TBPN - Slop vs. Steel Showdown w/ Delian & Everett, GPT-5 Backlash, Trump Eyes Intel Stake | Bill Bishop, Jimmy Goodrich, Lennart Heim, David Stout, Cameron Schiller, Cyriac Roeding, NFM Live

Starting point is 00:00:00 Watch the TVPN. Today is Friday, August 15th, 2025. We are live from the TBPN Ultradome, the Temple of Technology, the Fortress of Finance, the Capital of Capital. We have the debate of the century, the debate of the year, a showdown between former Founder Fund, Founders Fund, colleagues. Friends turned foes. Friends turned rivals. Everett Randall, he's been on the show before Delian Asperuhov. He's also been on the show.

Starting point is 00:00:27 They haven't been mincing words, John. They haven't. They have been throwing shots back and forth. Every TBPN appearance. Yes. They're calling the other one out. Yes. And so they will be debating and investing strategies.

Starting point is 00:00:39 We're going to settle it today on the stream. The slop versus steel debate, which is better, high margin software or CAPEX intensive reindustrialization efforts. We will bring in Delian and Everett into the studio. Welcome to the stream. How are you guys doing? I like that background. Very good. Energized.

Starting point is 00:01:00 Here we go. We're going to be breaking it down live here. Yeah, we're going to be breaking it down live. Every time one of you gets a point, I'll put a little point on more. If things get out of hand, we'll be banging the gong and bringing order like it's a gavel. But I'm sure, I'm sure this will be, I'm sure everyone will be civil. Oh, I'm sure. Keep the name calling to a minimum.

Starting point is 00:01:22 Good to have you both. Thanks so much for being here. Let's kick it off. Let's start off. What's your least favorite thing about the other person? I'm kidding. You could kick it off with the original story. Like how did this all start basically?

Starting point is 00:01:34 Yeah, yeah, yeah. Give us some backstory. Yeah. You want to give it? Yep, I'm happy to. So we were back at Founders Fund. We were starting to beef up our CRM and data science efforts. And so we were integrating some external data into our CRM, figuring out how we could filter

Starting point is 00:01:54 opportunities better to each of the investment team professionals. And we were looking and we were looking at the different data. I was like, ah, it'd be really nice if we could filter this by gross margins so that all of the negative gross margin companies that come into our CRM, we could give them all the DELION because it seems like those are the types of companies that he loves to invest in. The rivalry between the low gross margin side of the house and the high gross margin side of the house was born then. Okay.

Starting point is 00:02:26 And Deline, justify why, why? do you like these businesses? What are you in? Is that even a fair characterization? Fair characterization. I think, you know, my sort of one line would be, I'm not sure that, you know, sort of gross margin is actually like the right thing to focus on in a business, especially either sort of early on.

Starting point is 00:02:45 What you want to be thinking about is obviously EBITDA margin in particular terminal, you know, EBITDA margin. And so when I think about the like, at least founders filled ethos to, you know, sort of investing, we think that that terminal EBITO margin mostly is determined by ultimately how much about, you know, monopoly your company can either be. in the long term. And so if you look at, you know, sort of mag seven today, obviously there's a decent chunk of them that, you know, have some phenomenal, you know, sort of gross margins, and those tend to be the ones that are a little more software oriented. If you look at the one that is at the

Starting point is 00:03:09 biggest scale and has the best eBedat margins, it's the one that is the most, you know, basically hardware oriented. For sure, some of it propped up by like Kuda and they're like, you know, sort of software side of the house. But like, Nvidia is the one that is performing the best of all of those. And then even if you study within those, you know, which of those, you know, companies on the hardware side have monopolies versus, you know, sort of not, you see it's the one that with the monopoly, you know, clearly outperform the ones that don't, right? So Tesla, obviously, in that, you know, sort of mag-7, but a part of why they, you know, sort of suffer much worse margins than, like, an Apple or Nvidia is because, like, they actually do have, you know, you know, competition.

Starting point is 00:03:39 And so my general characterization of, you know, sort of SaaS is people always, you know, sort of study their original, you sort of gross margin margin, but weren't burdening in the, you know, sort of cost of sales, marketing, et cetera. And because you just have much less of monopoly, typically in SaaS, that ends up totally, you know, hurting your EBITO, you know, margin profile. So take like the like the favorite, you know, terminal scale thought of as a monopoly, you know, sort of SaaS company that, you know, I'm sure loves. Salesforce. Their market share and all things CRM is 25%. And so that's why you end up seeing like, yeah, gross margin profiles only burdened by like, you know, cloud, you know, sort of cost.

Starting point is 00:04:12 But they're even out margin profile that is like, you know, sort of 40%. And so, you know, the reason that I like these negative gross margin businesses is, yes, they're like tougher to start. They may be more equity intensive at the beginning, but end up with way better, you know, sort of terminal margin profiles versus, You know, Ev loves to, you know, invest in the, you know, AI slopcos that might have early gross margins in revenues. But you know, so, Ev, what's the bull case for software? What's the bull case for SaaS? What's the bull case for AI slopcos? Look, so to quote the godfather Neil Mehta himself, the laws of great businesses are the laws of great businesses.

Starting point is 00:04:47 The job of a business in the capital society is to maximize and find the efficiency frontier, three things. Roik, a.k, a return on invested capital, the amount of capital you can actually deploy and how long you can deploy that amount of capital at and above market Roik. There's a lot of different framings for the paths to do this and how companies can actually do this. The one that people like in tech circles is Hamilton-Henlars, seven powers. A company accumulates power in the form of scale economies, network effects, whatever power you want to take, and then uses that power to produce above market Roykes for as long as possible and with as much capital invested in the business as possible. There are great Adams-based businesses that do this. There are terrible Adam-based

Starting point is 00:05:31 businesses that don't do this. There are great digital businesses that do this. There's great, or there's terrible digital businesses that don't do this. I mean, you want to hear about a great Adams-based business that does this, listen to the acquired pod on Costco. Like, it's certainly not like a Adams versus SaaS thing necessarily. The advantage that digital businesses have is that in like in this process of producing above-market royke, for a long time is that their product form factor and the way that the distribute their product lends itself more to the process of creating power, I'd argue, than most Adams-based businesses. So if you think about like network effects, the best place to create network effects is in a

Starting point is 00:06:10 digital marketplace like an Uber and Airbnb or a DoorDash. And so there's a lot of these forms of power that naturally lend themselves to digital products and the scalability of digital products tends to be a lot greater than physical products. And so you can see these rapid growth trajectories like we're seeing from Open AI Anthropic and many others. When did you guys find common ground? Was it in the e-scooter era, the sort of 15-minute delivery era? Were you ever able to kind of come together and say,

Starting point is 00:06:38 like, yeah, this is, you know, we can both agree that this is not it? Or good, or good. I mean, to be fair, Kleiner Perkins, Founders Fund, have both invested in Figma, Stripe, Airbnb, There is some portfolio overlap. Rippling too, right? Rippling as well.

Starting point is 00:06:54 There's a couple, modern health, I believe as well. There's a few others. But yeah, to Jority's point, where else is the common ground and where else is the divide? Or the consensus in the disagreement. Yeah, is you to say, you know, we're texting before this

Starting point is 00:07:08 of like, you know, what are sort of two companies that I think, you know, both of us were enthusiastic about in, you know, sort of 2021, that actually both of, you know, sort of trended well, but are, you know, sort of counterpoints our two arguments. And the ones that we kind of came up with were, you know, in 2021, I was really, you know, high conviction on Hadrian in 2021. I was super, you know, high conviction on Rippling. Both those investments have, you know, performed quite well over the last couple of years. But look, you know, sort of wildly different in terms of, you know, profile. You know, rippling, like many other, you know, sort of SaaS companies does end up having, you know, an initial, you know, very high gross margin, but does still have to spend a lot on sales and marketing to bring in, you know, sort of net new customers. Hadrian on the flip side, deeply, you know, sort of negative gross margin to start.

Starting point is 00:07:49 But now as they've gotten to scale, they actually have like super limited, you know, sort of sales and marketing spend because there's only like, you know, 10, 15 customers that matter. And the moment that you're delivering for them, they just proactively start, you know, sort of throwing revenue, you know, at you. And so, you know, I think there are times where, you know, both of our, you know, stories, obviously, you know, can play out. The thing that I'd be curious to hear from, you know, sort of average is to actually, like, compare and contrast, you know, you know, you're bringing up, you know, some of these digital businesses, you know, that end up having these, you know, network effects. I would kind of argue that like, you know, the like 2010s negative gross margin businesses, like, you know, the like Uber DoorDash, you know, types. I think of as more as like, you know, Adams businesses, but there was a whole set of investors in like the mid-2010s that were generally unwilling to approach both Adams-based businesses that started with negative gross margin, but even some of these local marketplaces that started with negative gross margin that like swore off of the Uber's, the DoorDash, et cetera. you know, it's very clear that Uber DoorDash through, you know, lots of investment, through building out these local, you know, certain networks of, you know, both supply and demand, we're able to, and, you know, drivers were able to eventually get to a point where now they actually, you know, have very attractive, you know, sort of financial profiles. Today, the equivalent of that is, like, there's all these investors that, you know, back in the 2010s would have refused to invest into any company that had negative gross margin and are all now pouring cash into both the, like, AI application layer companies and the, like, you know, foundation models that all have, like, ridiculously. I mean, I forget to.

Starting point is 00:09:08 I think it's girly is nonstop, you know, not my favorite person in the world, but Gurley is nonstop talking about like, you know, what is going on here? They're telling a buck for 90 cents. So I think it's an important example because you had that, you know, plenty of examples of these chained losses during that era where a restaurant was selling something below cost to a platform that was selling something below cost to a logistics provider, an individual contractor that like maybe wasn't actually making money if you factored in depreciation and fuel cost of their vehicle. And that ultimately worked out, right? DoorDash is a massive,

Starting point is 00:09:43 fantastic business based on the power of the American consumer. But when you compare that to today, where a lot of the conversation on the timeline this week has been the margin profile of this new generation of software companies that has to pay a lot for sales and marketing, but also inference. And so I think like the debate should really be, you know, continue to be around just how quickly will the cost per token fall. And I think a lot of people have a lot of confidence around that. But I think that that is the key thing that Everett's sort of like broad investment thesis right now is dependent on. Yeah, like FD, think there's going to be that same path of like Uber for a while had a bunch of negative gross margin people going into it. Like do you actually think that a cool?

Starting point is 00:10:32 I want to pull this post up. Everett actually posted this. January 31 of 2024. So over 18 months ago, he said, I'm making a real effort to not take for granted the $3 Uber across town era of AI. And I hope you are too. And so I guess the question is, because then a bunch of people,

Starting point is 00:10:53 I thought it was a good point. I thought it was a hot take then. And I think then a bunch of people kind of parroted that take all over the timeline. Stole your whole flow, as you'd like to say. But I guess the question is like, Are we in some sort of different regime right now where the traditional gravity and like fundamentals of software investing have changed because we are out of the zero marginal cost era and does that impose risks to the strategy that, you know, have you sort of employed or like

Starting point is 00:11:23 we're kind of putting you in this whole in this box? But if the fundamental structure of zero marginal cost era is going away, that presumably forces like a rewrite of your logic around investing, I would imagine. Yeah, I think that the biggest variable that's changed from the 2010s, SAS era to today, is that in the 2010s, and you basically made this point without making it, Delian, though, is that the thing that was missing from your talk track is that the competitive intensity of SaaS during the 2010s was much, much, much lower than it is today. Like, during the 2010s, there was an entire crop of companies in the 2000s, but

Starting point is 00:12:02 especially in the 2010s, you could basically pick either a vertical segment, you know, like HVAC or card dealerships, or you could do a horizontal function like the CRM or, you know, some very niche workflow for like the finance team. You could build a software product around that workflow around that vertical. And you really only had to deal with typically like two to three competitors. Like there really wasn't that much competition relative to what there is today. And there was less just like general pricing pressure, competitive pressure, just the general pressure that you actually had a lot with some of the digital marketplaces early on. And so like I think there was a whole crop of investors then and like the SaaS investors then, we're like, well, we don't need

Starting point is 00:12:45 to, we don't need a bunch of cash burn. And it's actually, it's a really unhealthy indicator if these SaaS companies are producing a bunch of burn because they're not competing with anybody. So if they can't like sell their product for good uneconomics on day one when the competitive intensity isn't very high, then they're probably not a very good business. I think the thing that's changed now is one, you have the change from zero marginal cost to actual meaningful marginal costs in the form of inference. And it's also just a hell of a lot more competitive than it used to be. And so you are and by the way, there's an immense, it was probably 10x more capital than there was 15 years ago to go into these companies. And so like every single category now has become like mini ride share or like mini Uber market where it's like, hey, there's probably a really big pot of gold at the end of the tunnel. And we need to be the ones that get first to scale.

Starting point is 00:13:33 And in a lot of these categories, the ones that have gotten first to scale have gotten a lot of brand equity out of it and have gotten a pretty resounding lead. I think the only other piece I would say, I lost my chance of thought. Yeah, yeah. But it's going to be basically it's going to be like a capital fight now on the, on the, on. on the SaaS side, I wonder if the contrarian trade around hard tech is entering a similar era, where it's become consensus. And so we're going to see more capital fights. And when a founder goes out and says, yeah, I'm going to do something crazy, but I need to spend

Starting point is 00:14:04 a billion dollars of CAPEX. People are just like, yeah, this could be the next base. Yeah, it made sense to have a capital war in ride share, but now we have a capital war in like this niche, agentic workflow in some industry that most people have never heard of. And then also a capital war for, $200 million for funding. Military boats and UAS and UAP, like all these different sub segments are going to wind up. If the, if capital war start popping up there, that could potentially be a headwind to Delian's model. Is that, is that roughly correct?

Starting point is 00:14:33 How would you, how would you, how would you fight back against that? Look, I think it's always, you know, sort of important to talk about, you know, sort of specifics here, right? You know, one of Eves, you sort of major investments in the last year is this, you know, company called captions that basically does AI captioning of, you know, various, you know, sort of videos on social media. When I think about, you know, handing, you know, sort of two Stanford grads and $100 million to go try and, you know, sort of replicate that, yeah, feels like, you know, they could, you know, go do something like that. There's like, you know, clear, you know, voice recognition models. They can go, you know, sort of pay on ads on TikTok,

Starting point is 00:15:04 et cetera. And you could probably go and replicate that. And so, you know, our one line, our founders fund is competition is for losers. And so, you know, I think I was a loser for investment. You know, shots fire. This wasn't spicy enough and you just delivered Deli, so thank you. Now, you know, if you take, you know, sort of two Stanford grads in $200 million and tell them, hey, I need you to go replicate this manufacturing facility and go start building a bunch of, you know, sort of satellites, reentry vehicles, you know, bioreactors that can actually survive the environment of space. Most, you know, sort of stand for grads, you know, can't go, you know, ask chat GPT how to go do that. And yet, yet, yeah.

Starting point is 00:15:41 But I haven't really faced significant competition, irrespective of the fact that, you know, all things, space facts. are thought to be, you know, sort of the hot new thing. To be clear, we use captions here on clips. We enjoy the captions app. We thank for making it possible and subsidizing our token. And there is a YC, there is a Y, a Varda-esque company. So you're coming for you. You know, Indian Varda, I think will be a little bit less competitive than, you know,

Starting point is 00:16:04 so Indian captions. Also, if you're the caption CEO and, you know, founder's fund is trying to invest in your next round. Please still let us do that. A helpful counterpoint for me. Delian, you were, you were correct that it was getting, it was getting too friendly of a debate. I did want to make sure I could, I could pin this one on you. If you can recite the equation for a return on invested capital, I will victory to you and I will donate $5,000 to a charity of your choice.

Starting point is 00:16:31 Hopefully he's got Cluley running. Yeah, exactly. My like, you know, equivalent forever, it will be if you can explain, you know, basically why you can't create microgravity down here on Earth, I will also donate 5,000. to a charity of your choice, but I don't think of it, you know, I may not have the, you know, basic understanding of business physics, but you don't have the basic understanding of physics and one's more important about understanding the universe around you. Okay. I mean, I'm, I'm pretty fixated on the 2035 Midas list. That's really kind of the final. That's this.

Starting point is 00:17:00 Have you been on Middink yet or I forget whether or not you've made it up there. Oh, taking shots. No, not even on the brink yet. You know, Lee joins, you know, KP after you and she beats you laughing me. It's okay. It's okay. It's okay. Eventually we're going to bring back the extra names in Kleiner Perkins. It used to be Kleiner Perkins, Coffield, buyers. It's going to be Kleiner Perkins, Randall, Braswell, eventually. We're working on it.

Starting point is 00:17:23 We're working on it. We're pitching it. Where should we go next, Jordy? I guess, Everett, how are you, how quickly, like, how much should people be fixated on the cost per token with these frontier models over the next six months? like how long can venture capital sort of like backstop these chain losses? Yeah, I think that the way to delineate the whole so obviously like I think there was this

Starting point is 00:17:54 kind of consensus narrative that like every, you know, 12, 18 months token costs were going down an order of magnitude. I think that did hold for a while. I think what you've seen now is like actually for frontier models that started to peter out of it. And like pricing has actually started, it's still going down. It's not going down nearly as much as it, as it used. used to when when we were kind of in the in the like the meat of the curve of capability

Starting point is 00:18:17 improvements on frontier LLMs in terms of pricing curve. So I think that the way that you want to delineate it is like there's a certain like what I always tell everyone is that like there hasn't been a chat GPT query since GPT4 that like my mom hasn't been able to ask and have it answered by the model. So there's like the mom test of models where like there's a growing subset of tasks like economic or knowledge tasks that the models are tasked to do that no longer need frontier intelligence. And when you're not on the frontier, either through open source or just the cheapening

Starting point is 00:18:51 and disilling of older models, like the price still falls off a cliff. There's going to be a very, very large set of tasks that models do that are not on the frontier and those are going to continue to get dirt cheap. I actually think that at the frontier, you're probably going to see continued price decreases on a per token basis, but nowhere near what you saw before, which was like this order of magnitude decrease on a very regular cadence. And so I think for like depending on the company, it's going to depend on one, if you've actually built a company that has enough power where you have pricing power

Starting point is 00:19:21 where you can price above the kind of marginal token price from the actual model providers. And then two, like how much of your inference actually needs to be at the frontier? Like how much of your inference can be an older model that's much, much cheaper versus how much do you need to do on the actual frontier? I think that's what you're seeing like, you know, everyone loves to talk about. Cursor and Chris Pake over at Pace Capital had this really great kind of like mini essay, I think only like last night or a couple of nights ago. And he talked about like, no one knows if Cursor has power yet because, you know,

Starting point is 00:19:50 coders and developers, they're very, very, like they're tastemakers. They're very good at understanding the quality of the models and how much inference they're getting. And there's a lot of price sensitivity for them because they have a really good understanding of how much inference they're getting. And so no one really knows, I think no one can definitively say whether a lot of those types of companies have actual power with their users or if they're just drawn to an interface for frontier models or not.

