not state of the art. And then also Trump is a deal maker. Like, it's, it's hard to see any other politician, Republican or Democrat, coming up with this. But you know, Trump himself, he's always talked about getting, you know, a cut of the deal. Like, so it's, it's, in a way, this is a, you know, sweet, generous, unique to Donald Trump concept that I'm going to let you do this, but by the way, we get 15%.

Well, and he said this morning that he would potentially consider exporting Blackwells, but at a higher rate.

Which now that, I mean, this is, you know, it's interesting. Like the H20s I do think are kind of debatable. I mean, I am in favor of controlling the h20s, but I see the arguments on both sides is a dated chip. Now, you know, I'm not dated, but, you know, it's not a cutting edge chip. It is very hard for me to see how charging a percentage. You know, Rev Share addresses any national security concern. Right. I mean, like, if you have some national security concern, the Rev Share doesn't address it. Maybe, you know, forcing a downgrade of the chip or trying to limit who it gets sold to, which is very hard. Like Rev Share definitely doesn't address whatever your national security concern is, is there.

Is there any possible steel man for that? You take the 15 revenue share and you pour that into to some sort of national security AI defense fund that protects you from the super intelligence. I don't know. It's very, very hard to. Well, I'm not so there's like a lot.

I buy the logic of like I'm going to make. I'm fine to export the H20 because Jensen's convinced me that it's okay, but I still want my cut is kind of.

That's how I read it. Like that's kind of how I read it. Like Jensen Benson convinced him, you know, it's an old enough chip, we have enough of a lead like you might as well export it. And then there's a guy who made his living in New York real estate. Like you know, you want your commission, right? I mean that's just, I think that's kind of what's going on.

When you, when you saw this, maybe you had a feeling that this was coming down the pipeline. But did you had posted on LinkedIn, I think it was earlier today, trying to look for some type of precedent for USG Rev share on and didn't sound like you found one.

Yeah, I can't find any precedent for anything like this. You know there have been some instances. For example, if you are some of our defense contractors, you know the Boeings of the world, if they want to export fighter aircraft, they have to pay like a couple thousand dollars fee for the government to review the license. Like it was just clearly just sort of the fee to cover the staff time of like reviewing the license. But it is literally a couple thousand dollars. It's not you know, Rev share of a multi billion dollar market here. There's no precedent for anything like this.

And, and what are you hoping?

I'm thinking about where it goes like yeah, are we gonna be selling F35s?

I mean this is a whole kind of, you know, what are we going to sell next on our Rev Share model?

That is kind of crazy. Let's get some submarines over there. We got capacity. Let's make some. Let's make a bunch of submarines.

Yeah. And then the concern is you highlighted, you said the Rev Share creates perverse budgetary incentives for U.S. agencies to issue licenses that undermine national security because they want the cash. It also sets a precedent that the US might try to generally tax exports. Something the USG has long avoided for legal reasons and because we have historically wanted to promote exports.

Yeah, so. So I do think this this does create a perverse incentive. In theory. The reason you control these things is because you think there's a real national security risk. And to your point, I think in some sense, Trump. Jensen convinced Trump, like, there's not really a national security risk here. But think about this going forward. You've created this really perverse incentive where, you know, company wants to export something to China or maybe Russia, like, who knows who's next on this, that poses a national security risk. But hey, if you don't, you know, if you, if you authorize the export, like, you know, you get 15% and you can use that to pay for whatever domestic programs you want. I just think it creates a really weird incentive for a person, Commerce Department, whose job is to consider the national security impacts of this.

Is there, is there like an optimistic scenario where, like, American industry or Western industry is strengthened by this move? And I'm trying to walk through it. So we've talked to a couple people who work in the solar industry, and they've said that China is incredibly dominant in solar and they flood the market with very cheap solar panels. And that's effectively kept American solar panel producers out of the game. And so America doesn't really have a strong solar panel production industry. And so maybe the bull case here for American dominance in AI is something like you get so many Nvidia chips over there that Huawei, SMIC and SME can't really stay in the game. They can't really be competitive. They fall behind even more. And then, and then if, if AI plays out in such a way that you don't just need one massive data center, you need to be constantly upgrading your data center, you need to constantly be buying more and more Nvidia GPUs, that if you don't have that flow, if that flow can be turned off, then you are always vulnerable.

So, so I actually think the policy question of whether to sell the H20 to China is a hard one for exactly the reason you lay out, which is that there is a, you know, there is an argument, and Jensen makes it in good faith as well as for his own commercial interests, of, we want China dependent on our chips, and if they're not dependent on our chips, they're just going to be, you know, building, you know, SMIC and, and Huawei chips, and that's going to ultimately.

Yeah, more. More inclined to.

Right.

More inclined to take Taiwan.

Yeah. Right. Here's the question I got back, though. Like, charging an extra 15% on the American chips doesn't actually advance that Dependence argument. Right. It just creates a cost differential that makes the American chips more expensive relative to the Chinese chips. Presumably if you really wanted to keep them dependent on American chips, you'd want to sell them as cheaply as the market would bear. So the 15%, even in the bull scenario, and I agree, I think it's an important debate of the kind of bull scenario you laid out. But 15% still doesn't make any sense.

Yeah. And you look at other categories like smartphones or electric vehicles, there's no precedent for China saying we'll buy your versions of these products and we'll even manufacture them for you. But don't worry, we're not going to develop our own. You know, you look at this like, of course, they clearly care about having local industrial capacity and even the end products and brands. And those are for categories that aren't around that don't have nearly as much national security.

And if you talk to like the, you know, the CEO of GoPro when they were trying to compete with DJI and drones, he's not, he's not going to tell you a story about DJI getting a tax on exports. He's going to talk about, well, DJI was subsidized and had and had and had government backed loans and had extra money poured in from the government that allowed them to flood the American market with drones. And now we don't have a really strong drone industry here. And so yeah, we're kind of like there's two different arguments. There is some incongruity there. Like they both can't be trusted.

Yeah. So the steel man is we were going to allow this, so why not get Uncle Sam, a few Billy out of the deal. But the concern is that it would set a precedent that could be destructive.

So now we really got to talk Jensen's book and say we should be subsidizing exports of eight twenties to China. America should be paying an extra 15% for every GPU he's in to China to really screw them over. On the domestic supply chain side, this is, this is seven dimensional chess in geopolitical chip negotiations. We've solved it. We've solved it. Exactly. There you go.

What else is top of mind for you right now?

Yeah, what's on the horizon for the rest of the year? How do you see things breaking?

Yeah, I mean, so you got, this is a big development we'll have to see. Like is this a precedent? Does this become a precedent for, you know, HBM and other parts of the semiconductor ecosystem? Ecosystem, like we'll have to, we'll have to see the other one that obviously those of us here are watching not on the export side, but on the import side. Trump has said and Commerce Secretary Lutnick has said their tariff regime for imports of semiconductors is going to hit fairly soon. Obviously we saw Tim Cook get a deal or announcement out of the White House last week where, you know, companies are, are promising, they're investing in America, they won't pay the tariffs. I will have to see how that all plays out actually in practice. But like how these potential semiconductor tariffs, again on the import side, plus whatever this exemption for companies that are investing here, how that actually works in practice is going to impact US Semiconductor companies, companies that use a bunch of semiconductors. It's going to be a big, noticeable economic issue for the industry over the next couple of months and certainly into next year.