Starting point is 00:20:14 And so I think that's what everyone needs to be looking out for is those two things. Like, do you actually have power? Like, will people give you margin above the marginal cost of tokens? And then two, like, do we even need the frontier inference for the vast majority of your product or is that, is there a lot that you can offload to cheaper models? Yeah, I mean, I guess your counter, you said there, Everett, is that, you know, a majority of what the foundation models are providing in terms of the user value there in users is starting to be, you know, sort of obviated by the, like, you know, historical generation, even some of the

Starting point is 00:20:42 ones that are, you know, sort of open source. So it seemed to imply that where value is accruing and where you'd expect, like, the highest revenue growth wouldn't necessarily be at the foundation layer, but you'd see it more at the application layer since those folks can swap models out. But, like, in reality, that's, like, literally just not what actually is happening. Like, if you look at which companies are, you know, sort of fastest on, you know, sort of revenue growth, user growth, et cetera, it is the foundation model companies. It seems like a part of it is that they also have, you know, sort of the most pricing power where, yes, you know, your mom, you know, uses GPT4, but, like, she's not the one that's necessarily paying, like, you know,

Starting point is 00:21:14 $100, $1,000, $10,000 per month versus the true frontier capabilities on, like, you know, AI coding, the pro users, the one that actually do care about, you know, maybe your mom is fine with 115 IQ model, and that's, like, fine for the rest of her life because she's just, like, not asking it that difficult of questions versus the people that actually are willing to, you know, should pay are the ones that actually do care about the 140, 160, 180 IQ. Again, maybe at some point that gets, you know, should commoditize as well. But my sort of counter to you would be, you've made this argument that it seems to imply, hey, you know, things will accrue to the AI application layer, which if I understand your guys' portfolio is largely where you guys invested. But in reality,

Starting point is 00:21:48 that's not what's played out. The like places that have captured the most, you know, revenue growth, the most market share have been the ones that are actually pushing the true frontier, you know, of the, you know, so technology forward. And so far, at least in last 18 months, your thesis is not playing out at all. Well, to be clear, isn't, isn't it somewhat widely understood that Anthropic has negative gross margins as well? So it's, it's not like they're doing. Like, Ev's point was that you want to invest in these companies that have the, you know, sort of seven powers. And like, you know, in the, you know, days of like Uber, you know, DoorDash, etc. that did end up, you sort of translating. It seems like Invigia has the most power, then the foundation

Starting point is 00:22:24 model labs, maybe then the application layer. We'll see how much power develops in the application layer. But, Ev, we'll let, we'll let you respond. Oh yeah, I was going to say that basically what Delian said was just wrong because even though it is even though like if you if you think about okay, like let's take like whatever Open AI and Anthropics recently reported revenue run rate is, the majority of all of that or at least the plurality of all of that is chat GPT. And chat GPT even though it is served by a foundation model company is an application. It is a consumer subscription that has an immense amount of power. It has an immense amount of branding like, you know, it is the only is like the first billion plus user. consumer application that's been developed by a new company in a really long time. And so I think that, like, you could put whatever models you wanted through chat GPT at this

Starting point is 00:23:10 point, and it would not knock it off of its perch. I think that is power. Like, you could run Claude3 Sonnet through chat GPT, and I guarantee people, like, the average user wouldn't actually know the difference. And that, to me, is power. And just because the foundation model companies are producing apps themselves doesn't mean that it's not the application layer that it's, that is accruing the value. Okay.

Starting point is 00:23:29 my question is, you know, you've got, you know, opening eye with the best possible consumer application layer, you've got Anthropic that, like, shifted over to positive gross margins and those margins that are expanding, and yet, clienters not investing into either of those foundation model, you know, so companies. Why? I cannot comment on our current investment activities. Okay. Can you comment with, can you count on down, Donald? I mean, look, I just, do you like making money or do you like, you know, you're going to Can you comment on Donald Boat?

Starting point is 00:23:58 Have either of you bought anything for Donald Boat, the notorious e-begger on X.com, the Everything app? Like my little brother, you know, played the Uno reverse card and tried to get Donald Boat to buy him something. So smart. You know, contrary to a rohov nature. Let's talk about revenue quality, because I think that you guys run into this

Starting point is 00:24:18 in your respective domains every single day. Just like in AI, you can have low quality revenue. like that might be the explosion of like consumer prompt to app activity, you know, might not be the highest quality revenue. Meanwhile, on the hard tech side, if somebody gets like a random, like, cyber or like experimental, gets like experimental budget from some branch of the military and it's like a, you know, fixed length contract, it's not necessarily the right strategy to slap like a 50x revenue multiple on it. So like what's your view on both of those? And then I want to talk about if we

Starting point is 00:24:57 should get into if accounting rules even matter at this point. Yeah, yeah, for sure. Yeah, I mean, in hardware land, we think about this all the time of like there's clear differences in quality of revenue, everything from like, you know, defense, you know, program of record. You have to value that very differently than even like a $50 million, you know, SBIR. And so it has been interesting to see a bunch of investors coming into this field where I think there's a lot of pre-existing 10 years of rules around software of like what healthy revenue looks like rule 40 there's all these things that like you know even if you're somewhat unsophisticated infinite blog posts when you look at that in the world of like hardware and defense you know

Starting point is 00:25:31 sort of investing or aerospace there aren't like infinite blog posts for people to study and so I admit that I'm sometimes amazed when I watch people come in even for I should never you know sort of trash my own portfolio but sometimes even my own portfolio companies I watch people invest in them and I'm like wow like you just have a deep underappreciation for just like how long this company has until gross margin flips to like positive, how long it's going to be until they're actually, you know, sort of ready to go scale revenue. Even if it on the back end, it might be attractive, it may be years and years for them to, you know, sort of get there.

Starting point is 00:25:58 And so, yeah, I see huge variation on that. And then mostly what I end up, you know, sort of seeing is people just come in and like slap a 10 to, I even saw 100 X rev rate multiple on this, like, hardware company recently. And I was like, holy shit. Wow. People like not have an IRA for a long time. Oh. Yeah, I think so Delian's hero and close mentor, Bill Gurley, had an essay a long time ago called the 10x Revenue Club.

Starting point is 00:26:27 And I think it's like a good abstraction for kind of like tech revenue quality and like what makes up revenue quality. And it's things like, you know, how durable is the revenue? Like if you sign a customer, are they going to stay for a year or 20 years? You know, how much contribution profit is going to come off of that revenue stream over time? All the basics. And I think you can like take those same building blocks and apply to AI. I think there's several things that are worse for AI. at least than relative to SaaS for now.

Starting point is 00:26:51 So generally gross margins are lower, which means contribution profit coming off is lower. I actually think that like depending on the category, you could have customers that are more sticky or less sticky. Like I know the meme is that everything's experimental run rate and none of these customers are actually sticky. I think we see something very, very different among our group of portfolio companies.

Starting point is 00:27:10 I think that the biggest lever that didn't exist in SaaS, that exists in AI that could be a huge call option boon for the revenue quality of AI is the actual contract sizes as people start to eat into potential labor budgets. I know this is still kind of like inning one and inning two. And it's also like a little bit of a meme where everyone's like, oh, it's going to replace labor and labor's 10 times SaaS. And it hasn't really happened yet.

Starting point is 00:27:30 But I think if you look at some of these coding tools and you look at something like Claude code, that is the first place where you can really actually say like, no, this is replacing the labor that a developer would do. And it is paid for on like a metered consumption basis. And the monetization numbers we're hearing around developers using cloud code are pretty crazy in terms of like, wow, that's like you're paying like one-tenth of like a developer's full-in cost to a company on an annualized basis for this product. And so I think that the like, the thing to watch is like durability of revenue plus the amount of actual revenue that a customer

Starting point is 00:28:00 can give you. And I think that you're going to end up, or the amount of gross profit that a customer can contribute over time. And I do think as some customers crack these agentic products that look and monetize more like labor, AI revenue could actually exceed the quality of SaaS revenue just because you're getting so much more gross. profit per customer or like customer relationship than you would on the SaaS side, even though there are clearly things that are worse about AI revenue at this current point in time than there are about SaaS revenue. Dillon, how do you think about the, the moral imperative of a venture capitalist to invest in

Starting point is 00:28:33 positive sum versus zero sum markets, this idea that, you know, you're reindustrializing America, you're saving the West versus moving chips around, you personally, versus moving chips around the poker table taking taking from some legacy you know web 1.0 company and putting it into an AI company what's your what's your thinking and argument there is is is a a market beating ROIC all that you need yeah I you know I think Peter always reminds us like you know our number one job is deliver returns for is RLP's and so I actually tend to not try to you know sort of overly moralize when like analyzing the things that I want to you know sort invest into for sure when

Starting point is 00:29:16 it comes into like policy and I'm in DC and I like need to you know sort of report to you know the Security Council that you know Bill Gurley is at either sort of Chinese spy and like the investments that he's making should probably be banned from the United States. Yeah for sure there I have you know sort of moral imperatives and things that influence that may end up you know shifting RIC right so you know but when it comes to you know like which literal investments are we making I think of it is just like yeah you just have to you know sort of make the you know sort of best possible investments irrespective of you know sort of moral imperatives but in some ways I tend to think, it turns out actually if you, you know, go to immoral, then that ends up,

Starting point is 00:29:50 you know, sort of affecting R.C. So, it may, and the last thing that I would at least, you know, sort of, you know, close on, you know, for my, you know, sort of question, you know, for Everett, is, you know, one of the upsides of Founders Fund is, you know, we're very, you know, sort of, you know, let's say, non-centralized, distributed, you know, not many, you know, sort of rules, which, you know, ever, for some reason, you sort of chose to leave. And so I know nowadays, everything that he says publicly, you know, probably five comms people and five compliance people that need to, you know, sort of approve it. And so I have my only request to you is, you know, to bleak twice if somebody's, you know, got a gun behind the camera, threatening to shoot you,

Starting point is 00:30:26 never say anything that, you know, goes through an off script. That's all you got to tell us, brother. Yeah, let us know. Hey, our wonderful marketing partner, Allie, is behind the camera with a green and red paddle, and she has to raise the red paddle yet. So that's great. That's great. Well, thank you both for joining. Yeah. Are you worried about Uncle. Sam potentially having sharp elbows now that we're hearing about the federal government taking a stake in Intel. Any concerns about him going down the stack into the early stage game competing for those seed in Series A allocations? Look, if Trump Capital wants to mark up some of the reindustrialization companies, I'm all for it, baby, cheap cost of capital. You're all for it.

Starting point is 00:31:07 I'll say two things. I'd say one, I think that the EV of like the enterprise value of founders fund probably 3x the night that Trump got elected. So I don't think Delian would complain about that. And then as a parting gift, Delian, you know, I think this conversation's been great. And it's made me realize why you want to build factories in space because your math on Earth doesn't make any sense. Well, thank you both for joining. This is fantastic. You're both good sports. We'll have to do this again. I think it might be a draw. We'll have to have you both back soon. Thanks so much for hopping on.

Starting point is 00:31:39 Great stuff. We'll see you guys later. Cheers. Let me tell you about ramp.com. Time is money, save both. Easy Use corporate cards, bill payment, accounting, and a whole lot more. A whole lot more. Do accounting rules matter? Yes, they do. Yes, they do. And you can enforce accounting rules on ramp.com all in one place. Go to ramp. That was, that was beautiful. Two, two former colleagues barely holding back from saying things that they would ultimately regret. But they did. They did a good job. They did. Yeah. I like the debate format. We should definitely do more of those. I think that was a lot of fun. I think the chat and enjoyed it. My favorite rude comment in here, oh, we got Andrew Reed in the chat,

Starting point is 00:32:18 Everett versus Tellian, who can grow the most average beaer? Oh my God. Thank you for watching, Andrew. Let us know when you've selected an opponent and we'll have you on the show to debate someone. Yeah, I think we need it. We need to get the Holy Trinity. Yes, yes. If you're new to TBPN, the Holy Trinity is, of course, the three venture capital firms that have done a seed deal in a now hyper scaler or now mag seven company so that is Sequoia capital founders fund with meta originally Facebook and Kleiner Perkins of course and so the holy trinity are the three most storied venture capital firms in the valley much like the three famous watch brands the holy trinity vachron constantaq philippe and adamar paget of

Starting point is 00:33:05 course over in switzerland um if you enjoyed this stream and you want to make your own stream get on restream one live stream 30 plus destinations multi stream and reach your audience wherever they are if you're the backbone if your company and you're doing the launch it streaming is the way to do it you don't have to try to poach Ben or anyone else no you can just go to restream it's fantastic check it out anyway going back to the death star the vague post we're one week out from GPT5 how have how is your GPT? PT5 experience been. Also, Tyler, the chat wanted you to answer. Explain the formula for ROIC.

Starting point is 00:33:48 The formula. So, yeah, I think it's, I believe it's operating income divided by book value of invested capital, right? Oh, you got it. Clearly's running, clearly. That's off the dome. That's off the dome. I did not just look it up. And then, and then what did, what was Delian's rebuttal? He wanted why you can't achieve microgravity on Earth? Why can't you achieve microgravity on Earth? on Earth. Do you know that? You're a physicist. You study physics.

Starting point is 00:34:14 You should get this. I'll have to get back to you. Let me think about that. Let me think about that. I think it's just the, wait, I actually, I can't really explain it. That's kind of hard. I mean, I know that it's like Earth has a gravitational field

Starting point is 00:34:27 and we don't have the technology to reverse gravity, but I can't really tell you why we don't have the technology to create an anti-gravity chamber on Earth. Like I can't walk through the physics for that. I just know that you can't do it on Earth, but mostly because we've been trying to to get a gong in microgravity here on earth. I think it's challenging.

Starting point is 00:34:43 You can do it for like a very short amount of time, right? It's like when you see you ever see the planes. Yeah, that's not that's not actually microgravity. That's just falling, right? That's just falling in a pressurized capsule. Like you're still under the, you because you are, you are literally falling down to Earth during that. You just your surrounding environment is pressurized.

Starting point is 00:35:02 And so it feels like you're floating. But in fact, you're you're really just falling closer to Earth. Like when the plane goes down, you are, You are descending. There's no machine on Earth that will effectively like levitate something and reduce the force of gravity to zero on Earth or even reduce it significantly. Yeah, but like that's the whole thing. It's like from the observer. It's the same.

Starting point is 00:35:25 Yeah, but for for space manufacturing, for like growing a crystal. The reason that you can't do it on Earth is because it's too like it's not long enough. No, no, no, no. I think that even if you, even if you tried to do something in that in that plane scenario, like you're still subject to the force of gravity even though you're because you're effectively falling even though you're not even though you're not feeling like the relative the like the wind speed you still are under the force of gravity I don't know we'll have to figure it out we'll have to tell them back and explain it to us anyway what's your what's your

Starting point is 00:36:02 one week one week review of GPT 5 what's your takeaway Jordy I've been I've been it's been Fine. It hasn't been that drastic. I still find myself navigating between different models using the switcher. Have you turned on the legacy models? So if you go deep. I haven't gone into the hidden. No. So Tyler, you have though. Explain. Yeah. So if you just go into settings, you can turn on legacy models. Well, so all I have in legacy models is 4. So so 4.0 came back as a drop down. But you can go into the settings and turn on legacy models and then you can access 4.5. right? Yeah, 4-5 still. And 4-1-0-3. 04-many. Oh, so you can access them all, but it's tucked behind even more menu. I think that the Ben Thompson take was that they are not being bold enough as a consumer company and telling people, you know, the Henry Ford thing, if I asked people what they wanted, they would have set a faster horse. You know, I gave them one color of car, black. I didn't ask them for input on that. Steve Jobs did the same thing famously with Apple, made a bunch of bold product. decisions and then just said consumers I don't I'm not taking input you want you want a headphone jack too bad I'm taking it away you want an extra port on your MacBook you want what was the thing that they took away they took away all the ports for a while they had no ports for a while it's just USBC ports on the edge

Starting point is 00:37:28 and so and Facebook's been similar with the with the removal of the original feed and then the and then they moved away from a chronological feed to an algorithmic feed and there was a lot of pushback for that But Mark Zuckerberg channeled a mentor of his Steve Jobs and said, you know, I know that this is better for the long term. I know that this is better for everyone in the long term and that people will ultimately love this. And I think that's probably— I mean, the main thing is it's very interesting that this was reported by Alex Heath and the Verge last night. Apparently he got dinner with Sam Altman, I think, and Greg as well or some other executives.

Starting point is 00:38:04 And they said last night about an hour before the dinner started, Open AI. I pushed an update to bring back the quote unquote warmth of 4-0, which is what the Reddit, the Redditors of the world, the AI as my boyfriend, enthusiasts were clamoring for. So it's interesting to see how quickly they folded there. I think, yeah, clearly they made users distraught. I also think of, I imagine a lot of those users were paying the top tier subscription. And I, you know, again, it's hard to read too much into Reddit or really anything you see online, but a lot of people were canceling or threatening to cancel if they didn't bring that functionality back.

Starting point is 00:38:49 So anyway, Sam, I mean, I would expect going forward like constant changes and iterations, just like the YouTube algorithm is constantly changing. The X algorithm is constantly changing. These updates get pushed very incrementally. There's constantly tweaks that are happening. So Sam is quoted saying, I think we totally. screwed up some things on the rollout. On the other hand, our API traffic doubled in 48 hours and is growing. We're out of GPUs. Chad GPT has been hitting a new high of users every day.

Starting point is 00:39:17 A lot of users really do love the model switcher. I think we've learned a lesson about what it means to upgrade a product for hundreds of millions of people in one day. Yeah, it's always tough. He pegged the percentage of ChadGPD users who have unhealthy relationships with the product at way under 1%. I agree with that. That sounds right. But acknowledge that OpenAI employees are having a lot of meetings about the topic. Yep. Quote, there are people who actually felt like they had a relationship with chat GBT.

Starting point is 00:39:41 And again, in this article, this post that we read yesterday on the show that he posted eight years ago called the merge, talking about the inevitable point that humans and machines merge. He said the merge can take a lot of forms. We could plug electrodes into our brains or we could all just become really close friends with a chat bot. So he was aware. credit to Sam for calling this one pretty much perfectly because clearly, you know,

Starting point is 00:40:12 it's millions and millions and millions, you know, even if it's way under one percent, and this is, you know, tens of millions, right? Yeah. Yeah. Hundreds of millions of daily users. Yeah, probably like under a million. So hundreds of thousands of people, that's a lot that could have an unhealthy relationship. Yeah.

Starting point is 00:40:31 Okay. Way under one percent. So millions of people. Maybe. Yeah. They're not quite at a billion. active users. So and then and then a lot of those user international in terms of like the I mean not that really matters you want to be keeping everyone healthy, but yeah, we're talking about like hundreds of thousands of people that are potentially negatively affected. So they got to drive that to you know point one percent and then point oh one percent and then you know get as close to zero as possible. Yeah, there's go there's always going to be some people that you know use they read the newspaper and they go crazy. But yeah, you know, the more that you can do the better. Sam says you will definitely see some companies go make Japanese anime bots because they think they've identified something here that works.