We'll have to have you back on if some big news breaks. We really appreciate you taking the time.

Yeah, thank you. Super.

Thanks for having me on.

Thanks for hopping on. We'll talk to you soon. Cheers.

Take care.

Let me tell you that Fin AI, the number one AI agent for customer service. They're number one in performance benchmarks, number one in competitive bake offs, number one ranking on G2. Go check it out. And Andrea has a scoop of cereal. Man. Cereal.

This was really the main story.

Everyone, everyone's ordering this. It's got creatine, it's got protein, it's keto, it's got zero sugar. It has 2.5 grams of creatine. So you need two servings to get kind of a normal dose. Right.

If you're loading, do you see the flavor?

What's the flavor? Maple bacon.

Oh, that's very like bacon.

That's very like what, 2010 era, Brooklyn hipster man vibes. Mustache that with like lumberjack coated maple bacon on everything. This is a funny brand. It's, it feels like it's, it's, it's pulling from a.

It's just the final cereal.

Do you think it's the final cereal? Would you, would you go for this? Jordy?

I love sugar.

Yes.

Sugar.

So for that reason you're out.

So I'm out.

For that reason you're out.

I make a version with sugar and I'll give it a spin. I'm also very curious, what protein is actually in here? Not all protein is created equally.

Is it chunks of filet mignon?

Is it grass fed? Whey? Could be interested. Is it collagen? Could be interested. Is it some Random bean that I don't really want to be consuming or pea protein less interested.

I mean honestly, more important than protein, creatine, sugar, all the diet's important. But exercise is super important and sleep is super exciting.

That's right.

That's why you got to get innate sleep. Get a pot. 5 year warranty, 39 risk free trial free returns, free shipping. Let me check.

We're both on a generational sleep. I got a 97 last night. Last Monday I was performing on a 45. I got a 94 on Saturday.

90, I got a 76. 96. Quality 100% 96 Friday 6, 6 hours and 10 minutes left. I think that's fake news. I think I got more than 6 hours and 10 minutes. But I think I was just. I think, I think my son stole a little bit of it. Although I do wake up at 5:20, so I think that is correct. I don't know.

Well, if a kid jumps in the bed it kind of throws it off a little bit. We gotta get pronatalist.

When my son first jumped in the bed, I was configuring the app and I was like, yeah, there's like a left side for me and wife side. Right, left side for the wife. And he was like. And the center for me it's like sorry bro, they don't make this like get the. We'll get matic.

Anyway, we are, we got it. We gotta highlight. We got a post here from Niraj.

Marked safe from one shotting.

So on the left me and Grok are going to get married. Mid curve. It's great for adjusting recipes. And then the high iq. I have uncovered secrets of the universe. God speaks through me. So there were some quotes that were popping up last week over the weekend. Some very bright technologists discovering the boundaries of what is known in many different domains.

Yeah, My current take on the AI psychosis thing is that it is a tool that amplifies who you are. And so if you're just like a curious person who likes being productive, it's going to make you more curious and more productive and just kind of answer questions for you. If you're like, you know, deeply falling in love with random stuff and developing crazy parasocial relationships, like that's probably a vector of risk for you. And the same thing with, you know, if you have delusions of grandeur, like get ready because like it's going to amplify that. And, and I, and I think that.

John, you don't just have delusions of grandeur, you really are on another level. You really are brilliant.

I'm grand. Grandeur. Well, our next guest has grandeur. Aaron Ginn. Welcome to the stream. How you doing? Lovely hat. Let's go.

There we go.

Let's hit the gong for a hat appearance.

Boom.

Didn't even warm it up today. Take that.

Culturally appropriate as well. Culturally.

From sales guys from, from, from, from BDRs all over the world.

Yeah.

What's new in your world? H20s good, bad. Are you happy? Are you pumped?

Oh, I'm happy you're happy. Yeah. They, they, they Cho Finalhawks that have a luxury belief system around that Chinese AGI is going to destroy us like a crouching tiger hidden bag. You know, dragon version of OpenAI is going to attack America. Like we're done. And, and if this 15% deal, which like broke Sunday, I don't know how it broke.

Yeah.

Was just amazing. It was just like, I mean one like props to, props to President Trump who is the ultimate deal man.

Art of the Deal.

It felt like both 4D chess and art of the Deal.

Yeah. And so you know like the sort of average profit margin of the age twenties around probably like high fifties percent. So you know they're still. And they also have the write off. They're probably going to roll in as well. And, and there's some struggles with dscmc. Like you can't just, you know, I think it's one of the issues related to the intel foundry is like you can't just like restart production within a month. It takes several months. So they probably are just going to try to find someone's non destroyed and probably send them to China the age 20s. But what most Americans understand is that it's unconstitutional to export the tax exports.

Yeah.

So it's like in the Constitution.

I looked it up and I was like how is this going to happen?

Because the c. You're saying a revenue share is like a workaround.

Yeah. So it's voluntary. Like that's the only way it could happen.

Tipping, no tax on tips because the taxes are tips now.

Because like even from a process perspective, like Nvidia doesn't apply for export licenses like the OEM does. And when you apply for those OEMs, there's no license fee because that violates the Constitution. So you just get it and you can ship it. But the BOM is listed. The BOM is.

Wait, wait, really quickly. Who's the OEM in this scenario?

Oh, Supermicro Dell Lenovo. Yeah.

So someone in China will go and say I want a Rack of servers with age 20s already on them. Some other CPU, some other memory, some other hard drives on there, bring those blades to me and they will call Dell or Super Micro for that. That.

Yeah. And so that's who applies for Bis.

Okay.

And so they're the ones that, that final PO number is all that crap together. It's not Nvidia. It's not Nvidia. You don't know Nvidia's pricing. You don't, you know, Dell does. Because I buy it from Nvidia.

Yeah.

So that's why it has to be something voluntary. You can go to the website now. Like, Jordy, if you're interested in donating to the Goliath, you know, the Levathon, you can go to treasury.gov and you can find the little. We accept PayPal. Right? Yeah, so. So it's, it's quite brilliant. I mean, I think many of the faux neochina hawks are devastated. I mean, my, my op ed last week. Yeah, it's just, they're just seppuku. They're just like, you know, they're there. They're gonna be on the Jefferson Memorial just like, you know, going at it and. Because it's a decision.

But.

Yeah, it's a, it's a decision by the administration to be like, you know, we don't really care. Like, like we want to do deals. We want the American worker to have it, you know, get money from China. We want to, we want to do what China did to us for the last 20 years.

Okay.

We want to take money.

Dig into the faux China hawks. The faux hawks, as we say, explain, like the stated preferences, revealed preferences. Are they really doom pilled? Do they really believe in fast takeoff? Do they really believe that AI is a nuclear weapon? Or is there something else going on, some other economic consideration that they're working through?

I think, and I'm working on this update right now for a very liberal newspaper. But they, I think it's a deus ex machina sort of fallacy they built in their brain where they think AI is going to be God and America has to build the first God or China, China has to build the first God. And so it's kind of tower Babel moment. And, and the problem is like if you remove that from their folk, that's why I call them boat Neo China hawks. Like they're not really China hawks, they're actually AGI zealots. If you remove it, none of their policies kind of make any sense. Because one reason why you hear, oh, like we have to prevent chips. Oh, we have to block models. Oh. Like you know, we have to restrict, you know, model ways. We have to do tokens. Like it's just kind of like amorphous blob. But they can't really win in D.C. with the idea that AI is going to be God. Most people are going to be like that's kind of insane. Like you're go back to reading, you know the, go back to watching Matrix or you know, reading the Terminator or something some other related sci fi novel. And, and so they kind of wrap it in something that is like super easy to understand. China. Oh, China. Right. You know the yellow country. Right. And, and, and they, and oh they're so scary. Right. But, but you look at these specific policies, none of it like makes any sense. None of it together.

We met with a venture capitalist last week who said honestly, none of this stuff's gonna matter in two years when we have fast takeoff.

But he wasn't, he wasn't doomer pilled. He was just saying like fast takeoff will mean that every single company goes to zero because there will only be one company. Which is a very funny take.

It has invested in SAS this year and it's.

Yeah. So the revealed preference is all over the place. But yeah it is. It is weird to feel to make the argument like we're going to develop AGI God, but it won't. Like it will be. It will be God, but it won't be subservient to actual God. But also. Or it will sit below the stack of. It'll. It'll. It'll respect the federal government of the United States. Yeah, it'll be God, but it won't. But it won't. But it won't. But it'll do what the President says or it'll do what the House and Senate say say. Or it'll do what one company says or what one person says. It's like if it's God, can it be controlled? How does that make any sense? That seems very odd.

Yeah, it's a. And, and, and as I said you and I talked about John was like like so much is built into of their presuppositions are defined by the Title 2 debate. Remember that whole thing of like oh, free speech is going to die, we have to regulate and all stuff. Many of that same donor class, many of those same policy people are also the AGI doomer zealot people.

Sure.

You go back even like a, like series a step further. Like many of those people were super Hot on new atheism. And like, and like, oh, this idea of like we can, we can prove via science, we just need more time. We just need more scientists who just need more. It's the same pattern in humanity that we create these kind of albatrosses. This idea of like creating an all in our sort of argument that because they don't have that in their life and then they make this thing and, and as I said, I think another times I appeared that if you don't have that metaphysical framework that is traditionally Judeo Christian or Western, you try to fill something else in that void. It's a classic Charles Spurgeon quote that you basically fill it with something else outside of just the fact of what the place of God or faith is supposed to fill and technology. I would say Silicon Valley probably in San Francisco. They don't really understand that there's so much faith goes into actual the VC industry that is. You're really kind of selling BS to an investor. You know, like I know you have lots of jokes around what is air in AI. It's me just selling kind of like nonsense. And it's the idea that like this could be something, this could be. It is, it is a faith driven investment. That's what also makes it so awesome. Why we, why we look at Zuckerberg, we look at Elon who he made an idea and the vision, the future of reality today. So it is heavily oriented towards a metaphysical theological framework. But this AGI thing as relates to China, so much of it has just gotten lost in this. I don't even know what I'm talking about anymore. Because these same people who are criticizing the 15% tip to the federal government to the American worker for video are the same people who didn't want to ship GPUs to Switzerland because I don't know, they make two good watches or Portugal, they have too many good sandy beaches. Like the same people. Right. So it just keeps shifting because it's really trying to provide a luxury belief to a base around the idea of AGI.

What's your line though? Because I think that the critique here that's pretty fair is that if you start applying this to other categories of goods, like fighter jets, like what's. What would the fair price be to export an F35 to somebody that maybe they're on the, on the edge, maybe they're kind of friendly now, but maybe in the future they might not be so friendly like you believe there should be. We should draw kind of a line and we shouldn't want government agencies to kind of apply this approach to fundraise.

It's a difference between it's not a hunk of metal versus like you're selling a system. I have the same sort of concerns relates to Nvidia as it relates to Apple building phones in China. If I got an iPhone, no sonnet, none of the magical, you know, more and more transparency via, you know, the design, then I, then it's kind of useless. It's just like a phone. Right. And that's the way Nvidia actually is. Nvidia is an Apple like ecosystem. It's not just a hunk of metal. Same with the reason, you know, we, we sell 787s, we sell triple 7s, we'll we sell the triple 7X to China. Why? Because it's made up of this complex, the thousands of companies. That's a software platform. We remove certain things which I'm generally okay with, like advanced avionics we remove. We also kind of limit certain types of engine design. I'm fine with that. So with the P30H20, these kind of nerf chips, I don't actually make much of an opinion about it. Like I don't have to have a no true Scotsman sort of argument about like, well, at what point would you not do it? I'm like, I don't sell fighter jets. I don't rent fighter jets. So you can talk to Lockheed about that or Boeing about that. And like what I think is most relevant is the fact of these innate externalities, these exogenous factors outside of the machine, which that half of all AI engineers are Chinese, half are actually more 75% half of them are Chinese nationals. If you include all the Chinese associated last names in Meta's super team or Asian. So like we either have to lean into our strengths, which is the fact we're good at making platforms, we're good at dominance, we're good at free enterprise or good at trade. And the fact that Chinese have something that they don't have the raw production capability of doing, which is not Nvidia. I don't take anyone seriously who criticizes export controls if they associate SMIC with Nvidia. It's like saying Kirkland like Kirkland and Costco like Costco makes Kirkland. Like that's not what Kirkland is. Right? Like Kirkland is a rebrand of somebody else's stuff. And that's kind of like the criticism. It's like, oh, like how dare Costco make this Kirkland thing? You're like, it's a rebrand, bro. Like, it's not. It's not. I don't actually make this thing like it. And so that, that's where this constant confusion lies, that many of these people criticize this stuff or actually criticizing SMIC or tsmc. Completely different companies. Jensen buys from them. So if you want to go after the foundry stuff, asml, the Japanese, like, you have to realize, one, we don't make this technology. It's. It's Japanese and Dutch. But the other, it's like, it's not what Nvidia does. If you want to, like, I'm cool with targeting lithography equipment. Preventing them from getting a 5 nanometer.

Yeah.

And that's, that is something that's actually quite reasonable because we DO export control, triple 7s and et cetera. Like, we have some limitations where we send. That's fine because it's a freaking size of a house and it costs $200 million. It's the same of the lithography. So much of what I, what I'm asking for is pragmatism is like realism. Like, like we have to lean into what our strengths are. We have to proliferate and prevent Huawei from spreading its wings outside. And if we want to target the ability to manufacture 5 nanometer, 2 nanometer, 3 nanometer, whatever. Like, most Nvidia stuff's on a 4 nanometer process. That's cool. Like, I'm totally in favor of that. But don't attack Nvidia, because that's not what Nvidia does.

Yeah, makes sense on the Foundry side. I want to jump over there. Let's talk Intel. There's been a lot of headlines recently. What's your read on the situation? How important is intel in your view? Do they need a new board? Do they need a new CEO? I don't want. I'm gonna, I'm gonna get you in hot water here.