Starting point is 00:41:10 You will not see us do that. We will continue to work hard at making a useful app and we will try to let users use it the way they want, but not so much that people who have really fragile mental states get exploited accidentally. Well, as they continue to iterate on the product, they have to use figma. BuildFesFigma. Build faster. Figma helps design and development teams build great products together. Go make something I'm figure to make today. We have a question from the chat for Ben. Kohler, a producer, was recently followed by Reed Hoffman. And the question is, did Ben get any Hoffman chat? Did he slide in the DMs or did he just follow you for updates?

Starting point is 00:41:45 I think just for updates. Just for updates. If you're not following Ben, you've got to follow him. He posts constantly. He posts constantly throughout the show when big things happen, when crazy stuff's happening on the stream, he's kind of like the premium feed. Like, you know, we put a lot of stuff on the main account. Ben's the behind-the-scenes guy.

Starting point is 00:42:02 So go follow him. Boys back year too. Yeah, the whole, lads, lads, lads. So my final takeaway from the Death Star Vague Post is that there was this viral image when, in the lead-up to the GPP4 launch, is this data visualization, and we can pull it up as the first slide in the deck. It was visualizing the number of parameters in the model, and this went very viral multiple times. So it was GPT-3 had 175 billion parameters, and GPD-4 had 100 trillion. billion parameters and so you see the small dot and then the huge circle and a lot of people were afraid by this and it was kind of this indication of exponential take off and we really

Starting point is 00:42:45 did see a qualitative improvement in just scaling up the pre-training run from GPT3 to GPD4. But GPD 4.5 taught us that pre-training scale is in fact not all you need. And the way to make a great AI product in the modern era is a mixture of techniques, experts, and researchers, you need a whole host of things, particularly with GPT-5. It feels like they RLed on a lot of different problems. And so to me, the Death Star Post represents an even bigger circle from that GPT-4 circle. So the Death Star is the biggest possible circle, and it's an expansion of that GPT-3, GPT-4, GPT-5. But couldn't you read this that GPT-5 is the Earth?

Starting point is 00:43:32 No, no. GPD 5 is the death star, or the perception of GPT5 as just being a bigger model is the death star. OpenAI blew that up. They blew up the metaphorical big circle with a model that isn't just bigger. Some analysis said the release is the router. The router is the release. And what that means is that the gain in the value delivered by this product is not just a bigger circle. It's a more complex coordination.

Starting point is 00:44:01 kill you. It's all the different X-Wings working together in tandem. And then the Millennium Falcon comes in and saves the day. That's the, that's the, that's when, you know, the model router acts like triggers a reasoning step and it thinks for a long time. The Death Star is the end of the pre-training scaling law. Well, let's get into some more coverage here from Alex Heath. Sam says, you should expect Open AI to spend trillions of dollars on data center construction

Starting point is 00:44:35 in the not very distant future. He confidently told the room, we have to make these horrible tradeoffs right now. We have better models and we just can't offer them because we don't have the capacity. We have other kinds of new products and services we'd love to offer. So obviously, agent making that more widely available, I think is what he's alluding to. He also thinks we're in an AI bubble. When bubbles happen, smart people get over-excited about a kernel of truth. If you look at most of the bubbles in history, like the tech bubble, there was a real thing.

Starting point is 00:45:08 Tech was really important, but the internet was a really big deal. People got over-excited. Are we in the phase where investors as a whole are over-excited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes. Yes. So he obviously contributed to this exciting.

Starting point is 00:45:27 in a sort of yeah I would say that he he didn't invest in very many competitors and if there's a power law here it could play out like the social media bubble where there was there was an immense amount of excitement around Facebook cracked it it's on the way to be a trillion dollar company and there was a belief for a while that that it would be oligopolistic and Twitter and foursquare and Pinterest and Snapchat would also be trillion dollar companies yeah but that didn't happen and if you you invested in those companies at unicorn valuations, you have not seen fantastic return on invested capital as opposed to you invest in a base market.

Starting point is 00:46:06 I just think it's fair to say that Sam helped get people overexcited and that in many ways he was saying, you know, with with GPT6, we might be, you know, discovering novel physics and curing what? Who's the we there? Is the we open AI? Or is the we every company that's raising in the valley, right? now. It's an important distinction. Speaking for open AI, but I think people are going to naturally yeah, totally. No, I agree. Take that as, as AI as an industry broadly in terms of its potential.

Starting point is 00:46:41 You know, so people might start thinking, yeah, maybe the 20th best LLM has a good shot at curing cancer. Yep, but the 20th best social network was was not worth 1 20th of Facebook. It was worth 1,000th of Facebook or 1,000th thousandth of Facebook. And that's the nature of these power laws. But my take is that, so the model, the release being the router and this shift towards open AI potentially shifting into dominating agentic commerce, having a monetizable free tier. This is actually a bull case for super intelligence. A lot of people on the timeline were like, oh, GPT5 was supposed to be like, you know,

Starting point is 00:47:25 an order or magnitude gain, something really qualitative, like you use it and just feels different. It's just 100% on all the benchmarks, whatever. It wasn't that. It felt very incremental. And a lot of people were kind of, you know, we're plateauing, all of that. But I think that, I think that shifting to a, shifting to a freemium model of a monetized free tier is actually a bullcase for building the trillion dollar cluster. And my thinking goes like this. So you can, you can build the first GPT, to GPT3 cluster with non-profit donations. Like $100 million gets it done. And that advanced them to that stage.

Starting point is 00:48:04 But to do the GPT4 training run, they could not marshal the capital in the nonprofit space. They had to become the for-profit. They had to get venture dollars in. And yeah, and so basically like the Shogoth demanded capitalism. This is the Nicolland take that artificial intelligence was sent back from the future to invent capitalism. Have you heard this take?

Starting point is 00:48:25 That's great. And so the idea is, you know, like you could not get to GPD 4, GPD 5 without a for-profit company with the promise of return on investment. And so you pulled in all the venture dollars. The question is to build a trillion dollar cluster, I think Mossa is going to be tapped out soon. I think Mossa is a card you can play once. I think that there is a limit to how much capital you can marshal in the private markets,

Starting point is 00:48:54 even in the public markets. I just think it's impossible to raise a trillion dollars necessarily. And that cluster must be funded, must be funded, it must be underwritten by a company that can justify a return on investment from their direct product. And so we're seeing this right now with Google and Facebook and investing their free cash flow. Yeah, they're investing their free cash flow.

Starting point is 00:49:15 Yeah, they're doing some debt. But I think to get to the really, really big numbers, the trillion dollar cluster, it's going to have to be built on just continual free cash flow investment from a company. Tyler, what do you got, Tyler? What do you think about like situational awareness, like nationalizing labs? You think governments can, can fund the way. We're about to nationalize Intel, so maybe that's a path down the road. I don't, I don't know. I don't, I don't think it's on the horizon anytime soon. Mostly because we're, we're just not seeing capabilities that would threaten.

Starting point is 00:49:50 It comes actually insane. So, so a week ago, when that. reporting from the journal on Leopold's situational awareness. Everyone is just dragging him, dragging him, dragging him, dragging him. Why is he long into his fund is his fund is probably blown up already. It's up 21% in the last five days. Oh, my God. Wow. It's a gong for Leopold, Aschenbrenner and situational awareness.

Starting point is 00:50:19 Anyways, there's some more, there's some more interesting stuff in here. I, I, I, I don't see it happening until the labs pose a threat to the U.S. government in some way and are so dominant. I don't think we're at that phase. We're getting into the danger zone. We're in like new Google territory. It's a dominant consumer app. So there's some more interesting reporting here from Alex.

Starting point is 00:50:43 He says Sam confirmed recent reports that OpenAI is planning to fund a brain computer interface startup to rival Neurlink. I think that neural interfaces are cool ideas to explore, says Sam. I would like to be able to think something and have chat GPT respond to it. And of course, I think it was the Financial Times who was reporting that Sam Altman would be a co-founder of this company, Merge. Does Fiji-Simo joining OpenAI to run applications imply there will be other standalone apps besides chat GPT? Sam Altman says, yes, you should expect that from us. He hinted at his social media ambitions, quote, I am interested in whether or not it is possible to build a much cooler,

Starting point is 00:51:24 kind of social experience with AI. He also said, if Chrome is really going to sell, we should take a look at it. Alex says, well, Altman has a lot of interest. It's not clear. It's not actually clear that running Open AI over the long run is one of them. Sam says, I'm not a naturally well-suited person to be a public company CEO, he said at one point. Can you imagine me on an earnings call? I then, Alex then asked if he would be CEO in a few years.

Starting point is 00:51:51 Sam says, I mean, maybe, maybe an AI is in three years. That's a long time. I love it. That's great. Some other. What else? I love Vanta. Automate compliance, manage, risk, improve trust continuously.

Starting point is 00:52:04 Vantz'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. So a few more points in here. Altman had notes on making GPD5. We had this big GPU crunch. We could go make another giant model. We could make that. and a lot of people would want to use it,

Starting point is 00:52:22 and we would disappoint them. And so we said, let's make a really smart, really useful model, but also let's try to optimize for inference costs. And I think we did a great job with that. Obviously, we're clearly wanting to be more competitive with Claude and Anthropics API business. On the AI device with Johnny Ive, Sam said it's going to take us a while, but I think you'll think it is very worth the weight.

Starting point is 00:52:49 I think it is incredible. You don't get a new computing paradigm very often. There have been like only two in the last 50 years. So just let yourself be happy and surprised. Only two what? New computing paradigms. Oh, yeah, yeah. He says, so just let yourself be happy and surprised.

Starting point is 00:53:06 In the last 50 years, he says. In the last 50 years. PC era, mobile, cloud. I guess mobile and cloud are tied together. He's been Thompson-pilled. On the future of web and publisher, Sam says, I do think people will go to fewer websites. I think people will care more about human-crafted content

Starting point is 00:53:21 than ever. My directional bet would be that human created, human endorsed, human curated content all goes up in value dramatically. Let's go. Let's hear it for our live stream. Human curated. Human handmade. What AGI means, Sam says maybe the milestone that's most relevant to us is when most of our research cluster is allocated to the AI researcher instead of the human researchers. But I don't think that's going to be so binary because I think it'll feel like people get a little more help and a little more help and a little more help. He also said if we didn't pay for training, we'd be a very profitable company. That's a good question. So, Tyler, have you thought more about what happened to GPT 4.5? People, it's like, people always say like, oh, it was so bad. And same thing with GPD5. It's like,

Starting point is 00:54:08 okay, do you remember when we had Jack on, and he talked about his blog? It said, GPD, or 4.5 is like as good as we should expect. Yep. Five is the same thing. I think it's a good model. Yep. If you, uh, wait, Okay, if you would please consult the graphs. Okay, pull up the meter. So my question with GPT 4.5 is I understand that it's as good as we should expect. The question is just, is it in the money? Because GPT2 was also on that curve and deeply unprofitable, right? They had to pay not a ton of money to train it.

Starting point is 00:54:43 And it made basically no money because they didn't even sell it as an API. Remember we talked to Greg Rockman. He was like, we had to pay people to use our models. Then all of a sudden the 3.5 DaVinci came out. Some people were using that. They might have spent, I don't know, $10 million training GPT3, 3.5, and some people paid for it. They probably made their money back. Who knows?

Starting point is 00:55:03 Then GPT4, they do the big training run, the 100 trillion parameters, the big circle. And that has to be one of the most profitable training runs ever because they maybe spent $100 million, but they are making a billion dollars a month inferencing it. And like the inference cost is probably gross margin positive more or less. But 4.5, they probably, it sounds like they paid a billion dollars to train it, something like that, a lot of money to train this big model. And it's expensive. And yes, it's better, but it's not better to the point where people are willing to bear the cost of inference for it.

Starting point is 00:55:36 So it's kind of mothballed. And you can see that in the app. It's like, yeah, but it's like, okay, it's worth one AI researcher. I mean, I think that the value of that like R&D, like the knowledge that they now have. training the next model is probably worth a billion dollars. Totally. If you're considered AI researchers worth billion dollars easily. 100% worth doing 100% not a big deal for their financials, not a big.

Starting point is 00:55:59 It's just interesting that we went from a paradigm of like the big, like the big training run was unprofitable. Then it was massively profitable. Then it went back to being unprofitable. And the profitable research that they were doing shifted to some RL that they did on, you know, hallucination to reduce the hallucination rate. Like that RL run, probably not a billion dollars in training cost. I don't know. But it's clearly making the product better.

Starting point is 00:56:26 People are going to use Chachapagip more. They're going to be more likely to upgrade. And if the hallucination rate is lower, people are going to trust it to go shop for me. And they're going to make a ton of money off of that, right? Just an interesting dynamic. I don't know. Yeah. I mean, I think a lot of people were like singling out this quote.

Starting point is 00:56:42 If we didn't pay for training, we'd be a very profitable company and just, you know, obviously it's easy to poke a little bit of funny. You're saying he's gross margin positive. No, no, yeah, no, I know. I know. But still, anytime you have a CEO being like, if we didn't have this cost, we'd be profitable, it's always funny. Yeah, I mean, the follow up on the earnings call would be Sam, any place to stop training then?

Starting point is 00:57:01 Well, well, yeah, exactly. So we said, net profits. You know, this is whatever it said and this is what we said, you know, in response to the GPT5 launch is that the product is now the most important thing. and what Everett said just now was that, you know, you could swap in much cheaper models, even open source models, and people would still be using the product in the way that they are. Last quote from Sam from Alex's coverage, he says, I don't use Google anymore.

Starting point is 00:57:29 I legitimately cannot tell you the last time I did a Google search. Mogged. Yeah, and this is, so this is the interesting thing, right? When you think about the browser wars, which we went from the browser wars a month ago, being like everybody's making their own browser to now, everybody's just trying to buy Chrome. And it's still very much up in the air, whether they'll be forced to sell it.

Starting point is 00:57:53 Google's not going to sell it by choice. But it just does feel that chat GPT with GPT or chat GPT agent is effectively a web browser already. You're just browsing the web. And so I think that the real browser war is the fact that chat GPT functions as Chrome plus Google search in a single product already. Yep, yep. No, this is a great take. I completely agree. Wildcard, truth social buys Chrome. Truth Browser. This would be the

Starting point is 00:58:28 most aligned. Yes, this would be the most aligned with the current administration. Tyler, what you got for me? Okay, so yesterday, it's not here anymore, but if you went to opening ad.com slash new tab page, they like leaked this page, like not on purpose. Someone just found it. But it's basically like very close to like a browser style where you would type in and then it would like auto fill some possible questions then you could like save like links and stuff so like it very much looked like the the Chrome like homepage yeah I wonder in the context of mobile I mean using using generative AI to generate code and HTML has just completely pilled me on the the generative UI elements and I feel like like I I would probably be

Starting point is 00:59:12 less interested especially since I mean I use chat GPT mostly on my phone I I use Chrome mostly on my computer on my Mac and so and so I wind up like it's a very different style of working and I can imagine that the evolution here is not the chat GPT app likes opening eye frames and and Safari web views and surfacing something that actually renders the native HTML. It's more like it scrapes all the HTML from a website into the reasoning chain. It gets all those tokens.

Starting point is 00:59:52 And then it kind of just like reinstantiates the UI in like native elements and kind of cleans it up for me. And so I'm getting like a hybrid of like chat GPT used to just be pure text response. Then it became text response. And it also has links in there now. And it also has commerce. Yeah, it also has, um, yes,

Starting point is 01:00:11 tables and it can it can put in images now and if you search for a product it can share like little preview images with a link and so they're hydrating like the tokens and think about how bad 99% of websites are i completely agree and they're so bad right it's hard to it's hard to navigate them there's pop-ups and things like that there are people that that deliberately browse the web with JavaScript turned off because it forces websites into a more like usable plain text experience. And most websites have a have a have a have a fallback in case JavaScript's not working or blocked. And so you can wind up going to the United Airlines checkout and and it'll be just like normal buttons instead of like the page is jumping around, refreshing pop-ups, all that

Starting point is 01:01:00 stuff. All that stuff gets turned off and some people like it. Cookies. Cookies. Yeah. Anyway, if you're trying to improve your website, you're managing, you're GitHub installation. You got to get on graphite code review for the age of AI. Graphite helps teams helps teams on GitHub ship higher quality software faster. Well, pull this up in the timeline. Boys, we have a post here from the New York Stock Exchange, otherwise known as the New York Style Exchange. And here we are.

Starting point is 01:01:26 Let's go. Nicey president, Lynn Martin stunts in the TBPN spring summer 2025 collection. Looking fantastic, Lynn. Collection. Thank you for acting. as our model for this season's TBPN collection. We really designed it for tech and finance leaders that are dedicating their lives

Starting point is 01:01:49 to improving capital markets and maintaining American dominance globally. That was the North Star with the collection. Patagonia is like, oh, this was designed for your next hike. This was designed for Everest. Well, this was designed for the trading floor on an IPO day. This was designed for the trading floor on an IPO day.

Starting point is 01:02:05 This was designed. designed for the hike up to that bell. Exactly. And for the Mount Everest of capital markets, the New York Stock Exchange. For the gong hit that retires the next TBPN gong. Yes, exactly, exactly. We have, I think we can skip over this coverage from the Wall Street Journal. They said opening eyes rocky GPT5 rollout shows struggle to remain. Yeah, so that the art, this article, which was released a couple days ago, the title is opening eyes rocky gpt five rollout shows struggle to remain undisputed AI leader and it's basically just coverage from a bunch of people complaining and it doesn't capture any of the actual underlying doesn't feel like they're struggling to remain

Starting point is 01:02:47 the undisputed consumer AI leader I think you could argue that yeah there's certainly a much closer race in co-gen yeah so I yeah I I put that in just as a reminder to talk about GPT5 yeah the the real news is The Financial Times has a story on Deepseek that isn't super deep in terms of the coverage, but there are some interesting tidbits in here. So the article is DeepSeek's next AI model stalled by Beijing push to take up Chinese chips. We talked about this a little bit with the Nvidia H20 now available in China and what that means for Deepseek. So Deepseek, obviously, everyone should know is the disruptive Chinese. open source frontier reasoning model maker from high flyer they were in the high

Starting point is 01:03:39 they were in the high frequency trading business then they decided to go into foundation model training and they developed and they developed a very very solid open source language model very quickly and it surprised everyone we started talking about Jevin's paradox and the idea that cheaper AI will just wind up driving more and more adoption we've certainly seen that and the sell off that happened in the AI trade in the public markets came rip roaring back and video rocketed to over a $4 trillion valuation after they'd sold off slightly after the Deepseek news. So apparently they've been trying to get this R2 release out the next version of their reasoning model.

Starting point is 01:04:20 And they're having a hard time because allegedly they're using chips from Huawei. So China, the CCP, and Beijing has pushed Deepseek to switch from Nvidia to Huawei. Everyone suspected. And it was in it and it's not technically illegal to use Nvidia chips, but it is politically incorrect according to one person familiar with the conversations. Currently, yes. And and and there were export controls.