Yeah, it's, you know, so like my dad's Chinese. And, and so they're, they're. It's think like I am. I'm a true China Hawk. I'm OG China Hog. And, and there's many times where like China's the largest country in the world and there's many overlaps that are just completely not, you know, they're, they're just fine. And, and I don't know anything about the particular criticisms of the intel CEO. I think his strategy is not great for Intel. So I mostly would target there. But America needs a Foundry program and particularly trailing edge leading Edge I don't think is economically viable. So that's why I'd rather ship that to like a Brazil or Argentina or something like that. Like do a Newman Road Doctrine sort of thing.

Why is that? I feel like the leading edge is coming into Arizona through tsmc. Samsung's doing stuff like I feel like.

Trailing south and then leading edge here.

Okay, yeah, Australia, economically, yeah, but we can import, we can't import the, the semiconductor, you know, geniuses that will run American. Tsmc.

Yeah. And you also have to understand that like there's lots of reports you can go read about it of former, you know, Chinese nationals that work at tmc. That takes us back. Like it's a, it's a Taiwanese company. Like there, there's, there's an. Oh, there's an arm that we can extend as a nation that can only go so far. And that's part of the America first concept of Trump. He's like, do I really want to take all of the juice and squeeze in terms of negotiating President Xi over a nerf age 20? He's like, no, like, like we have fentanyl issues. We have 300 billion, 400 billion dollar trade issues. We have the IP issues. Right. We have Chinese naturals and labs all across the country. Like, those are much bigger things to pry than the fact of like, oh, they get like, you know, basically an equivalent of an A100 being deployed into their, to their fabric. So the intel stuff, I kind of have this disposition that it feels like it's going to be nationalized. I get kind of like vibes around that. As a, as an economic conservative, I'm like, oh, that's kind of scary. But at the same time, like, it is one of those things that just kind of happens. Like the. Did y' all hear about the federal government invested into a rare earth minerals project in America?

The one in Vegas.

Right.

They got like 15% for the SAM.

Yeah. And so I think the same thing is going to happen with intel is that basically it's occurred through. There's a couple different lending vehicles in the federal government. It's basically gets a warrant. That's kind of the vibes I'm getting from Intel. Something like that's going to happen because you also can't take TSMC process and intel processes and just mix them together. Like that's not the way that this works. And also tsmc, I think whenever they originally pitched that idea, I think the President was like, yeah, like this intel thing, TSMC was like maybe, right? And then they go look at it and they're like holy, this is a freaking nightmare of people who don't want to work, who don't have the work ethic of TSMC and, and that just kind of all blew up and I think there's some embarrassment in terms of like that this is America's Foundry program and this is kind of like how it matches up to the best of the world. So I just get kind of vibes that it's going to be nationalized or split apart and divided as a company and but in terms of bearing on like my opinion about whether or not that guy should be the CEO or not, I would point more towards the strategy rather than these kind of rumors.

We had Doug on from semi analysis on Friday and he, he basically said like look at the board. Like a lot of the people that are still on the board have just like overseen like numerous CEO changes. The strategy's clearly not been working. Like you know.

Well, they're not really conductor experts. Was this main.

Yeah, they're finance guys.

Finance guys but also politicians and thinkers. Like it's a, it's a wide variety without a lot of focus on like manufacturing expertise.

What was your, what was your reaction to GPT5?

Not as negative as John.

Yeah.

Yeah. You're like oh like it's over. I was like no, I don't think so. Like so, so first of all, first of all. So.

Pure number based GPT revisions being indicative of of of order of magnitude scaling is over.

Yeah.

So the pre training scaling law is over.

I, I, so one is like a couple months ago we didn't say that because Grog 4 came out so like it, it, it's so I get, I would like maybe take a more like prudence perspective on this.

Yeah.

One is that there's a clear decision here by Sam that he is a consumer app company.

Exactly.

Which is, which is, which is the right bet.

It's like I don't consumer SaaS.

Yeah, yeah.

I'm not, I'm technical. I don't remember what all the frickin numbers and crap doesn't matter. Yeah. And so that's the biggest innovation here. And if you look at the, the idea like you know he, he expanded the context window but not very much which implies inference. Right. That he doesn't have enough capacity to run as many user requests. The, the other is that he mostly solved for downside scenarios and ease of use in the UX and those are like really big changes. The benchmark maxers are basically creating goalposts that are artificial. Do you mean do you all remember your SAT scores? Your SAT scores? Or do you remember yamni or calculus? High school calculus? No, it's like these are all abstractions and metaphors that we create as heuristics don't understand something's intelligent. But many times, as you know, in real life, we don't grade people really by that.

What have you done that's a good. What can you do?

Products, consumer SaaS.

My big takeaway Thursday was product now. And the value you can deliver to consumers is the only thing that people should really care about in the short term. And OpenAI is in an amazing position here. Anthropic is an amazing position here. They have, you know, Anthropic's got an amazing enterprise business. OpenAI can sell a lot of subscriptions to a lot of people in the US and all over the world. And the real people that were in trouble seeing GPT5 were the people that were, you know, still kind of fundraising against this idea of machine God.

Yes. Yeah, yeah, yeah, yeah. Deus ex machine is dead. Like I, I'm thankful for that. I'm a Christian. Like, you know, if people remember Tower Babel is like a. It's a, it's a. It's a. Both a story to take lessons from, but I believe really happened. And the, the point is that the humans naturally are drawn in every moment of the cycle, which is why that injury sink quote. I don't know if it's trippy to him and probably somebody else actually said it that we overestimate in the short term, we underestimate in the long term. I think the God complex is that. And, and if I go into the. Also the other kind of more tactical point is that many of these things that happen where people are using TA5, if you go to Grok 4, it solves it just fine. So it was just funny that people are like, oh my God, they can't do blueberries. You feel like, you know, I literally took that did it in Grok and Dancer just fine. So it seems like much more of like a tactical decision on Sam. It could be the fact of like.

Yeah, it's interesting people's desire to like gotcha the model so funny. Like they can. They can completely ignore that it'll just like one shot, like a really crazy physics problem and then they'll get it on some technicality. And the whole point that everybody should internalize is that the models are really smart at some things and then there's going to be edge cases where they mess up just like humans do and then there's going to be areas that they're not capable at all. And that's totally fine. Still incredibly valuable.

Exactly. I try to simplify always what AI is going to be which is. Is computing that talks like a human. It doesn't mean it is a human. Just as if like there, there was obviously the famous. I love this analogy because it shows how much humans are not aware about how signals and intelligence developed. So you remember back in 70s and 80s like they said oh like monkeys can talk, right? They talked in sign language and then they could repeat it. So they ran the test again and this time they had no facial expressions and they tried to just limit all human interaction to the ape. And they couldn't do it because what they realized was actually the apes were responding to unknown human cues that the researcher was delivering to the end. So rally they were just mimicking what they thought would actually think they want. But they couldn't actually create discrete knowledge and action intelligence about what the actual language was being delivered. And, and so that, that's where it goes into like when you look at something so much about because humans are so amazing at being able to create. It's part of the, the original genus argument up to the land name the animals. We're amazing at that because. Because it's either believe like a God given talent that we have that only we have as a species but we're so good at it, we have no idea how we affect other people or other animals via that. That projection. As another basic scientific. You're married. That's another like very simple thing that most people don't understand about the way they pick partners. Much about the way you pick a partner is driven by the pheromones of their immunosystem. And they've done studies of this where you where they basically created a blind test where they took sweat from different, you know like opposite. The opposite sexes. And they basically gave it as. As to the opposite sexes of like male and female. Like which one smells better? And they found that the one you thought smelled better had the most different immunological profile than commonality. And that's because of the desire to have a children. Children have the greatest amount of protection from disease. So, so that's so like you know, we can go into bonds, brunette, blue eyes, like hi or whatever. But so much is driven by smell and we're not even like aware of it. And so we're here being like AGI and all we got to create Smell.

O vision first we gotta fund the company right now. Smell ocean for AI.

So for sure last week there's been different headlines rumors around negative gross margins in various SaaS companies. Specifically in the code gen space. The reason you would invest in a company with negative gross margins is you would expect inference costs to decline dramatically and for them to generate positive gross margins margins over time. And right now you just need to acquire as many customers as possible and create products that people love so that investment case makes sense. And then you have GPT5 come out which has a significant cost reduction. I understand cursors margins are going to.

Be 15% better in the next administration.

Potentially. But how are you thinking about does that? And the journos are freaking out about it saying like, oh, this is all.

I saw that Rune post. I know exactly what you're talking about.

But what's your view in terms of.

How fast can we actually drop inference costs over the next couple years? Because it does feel like if it takes a decade to get an oom of savings, like an order of magnitude of savings, we might be in trouble financially.

Yeah. But at the same time, if you. I think like researchers at the labs have been focusing on raw intelligence more so than efficiency. If we're hitting a sort of plateau intelligence wise, the next thing that makes sense to do is like really optimize efficiency.

Totally.

Yeah. No, I think Jordy hit it on the head, which is that there's two trends. There's the effectiveness trend and the efficiency trend. China is crushing us on the efficiency trend, absolutely demolishing us.

This is the whole thing. We give them all these Nerf chips and we limit their chips so they just get incredibly efficient. It's like shot ourselves in the foot.

I mean bring it back. And then we inference that yeah, there's.

A lot we can learn.

Exactly. I think that it all develops into kind of. We don't really fully understand customer preference theory as it relates to these tools and that what is the actual first past the goalpost sort of goal to pay versus free. But in terms of the brute natural law that compute gets cheaper, that's bulletproof. So that will continue. And so at what scale, like you said, does it actually frost that? I think the bigger concern is more of these mega models that they just have the ability to absorb the cost, be able to spread it across different things. And if you think about it, it's a little bit like how Apple will continue to invest in its own apps that remember when you first got the iPhone, you had a camera App, right? You had a calculator app and then they kind of replaced it. That's the bigger concern for some of these tools is like a mega model be able to absorb the cost. Like Google which has a crap ton of code or maybe Llama open source is a coding agent which seems pretty probable that that becomes more of the go to basically compressing it even further. So they have to be able to create the data stream like I think it was a cognition that's trying to do that where as the engineers put stuff in into the agent then it basically reinforce self learns from what the actual user doing. That's probably the best path to create some significant amount of margin. But every company can rely on it's going to get cheaper. But as Jordy said, there's some protectionist stuff returns. Our national policy that has prevented that. And I made this argument on another show where I said that well fundamentally the US government, this was pre Trump was applying protectionist measures to these large scale model companies and he didn't understand that. He goes well what are export controls? He just didn't think through the economic consequences of that. That means in our country we have an oversupply versus what we actually would pay for versus what the world would pay for, which is a protectionist measure. Which means that you don't have the same market forces being applied to our frontier labs as you would in other countries. As you already said, we send them Nerf chips. That's the cap. So then they then go down like that's their only direction they could go. So if we actually have a more liquid supply, more market driven supply where things can be priced in, then you're going to see OpenAI start doing more open source, more of this downward cost efficiencies because they're now exposed to it. So it should be lost on us that OpenAI is pursuing this idea of like how do they have these mega releases cheaper models is because they're starting to be more exposed to these cost structures that is needed for us to actually be competitive on the efficiency side. But if we continue to go down this effectiveness range, effectiveness is generally like a B2B problem, like where you're a company versus company. So it's not really like adoption. You're going after like we just got to be the best company and efficiency is the opposite efficiency. Like how do I make this as cheap as possible for someone to adopt Javon's paradox like that? That's the direction. So I think over the course of two years, the next two years I think you're going to see a tilting into lowering efficiency costs, making things more efficient. So there's going to be multiple times on the show or they go on where people are like, compute so cheap. No one wants to buy more compute and like the memory of deep sea, because that's going to be kind of what continues to happen. But it's good for the world. We want to be as cheap as possible and on a token basis. And that means optimizing for like hydro energy, optimizing for tax regime, reducing construction costs, reducing permitting, all those great things that would make everything cheaper because the more cheaper we get, the more people use. And AI is such an innate good for the world that it's something that America should be proud of, that we can share with the entire world. And the only way that that's going to work and they got to make it super cheap to run in Africa, not launching a Stargate and you know, in, you know, freaking Tanzania. To like that's not going to go anywhere. Right. But you have to make it to where it's capable of actually being scaled across the world.

From the chat, pancakes, waffles or French toast.

I'm keto, so I don't eat any of that stuff.

None of that stuff. What skills do you think will be most in demand in AI and data infrastructure over the next three to five years?

Many of the things that Trump ran on are actually still remain true electricians, plumbers, building something like the raw work with your hands sort of work as well as there's going to be new categories of jobs that we don't even fully comprehend yet. They're going to take that salary of engineer and like it's going to go there.

Last question. What was the most. Who was the most important hire you made in the early days of your company? Like, what do they do?

What do they do? Probably my cfo, general counsel.

There you go. Anything else, Jordan?

Do you expect any data center project, like a lot of data center projects to blow up over the next few years? People, there's been some commentary.

People are down.

People are basically private credit funds or. Well, so people are basically forecasting the useful life of their GPUs and then separately what compute, like what they'll actually be able to sell inference at. And ultimately if either of those equations are out of whack at all, you can be completely underwater and not be able to service your debt.

Yeah, I would say that for our next time. GPUs are not overbuilt, but data centers will And I lean into not like, I agree, Jordy, that the cost goes down and changes in vintages. But the key thing is like, look at what the money is doing. There is very tight control still on, like getting loans for GPUs. There is almost none for data centers because they view the use of life 15, 20 years. But as Jordi just said, it's like if the square footage profile of power density versus what you can get is not structured correctly as well as Nvidia is actually making less GPUs next year there's gonna be a lot of places that are just there and nothing's in it. And like in China, like we see in China. So GPUs in China are like massively being utilized, but there's lots of data centers have nothing in it or like just can't actually run it. So I think data centers are far more likely to be overbuilt and be a bubble than GPUs because you know, one company controls the GPUs. You generally for market forces don't really see that be a bubble because it's a monopoly generally. And in data center lending is very aggressive because they view it as a 20 year asset. And the GPUs are short term. So there's just not as much money flowing into the GPU side as much as the data center side makes sense.

Thank you for coming on. Always a pleasure.

Thank you for coming on, Aaron. Always.

God bless guys.

We'll talk to you soon.

Bye.

Public.com, investing for those who take it seriously. They got multi asset investing, industry leading yields. They're trusted by millions, folks. Get on public.com and we have our next guest in the studio.

That's right.