Starting point is 01:04:48 There were never any import controls. So if you're high flyer or deep seek and someone comes to you from Malaysia and says, hey I got I got a hundred thousand H-100s right here. You want to buy them? You want this rack. You're welcome to buy those. at least you were, now it's politically incorrect to do so. And so,

Starting point is 01:05:08 Huawei hasn't really gotten the job done. Lots of recent model releases have failed to live up to expectations. This is what happened with GPD 4.5, Lama 4 Bohemith. Like the models are getting more, they're getting bigger. There's more and more integration points in the training cluster as you're actually building these out. There's power management issues. There's memory issues. There's all these different things.

Starting point is 01:05:28 And that's why the AI researchers are making such high salary. and the trade deals are happening because if that there's one researcher who can tell you that line of code is going to result in 20 million or more or 200 million that's really valuable and so this case shines a light on the on the exact nature of the gap between nvdia and wawaway so when the wawaway ascend cloud matrix 384 came out everyone was kind of saying okay wow like wawa is basically caught up It's not as efficient on a dollar per flop basis. It's more energy intensive. But if you're willing to spend a little bit more energy, you basically get the same capabilities.

Starting point is 01:06:09 That might not be the truth. Like there might be actually some qualitative value to CUDA and the reliability of the drivers and the software on top of Nvidia and actually the underlying chips as well, such that even if you have the Three Gorge's Dam, you have cheap energy, you have nuclear power, China's developing more and more energy. It's getting cheaper and cheaper.

Starting point is 01:06:30 Even if you have cheap energy, if you go to set up the massive data center to do the huge training run on Huawei Cloud Matrix, 384, you might still be in trouble and you might not be able to get the model out the door. It could be something else, though. We don't really know. This is all kind of just like little tidbits here and there. It's notable that all of these have led to,

Starting point is 01:06:52 they originally wanted to launch R2 in May, and it's still delayed. Yep, yep. And so it'll be interesting to see how Deep Seek reacts. They could potentially say, you know what? Like, Huawei's just not getting the job done. We'll deal with the pushback from Beijing. We're putting in a huge order for H20s from Nvidia.

Starting point is 01:07:14 We want the best, or at least the best that's available to us, even though the H20, of course, is four years old at this point. Yeah. And severely nerfed. So we'll see when will they get R2 out, how powerful it will it be? And most importantly, what will the cost per million tokens be? Because if we get an 03 level model from DeepSeek and it's 100 times cheaper, even if that

Starting point is 01:07:38 doesn't displace Open AI meaningfully because Open AI is operating at the application layer, it will be incredibly bullish for every wrapper company because all of their gross margins will flip positive very, very quickly because they'll need to do some fine tuning. We'll need to see what perplexity did where they made, instead of Deep Seek, they made it like... 1776. 1776 Seek or something like that. They did a fine tune on it to kind of make it more American. But the most important thing was that the Deep Seek researchers figured out a bunch of interesting hacks to make just inference way, way, way, way cheaper.

Starting point is 01:08:18 Anyway, speaking of rapper companies, application layer companies that we love, Julius, what analysis do you want to run? with your data and get expert level insights in seconds. Ask Julius. Look at this view. I love it. Ask Julius to analyze your data like 2 million users have already done. Folks from Princeton, BCG. Love,

Starting point is 01:08:38 Julius. I wish, my only, my, the only thing I would change with Julius is I wish it was Rahul. Yeah. Or sunwalker. AI.

Starting point is 01:08:47 But it's always time for just like the Ford, just like the Ford Motor Company, you know. The Sunwalker, artificial intelligence. company. Maybe. Maybe it could happen. So T.R. Taxes has a take on the, on the, the, the, the, deep seek story because they think it's a confused narrative with no sources of deep seek confirming it. So T.R. Taxis says, this story is so insane. Dream narrative for burgers and

Starting point is 01:09:15 their I think that's an American vassal. Yeah, I guess. That I might well, that I might well cook up my own. Also based on half-bake rumors. experience in an authoritarian society and just a little bit of sleuthing. As expected, the plot thickens. Siegeng Ping's heavy-handed central government approach, a stalling development is the take that TOR taxes might be debunking here. Deep Seek was a breakout hit, but patronage networks don't reform overnight. The actual Chinese national champion in AI is, as we know, Huawei, they get unconditional

Starting point is 01:09:47 subsidies and the nation's hopes are pinned on them. On a software side, it's also Singwa, the university and their brainchild ZAI with GLMs, but Huawei does everything. In February, the party asked Ren Zheng Fei to partner with major AI labs, including Deepseek, and beat America at AI. They approached Deepseek, sending personnel to adopt Ascend clusters for V3 inference.

Starting point is 01:10:13 We've seen papers following from that, and we know these clusters now work at Silicon Flow and elsewhere. They also suggested training the next generation models on Huawei, but were privately told by probably after some experiments that the Ascend ecosystem is not yet mature or reliable enough and will go with H-800s. Thank you very much. With the knowledge gained, they had set out to train Pangu Ultra MOE mixture of experts as a reproduction of V3R1 and may or may not have failed at that due to interconnect issues and broad lack of competence resorting to repackaging R1

Starting point is 01:10:49 with the intent to report to the party that DeepSeek had proven uncooperative. But there's nothing special there they can do equally well and will soon surpass saw now as deep seek is not releasing any rumored r2 the timeline never once made sense and that's a big issue you need to have your timeline straight there's renewed discussion about importing invidia they are trying to spin this to to their benefit leaking to journalists that it was deep seek that had failed at r2 while wawa's noah's arc small model lab is moving smoothly they may know that V4 is planned to come out late enough that they still have some hope of producing a more persuasive internal result. For now, they are probably optimizing Cloud Matrix hardware

Starting point is 01:11:33 and CINN testing 910D and 920 and hiring people with LLM expertise. The above is an educated guess. The serious argument is that if you want to talk about the failure of Huawei's hardware, it's important to focus squarely on Huawei and not a fanciful and unprecedented narrative where a historically independent startup is forced into changing their training stack by heavy-handed politicians. And so the takeaway here is that Huawei might actually be significantly behind NVIDIA,

Starting point is 01:12:01 and it's less about the CCP saying, you know, we want, I mean, of course, the reason the CCP is saying by Huawei is because they want to improve Huawei and give them as many advantages as possible to get to the frontier and provide, you know, the best, AI training hardware possible. But the flip side is that deep seek is down to use anything and train on a

Starting point is 01:12:26 bunch of different stuff. And really they are, they probably, at least according to this, they really are just having trouble training on large Huawei clusters. And so they're like, let's get back in the Kuta ecosystem. Anyway, let me tell you about profound. Get your brand mentioned by ChatGPT, reach millions of consumers who are using AI to discover new products and brands. Get a demo. Go to profound. Be like the Mag 5. What did James say? The Mag 5?

Starting point is 01:12:56 Well, you were saying, like, I can't say who's using it. Oh, yeah, the Fortune 5. The Fortune 5. He's got a Fortune 5 client. Yeah. So it's like 105 companies. We'll leave it to you guys to guess. 20% chance you just to guess it correctly.

Starting point is 01:13:08 Anyway, lots of people making money on the Intel story. So the story today is that the Trump administration, just last week called for the resignation of Lipp Bhutan said that his ties to China were too much for an American champion like Intel. But Donald Trump has reversed course

Starting point is 01:13:29 and called TAN a success and the idea of the U.S. government buying a stake in Intel is now floating around. I don't love the idea of the people that brought us the TSA running the most advanced manufacturing process humanity has ever produced.

Starting point is 01:13:44 Or the folks behind the DMV. The folks behind the DMV getting in the fab, getting into the clean room, might be a little bit of a stretch for me. But Intel does need better shareholders. There was a few years ago before the Chips Act. Long-term patient shareholders. Exactly. People were talking about,

Starting point is 01:14:01 oh, we need an American semiconductor champion. This was during like the reindustrialization meme kicking off. Everyone was saying this, like, we need American chips. And yet no one was like, I'm going to actually go build a position in Intel. And so everyone was like, yeah, we need this. It was a, what is it, a cocktail position? It was a cocktail position, meaning something people like to talk about at cocktail parties, but they don't actually put their money where their mouth is.

Starting point is 01:14:27 So you sound smart saying we need to, yeah, we need to make Intel an American champion. We need to make chips in America. But I'm not willing to put any money on the line to actually do it. And so Intel's share price has been kind of in the dumps. Well, until recently. Until recently. So Dan Gallagher in the journal says federal support could get the trouble chip maker over some hurdles. but risks great harm to the U.S. tech sector and get into it.

Starting point is 01:14:49 So Intel definitely needs help, but the government support always comes with strings attached, and those strings in this case could ultimately trip up the Silicon Valley Pioneer and the broader U.S. chip industry. The Trump administration is discussing options with Intel that would involve the federal government taking a financial stake in the troubled chip maker. The idea came up during President Trump's meeting with Intel CEO Lipbutan on Monday, and the discussions are still in an early stage.

Starting point is 01:15:16 It's so funny after talking to fabricated knowledge over at semi-analysis about, like, his main thing was like, the problem with the Intel board is that there's too many government-type people on the board. Politicians. There's too many politicians, too many, like, famous people, writers, thinkers. There's not enough, like, just semiconductors, like, scientists. We need, like, physicists on the board and, like, technologists. We need, like, an Elon type or, like, you know, someone who understands the actual tech. Yeah, I mean, just Steelman. It's like Trump has, you know, a.

Starting point is 01:15:46 multi-billion dollar digital asset business. I was about to say, yeah. He has a multi-billion-dollar social media company. He is a founder of a tech unicorn, so it's on his first rodeo in the tech industry. That's right. So you got to give him some credit if he was able to get in there. So Dan says that marks a fast turnaround, given Trump was calling for Tan to be fired just days ago. The news was encouraging for Intel's beleaguered investors who have watched the chip industry's

Starting point is 01:16:09 once undisputed leader lose more than half its market cap in less than two years. The stock jumped 7% Thursday on the initial reports of the talk. and gain more ground early Friday morning. But investors should still be wary. Intel's problems are such that even a big check from Uncle Sam won't fully solve them. The company has burned a total of nearly $40 billion in cash over the past three years trying to regain its manufacturing lead from TSMC. Intel has also been granted up to $8 billion so far in direct funding through the Chips Act.

Starting point is 01:16:39 But that hasn't been enough. Intel's most state-of-the-art production process called 18A was supposed to close the gap with TSM. But the company admitted on its own second quarter earnings call last month that 18A will be used mostly for its own products, meaning few outside chip designers have found the technology compelling enough to sign on as customers of Intel's contract manufacturing service. Wall Street expects another $7 billion in negative free cash flow this year, according to estimates from visible Alpha. Sorry, taking me a second there on the soundboard. Tan told investors in the same call that he won't commit major. capital spending to Intel's next process called 14A without commitments from external customers.

Starting point is 01:17:21 Smart. We couldn't get a customer for that. That was widely seen as tan drawing a line in the sand, a line by which he would determine whether to keep Intel in the business of manufacturing chips. But Intel pulling out of that business would be detrimental to the government's efforts to shore up domestic chip making for national security and supply chain stability reasons. Doug over its semi-analysis was talking about how the chip design business seemed like it could be a target for like PE in the sense that if you came in and kind of

Starting point is 01:17:53 overhaul. Hawk Tan is going to come in. Yeah. If you, you know, dramatically cut cost and really focused on serving customers and of course raising prices, there's probably a good business there, but that the foundry business was critical and we don't want to risk losing that. Yeah, Hawk Tan is the CEO Broadcom. I just wanted to say, have a great flight, John Exley. He says he's taking off he's landing in one hour. And if you're trying to set up a semiconductor line, if you're trying to build or plan products,

Starting point is 01:18:23 get on linear, linear.com. Great idea. There's a purpose-built tool for planning and building products. Meet the system for modern software development, streamline issues, projects, and product roadmaps. So I this chart chipped away.

Starting point is 01:18:37 Chipped away. Intel and TSMC revenue. So I'm still sort of split on this. We have some guys. on the show today to talk about the dynamic between the US and China in the semiconductor race I'm I'm sort of open to the idea that we sort of solve the

Starting point is 01:18:56 US based manufacturing of semiconductors through partnerships with TSMC and Samsung even though those are not American companies if they set up fabs here in some sort of you know negative conflict scenario it's like well we still have the the factory is here even if it's run by a company in in South Korea or Japan or Taiwan, like the factory should continue to produce for the most part because most of the team that would be building And I think the pushback there is we don't have the talent,

Starting point is 01:19:26 which is key to staying on the leading edge. Yes, that's true. But, I mean, it yields at TSM are Arizona have been good so far and it feels like we could continue to scale up and it feels like a lot of the talent will be coming over. And so there's a little bit of like, you know, you start to ramp up that supply. But it does feel like Intel is particularly good at the trailing edge, but maybe that goes international, but to South America.

Starting point is 01:19:51 Unclear where it goes. Unclear how, at least to me, how important Intel is as like a strategic company versus just like it's one of the greatest technology companies America's ever produced. It's a crown jewel. It should just be protected because it's like good for the America brand versus like if Intel disappeared tomorrow. like how bad would things be in America? Would we be able to get by with other suppliers? Yeah. Because we do have AMD.

Starting point is 01:20:21 We do have Nvidia and then TSM and Samsung are not in China. That's not where we buy our chips from. Well, Dan, in the journal says the government might, for instance, pressure chip designers like Nvidia, AMD, or Qualcomm to manufacture with Intel, perhaps as a condition for getting export licenses for China. And that could easily go wrong. If companies are forced to use Intel's factories before they can make chips with production yields that match TSMCs, it could result in inferior products and wastage by Intel because so much silicon has to be thrown out to make a working chip.

Starting point is 01:20:55 More broadly, if chip designers are using Intel fabs, even though they aren't the most advanced or efficient, the entire U.S. chip industry could lose competitiveness. That would undermine the ultimate goal of government intervention in the industry, which is to maintain American technological supremacy. Yeah. Rubicon of state intervention in chips has already crossed the administration, already signed significant leverage over Intel, has already has significant leverage over Intel, thanks to government factory expansion grants that place limits on how it can restructure its chip design and manufacturing arms without government consent. But the federal government must take care not to go too far lest it undermined the market model that made American technology. Slippery slope.

Starting point is 01:21:39 It is. But at the same time, there is a case to be made. And I guess, you know, like if you put the U.S. sovereign wealth fund under the direction of Leopold Oshenbrenner, there is a case just for make money for the taxpayer. And this is actually the opposite of a bailout. This is what Tim Geithner got in trouble for during the, not in trouble. He was ultimately vindicated during the financial crisis in 2008. He went and made a bunch of loans to banks on the order of like billions and billions of dollars.

Starting point is 01:22:09 And everyone was like, this is a bailout for Wall Street. This is a bailout for the banks. But he actually only invested in the banks that made it through the crisis. And so those loans, those backstop debt instruments, were paid back with interest. And so the U.S. taxpayer actually made money on those deals. It feels a little weird. But the same thing could happen here. Like Intel's a $11, $110 billion company.

Starting point is 01:22:35 if the U.S. invests and is able to do things to turn it into a $300 billion company, well, like, that's an extra 3x for the U.S. taxpayer. And it doesn't actually result in any, like, lost money. It's not a bail. And if Trump can just get three Xs and string like a hundred of those together in a row, we'll solve the whole federal debt crisis. That's a high watermark. I think he should be targeting, you know, a nice 5x fund.

Starting point is 01:23:05 for the first run, then raise 10x more, and then scale up, and then, you know, start, start deploying the big money, big money. You should buy 100% of Intel. What you got, Tyler? This is like we should give, put Jane Street, you know, high frequency lawmaking. Yes, just optimize for GDP. Direct right access to the legal code. Yeah.

Starting point is 01:23:27 So anything they can do to just maybe one of the foundation labs actually. Instead of nationalizing the labs, we need to, you know, corporatize the government and do a reinforcement learning environment with a verifiable reward. The verifiable reward being the stock market price. Orl for business, but the business is the government. Exactly. So what can you change in the legal code to make the stock market go up? And so you're just feeding off of that, constantly rewriting the legal code. Well, Zoomer at ZumiZoom has been going viral again.

Starting point is 01:24:05 He says why AI is a house of cards. He has an entire thread breaking down these sort of chained losses that we've been talking about. And he's getting community noted quite a bit. Someone added 1H100 can serve thousands of users at once, depending on batch size and model. For inference, you don't dedicate a GPU to a single user. you load the model then stream requests for many users in parallel other numbers in this post are also wildly widely exaggerated yeah this might have come from a group chat uh originally that was uh maybe not fully fact checked but uh you know told told the compelling story so yeah it's a fun with it

Starting point is 01:24:45 and nick carter says compelling threat if you ignore that inference gets 10 to a thousand x cheaper every year if you're willing to pay 200 dollars a month for a i and bc funding sub and bc funding subsidizes half of that simply wait six months. I would say, you know, Mr. Randall earlier on the show said it's actually not all of a sudden getting 10 or 1,000 X cheaper, at least for frontier models. But the point he made is that a lot of prompts could be served with older, cheaper models, and that's going to be a big focus. I mean, clearly that was a focus for OpenAI with the recent launch. Yeah, I dug into this to see, you know, how would we really hit a thousand X cheaper this year on the models? And would the gross margin profiles of these AI companies flip extremely quickly, or is it more like a five-year change to really optimize this stuff?

Starting point is 01:25:39 I'm kind of split on it. You know, the charts that Nick Carter shared here are pretty compelling. I just hope that the trend continues because the dynamic of reasoning models and test time inference is, slightly different like you are just it's it's less algorithmic driven it's more just throwing raw compute at it and generating a ton of tokens so I don't know Tyler what do you think you're closer to 10x cheaper inference every year thousand X cheaper inference every year what type of gain in cost per token do you expect over the next few years don't make mistakes do you mean in what models are

Starting point is 01:26:19 talking about like frontier level models all the models but I think frontier will probably stay similar price and then you'll just see like over time like now we have open source models that are easily as good as like two years ago yeah right you have like 4 oh which is super cheap doesn't mean free like it means free no license but you still have to inference it on an invidia GPU that costs money and you have to spend electricity that costs month just open sourcing a model does yeah but but but when you open source something you You can distill it even further. You can get some optimizations there.

Starting point is 01:26:53 That's cheaper. So O3 Pro, let's call that like an expensive frontier model. How cheap do you think that is next year? Do you think it's 10 times cheaper? Two times cheaper? A thousand times cheaper? Like an equivalent model. Yeah, O3 Pro, heavy reasoning.

Starting point is 01:27:13 Thanks for 10 minutes, generates tons of tokens. In a single year? I'm probably closest to 2x. 2x. Yeah, I wouldn't say massive games. Well, guys, I hate to interrupt, but there's some breaking news. William Jong is proud to announce that he's the reigning OpenAI McNugget champion. He ate 55 nuggets in 10 minutes.