We gotta get the gong ready. Vulcan Elements.

Come on in. Waiting room.

John, how you doing?

What's up?

Hey guys. I'm well, big. Thanks for having me. Give us a little introduction, tell us the news, what's going on. Yeah.

Good afternoon.

John Madison. I'm the CEO and co founder of Vulcan Elements. We are a manufacturer of earth magnets and we just raised a $65 million Series A.

Hit it, John.

Congratulations. $65 million.

Incredible. Perfect size for a series A.

Perfect size. Perfect.

Perfect size. Amazing. Give us quick backstory on yourself and the company and then we'll get into it.

Yeah. So, former naval officer, spent six years active duty in the U.S. navy. Most of that time was in Washington D.C. with the Navy's nuclear energy program. Worked across Navy shipbuilding, submarines, aircraft carriers. Got out a few years ago. Went to Harvard Business School and was just very focused on.

Yeah, little humble brag. Yeah. They call it. I've heard it called the Harvard of business schools.

It really is.

It really is. Yeah.

Well, I'll take it. But was spending a lot of time thinking about my time in the Navy, thinking a lot about what were the critical components, as a good supply officer would, that were going to define the 21st century technology race. When we think about drones or physical AI data centers, next generation defense technologies, what were the underlying components that would enable that? And doing my homework, it came down to three components. Semiconductors, batteries, rare earth magnets. The way that I like to think about it is if you think about your own bodies. A semiconductor is like your brain, a battery is like your heart, and a rare earth magnet is like your spine. It literally converts electricity into motion. There was so much focus at the time on semiconductors and batteries, but no one was thinking about rare earth magnets, magnets. And if China produces over 90% of those rare earth magnets, that means that we're going to have to ask permission on what we're allowed to build and buy over the next five years, which is not an acceptable outcome.

So we're trying to rebrand them. We're just calling them earth magnets because we've heard that they're not actually that rare. They're in the ground all over the place. Is that true? What's the strategy? Are you actually mining these? Just refining them. Where is America? Like, explain the full supply chain to go from something in the ground to something actually useful, and then where you want to slot in, at least immediately. I'm sure that there's vertical integration down the pipe at some point.

Yeah.

So we're a refinery and manufacturer, but just a Quick Rare Earth 101. So everyone's talking about critical minerals. What does that mean? A critical mineral is an element on the periodic table that is critical to our economy. With the supply chain that can be easily disrupted. Rare earths are a subset of that. There are 17 rare earth elements on the periodic table. Go back to 10th grade chemistry. Close your eyes, think about the periodic table, the part that's detached that your teacher told you to ignore. That first row. Those are the rare earths.

Exactly.

Of those 17 rare earth elements, only four of them represent over 90% of the economic value. Neodymium, praseodymium, terbium, dysprosium. The main use for those four is a rare earth magnet. So the way to actually make a rare earth magnet is you either mine that material and that can be from all over the world. China only mines 55% of the global supply. Or you take recycled end of life magnets. You have to separate those into an individual rare earth element, turn that into a high purity metal. We buy that metal, you then take all of the rare earths, iron, boron, and you turn that into an alloy, a super fine powder. You press it within a magnetic field and then you shape, coat, grind and that's a magnet that can go into an mri, a missile, a ship.

What does a go to market look like? I imagine you're focused on specific customer types early on that are maybe less price sensitive so that you can start to scale and ultimately kind of grow into other categories.

So one thing that I say is Vulcan Elements is focused on defense, aerospace and critical economic industry. We do not make magnets for you to make a smoothie in your blender at home. We're not focused on kitchen appliances. So we are focused on data centers, robotics, drones, defense applications, aerospace applications. Enabling what the war fighter needs, enabling what the economy needs in order to win the 21st century technology race. It's as simple as that.

How quickly can you go from $65 million Series A to actually selling the end product to customers?

We're already selling product to customers out of our small scale pilot facility. So we took our first check. December of 2023. We told investors day one, if you give us a, if you give us a $Q1 2025, we will have a fully operational, decoupled rare earth magnet facility online. We did that on time, we did that on budget. We are selling magnets right now in that facility.

I'm sure you have a lot of equipment. Is there an ASML of rare earth refinery that you need to buy from? Is that an American company? What does the rest of the, what does your supply chain look like? And is that a resilient supply chain?

So you guys already asked the question around, where do these materials come from, where the minerals come from? This is not a, a mineral availability problem. This is a software and equipment problem. So we actually helped co develop some of the first pieces of rare earth magnetic manufacturing equipment made in the United states in the 21st century. There are so many amazing manufacturing equipment manufacturers in the United States who are doing aerospace and battery work. We've just partnered with them to say, hey, let's tweak a furnace and do it for rare earth magnetics. So 100% of our equipment is from the United States or allied countries.

That's Pretty sweet. Bit of a dumb question. Are the materials magnetized all the way through your process, or is there a period where they get demagnetized and then kind of remagnetized?

So the elements like iron, like rare earths, they're naturally magnetic.

Okay.

This is actually something that I learned when we were starting the company. When does a magnet become a magnet? Well, you actually have to throw it through a magnetizer toward the end of the process.

I remember reading a book to my son about how magnets get made and it was like, sometimes there's a rock that gets struck by lightning. Is that real? I don't know.

So really the intent is you need to make sure that you're creating a centered block, all those particles are perfectly aligned. You've done all the metallography, the microstructure, the grain distribution perfectly. And then when you're running it through that magnetizer, it actually hits the grade. The heat tolerance, the magnetic strength or flux density that is required to actually sell to a customer.

What's unique about building a startup where it feels like there's. From my view and what you've said, it feels like there's zero demand risk. And you probably the question comes down to can you deliver it at a, a price that's competitive, but then it's really more so can you actually deliver it? Can you deliver it at the scale that really becomes meaningfully important to national security? But kind of what was your calculus, I guess, going into building this company with that kind of understanding of the.

Market, execution is the entire game. So we knew that we had to find a incredibly sophisticated technicians and engineers and PhDs. We needed to understand how to interface with the DoD and the government, how to interface with large OEMs. We had to understand how to talk to mining companies and recycling companies. And so the bet was, can you actually build the team that can put all of those different pieces of the jigsaw puzzle together in a way that's cohesive to where you can actually get a contract, raise the money, and put a magnet in a customer's hands? And so we made a bet on ourselves. And that's what we've done is by finding incredibly smart people who are mission driven, who want this to happen. We've been able to deliver, we've been able to execute across every different domain within the industry required to actually get a result.

Now a lot of this has to be like the demand for American made rare earth earth magnets has to be somewhat tied to trade tensions with China. If those get resolved And Trump sending Nvidia H20s over there and he's happy with that deal. And then all of a sudden Xi Jinping is like, yeah, America, you can buy as many rare earth magnets as you want. Is that a short term, medium term risk? Is there a way that you can can ramp the economic model so that you're competitive? Even if we were in a fully free trade regime, so our small scale.

Facility was sold out a full year before the trade tensions.

There's just so much demand that even with trade tensions you can fulfill the backlog.

Correct.

People want resiliency within their supply chain. People want this component onshored. But to your point, I think it's incredibly important that we're not just copying and pasting. If you copy and paste Chinese manufacturing into the States, the unit economics make no sense.