Starting point is 01:27:36 Absolutely incredible. Some other breaking news. Numeral HQ, sales tax and autopilot. Spend less than five minutes per month on sales tax compliance. Go to Numeral HQ to get started. And we have our. third guest of the stream since we already did the debate. We have Bill Bishop from Cynicism. How are you doing? Signed. How are you? How are you? Thanks for

Starting point is 01:27:55 having me. Thanks for hopping on. We really appreciate you taking the time. Give us, I mean, I'm super familiar with your work. I think everyone should be, but give us the high level of the pushback that you saw yesterday. We enjoyed having you in the comments. And I want to know how you would frame the counter argument to what Aaron Ginn was making yesterday. Okay. Well, thanks for having me. And I'm impressed that you guys respond to jerk comments. That's good. I don't see it as a jerk comment. Well, yeah, we don't see it.

Starting point is 01:28:22 I think it's extremely fair pushback. 100%. Okay. When I have Aaron Garon, I see it as having Jensen Wong on the show. Okay. Well, that is, that is how I see it too. So that's great. And, and I can't have, I can't have Jensen come and whiteboard out pros and cons with me. And I can't throw random jokes at Jensen all day long, but I can't. Aaron. And so I enjoy, uh, having him sit there and I can throw stuff at him.

Starting point is 01:28:47 And of course, he has his opinions and his arguments. But it's great to have somebody that can speak the language of invidia. Yeah, no. And again, I think thanks for having me. I'm psyched to see you guys on Substack 2, which is great. And I see your running show today. You've got two great guests, Jimmy Goodrich and Leonard, who are going to be way better on sort of this discussion.

Starting point is 01:29:13 So I think that what I would say is what's been interesting to watch. And I think talking about the reversal on the H20 trip, we got H20 chip from the VD we got to remember, right? There was a proposal teed up for the Biden folks to ban the H20. And they never did it for whatever reason. Trump comes in, actually bans it. So it looks like he's being more hawkish. And then a couple months later, because of really effective lobbying from NVIDIA

Starting point is 01:29:39 and specifically from Jensen Huang at the principal level, to the president, to David Jackson. Yeah, I mean, it's not even, it's not even traditional lobbying because there only, They're spending like seven, low seven figures on lobby. Probably 10 times that on Jensen Wong's like flight schedule. Flight school, right. No, it's brilliant. They say sort of, you know, disrupt the lobbying business, go right to the decider.

Starting point is 01:29:59 Go direct. And how the narrative is shifted. And so Trump on the one hand looked like he was tough on China. Then he backs off and then of course he gets a bunch of flack for, oh my God, he's caving to China. He's caving to Jensen, right? Yeah. When in fact, he sort of did what Biden didn't do.

Starting point is 01:30:14 And then because of this personal interaction and And the personal effort from Jensen Huang, he reversed course. And I think it has to be seen in the context of the broader, where I think there was a correct decision, the rescission rule, right? Which was really, you know, David Sachs was, I think, a big advocate of getting rid of that rule with a Biden administration rule that was going to limit countries that could buy Nvidia GPUs, right? The idea that the U.S. needs to lead, the U.S. needs to be out there competing with China, not just limiting.

Starting point is 01:30:43 And the way to do that is to get people hooked on the U.S. AI stuff. Dax, specifically in Vidia globally, X-China. That makes a lot of sense. The China decision on H-20, I think, is flawed. The idea that, oh, we're going to keep China hooked on Nvidia by selling them H-20, and then I think, you know, the Jensen Huang is lobbying to sell sort of the next somewhat nerved chip, right? A little better, but still not near the top of the Nvidia product suite.

Starting point is 01:31:12 The idea that just Nvidia is better, Kuda is better than the Chinese, the Chinese, Chinese hypers, the Alibaba, Tencent, Bite Dance, Deep Seek, they're going to want to stay with Nvidia, makes sense in a world that is sort of a normal political economy, a normal competitive world, doesn't make sense in the context of a world or market that's run by the Chinese Communist Party of China. And Xi Jinping made very clear in April at this Paul-upier study session that was about AI, specifically said he wants China to build its own indigenous AI stack. And so they understand very well this idea that, you know,

Starting point is 01:31:51 Nvidia wants to hook their companies on Nvidia hardware. The party wants self-reliance. And so this idea that we need to compete, America needs to compete by selling into China and therefore they're addicted, they won't break it, that I think is fundamentally naive and misunderstands how the party operates. I think what it does is it helps China fill the gap between where they are now in terms of lagging capabilities and lagging.

Starting point is 01:32:17 output or quantities in terms of the Huawei chips, it gets them to where they need to be over time, but they're going to be only intensifying their efforts to strip out Nvidia, to break any reliance on the USAI tech stock. And so by selling H-20 chips now, we're just helping China keep in the race when in fact if they weren't able to get the H-20s, it probably, I think, would help the U.S. at least maintain a lead, if not start accelerating and accelerating some amount of separation. What's your reaction to the commentary showing that the CCP is actually like actively pushing back against Deepseek, which is, you know, the clear open source or maybe not totally clear, but an open source. An open source leader. But yeah, your reaction to the news that Beijing asked Chinese Foundation labs do not buy NVIDIA, we're just to find.

Starting point is 01:33:16 Effectively, even if it means delaying like deep sea cards too. So there was a financial times report yesterday where it said they were encouraged. We would love to know what the encouragement really was. But yes, they were encouraged to use the Huawei. I think it was the latest, the send chips, which were all based on dyes that were illegally fab, the TSMC, where Huawei used a cutout company that was related to BitMane called SoftGo to get, I think it was two million dyes from TSMC that they can't make on their own in China. And even then, they're still not where they need to be.

Starting point is 01:33:49 And so I think that this goes, you have that news. You have the news since the announcement by the Trump administration that they were going to allow, again, allow licenses or give licenses to sell H20 to China. You've had significant pushback from some of the regulators in China. You've had talk about the chips are unsafe. Maybe they have back doors. They're environmentally unfriendly. There's security risks. And so I think what you're seeing, you know, there's different hypotheses about what's going on on the, on the push-

Starting point is 01:34:16 back on the H20, it's maybe they're trying to negotiate for the better chip. I personally think it's actually more of a manifestation of parts of the system really just are like, we need to stop this reliance on American ships. We need to make sure we're focused on building our own pathway to self-reliance. And that I think is related to that news that DeepSeek is being encouraged to use these lesser chip, even if it delays them, because all ultimately when they figure how to use them and you know the report said Huawei has engineers on site trying to work through it. Ultimately that will help China because that will help them solve over time the various bottlenecks they're facing. But I think you can read into it and and one way you can read into it is that the CCP doesn't believe in this sort of fast takeoff scenario, you know, like runaway AI in the next, you know, two years, right?

Starting point is 01:35:10 Which I think broadly, I don't know a lot of people that still believe in that, but it's notable. No, I think that's right. I think this is more of a, we're going to, we are going to set the foundation for doing it in a self-reliant way, even if it takes us longer. And so that's where I think what's been interesting to watch in NVIDIA is how the narratives have shifted in D.C. Where the NVIDIA line of we have to compete, we have to compete, we have to sell in the China, have to addict them is basically everywhere now. There's just like, there's barely any pushback. And certainly in the government, from what I'm understanding is there's no longer any process, right? back to the whole sort of how did Jensen lobby to get decision made.

Starting point is 01:35:49 There's no process. There's no, like many of the people who worked on these issues were fired. And now it's basically like he gets to the principal, he gets to Lutnik or Sachs or the president, and that's the decision. There's no like national security discussions. There's no process anymore. Yeah, it feels like interestingly,

Starting point is 01:36:07 people often project like a monolithic culture upon China, but then severe division, within America and it feels like there might be some division on both sides in the sense that in America there are arguments for let's export all the GPUs keep them dependent on us versus let's let's hold it back and hurt their ability to scale and then in China they might be saying the same thing hey we need to just buy buy buy and stay near the frontier this will actually help us accelerate and then there'll be a different argument for for maybe maybe we need to just there's also There's also the, you know, in many ways, Deep Seek, the original Deep Seek release was,

Starting point is 01:36:47 was in some ways economic warfare on, on Nvidia, right? You saw this massive sell-off immediately. And there was clearly somewhat of a coordinated hype. Yeah, I mean, the App Store, the App Store chart, you know, Deep Seek getting all these downloads was completely fake. Twitter, Twitter bots, Twitter trolls. And so I think there's, you could also read into this and think, uh, Bay, doesn't think that the next deep seek release, regardless of how much progress they make around efficiency,

Starting point is 01:37:16 will have the same effect on, you know, making Nvidia sell off, you know, massively. Do you have any context on previous technological revolutions and the history of the U.S.-China relations? Like going back to like the cloud or mobile, like was there ever any similar considerations of don't sell iPhones? It felt like in the previous era, every big tech CEO was like, I'm going to massively Tam expand by getting into China. And then they got blocked. And this is kind of the opposite where Jensen's been playing that. And then now he's having to pull back. And the government's like the U.S. government's the one that's saying don't sell the China.

Starting point is 01:37:56 Whereas in the past with Uber and Google and Facebook, it's been the Chinese government saying don't come here with your technology. I think this is fairly unique as far as I in my memory, certainly with this sort of important technology. There have been certain types of things that the U.S. government hasn't allowed to be sold into China, but not at the scale or the sort of economic importance or frankly the market cap importance. Zooming out, how do you feel like U.S.-China relations are just going generally? I feel like two years ago, there was a ton of saber-rattling about we need to get sharp on Taiwan. Everyone needs to learn what TSM is. We need to talk about defense technology and Taiwan invasion in six months, 12 months.

Starting point is 01:38:38 It's going to happen. And then it feels like we've been in a bit of a lull, a little bit more economic, you know, economic warfare. But it feels like we might be coming out of a period of high tensions. Just give me like the general pulse check from your side. It's a great question. And it's one that it's still quite unclear. I think that you see the beginning of the Trump administration, the economic tensions rose pretty high. those have come back down to where there's now a sort of a tariffs are high, but there's a, it's calmer,

Starting point is 01:39:15 although the Chinese, in part, it's calmer, I think, because the Chinese pulled out their export control trump card, so to speak, around rare earths and rare earth magnets and really, I think, showed the U.S. that they had actually a lot of leverage that the U.S. didn't necessarily appreciate. And so I think you're in a bit of a law on the U.S. side because there are things, like, for example, on the technology stuff, you know, they were a bunch of new actions around export controls, around chip-related stuff. They're all tabled, right? In part, I think, because of how the Chinese were able to push back on the, initially, the beginning of the sort of the trade war, using their rare earth's card. Generally, though, when you look at the broader, you know,

Starting point is 01:39:59 you look at Taiwan, you look at sort of things like the South China Sea, You know, the, you look at the other economic issues around, you know, what the U.S. states overcapacity. The structural issues are not going away. We are, I think, as you said, we're in a bit of a lull. And the Trump administration seems to be more focused on the transactional bit, parts of the relationship for now. But, you know, there are some people who want to talk about it. Maybe there'll be this grand bargain.

Starting point is 01:40:31 You know, Trump and she may meet this fall and they'll have some great grand bargain. you know, it's hard to see how that would happen and how it be sustainable just because of the real structural issues and relationship. But there's no question that the narratives have been shifting. The Chinese have been working really hard on people-to-people stuff that has, I think, pulled us back from the sort of peak of China Hawks. I said, you know, my Sharp China podcast last fall or the end of the year, I just, we're joking. I said, you know, I think we've hit peak China Hawk, right? No, seriously, right? It's going to, it's going to sort of moderate at least for the time being.

Starting point is 01:41:09 I think that's what we're seeing. Is the rare earth element stuff an ace in the deck of cards, or is it more like a jack or a queen? I think about, you know, we haven't even gotten to the, obviously, Taiwan invasion feels like more of the ace in terms of just like how much pressure that would put on the relationship. But also Apple, it feels like if China were to put pressure on Apple, that would potentially be more. disruptive to the American economy just because it's such a huge company. It's so critical to American technology than rare earths. Or is there some other dynamic at play there? I think the Chinese, you know, Apple is one of those companies that every time their tensions, it comes up, well, China could do something to Apple. And they have, you know, Tim Cook has

Starting point is 01:41:55 been brilliant at managing President Trump and brilliant at managing Xi Jinping. And, you know, Apple, there was a great book that was written about Apple by Patrick. I mean, Apple has, does a lot for the Chinese economy. They employ a lot of people directly and indirectly. The Chinese so far have not really bothered them in any direct way. The rare earths is one where they have the ability to effectively disrupt significant parts of U.S. industry and European industry. And they did that. And that, I think, is why you see, you saw the U.S. sort of pull back pretty quickly in the trade discussions. And, you know, the way the U.S. did it is after the first meeting, in where was it it was in um shit it was london geneva at the first meeting all of a sudden the

Starting point is 01:42:41 u.s added these new export controls on like jet engines and and other things because the chinese weren't giving the rare earth magnets that the u.s thought they were yeah that that is the one where the chinese can cause pain immediately yeah yeah yeah so with apple if china does anything to apple that's like a million people unemployed in china very disrupted to the chinese economy whereas with rare earths, like, you could stockpile them. It's not as critical of like a labor market in China. No, it doesn't. If they can't sell to us, it, it's basically it hurts the couple, like one or two state-owned companies effectively. Got it. Got it. So it truly is more leverage for them. But it's, but it's the card that you can only play for a certain period of time. And if the U.S. government

Starting point is 01:43:21 and allies get serious about solving that bottleneck, it can get solved. The problem is, it may be the Trump administration now is serious. This is not an, this was not an unknown issue. The Chinese threatened this in the first Trump administration. The Trump administration, then we had COVID, nothing really happened. Biden administration admired the problem, wrote some papers, had some meetings, didn't fix it. Now maybe there's the urgency to actually address it. Yep, that makes sense. Last question from my side.

Starting point is 01:43:46 What is general sentiment from on the ground in China? What's your read on sentiment among business leaders today? Not being there, that's a harder question to answer. When you look at some of the data and the surveys, you know, you look at some of the multinationals. I think there is the surveys from various foreign chambers of commerce tend to be generally pretty pessimistic. More pessimistic than been in years. When you look at some of the surveys around Chinese business confidence, it is maybe bottomed, not particularly positive. Certainly there are pockets that are positive.

Starting point is 01:44:24 We talk about the deep seek moment. that had a real catalytic effect on certain tech sectors, and you certainly see the Chinese stock market. Chinese stock market's up pretty big this year. Things like AI stocks are up, AI concepts stocks are up big, some of the chip stocks are up big. You know, the H20 news and the fact that the Chinese maybe not want the H20s was good for some of the domestic chip companies.

Starting point is 01:44:46 So in those sectors, you know, robotics, I think they're feeling quite confident because both the markets there and then they've got massive government support. So it's a mixed bag. Well, thank you so much for hopping on. We are going to jump on with Jimmy Goodrich, but we'd love to have you back. I mean, this is great. No, thank you.

Starting point is 01:45:02 Any time. Also, if you're ever in the chat and you have a comment, you want to, you want to extrapolate on, we'll just drop you. You just join the same link that you have. Really, you can join us. We're alive. So anytime. Now you've got to do it. Appreciate it.

Starting point is 01:45:12 You'd love to see you. Cheers, Bill. Have a good one. Thanks. Cheers. Have a good weekend. Cheers. Let me tell you about fin.

Starting point is 01:45:16 .AI, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bankoffs, number one in ranking on G2, fin. Absolutely legends. And we have Jimmy Goodrich in the re-stream waiting room. Let's bring him in right now and continue our conversation on chips and China. Welcome to the show. Welcome to the show. Hey, good to see you guys.

Starting point is 01:45:36 Good to see you too. I'm not sure if you were if you've been tuning in or Bill Bishop gave you a highlight, a summary of the debate. We've been debating the pros and cons of exporting H-20s to China. And the back and forth America has had threatening to ban it, actually banning them. then pulling back on the band, would love for me to tell, for would love for you to tell us how you've processed that story, where you've sat on the issue over time

Starting point is 01:46:05 and where you're sitting today. Yeah, no, I caught the tail on it. I think it was a great discussion with Bill. You always got really good insights to add. I mean, clearly it's been a roller coaster. I mean, US export controls on China have typically been this sort of, the government thinks about doing things.

Starting point is 01:46:21 It leaks out in Reuters or the Wall Street Journal that there might be doing an, export control China learns about it about nine, 12 months in advance. They stoppile everything they need. Then they watch the Americans sort of debate openly, you know, in our democracy, which is messy. I think they look back all this is kind of silly. And then they kind of half impose a restriction.

Starting point is 01:46:41 Then they undo it. I think it's just all kind of comical for Beijing. So how big do you think the age 20 issue really is? You know, some people are talking it up is like the most important chip. It's going to completely unlock deep seek R3. It's going to be this amazing moment for them. On the other side, folks are saying it's a four-year-old, it's a four-year-old chip, it's heavily nerfed.

Starting point is 01:47:02 Like, yes, Deep Seek figured out a way to optimize around some of the limitations, but in general, this is not a real threat. How are you feeling about the actual value of the capability provided by the Kuta ecosystem on top of the H20? I mean, I think it's still a very valuable chip for China and for China's AI model developers for two reasons. One, in the AI world, obviously, you've sort of gone to digital. this depth on your show. It's about training and inference, and particularly for inference is where you

Starting point is 01:47:29 need memory bandwidth, and that's where the H20 excels. In fact, on a cost per token basis, is probably the most competitive inference generating chip in the world because it's the same memory bandwidth of a hopper, but at a reduced price. So it's a great value chip for inference. And that's, another key factor here is quantity, is that Nvidia can provide them in millions of units. that is something that Huawei and no indigenous producer today can do. Because of the export controls, because of the complexity of advanced no chip manufacturing, China's indigenous chip manufacturing ecosystem might in the future, but does not right now have the ability to produce enough to satisfy their own domestic demand.

Starting point is 01:48:10 So at least temporarily in this sort of one to two year window, the H20, and then possibly a downgrade of Blackwell, still going to be very useful China to China. And on top of that, of course, there's the Kuta Advantage, advantage. And if you talk to any AI model developer in China, they want to develop their model on the Nvidia STAC. They've been doing it since college days. Everybody knows how to code on Kuta. It's a big pain to move to another supplier. I mean, just moving to AMD, for example, is difficult. So, you know, let alone a much smaller, much more nascent developed Chinese competitor. So I think it's actually going to be a big game changer for the deployment of AI for the scaling up of Chinese AI models.

Starting point is 01:48:51 And if you think about, you know, reasoning and inference, if you want to develop more capable AI agents that are going to be doing more task-free autonomously, that's where H-20, high-memory bandwidth, good inference chips are going to come into play. Do you think it's smart for Beijing to take maybe a more long-term view here and say, we're going to throttle development in a short term to really develop the industry locally? Well, I think they've got two sort of interest groups they're trying to take care of. On the one hand, they have their AI upper stack companies, the model developers, DeepSeek, Moonshot, Emmy, Baidu, Tencent. They want just to be able to put out competitive models.

Starting point is 01:49:38 And frankly, having spoken to many of them, they'd much rather use a better, more capable chip, regardless of where it's from. And Nvidia certainly wins out in that right now. On the other hand, you know, China has a self-sufficiency national target that's Xi Jinping set as part of the 20th Party Congress, called it Putzili-Zest Xiang, or National Technology Self-Sufficiency. And he's talked specifically about using its secure and controllable indigenous chips. And there are set aside Huawei about a dozen indigenous GPU suppliers in China who want to take advantage of Nvidia not being in the market and expand their market share. And so on the one hand, Beijing is welcoming NVIDIA back in.

Starting point is 01:50:20 They're rolling out the red carpet when Jensen comes. They also, it doesn't matter why, they want an executive who's actively lobbying against tech restrictions in Washington. They want to reward that behavior. But on the other hand, they want to create a space for these indigenous GPU players. It's going to be in things like state-owned contracts, China Mobile, Telecom, procurement contracts are going to go mostly to those kind of hallway and other firms. But I expect sort of the Baidu, Alibaba, Tencent, hyper-scaler contracts are still going to be majority in Vida, particularly if they can get the licenses.

Starting point is 01:50:52 So Beijing sort of balancing both of these constituents. In fact, within China, there are many who actually don't like Huawei. There was a Chinese Academy of Science, very senior computer scientist, who's a vice minister in the Chinese government and party. And a talk of his leaked earlier this year where he was criticizing Huawei and saying, Beijing should not let Huawei dominate the AI stack in China. It's not healthy. They can't have a single large monopoly that the government should support and that they should be supporting competition with inside the Chinese system.

Starting point is 01:51:23 So, you know, China is not a, let's give everything to Huawei. There's a lot of people who think they're too aggressive. They're kind of like the apple of China. Nobody wants to really do business with them because they're cheap on price and very aggressive. They know it's like the long or the wolf culture. So, you know, Huawei has its own enemies with inside China, Interesting. Yeah, so let me walk through the current thinking and you can kind of push back on my reasoning chain here. So we are now maybe in an era of plateauing. We're not on the cusp of super intelligence by merely scaling up, you know, a bigger large language model.

Starting point is 01:52:01 And so what really matters is that inference is the actual deployment of AI getting AI all throughout the every crack in the economy is souping up the various SaaS systems and putting agentic workflows all over the place, increasing GDP, not to 20% overnight, but maybe just bumping it from 2% to 3%, 1 point or something like that. And so giving the H20 to China allows them to do that, allows them to scale inference, distributed inference nationally into all sorts of businesses from DJI will benefit from this marginally with slightly more AI all over their organization to some small machine shop that might be using it to run their HR software more efficiently. And so although it is somewhat of a more level playing field, we are still in the domain of just economic competition. And so it's not a major national, it's not perceived as a major national security risk.

Starting point is 01:53:09 it's mainly an opportunity for an American company to just play by the traditional rules of free market capitalism and export their goods all over the world. Is that like roughly the modern thinking, you think? I'd say I agree with you on that first point. If we want to help enable China to be competitive in AI, we want to help their AI model companies get access to the best infrared chips, want to help them scale up their deployment, win in the market at home and possibly also export their models globally, then absolutely we should be selling, you know, more H-20s to China. I just don't think that's in our national interest. You know, of course, it's in it's in Vidiya's interest.

Starting point is 01:53:47 They want, you know, they are agnostic to who wins in the AI race. Because at the end of the day, whoever wins is still going to be buying a boatload of Nvidia chips in Silicon. And so whether they're Chinese, whether they're from the UAE or from the United States, you know, it's, you know, multinational company that's selling Silicon is really not going to care where their chips are going to and what they're enabling from a sort of flagged country perspective. I do think, though, if you look at disinformation and cyber warfare,

Starting point is 01:54:17 and you look at the capability that autonomous agents are going to be able to, even at current GPT5 or future R2 level coding capability, if you think about scaling that up with, you know, 1,000, 2,000 autonomous, agentic AI coding capabilities are going to be doing vulnerability scanning, cyber offensive warfare, you really start to get an exponential capability increase. And so I do worry that, you know, the Chinese state with, you know,

Starting point is 01:54:46 two dozen H20 capable inference data centers could use that to do more autonomous cyber activity that's nefarious. And on the same side, disinformation. If you can have models that can reason for longer and on an agentic basis, interact with people online, shift populations, opinion in places like Taiwan, that's, incredibly dangerous and we've already seen the United Times reported about 10 days ago that state-owned companies connected to the Chinese state are using DeepSeek, which is going to be inferenced on you guess what, the best silicon possible to do exactly that, which is

Starting point is 01:55:21 disinformation campaigns against Taiwan and the United States. So actually I do think there is a national security concern here. One, there's an economic security leadership concern and then there is a, you know, enablement capability down the road that is actually going to be, you know, I think happening relatively soon. Last question. Another concern people have had is just like giving giving the party in Beijing broadly access to more compute and the potential applications of that in a military context, specifically drone warfare.

Starting point is 01:55:53 Is that something that you worry about very much? Yeah, I mean, there's, you know, there's like traditional applications of high performance competing, super competing, which is useful for weapons modeling simulation. You don't need multiple large systems to do that. You might have a couple of boutique, standalone, government HBC systems. Where more of that model data is going to be useful is if you're using large distributed systems of federated drones, collecting data, acting autonomously. For example, think about a world where you have your PLA signals intelligence communications battalion. That's in real time collecting all the battlefield communications in a Taiwan operation.

Starting point is 01:56:35 Then they're transcribing that in real time into a written product that's being analyzed by autonomous agents in real time. And then getting field reports into their commanders in real time telling them, hey, you know, there's a – you have a squad that's hit counterfire on this beach north of Taiwan. They haven't even reported it up to their superiors, but the autonomous agent AI system might actually be able to get that, deploy a drone. If you think about just those capabilities in the future, that's where inference really matters. And that's where it's going to scale up and create tons of economic opportunities for Chinese companies and e-commerce and all sorts of other areas, finance and SaaS. But also on the military side, it's really endless if you could think about the applications as well. Great, great answer. Last question for me, what's going on with TikTok?

Starting point is 01:57:26 It was the talk of the timeline earlier this year. Everybody seems to have forgotten about it. Any updates there? You know, I don't have a whole lot. I think it's one of these things where it's pretty obvious what happened to it. You know, the president likes the tool, thought it was useful for his election. You know, they've continuously renewed the clock on that 90-day extension. Unfortunately, you know, there's no longer really an operating national security council inside the White House like you would have traditionally to kind of figure out and coordinate the interagency on a solution. So I think that that feels like something that, You mentioned earlier, Beijing kind of laughing about how our, you know, we have our democratic system just publicly debates all these issues, creating this ability for them to, you know, make changes in advance. But this feels like one of those things, I mean, they have to be just laughing, laughing about

Starting point is 01:58:23 how we've dealt with this entire issue to date. I mean, like with many of our things, I think they look back and just don't think we're a very serious country. I mean, maybe with the exception of parts of our military, they think the U.S. is sort of a, you know, badass that should not be messed around with. But, I mean, look at us on rare earth. We can't get our act together. Export controls. We're moving back and forth. Whether or not we think we should actually, you know, get our act together in onshore ship manufacturing. We're, you know, interested in Intel here a little bit and then TSM there a little bit. You know, the Chinese government, I think from their perspective, like, look, we've got a 10-year plan, a 15-year plan, a 50. There's actually a 100-year plan. And they're just sticking to it. And they see us just kind of all over the place. And I just don't think they take us very seriously, unfortunately. Well, in 100 years, we know the plan in America celebrate the 350th anniversary.

Starting point is 01:59:15 You know, there's going to be a party and maybe a UFC fight. That's what we can play for in the White House line. Anyway, thank you so much for- Thank you for joining, Jimmy. Very insightful. We'd love to have you back and talk more as the stories develop. This is great. Yeah, happy to chat more.

Starting point is 01:59:29 You guys are always welcome. We'll talk to soon. Thanks, Jimmy. In other news, a Rune post has hit the timeline. Rune says, agree with Delian. Rune. T.S.Rune has spoken. Agree with Delian on the Maoist perspective that data centers should be turned into steel plants.

Starting point is 01:59:49 I love it. And you know what else? I love Adio. Customer relationship magic. Adio is the AI Native CRM that builds, scales, and grows your company to the next level. If you have a free calendar this weekend. fill it up with onboarding to Adio and spend 48 hours just playing around in there doing some deals in other news we have a we

Starting point is 02:00:10 our next guest is in the stream rating where we will bring in Leonard Heim the second time on the show I want to talk about Mr. Beast he says he plans to take a hundred software engineers and lock them in a room with no cursor subscription with the first person to ship something that compiles taking home one million dollars wow this is from Vass absolutely fantastic I I originally read this not as a joke. It's like, oh, he's actually doing the PMF or die thing because that kind of would work. And I thought it was with a cursor subscription. And then I was thinking about like, wow, he could wind up,

Starting point is 02:00:43 if he covers the bills, he could wind up spending $2 million on this challenge. But Mr. Bees should get into software-based challenges. I would like to see him have a different Mr. Beast of software. He should have a horse in the code gen race. Yeah. It was clear that PMF or die was on to something, but it required someone to make it their life's work. Exactly.

Starting point is 02:01:06 And if someone was really, I'm going to be the Mr. Beast of tech, I'm going to be doing crazy challenges all the time, live streaming, experimenting with all the different formats. There's clearly something there. I think Cluelly should run it back. Cluelly is a good candidate. Or Cluelly Hackathon. Anyway, we have Leonard back in the studio. Welcome to the TVPN Ultramd, Leonard.

Starting point is 02:01:23 How you doing? Hey, happy Friday. Happy Friday. I love to talk about Mr. Beast instead of age 20. again, you know, why we're already on it. Yeah. I mean, on that note, the Mr. Beast, if he was going to lock a bunch of, a bunch of programmers in a room with cursor, how bad do you think the gross margins would be?

Starting point is 02:01:45 Do you have a take on gross margins of application layer companies? We've been talking about that all week. Do you have any insight in there, anything that is muttering through your whisper network? Unfortunately, I'm not in San Francisco, so I don't know how code. nowadays work. I'm out of the old breed. When I did software engineering, it didn't have AI. But I always notice when I use my clock plan and I run out of like queries and was like, oh, damn, I need to write on my own and think of my own, who do I query? And then I spin up my second chat chad chivity subscription. So I think it would already apply to me. Being stuck without AI is quite

Starting point is 02:02:20 a problem nowadays. Well, give me the current read, your current take, the latest and greatest on the H20 debate. Where do you stand? pro-exports, pro-banning the exports. How, how, has anything shifted your thinking around it over the past? Do you want to nationalize Nvidia? Do you want to invest in Intel? You're trying to buy low. What do you think?

Starting point is 02:02:45 Well, the last time I was listening to the president speaking about Nvidia, he was more talking about initially wants to break them up. I know national, right? Because they were so big. Break them up, but then roll them up later. We've seen this playbook like 20 times with Trump and we get shocked He's knocked every single time. He comes out and says something,

Starting point is 02:03:02 this is the worst thing ever, and then a week later, it's the best thing ever, and we're partnering and we're doing a deal. Yeah, yeah. Yeah. Well, what he said during the, what was it?

Starting point is 02:03:11 It was the action plan launch, right? Yeah, I was sitting in a room. He was talking about Jensen, pointing to Jensen, and he was just saying, I wanted to break them up, but it's so complex. And I think just basically somebody convinced them, it's really hard, and therefore you're not supposed to break in video up.

Starting point is 02:03:24 Yeah. Right. So fair enough. And V-Vidia doesn't have as clean of a line to break up as Intel where you could you could gaming we're taking you over here yeah yeah we need to see you focus are going somewhere else yeah what would you do you'd open source kuda or spin that into a separate company like if you were even to break up in video like what would you actually do do do you have any yeah i think the soft vehicle system might probably be the strongest one here but again

Starting point is 02:03:50 it just goes hand in hand with the design right so again yeah i think it's i think it's a fairly hard one but for what it's worth i think their market share is only going to go down like there's more more competitors. I mean, the total, totally evaluation will go up. Don't get wrong. I'm like, I'm bullish on AI and Nvidia, but like all the other companies, all the other chip designers, they're just getting better. Is that driven by AMD catching up or new? What does Aaron call them? They're like the new types of chips like Cerebrus, Grock and etched. I forget what they're called. There's a new name for these crop of ASECs that are designed specifically for AI. And they could potentially pose a challenge, but they're certainly not taking market

Starting point is 02:04:29 yet it feels like it's it's mostly Invidia then AMD then maybe some Huawei the hypers right the tpherers the TPUs the APU AWS with the Trenium Google has the TPU since forever I mostly think about them right I think it's pretty clearly the case they have all the incentives in the world to build their own AI chips and reduce Nvidia's margin sure on the startups let's see how they're doing right I think hardware is hard hardware startups generally fail but if they find the right niche you know it's pretty hard And V-Divis is a more general thing.

Starting point is 02:05:00 And if you're like a hardware startup, you want to find a more narrow niche to be like more application-specific. And if you hit the right point, whatever next big thing is in AI, they might succeed. And we just see more and more of them getting there, right? And like we see Anthropics and other companies

Starting point is 02:05:14 using Google GPUs using Traneum. And again, the debate of the showy of Huawei is also getting better. They will also just, the market share can only increase, right? Okay, help me reconcile this. Google DeepMinds been seemingly fine with TPU and not having Kuda in their back pocket. They're on the parade of frontier.

Starting point is 02:05:35 The Gemini models are great. V-O-3 is great. The new Genie model is great. It seems like they are not suffering or falling behind despite not having Kuda access. But then simultaneously, we're hearing that Deep Seek, High Flyer, Alibaba, they want to train on NVIDIA.

Starting point is 02:05:55 They're not satisfied with Huawei. Why is Huawei? behind Google's TPU business? I think that's an example always spring up, and people say it's going to be so hard to switch Huawei. Google eventually succeeded, but also Google struggled. The TPUs are pretty, pretty odd. This was way before an AI hype.

Starting point is 02:06:13 And they actually also struggled. I'm not sure if you guys remember TensorFlow. This was originally what they did, right? And then later they switched to Jax and right, and they have the pie torch and everything around that. So I think over time they were struggling with software, but I generally see this as a one-time investment, and Google is a big enough of a company

Starting point is 02:06:29 that can just pay this one-time investment, then you develop on top of it, and eventually you will be there and you break even. You could probably do a survey if people are doing fine, like probably people still prefer CUDA because it's a bigger ecosystem, but it's just saying Google's doing fine. And again, Huawei will struggle,

Starting point is 02:06:44 it will take some time, but eventually they'll get there, in particular if they can use cursor along there, right? Yeah, yeah, yeah. And I guess to some degree the flip side is like, TPUs have full access to TSM, ASML and Huawei's restricted all throughout the supply chain. And so, yeah, and so like the latest Huawei chips, we were just talking to Bill Bishop.

Starting point is 02:07:07 He was saying that, like, that was from like two million, was it, dyes that they got from TSM through a shell company. And so they have this like one-time batch of supply chain, like ease. And then they were, and then from then on, they were supply chain constrained again. So then they had to go back to doing everything themselves. And that was a lot harder. Whereas Google just calls up at TSM and says, hey, do everything that you do for Nvidia. Just do it with our design, which is like probably slightly different.

Starting point is 02:07:34 Right, right. Anything else you want to dig into? No, I think we've come in. I think we've covered. Yeah, yeah, we've hit this pretty hard. Well, actually, have you guys covered the semiconductor supply chain and how good Huawei is? Because there's this quantity thing is that the, like in the middle of the debate, in my opinion, which I think is being missed here.

Starting point is 02:07:50 Please. So we, everybody compares the NVDAH 20 to the Huawei A's and 910C. and that's the best chip that's been putting out there. And if we look at the Huawei's 910C, it's like 80% there when H100 is. So like two or three years later than Nvidia, they're finally slowly getting there, approaching on the hardware specs alone.

Starting point is 02:08:09 And again, we can look at the specification sheet, compare them one by one, but it's never tells a real story, right? If you would do this with AMD and Nvidia, EMD is on paper on the chips as good as Nvidia, though one of them is 95% market share, they have one less than five, right? So looking at the spec sheet,

Starting point is 02:08:25 is never enough. That's where the software ecosystem come in, where we just talked about. Huawei is definitely struggling there. I think they will eventually get there. They just need to have more developers. And I think that's exactly one argument favor of letting the H20 go there, right?

Starting point is 02:08:40 The more people use the H20, less people use to Huawei, so less developers are developing the ecosystem. But where then comes in is, how many chips can I produce? We got one number undersecretary Klessar testified, so he's supposed to tell the truth, 200,000 ASE and chips this year

Starting point is 02:08:57 versus we're trying to sell, I think it's 700,000 to a million H20s this year to China. So this is where then the debate struggles, right? So if we wouldn't sell them the H20s, it's not like they have more ASE and 910 C standard, maybe they have more developers then, but they share the limited number of GPU resources. And that's the thing, which I think needs to be debated here.

Starting point is 02:09:16 And you can fall on both sides of the debate here. But like we need to understand that China is struggling and they cheated these, it's not even $2 million, it's 2.9 million dies, right, because they're struggling so much with their own production. Interesting. Yeah. So, yeah, I mean, maybe 200,000 is enough for people to actually bootstrap that software ecosystem. Certainly something to keep tracking. I think people consistently overestimate China in a bunch of different areas. Yeah, yeah. I mean, they can do software. I think they will eventually get there. And I think the idea they're just like Huawei's ecosystem will

Starting point is 02:09:53 always be terrible. It's just, look, don't go wrong. I would hope it's true, right? But they got good coders. They got good designers. They will just get better on all of these kinds of things. And maybe Kuda will always be better. But like Huawei is probably at the bottom right now regarding how good the ecosystem is. And when the deep seek engineers, you know, they're getting to it and they're struggling, well, but we'll get better, right? I think that's for granted independent of 2000 ships or million chips. Yeah, I think I still fall on the camp of probably H20 exports do put

Starting point is 02:10:23 enough pressure on Huawei to justify it, but it's tricky. This is a thorny one. It's not extremely clear cut for me. We should do a case study on Amphropic because they're the most beautiful example because they got so many different chips they're using. Yeah. Right?

Starting point is 02:10:39 And how's it going for them, right? How long did it take to train on trainium? Are they training on trainium? Are they deploying on trainium? I think this would give us an insight. How many engineers are they spending there and how bad as it still is. I think one question, one question is,

Starting point is 02:10:52 you know, is Beijing playing 4D chess by leaking out, you know, don't use the H20, don't use the H20, so then they, is that we just pile them into the country, right? I think so.

Starting point is 02:11:04 Is it for D chess? It's just like, you create an artificial demand. You'll say like, look, guys, you better buy some 910 Cs. You really don't want to. So we tell them, oh, if it's for sensitive government use, you'd rather use 910 Cs.

Starting point is 02:11:16 So we only see strong encouragement. not full-on bands yet. And we've seen the same game with CPUs. Ask Intel how it's going. I mean, Intel in general, but also Intel CPU market in China. The government also started encouraging there, basically, can you please use homegrown CPUs? Yeah.