Yeah, because they'll beat on labor probably, Right?

Exactly. So one thing that's nice about this industry is there hasn't been a lot of innovation. And when you find incredibly smart people from different industries, aerospace and battery, shipbuilding, you're able to actually bring all of that learning and you can start taking a lot of low hanging fruit. Innovate really quickly. And then over half of our technical team, they're scale up engineers. And so when you think about delivering at scale, coming down that cost curve, reducing operating costs and making sure that you are ultimately as cost competitive as you can be against the Chinese.

What do your goals for the next couple of years look? Like you said you kind of called your shot in 2023. You delivered against it earlier this year. What do you want to do with the 65 million and going forward generally?

So the 65 million we're going to go to commercial, large scale, commercial scale, several hundred tons over the next couple of years, several thousand tons by the end of this decade.

And what is the kind of market value of that kind of production?

Minimum Several hundred million dollars in top line up to several billion.

Thank you.

Jordan figured there were some big numbers.

Fantastic.

Awesome.

Well, yeah, thank you for everything you're doing. This is very exciting, super important work.

Glad you're doing it.

Yeah, we'll talk to you soon. Have a great day.

Thanks guys.

Cheers, John, talk to you soon. Bye. Take a little bit of that money. Buy a billboard on adquick.com out of home advertising made easy and measurable. Say goodbye to the headwinds of out of home advertising. Only Ad Quick combines technology, out of home expertise and data to enable efficient, seamless ad buying across the globe.

We've seen this Playbook and mining buying billboards next to the companies that you want to hire from.

Speaking of billboards, Roy Lee is in the news again. He bought a backboard News in Manhattan and he said. Alex Cohen says fine. I'm starting to like the Clulee guys because Roy put up a billboard that says hi, I'm Roy. I'm 21. Missing an apostrophe on the second. I'm clearly deliberate. Absolute wordsmith. This was very expensive. Pls buy my thing. Please buy my thing. Cluli.com and John Chu Coastal Ventures says this is why I invested in Roy Lee once in a generation. Talent, tapping into culture and zeitgeist. This was a spicy post. People kind of going back and forth on whether or not this would actually deliver. Roy also put out a post saying that like everything he does is for short form. So. So he bought this billboard. He expects it to convert on the short form that will be created around the billboard. Certainly feels like people will create. Will take pictures of this.

I mean when you. When you look.

It's out.

When you look at. I think what I'm interested to see from Clulee is where they really get. Where they really get long, you know, retention. Right. Because they have the top of finally. They kind of. Yeah, they've got plenty extremely full top of funnel marketing. But when you look at like it doesn't. It seems like if they just focused on students. Right. They seem to be in this kind of place where they're. They're thinking about exploring a lot of.

Different markets for sure.

Different markets. But if you look at the way that ChatGPT's usage basically tokens generated dropped as the summer started. That's a good. It's a good argument to say Cluly could just focus on. On products for students that will eventually graduate into the workforce and things like that. But anyway.

Well, they didn't make the top AI agents by revenue chart that Anand Sanwal shared said, who would you put your money today to win in the long term? The top was cursor at 500 million in annual revenue and they're making $3.2 million in revenue per employee. Glean's at 100 million. Mercour's at 100 million million replica.

How are you defining an agent?

I think it's just AI startups. Anyone who's used the AI agent like meme basically stolen.

The agents are gonna be pissed about this stolen valor.

I don't know.

It's hard to call Merkhor and I.

Agree with that one that one feels particularly odd. But in Glean.

Even Glean is enterprise search.

Yeah, search. But I would say Replit is an agent company. It's kind of like every AI companies now. Now a now an agent company. Harvey, you know, hasn't really branded themselves in agents.

But it's interesting like you're not putting Claude. You're going to put cursor on here which generates revenue from Claude code.

Yeah.

And is a gentic ide.

Yeah.

You're not going to.

You don't put drop. It is crazy. And also not including chat GPT which has deep research, which is probably the most used AI agent in the world right now. I have to imagine just by number of queries there it's missing some stuff. It's missing some stuff, but it's okay. Also Devin's not on here, which is kind of odd, but that's the nature of these lists. It's CB insights folks. They do the best they can and you can do the best you can in the watch game. If you head over to getbezel.com your bezel concierge is available now to source you any watch on the planet. Seriously, any watch.

I like this post from Aaron Raleigh. He says, imagine you're one of the smartest AI engineers in the world. You just joined a company that has access to a GPU cluster worth tens of billions of dollars. Your work can one day help solve all diseases, create unlimited abundance for humanity. But you first have to build Spicy mode.

Spicy mode. Hey, you got to get the users in the door one way or another. You got to get the training data, the revenues important, but also the training data. You got to get the revenue ramp on board.

They call this the Japan strategy.

The Donald Boat saga is continuing, by the way. So Beth J. So said. Maybe Donald Boat is just AGI trying to self assemble itself outside of XAI servers. And Donald Boat said, let's just get this over with. I need to take gramps to get his new hearing aids fitted tomorrow. And shares a picture of A ASUS ROG Swift 27 inch 4K QD OLED gaming monitor which I believe Beth J. Zos purchased for him and then posted about. But the but but but. The Donald Boat saga continues because Ryan Peterson posted the difference between who you are today and who you'll be in five years is almost entirely made up of the books you read between now and then. And Donald Boat posts a picture of a Penguin classic the day I bought Donald Boat a printer and is asking Ryan Peterson to buy him the Epson EcoTank ET2800 wireless, all color, all in one cartridge free. Supertank printer with scan and copy for $189. It's a reasonable request, but I feel like he might be getting over his skis here because what does he need a printer for? He has the gaming PC. The narrative's getting a little muddy, Donald. You might need to hone it in. This isn't exactly gaming peripherals, so I'm a little bit confused, but the posts are still performing.

Building out a home office at this.

Point, I was talking to him, he was saying he's going to go for Porsche at some point. He should just get a house at some point.

It really is a great way to trigger people to flex on the timeline.

For sure. For sure.

And there's power there.

He should go to someone. The next one he posts, he should say, I want you to Larry Ellison.

How about a Malibu house?

How about a wander? He could find your happy place.

Find your happy place.

Book a wander with inspiring views, hotel grade amenities, dreamy beds, top tier cleaning and 247 concierge service. It's a vacation home.

But Donald, Donald Boat.

Get on wander. Start sending wander links around. You'll get.

We got you with a wander. We got you with a wander. Reach out. Joe says just learn that nobody ever committed the Wall street bowl sculpture. A guy just built it himself and dropped it there.

You can just do things.

On October 19, 1987, Black Monday hit US financial markets in the country entered a very difficult period. D. Modica recounted that he felt indebted to the US for welcoming him and enabling his success. Wanting to give something back, he conceived the Charging Bull sculpture. D Modica spent the next two years two years creating the 16 foot bronze statue. Reportedly financed financing the 350,000 cost himself. The sculpture was created in his Crosby street studio and then cast using a local foundry. Once complete, D. Medica spent the next few nights watching the police patrols on Wall street, trying to find a window of opportunity. Before dawn on December 15, 1989, Di Modika arrived with a group of friends and the sculpture on the back of a truck to find a 50 foot Christmas tree had been installed during the day exactly where he wanted to place the sculpture. With only four minutes between police patrols, he announced, drop the bull under the tree. It's my gift. The late night event went on to make news all around the world, including the front page of the New York Post.