Starting point is 02:11:32 Right? So we've seen it all over. Yeah, I mean, it would be super easy for the Chinese government to import some crazy tariff, level the playing field more that way, like even ban the H20 importation. Like, there's so many different levers that they could pull. And the fact that they've stopped, They've only gone as far as like strong encouragement. We've kind of inflated that to be like,

Starting point is 02:11:51 it's political incorrectness. Yeah, if people inflate to be like it would be insane, it's suicide to not to go against a recommendation from the, from the CCP. But it does feel like they could have gone a lot further very easily if they wanted to. So we'll have to see. I mean, it'll show up in NVIDIA's earnings, right?

Starting point is 02:12:11 We'll see. Or we'll probably get data. So we'll have to have you back on that. But thank you so much for stopping by. Great to see you. Hope you have a great weekend. We'll talk to you soon. Tim, do you talk soon? Take care. And if you're looking to get some sleep this weekend, get an eight sleep, a pod five. They got a five-year warranty and 30-night risk free trial for returns free shipping. Jordy will rebut the code. TbPN. I was going to say John Exley has landed.

Starting point is 02:12:33 He's landed. Yes, let's hear it for John Exley. Welcome back to the stream. Thank you for joining. We have, we got to pull this up. Trump and Putin are meeting right now. Okay. And I have a video in the chat team if we want to pull this up very cool look at this john so trump and putin are walking and what do you see whoa oh that'll fly over that's a show force where are they that is wild yeah what what a display of force um i i wonder i wonder where they're they're in alaska alaska okay that's sort of neutral ground i guess it's technically america but yeah fly in the Well, we got our bears there.

Starting point is 02:13:18 Yeah. That's another. They should have a bunch of, you know, Codiac bears stamp key. What a fun place to meet, Alaska. Anyway, it'll be interesting to see what comes out of that. Hopefully it's a resolution of the Ukraine war. I mean, like, Trump has been, you know, talking a big game about being anti-war,

Starting point is 02:13:34 wanting to know more foreign wars. No, don't send all the money overseas and save that for the taxpayer. Say that for real estate deals, maybe. We could be building, you know, golden skyscrapers in America with all those, with all those drones we're sending, but we'll see where it goes. But hopefully a peaceful, a peaceful resolution. We will see. Well, next up we have David from...

Starting point is 02:13:55 Addquick.com. Out of home advertising made easy and measurable. We'll say goodbye to headaches of out-of-home advertising. Only AdQquick combines technology, out-of-home expertise and data to enable efficient, seamless ad buying across the globe. And we do have our next guest here. David, welcome to the stream. How you doing, David?

Starting point is 02:14:12 Thank you so much. What you got for us. Jordy's warming up. He's got the mallet ready. He wants to hit the gong. It's the first one of the stream. You got some good news for us? Yeah, yeah.

Starting point is 02:14:26 Yeah, WebAI. We're working on some pretty interesting things. Just recently, we announced our new knowledge graph mechanism, which is out-benchmarked all of the best models year to date. By how much? 7%. Seven percent. Let's go.

Starting point is 02:14:45 There we go. We like to hit the gong for big numbers. We like to hit the gong for big fundraisers. We also like to hit the gong for improved. Benchmark maxing. Yeah, no fundraising announcement today soon. Soon. Soon.

Starting point is 02:15:01 I can't leak it today. Well, we'll be refreshing our For Rock's account. But you have a fact raised money. Yeah, talk more about the genesis of the business, why you started it. And yeah, what got you guys here? Yeah, yeah, absolutely. So WebAIs really focused on building models that can live on devices, like the ones on your desk. Right?

Starting point is 02:15:26 So the Genesis, the company really started working on a watch or? Yeah, absolutely. Yeah, absolutely. All of it. Yeah. So companies started by working in computer vision. And we were working on how could we take like the YOLO models, if you guys are familiar with those.

Starting point is 02:15:44 It was in the early 2016 era. were like the biggest models because language models weren't really mature yet and did early work there, ended up creating our own runtime engine, so our own AI library and our own network protocol. And what this enabled is us to run state-of-the-art AI models across devices distributed. So when you think about the future of intelligence, we really believe that civilization model is the most likely outcome for superintelligence. And what we're building is the rails for that. So we serve and distribute models across hardware. So we're running some of the world's largest models today on things like a laptop.

Starting point is 02:16:22 So when we say we out benchmark like Opus 4 or GP5 in knowledge retrieval, that's happening on a laptop. So it's not like it's a pretty significant breakthrough in modeling. And we're doing this in lots of different industries. But what we believe is going to be a big step change in, you know, unit economics for AI as well. It's just not there in the cloud model. It seems very important because all week we've been talking about gross margins or the lack thereof in a bunch of these different applications. In France is basically free when it happens on device, right? That's the goal.

Starting point is 02:16:59 Yeah. And you can do some things that cloud players can't do. Right. So part of the way we're getting this accuracy, like there's always like this no free lunch, right? So why wouldn't, you know, Anthropic do what we're doing to get this huge accuracy retrieval bump? Well, it's RAM intensive. So if we're distributing across devices, we can arbitrage, right? So we can say, okay, we'll pull more RAM because we're inferring on a device.

Starting point is 02:17:22 But if you're hosting this for a million users on Nvidia, you can't do that. You can't load additional RAM resource for every user. It's just not efficient. But there's real things that happen on the edge that unlock, I think, technological paradigms in AI. that are more meaningful, like more accuracy, more context, all of that. And we're seeing that real time. What about privacy, too? Absolutely, right?

Starting point is 02:17:49 So in our stack, everything's downstream only. So when we partner with the group, like, we work with the ORA ring, if you know that company, we're doing the AI for them. And think about health data. Like, you want that to be private. So the dream there is how can we facilitate personalized models for millions of users that never leave their device? React to this post from Take

Starting point is 02:18:12 Kim, author of the NVIDIA way, he says, here's what I would do if I was the CEO of Apple. Quadruple the RAM and iPhones to 32 gigs, have the max model at 64 gigs. Memory is oxygen for local on-device AI. More equals smarter and more powerful. Take the margin hit. Memory isn't even that expensive. What do you think? I think memory, I think he's right. I think memory is fundamental in these models. I also think we need to tread lightly on this idea that. that we're retooling infrastructure and we're making all these big bets on hardware with frankly a pretty immature algorithm.

Starting point is 02:18:50 Transformers are not necessarily the winning algorithm. So I think we need to be cautiously optimistic, but we need to continue to work on what's next. Like I just, you retool based on all of these factors and an algorithm changes. And we don't know what the long tail of hardware is gonna look like. And in video was really relevant,

Starting point is 02:19:12 because pre-training and all this, but now pre-training isn't really happening at the same level it used to. And I think generally more RAM is a safe decision. But also, I don't know if I would jump in and totally rewrite how we're building chips until we know that this is the architecture we want to stick with. Would you recommend someone buying a new Mac, max it out and get the most memory possible? Yeah, absolutely. Absolutely. Yeah, why not? Why not? What about diffusion models? Do you think that there's a chance that they have a comeback. We saw that demo from Google where they were doing text like token generation through a diffusion model. It felt kind of like a wild card scenario. I don't know. Yeah. It's actually performing on benchmarks, but it seemed like a path in the tech

Starting point is 02:19:57 tree that was kind of, you know, more or less forgotten, relegated to image generation, but then kind of making a comeback maybe. I think I think there's lots of things that have been unexplored, relatively speaking. We spent so much time on Transformers, but we haven't spent equivalent amount of energy in dollars on other architectures that we know work. And we know they work at specific things, but there's typically a broader application. I think it's really interesting. I mean, we're working on new architectures today

Starting point is 02:20:24 with both like the public sector as well as the private sector. And we're seeing a lot of breakthroughs that I think make the transformer look a little old. Oh, interesting. How do you think about the business model here? Because you're not going to be selling hardware to an OEM in the supply chain, but you're also not an API, so you're not pricing on consumption basis. It feels like there's a world where companies are comping you to an open source

Starting point is 02:20:56 thing that they have to implement. Like, how, how, like, what does a great relationship with a big device manufacturer, like edge computing provider look like for you? Yeah, I think, I think WebAI, one, we have a license, right? Because we have a proprietary tech stack. We're not a rapper. We appreciate rapper companies. We think they're doing cool things. But we own our stack pretty vertically. So we own our runtime, our AI library, and our tooling around that.

Starting point is 02:21:23 And so when we work with a partner, we typically structure a base license minimum. And when we have that, when we have that license, we can inject forward deployed engineers to work with these companies that honestly just don't have the AI talent quite yet. And they need help. And I think that's something that people aren't talking. about is, you know, like these products don't necessarily solve a problem out of the box. A lot of these enterprises like Fortune 100 need help. And so we do that. And additionally, there's a way to take part in the success and the deployment.

Starting point is 02:21:54 So the usage fees we can get, even though we're running on device. So because our network is managing that. So you can imagine WebAI, you have two and a half million custom devices or maybe it's an iPhone. And we're shipping across that. Our network manages all of that. So we collect fees on that. So it sounds like it's somewhat case by case, but you could imagine charging like a per device license,

Starting point is 02:22:18 but also like a per token license in the future. Her answer is typically how we structure it. So it could be a book. It could be a one word answer, as long as it's an output that's solving a problem. We mostly work in mission critical use cases, like things like reassembling engines with multimodal AI, you know, health diagnostics,

Starting point is 02:22:40 public sector work. Yeah. How quickly are you going to kill Jordy's battery if you're doing test time inference on device? He's been already complaining about the iPhone not having enough battery life, but it feels like there was a glimmer of hope when we were just like, let's

Starting point is 02:22:56 just distill the models and it'll just be like a pretty short inference chain, but if you're, even if you distill the model, if you're inferencing for 10 minutes, that feels like a lot of heat in my pocket. Well, I don't know what Jordy is using.

Starting point is 02:23:11 I would assume it's a pretty nice phone. It's like an iPhone pro. Yeah, yeah. So, I mean, you mentioned quantizing. So I'm going to talk a little bit about that and what we're doing there. So we released an open source paper around a tech that we were building early that we've now expanded. And it's a little, it's more sophisticated now, but the principal's still there. It's called EWQ.

Starting point is 02:23:34 And instead of just quantizing and tell me if I'm going way too tactical here, Quantizing traditionally, you have like a fixed value. So we have, let's say you have a full precision model. And when you quantize something, you say, okay, I'm going to quantize it to four bit, or I'm going to go to 16 bit. And so you're just drastically chopping the model down, right? So from the float values that it can pass through.

Starting point is 02:23:55 With EWQ, what we do is we have something called device profiling. So when a WebAI model hits your phone, it's running our webframe library, and it profiles your hardware. And then what we do is on inference. we run EWQ. And what EWQ does is it does real-time quantization. So based on your question and the inference,

Starting point is 02:24:17 and what it leads to is close to 30 to 40% model reduction size in RAM while retaining accuracy. So what that means is we get bigger models inferring. And instead of like this one size fits all quantitation, we dynamically do that on inference. And what that leads to is less energy consumption, higher accuracy, less usage on the device. Yeah, so somewhat similar to the model routing that we're seeing in chat GPT now.

Starting point is 02:24:43 What were your overall reactions at GPT5? It's just an M-O-E router. I was kind of hoping it was a new foundational model. And when you interact with it, it's really clear that it's just a way to dynamically control price based on a question. So like you ask a question, they route you to a different model. If it's coding, it will route you to a different model. I can see where that's valuable. I have a lot of people that are non-technical that are in my life,

Starting point is 02:25:14 and I've watched them now switch off of GPT after the five release to things like ROC, which is kind of shocking to me. But I think people were used to a certain standard of response, and now the lack of transparency in picking the model you're engaging with, I think created some whiplash. But I'm sure there's areas where it's amazing. I haven't really gotten to tap into everything there. been enjoying a lot of the anthropic releases and typically probably tend to lean that way.

Starting point is 02:25:44 Cool. Well, thank you so much. Congrats on all the progress. And I hope you have a great weekend. Come back again soon. Sounds like you'll have to news. Yeah, absolutely. Yeah, we're excited. Yeah. Yeah. Absolutely. Great to me, David. Thanks for, thanks for joining. Let me tell you about public.com investing for those who take it seriously. They got multi-asset investing, industry and yields. They're trusted by millions, folks. Should we go to? Billions soon. What did Samma mean by this? If we didn't pay for training, we'd be very profitable. We talked about this.

Starting point is 02:26:13 Kristen Culver says, most successful coups in history. Napoleon Bonaparte's coup of 18 Brumere. October Revolution in Russia, 1917. The Nazi seizure of power in 1933. Egyptian coup d'Ate-ta, 1952. The Chilean coup in 1973. Big one. And the Open Door Retail Army at Open in 2025.

Starting point is 02:26:36 They could. Chris worked at Open Door, correct? I believe. Must have. And so she's having fun. It'll be interesting to see in other news. Yeah, so the CEO stepped down. This morning.

Starting point is 02:26:48 Yeah, this morning. So Kerry Wheeler posted on X. Today I'm stepping down as CEO of Open Door. When the board of directors asked me to take on this role at the end of 2022, the company was in crisis, the real estate market was punishing, the business needed a reset, and the path forward was uncertain. My mandate was clear. stabilize the company and do what was necessary to survive. Of course, I said yes, because I believed in Open Door. It wasn't easy, and it wasn't about glamorous headlines, but we stopped bleeding. We restructured the business, rebuilt an exceptional leadership team, got an NPS of 80.

Starting point is 02:27:20 And she says, I'm pleased the leadership team will continue to execute on the vision strategy. I'm closing this chapter with pride, clarity, and gratitude. So good luck. It is wild. Open Door is up 200% in the last 30 days. and the retail army said, no, we want more. They want more. New leadership. Retail traders are crushing it.

Starting point is 02:27:42 Well, I'm excited to see where Kerry Wheeler goes next. We should watch the new Jason Carman film. Let's do it. We have Cameron Schiller coming on the stream in just a few minutes. Let's pull up the latest work from Jason Carmen. I'll be right back. Dear son. Some time ago,

Starting point is 02:28:07 One. We have the desktop. The machines roared. The steel bent to our hands. We built for the stars, for our land, and for your future. We went fast. We went far.

Starting point is 02:28:47 It made us strong. It united us. The sound faded. to build, drifted away. Those who knew grew tired. But now, the fire returns. Steel needs you. Will you be, take us.

Starting point is 02:30:23 The stars are calling. Our future is waiting. So will you answer? It's a sick fit. We need one of those. When we're discussing hard tech, we should have gotten the suits today. Huge miss. Anyway, new video from Rangeview.

Starting point is 02:30:57 We have the CEO, Cameron Schiller in the studio and the TVPN Ultramm. Welcome to the stream, Cameron. How you doing? What's happening? Hey guys, good to see ya. I have so many questions. Did you act in that? Are you in that?

Starting point is 02:31:09 I am not in that. You're not in the suit? What? You got it, how, I thought you financed this whole thing. I thought you made this happen and you didn't, get a camio? You got to put yourself in. The machines got cameos. I know.

Starting point is 02:31:23 Right? The machines got cameos and I believe that last scene takes place at Rangeview HQ. Is that correct? Did I clock it correct? That is correct. That takes place in our technology demonstrator factor. We're running too. We got a production facility down the street, which we'll see in some new videos coming out.

Starting point is 02:31:36 But that was at this facility, which we've been at an hour just busting at the seam. So we've got to move and you'll see a bunch of content from that new one. That place is sick. They used to build space shuttle engines there. That's awesome. Wow. Yeah, the space shuttle shot was fantastic. I mean, Jason Carmen, he puts on a clinic every time he drops a video. What inspired it? What was the message you want to send? Is this just, is this a recruiting film? I noticed like the follow-up post was like, come work for us. This is not like an ad that you'll be running to get customers necessarily, or is it just kind of like vision film? What was the thinking? Yeah, I mean, it's really a message to America. I think it's a wake-up call. It's a question of who we really want to be as a country. What do we want to do? Right. I think, I mean, that was a big part. I mean, that was a big part. part of my life growing up going back and forth to China.

Starting point is 02:32:21 And I saw the American dream in China when I was there. I saw people from the center of the country, moved to the coast to work extremely hard and make a life for themselves. And when I came back to America, growing up, I just didn't see that here. And I think we have to bring it back. I think for national security reasons, I think across the globe. So the real question is, you know, can America bring it back?

Starting point is 02:32:46 Because we need to make a lot of parts very soon. And this is less so about range view. I mean, America needs a thousand range views. This is about people that are considering making a big pivot in their life to work on something that matters to the world. And when you say make a lot of parts very soon, is that specifically like defense tech and warfare? Are you seeing, are you worried about great power competition or is it more like we won't get the next generation of 9-11? or like the next great physical product won't be made without this happen. The one that I care about the most is the second one.

Starting point is 02:33:23 But the first one is definitely very real. If you think about it, you know, America did some amazing stuff. The F-117, we invented stealth technology, the SR-71. All that stuff happened in America. And that happened because I think of factory towns. I mean, I'm calling him from El Sigundo. This is a town that literally runs on jet fuel. Like there is a refinery that's right next door.

Starting point is 02:33:45 this pumping jet fuel out, you know, under this city to feed to LAX, which is on the other side of the city. And you feel it in the air. Like there's something that happens when a community wants to be a part of something great in the world. And when we look around, everything around us has been made in China now. And with it, you know, with it slipping, I think a great calling to be a part of something amazing has slipped as well. So we really need to bring that back. We're going to do that with, uh, with parts. We're going to do that with new technologies that enable factory towns all across the country. Yeah, I mean, I think it's super. What's going on behind you? There's like some scrolling image? What is that? Might be, I mean, we've got a lot of screens. Is that a screen? It's like a

Starting point is 02:34:31 TV or something? Yeah. I mean, there's, there's a lot of lights and there's a lot of this camera reacts. Yeah, yeah. Yeah, yeah. Right, right, right. Yeah, I think, I think what you're, one way to kind of summarize what you're kind of getting at. And from my view is it's important for American dynamism to not be like a venture hype cycle. That's sort of, it's not something that can be accomplished in two years since, you know, moving to El Sigundo became popular and a meme. And it needs to endure. Who helps with that?

Starting point is 02:35:06 Was it me, you, Jason, camera? Carmen. Yeah. We might have played a role in that. A little bit to do with it. But I mean, I guess the question is like, you say this is like, you know, a wake-up call for America. Like, are we not awake? I feel like, I feel like a lot of these, a lot of this message is broken through.

Starting point is 02:35:24 Like, like, what, what's left to say? What, what do we need to- I say it's broken through in the bubble? Yeah, yeah, maybe it's the bubble. Maybe it's the bubble. Yeah, what, I mean, what is your take on like the re-industrialist summit's huge? Like, you know, the American Dynamism Summit is huge. Like, like, people seem to be beating the drum.

Starting point is 02:35:42 People are at the White House. They're in D.C. They're a fraction of the size of Salesforce Dreamforce. John. That's exactly it. Yeah. Are you going to be there? Dreamforce?

Starting point is 02:35:51 Let's get you there. This man loves Enterprise SaaS. He won't, he won't admit it on camera. He plays this character that he likes reindustrialization. But really, he just wants to, he just wants to code. Yeah, it's all I want to do. That's all I want to do, John. No, we need to, we need to have more.

Starting point is 02:36:09 people where it's a it's inside the bubble and B we need to encourage the people that are working on the problems to focus on the things that matter and that's actually making parts we need more factories we need more metal moving moving metal is the problem right now there's a lot of people building tools for factories there aren't that many factories yeah yeah so if you want to join build a factory make parts move you know do real stuff in the supply chain I'm not talking like screwdriver factories bolting on imported components that's the vast majority of you know assembly in America is like what's left so we don't need

Starting point is 02:36:38 that we need people working on hard problems we also need you to a thousand X we need you to thousand X range view you said earlier we need a thousand range views but why don't why don't you just be copy and paste your yourself working as hard as I can yeah what is well what can you tell us work harder well what can you tell us about the state of the art in in manufacturing I know you there's there's additive manufacturing subtractive manufacturing there's CNC we've talked to people that are 3D printing metal now there's casting what what are you excited about what do you focused on where do you think there's still pockets of opportunity? Yeah, great question.

Starting point is 02:37:13 So we are working on casting and we are we're trying to give casting at CNC moment. And explain casting for the for those who don't. There's a casting is liquefying molten metal pouring it into a mold. It's solidifying and you're getting the you know the part that has that shape and almost everything is cast. It's you know even CNC shops buy castings. Today castings have have eroded so much that machine shops are just buying cast, locks and then they waste a whole bunch of time cutting apart into you know into its final part and a lot of chisling but if you get really good at casting you actually just cast 99% of the way and then touch the final bits up with

Starting point is 02:37:50 the drill bit so there's no one-size-fits-all solution in manufacturing it's one of the first things you learn it takes like it takes like a hundred humans from you know mining the ore out of the ground to installing the bracket on the end of the thing to actually make something happen and there's a ton of folks I think for for people looking at technology they're used to looking at manufacturing is just another sector. You know, there's FinTech, Health Tech, manufacturing, the truth is manufacturing represents more of America's GDP than all of tech combined.

Starting point is 02:38:19 And so it's huge. And so inside of manufacturing of all these sectors, and so many have just not been looked at yet, or not been touched. And so we're seeing resurgence. The other thing is, you know, you maybe shouldn't finance these things exclusively with all venture capital because the risk profile just isn't the same in the factory, right? Like if my factory burns down, I'm going to still have, you know, a thousand pounds of super alloy.

Starting point is 02:38:42 You know, maybe the crate that it wasn't caught on fire, but like they're not going to move, right? So you should have, it's not risky. It's not a risky bet. So you shouldn't have buy that stuff for that. So I think there's a whole new level of financing that's going to come in. And you see this happening with a few of these big factory companies where you're getting really smart finance deals,

Starting point is 02:38:58 where you buy the technology improvements with venture capital for those returns, but the rest of the factory is financed in a different way. Yeah, that makes sense. If I were to pull a hot take out of you, based on what you just said, it sounds like potentially the American manufacturing industry has over rotated towards subtractive manufacturing and needs to rebuild additive manufacturing or casting capability. Is that like roughly a reasonable take that you-

Starting point is 02:39:23 Additive and casting are not the same. Additive is like this SPAC machine that's blown up actively as we see. There are a few amazing people doing additive. But for the most part, like it's missing on qualification and it's missing on real unit economics at scale, which is as a whole, that's what America's missing, like really being able to build stuff at scale. If we had to triple the manufacturing output of the country, we'd be cooked, totally cooked. Like it would take us five years to get the factories up to do that.

Starting point is 02:39:47 And all the factories that would start would be sending them money overseas because none of this equipment is made in America anymore. We lost the factory industry, but we also lost the factory and machine tool industry. So all this stuff just goes overseas. So I wouldn't say that additive is it, I think casting is really important. A lot of these traditional forms of manufacturing are really coming back and making a big play. But we should just be encouraging everyone to make a lot of parts. Like we need so many parts and we need to get started immediately.

Starting point is 02:40:14 Parts maxing. Talk about your dad. Talk about the influence there. Jason Tanteased it a little bit, but I haven't heard the story. Yeah, yeah. He's a big part of my life. He's always encouraged me to be very, very honest and real about this world, which I think is really important in venture.

Starting point is 02:40:33 And he was a maker himself. You know, his family was Pittsburgh and he was, you know, Midwestern family values. And they, you know, they, he came here to work on the B1 Bone. The Super Sonic Bomber was pretty sick. And then I ended up growing up next to where Skunkworks was founded. So, you know, Bob Hope Airport actually, yeah, for once it, all that stuff happened there. And then it went out to Palmdale. And then it became a service-based industry.

Starting point is 02:40:58 Lots of B-to-B SaaS and entertainment happened in the area. And it really changed. But he always, you know, kept me, kept me centered. and he's a huge influence on my life. And actually, same with Jason's dad. So we bonded over that a lot. And they're becoming increasingly large parts of both of our lives. Can you raise like a billion and then run this ad as a Super Bowl ad?

Starting point is 02:41:21 You guys want to help me? Yeah, we got to get him back on the venture training. Cameron's always very like, oh, anti-venture. But we need to run this film as a Super Bowl ad. Spack Rangeview, come on. We can bring your own. Just to get the capital for a Super Bowl ad and then you can figure out to take private. 10 years of Super Bowl ads.

Starting point is 02:41:40 Run it every year. This is not going to happen overnight. Yeah, I'm in. Let's talk about it. Let's make a game plan. Fantastic. Well, thank you so much for hopping on. Congratulations.

Starting point is 02:41:49 We'll talk to you soon. Congratulations on the lunch. Cheers. Where can people go to apply for jobs? I know that that's important right now. Rangeview.com. Scroll down careers. Rangeview.com.

Starting point is 02:41:58 You heard of here. Thank you so much for having on. Have a great weekend. Cheers. We'll talk to you soon. And I will talk to you soon. to you about Bezell. You want to manufacture yourself a watch

Starting point is 02:42:09 on getbezzle.com. You Bezell Concierge is available now to source you any watch on the planet. Seriously, any watch? Any watch. They got them all. Well, I'm very excited for this next. Well, okay, so there's, I was gonna,

Starting point is 02:42:23 I thought we had our friends over at NFM live. They will be coming on in just a few minutes. But we will be joined by Surak or Surak. Syriac. Syriac. Good to meet you. Thank you for joining the stream. Why don't you take us up with the introduction on yourself and the company?

Starting point is 02:42:39 All right. Well, I am Syriarch from Early. Early is an early cancer treatment company. And essentially what we do is we create genetic constructs that are injected into your body. And they disperse everywhere in your body. They enter healthy cells randomly. And if you happen to have cancer cells, they will also enter. those but only if it's cancer these genetic constructs will switch on like a light switch and then they

Starting point is 02:43:13 turn the cancer cells into little factories that are forced to make any protein of choice in other words you can make something that makes the cancer visible or you can make something that activates your immune system to attack and kill the cancer so So the whole thing is relevant because in the last 50 years, we've always tried to find some markers on cancer cells that make them detectable or drugable, right? Billions of dollars have gone into that. And yet, we still have 600,000. Yeah, we still have 600,000 people dying from cancer in the US every year and 10 million globally.

Starting point is 02:44:00 So something needs to change. What's the background of the company? Is the tech transfer that's come out of an academic lab? What is your background? Yeah, I'm actually not a biologist. Out of 35 people, I'm like one of two or three people who don't have that background. I'm an engineer. I'm a serial entrepreneur, and the idea came out of Stanford University.

Starting point is 02:44:20 And it was one of the world's top people in early cancer detection, who then himself sadly passed away from cancer. Wow. Including his own son died at 16 from cancer, and his wife died two years after him. The whole family is wiped out. So I met him. Was that, out of curiosity, was that environmental exposure?

Starting point is 02:44:42 No, no, no. It's mostly genetic. And the mother had a genetic mutation that then got transferred to the sun. And what Sam Gambier died from the inventor of the whole thing is unclear to this point. It was cancer of unknown origin. You didn't even know where the primary tumor came from. So a very tragic story, but he was committed to flipping the tables against cancer. So I don't know if you guys have Jordy or John, whether you have anybody in your family or in your friends' circle that has been affected by cancer.

Starting point is 02:45:21 Yeah, of course. You know, it's just kind of crazy that we are always behind, a step behind or two steps. behind. We're always trying to find the next marker that we could hook onto. So what if we could stop looking for any marker altogether? What if instead we could force the cancer to reveal itself and make its own therapy to kill itself? So what's the pathway to commercialization? Imagine you have to go through FDA approvals at some point? Yeah, we have to go through a phase one, two, three trial. and then to commercialization. And we have spent the last seven years cracking this really hard problem.

Starting point is 02:46:09 You know what the biggest problem is in cancer? What's a cancer cell and what's not a cancer cell? Yeah, of course. Because, you know, different from a virus, this is your own cell that has changed just a little bit. And so differentiating that from a normal cell or from something that looks like cancer, but it's totally benign, it's really hard. And that's what we've spent so much time and energy on with AI. We're essentially producing AI results, liquefy them, put them into the body into a cancer

Starting point is 02:46:43 drug that then forces the cancer to produce its own therapy against itself. And what's the latest news with the company? Well, we just raised $44 million. Congratulations. Sounds. It's not cheap work you're doing. Sounds expensive. Yeah, biotech is not cheap.

Starting point is 02:47:07 So, you know, I don't know how much you know about the biotech world. It is in the biggest funding crash in the last 20 years. On both the public side, the private side, I know that the government funding is certainly at an all-time low. That's right. Across everything. Actually, you named them. The private funding is a. extremely low because of two reasons, high interest rates, which immediately affect a long running

Starting point is 02:47:33 product like bio, takes 10 to 12 years, right? And then AI is like a vacuum cleaner for money. That makes sense. It sucks up all the money that goes to tech firms because for VC companies, many of them believe they can make a faster return by putting it into AI classic tech. Of course. But bio and AI is a great interface that is now coming. to fruition. And then the pharma companies, they are concerned about tariffs. They are concerned about China catching up to the US and they start buying ideas and drugs there. And then the government is not stepping in to flatten out the curve. And here I would actually say, we really got to make a national commitment to biotech to flatten out this funding curve because, you know, at the end

Starting point is 02:48:23 of the day, would you like to be dependent on China providing the most developed, life-saving drugs for cancer, for autoimmune diseases? Do we really want to depend on that? I mean, it's good if they supply them, but what if they don't one day? So we should actually have a national commitment to biotech to maintain the world leadership that the U.S. has had for the last 50 years. Yeah, good point. stopping by have a great West rest of your week and have a great weekend and congratulations we'll talk to you soon thank you talk soon bye bye cheers we have some major guests in the restream waiting room let me tell you about wander first wander find your happy place find your happy place book of wander with inspiring views hotel great amenities dreamy beds top tier cleaning and 24-7

Starting point is 02:49:14 concierge service it's a vacation about better well joined by this is the moment you've all been waiting for they're calling it the brothers welcome to the show you guys look fantastic we got headphones on let's can you hear us how you doing yeah yeah give us some energy come on come on what time what time is it in us yeah it's five oh five 50 in the morning in the okay so we're just waking up here here i'm gonna help you guys i'm gonna help you guys I'm going to help you guys wake up. With a little gong.

Starting point is 02:49:49 Basically. For the launch of the team. There we go. There we go. Let's. Thank you. Thank you. So talk to us.

Starting point is 02:49:58 How has it been going? How has it been running the show? What inspired you? Obviously, you know, TBPN inspired a little bit, but have you been working in media? What's the background story here? Yeah. Give us, give us life stories.

Starting point is 02:50:14 Yeah. So both of us coming from, venture capital backgrounds. We've been based in Seoul for over like four or five years. So yeah, we love this industry, you know, private market, VC tech, startup, this whole industrial complex. And we thought that it's something, you know, we could do above just, you know, pure investment and, you know, media could be the like perfect complementary medium. So, I mean, you know, anyway, we were doing our own thing, you know, like running our own, like, blog and writing essays like since a few years ago and this guy here,

Starting point is 02:50:47 Song Zhong, he actually with his friends is running one of the biggest VC newsletters in Korea and you know I'm running my own blog too. So earlier this year we thought that oh you know podcasts could be the most optimal medium to reach the wider audience and also to reach the global markets and we found you guys randomly on the feed. Yeah. Oh, yeah, this is the shit. This is the, this is the moment we got a new media.

Starting point is 02:51:16 Yeah, so, you know, of course, you know, we, if there's anything we want to benchmark from the U.S., it's not all in, you know, it's not the boomers, it's a decision, you know, like, the medium, so. Yeah, so what's your guys' schedule? Are you, have you quit the other stuff yet? Are you going all in? Are you just putting, how much time are you putting up? Is it three hours a day? Do you have multiple guests? Like, like, like, what?

Starting point is 02:51:40 What have you taken from the show? What's working? What is in what needs to be different to succeed in Korea? So I will say, so okay, first of all, so we're like running like three times a week. So we will ramp it up. You gotta get those numbers up. You gotta get those numbers up.

Starting point is 02:52:00 What are people gonna do on the other days a week? Yeah. They respect to just twiddle their thumbs. Does the Korean tech economy not function five days a week? 24 seven? Yeah, but we're going to ramp it up. I'm letting you check us out in like six months. We might be running like seven days a week.

Starting point is 02:52:17 Who know? Yeah, yeah. So anyway, yeah, you know, the Korean tech market is as vibrant as, you know, as the US, I would say. But, you know, like, but, you know, we are just kind of like, you know, being the frontier in this like, you know, hold the new media and like this like, you know, like, you know. Well, in many ways we're old media. Yeah, we're old media. This is just television. You can say it.

Starting point is 02:52:41 It's just TV. Is it? Talk to me about the guests. Do you have dream guests? Who's the Palmer Lucky of Korea? Who's the Elon Musk of Korea? Who do you want to get on? Who have you had on?

Starting point is 02:52:53 What we have a lot of venture capitalists, but then we have analysts, politicians. We've kind of gone all over the place. Where have you had success doing guest interviews or are you even doing guest interviews yet? You believe you are, right? So we're in the very initial phase. Sure. We've only embodied, so, you know, only a limited number of guests, but so far we have some, you know, like, DMs, engineers, also venture capitalists, also authors who just published books. But, you know, we want to have, actually, we want to bring in everyone, you know, everyone for EIP, you know, even the president, you know, like, like, you know, like, even Trump, who knows.

Starting point is 02:53:31 So we want to bring you guys in our show. Of course, yeah, we'd be happy to. Let's do it. Let's do it. Well, you know, make us wake up at 5 a.m. I guess. We do. We're ready. When are you guys live? I mean, that wouldn't work time zone-wise.

Starting point is 02:53:44 But it'd be more like instead of going to bed, we'll pop on your guys' show. Yeah. Are you guys live at 11 a.m. local? It's like 6 p.m. on L.A. 6 p.m. in L.A.? Oh, yeah. Yeah, we can do that. Dinner time.

Starting point is 02:54:00 Our wives will be very happy we're bailing on dinner to go on the N. What's the OpenAI of Korea? Yeah, what's the hottest company? What's the one that everyone's focused on? SK-Hinix is obviously like later stage, but what, well, who is ascendant? Okay, open air of Korea. Okay, we got a best treasure here. It's tough. I would definitely pick, you know, in terms of like, you know, semi-conductor business, SKHinex, Samsung, semiconductor, of course. And also we got some,

Starting point is 02:54:30 you know, like, Hasha Korean developers at Open AI. So, you know, like how do you say, how do you say cracked engineer in Korean? like, M-Cin Engineer. Like, you know, like, a M-Chineerineer is like cracked, you know. I can hear that. Yeah, yeah. The team loves it.

Starting point is 02:54:48 The teams loves it. That's great. Anything else, do you? No, this is great. We love what you guys are doing. And happy to come on the show. Yeah. And have fun.

Starting point is 02:54:59 Have fun out there. And you guys look sharp, too. Thank you for, for, for, you know, copy and pacing the suits as well. Yeah. this is on an info from you guys but you know like uh as as we want to you know like a make our own path and only you know from here on so um yeah we're gonna build our own brand this is nfm live and fm live love it well we support you guys enjoy thanks for coming on we'll talk Cheers.

Starting point is 02:55:26 Thank you. Good stuff. Good stuff. Lads. Lads. They're also pretty well positioned to cover defense tech. Korean, South Korea obviously has mandatory military service. I think most people kind of interrupts college.

Starting point is 02:55:44 They kind of take a break from school, go serve, then go back. So anyways, glad that we have contact with the Korean market. Yes, yes, definitely. Last post, close it out from Andrew Reed. I knew you were going to pull this one up. These shoes have gotten an obscenely high market share while accumulating zero aura. What are these? These are, I think, velas.

Starting point is 02:56:07 I never heard of this. Wouldn't, wouldn't, definitely came, you know, kind of a. He's just taking shots left and right. I think they're vea. Vejas. Vejas. I don't have a pair of these at one point. You did?

Starting point is 02:56:22 Yeah, they were very much just like shoes to me. Number one question on Google, why is Veja so popular? I mean, I wonder if the business is doing well. I wonder if they've figured out some sort of distribution, some arbitrage, maybe some, I don't know, are they more D to C? It does seem like a newer brand. And I certainly do. Founded in 2004.

Starting point is 02:56:44 Okay. So, headquartered in France. France, interesting. And yeah, I don't know. I mean, I think they just kind of tapped into the common project. projects, sneaker era. What is common project? What are you talking about?

Starting point is 02:56:59 Well, that was just like the definitive white, like a sneaker, right? But common, comments. That was the gap in the market. No one thought to create a white sneaker. I mean, a white leather sneaker that was not from a, no, no, no, but not sportswear. That's a key thing, not like basketball themed streetwear, right? Something that was versatile. But yeah, I wouldn't, uh, wouldn't be, wouldn't be cut.

Starting point is 02:57:25 dead in them. Those would be oara farming you if you put them off. Yes. Yes. You got to be careful not to get aura farmed by your own clothing. It happens sometimes. It happens to the best of them. But I'm happy for Beha's success. I'm happy for their success. Overnight success, 21 years. Keep it going. Anyway, that's our show. Thank you so much for listening and watching and enjoying the debate. We will see you on Monday. Leave us five stars on Apple Podcasts and Spotify. Can't wait. And thank you. I cannot wait. I cannot wait for Monday. I was I can't believe we get to do this again. I figured it was Friday, and I figured out it was right.

Starting point is 02:57:58 No, you really did think we still had, you thought today was Thursday. I thought we were a day behind and I thought we still had more time. Well, have a fantastic weekend's folks. We love you. See you. Bye.

TBPN - Slop vs. Steel Showdown w/ Delian & Everett, GPT-5 Backlash, Trump Eyes Intel Stake | Bill Bishop, Jimmy Goodrich, Lennart Heim, David Stout, Cameron Schiller, Cyriac Roeding, NFM Live

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.