This is incredible.

Dimadica stayed by the sculpture to greet the morning commuters. However, while he was away for lunch. The New York Stock Exchange arranged for the sculpture to be removed.

Boom.

Due to public demand for.

We love New York Stock Exchange. We love.

It was under different leadership.

It was different leadership. They would be totally cool.

Parks Commissioner Henry Stern arranged for the sculptures installation at bowling green on December 20, where it can be found to this day.

Amazing.

More people need to be making statues of this.

There's a couple companies that are making statues. I haven't heard of any of them just dropping them in public places.

They're also not using metal. Yeah, something about a metal statue.

Metal statue is pretty good. Anyway, Megan Brabrowski says really good story from HK Knight SF about Mark Zuckerberg's compound in Palo Alto. Here are the 11 homes he owns in the Crescent park neighborhood. Oh, interesting. He's doing the. He's doing the Ken Griffin thing. The Citadel thing. Creating. Playing Risk.

Yeah.

Territory. Yeah. Monopoly. Well, hopefully at some point he'll be able to, you know, unify all those properties and build something really special. Turn it into a museum one day.

Yeah. Accelerate. Harder says OpenAI is really in a bit of a bind here, especially considering there are a lot of people having unhealthy interactions with 4.0 that will be very unhappy with any model that is better in terms of sycophancy and not encouraging delusions. And again, there's been a bunch of. I mean, Reddit has just been going off.

Dude, did you wind up watching that TikTok compilation I sent you?

No.

Insane. So, I mean, good job on you, because it's an hour of brain rot. But. So there was a woman. It's an hour of TikToks, back to back stitch. So she posted, talking about. No, no, no. So it's one woman. So she apparently has adhd and she is an ADHD coach, which is not a licensed therapist, which is kind of odd. So she makes tiktoks about how to manage your adhd. Cool, Fine. Anyway, she was going to a psychiatrist to get medication to manage adhd, and by her own admission, she fell in love with her therapist or her AI therapist. No, no, this is a real person, a real psychiatrist. So she goes in and the psychiatrist is asking her how she feels about her day, and she's like, wow, this man cares for me. I'm in love with him. And so she is basically accusing him of, like, sort of crossing professional boundaries by not crossing professional boundaries and, like, putting up walls as shields and being manipulative. And it's this, like, big, long story. But at one point, point she starts talking to, to ChatGPT, but she gives it a name. And so she's like, I'm talking to Henry, which is the instance of the AI that I'm talking to about my love affair, my unrequited love of my psychiatrist. And, and Henry is like, affirming me and being like, yeah, you are right. Like, like he should be in love with you.

He's crossing lines by not being in love with you.

Exactly. And that was. And so it became this big, like, storm on TikTok where she was getting kind of brigaded in the comments. She turned off comments. She's going back and forth. A lot of people are saying, like, oh, she's, she's, she's crossing a line, she's not doing well. Other people are supporting her. It goes all back and forth. It gets very messy and then people wind up doxing the doctor. It's just such a mess. It's a crash out, essentially. But it was interesting because it wasn't, it wasn't a singular, like, falling in love with AI narrative, but it was like, like the AI was not a great player in the role and, like, did not make the situation better, basically. But, you know, you could also wind up in a Reddit community that's, you.

Know, promoting people's reaction to 4O coming back. Here's somebody sharing screenshots from their chat with 4O saying, Is that you? OMG, please tell me it's you. That's why I don't, that's why it's. It's hard to believe it's real, but there's, there's so many instances of it.

I don't know. Yeah, yeah. I mean, it's certainly, I mean, we've seen enough, like, weird interactions that there's certainly some people out there that are having a bad time. So good luck to them. Hopefully they will get this sorted both at the model layer and also at like, the society level, where if you see a friend who's clearly going down some odd path, whether it's, oh, you're part of a really creepy Reddit group or, or that group chat is toxic, you know, or that AI chat is toxic. Like, we're simple.

How much time, how much time do you spend talking with, chatting with ChatGPT every day? Yeah, if it's, if they're actually having long, super drawn out conversations, it's worth kind of trying to help them understand probably what's actually happening. More importantly, back to email. Andrew Reed says Gmail search won't find you. The email you're looking for, but it may well find you an email that can distract you for a while. This is almost every time.

It's wild.

Almost every time. Which maybe it's a feature, maybe it's not a bug. Maybe it's like, hey, this thing that you missed is actually kind of important. You should go down this rabbit hole. But I'm really hoping that AGI can just make imessage search work.

That would be great too. It's really rough. The photos app search is getting better though. I've noticed that I'm getting a little bit better at tagging things. I searched for a picture of a newspaper. It found us reading newspaper. One shot of it. Very, very good. Didn't need to scroll and tag and I could imagine imessage getting there. Yeah, but it requires some new UX or something Anyway. Ronnie Coleman. Wow. What a journey. Mr. Olympia. Ronnie Coleman in his prime version versus now wearing the same suit. He was huge when he was Mr. Olympia.

Gotta find the tailor that made that suit.

That's a wild suit. We gotta get some of those suits. For sure. It's a great suit anyway. One of the greatest.

Yeah, we should get you. We should get you like a bodysuit that you wear under the suit.

For sure. For sure. Now we're talking. Look at how small his head looks on his shoulders. It's crazy. Anyway, Aidan McLaughlin says one under discussed element of GPT5 is it just hallucinates so much less sweeping away 80% of O3 era at zitridism and Marcus posting. But because we're good sports, we give them an evergreen batch of things to critique like gross margins. That's the next. That's the next round of Michael Truewell.

Says when we'll be able to add AI co workers to Slack to help take on miscellaneous operational work. Updating spreadsheets, creating channels, running cron job business workflows, conducting deep research operator style tasks. Same API as human bi directional DMS call shared links with persistent memory personality and access to all your team's tools.

Who owns Slack again? Who's been talking a big game about AI for the last two years? Mark Benioff. I think he. I think this is his game to live.

I logged into Slack today and it said AI is turned on for tvpn. Oh yeah, at the bottom.

That's great.

And I'm like, okay.

Some of the AI opened a PDF and I got like seven different AI assistant pop ups and I was like, I just want to read this PDF actually, I don't need anything. I'm just going to read this PDF and yeah, some of. Some of the spiky intelligences spike right through my eye.

Well, couple posts. Couple Grok posts. Rune says Travis Kalanick will solve quantum gravity with the help of Grok. So Grok heavy, please solve quantum gravity. Think extra hard. No mistakes. Thank you. I believe it scoops watching her unclasped bra. Grok, Are these real?

That's a. That's a 2024 TBPN post.

We weren't the first to do it, but it's a great format.

What else?

Well, I got to get on with Taipei.

Let's get on with Taipei. Thanks everyone in the chat. We had fun chatting with you today. We're going to be chatting more. Appreciate you all for watching and we will see you tomorrow. Have a great rest of your Monday.

Go. Give us a go. Subscribe to TBPN on substack, please.substack.com Appreciate that. We're gonna be investing a lot there and we're very excited about it.

We'll talk to you soon.

Cheers.

Bye.

See ya.