"The Information" Profiles TBPN, AI Ups and Downs, Nuclear Regulation | Zach Weinberg, Leigh Marie Braswell, Stephen Balaban, Sam Lessin, Bobby Goodlatte, Auren Hoffman, Doug Bernauer, Isaiah Taylor

36:14They've all done podcasts, but there's no podcast that has done all seven. And it says Coogan. That's me. On the drive over, Hayes mentioned a dream prop for their Hollywood set. He and Coogan have already, already have a small gong in their Jonathan Club that they ring on the show when talking about a company's notable achievement, but he would like to buy an even bigger gong. Looking around at the Hollywood studio, he could see how nice an extra large version might look. Yeah, we finally have the space for it. Oh, what a fun piece. I can't wait. We need a 50 foot by 50 foot gong that requires a group of people to stop it from ringing when you drive a car into the gong.

36:53We do. We're working on it. What should we move on to? I mean, the main thing we have to cover, so yesterday X was a nightmare. DMs weren't working for a lot of the day, which was rough for us on the show. Apparently there was literally a fire in the data center. So yeah, a fire broke out Thursday morning at a data center in Hillsborough, Oregon, leased by Elon Musk's ex, forcing an extended response from emergency crews, according to multiple sources who spoke to Wired. The sources required anonymity as they weren't authorized to speak publicly about the company. Firefighters arrived at Hillsborough Technology Park in a suburb west of Portland at 1021 a.m.

37:31man john portland oregon vandalism uh uh you think this is you call an arson i have no idea i don't even know how you how you vandalize or commit arson in a data center they seem the only the yeah but the every time i see portland in a headline in tech maybe it's some type of this is pretty advanced stuff to figure out where the data centers are okay i'm sorry if you're uh uh extremist and you want to bring a bunch of gas will monitis posted that whole thread about shooting the transformer and taking down all the electrical infrastructure and how like you can basically take out the energy infrastructure with just small arms so like an ar-15 shot at a transformer can take off a whole a whole whole piece of the grid and so you would think that they'd go after that i don't know it's very very very odd hopefully they sort it out quickly yeah so it's interesting so before i mean tesla's have been setting on fire so maybe x data center so this is interesting so before elon bought twitter the company had three data centers in sacramento portland and atlanta this ensured that if one data center went down traffic could be shifted to the other two and split so no single data center was overwhelmed around christmas eve 2022 musk shut down x's data center in sacramento in an effort to cut costs the company experienced It's a major outage in the wake of the shutdown.

38:56Over the next six months, the company moved more than 2 ,500 server racks from the Sacramento facility to data centers in Portland and Atlanta. So it sounds like they have two core data centers now. Okay. I want to rip through a bunch of stuff really quick. Let's do it. I'm just going to rip through this stuff. So Donald Trump just signed an executive order about nuclear. And so we're going to try and have some folks on very quickly and slot them in. might be 2.30, but I'm talking to some nuclear folks about coming on and breaking it down. I haven't even had a chance to read it yet, but it's out apparently.

39:30I don't know if you want to dig into that. I also want to give you the final analysis on Apple's AI struggles. We've been dipping our toe in and out of this story all week, but Ben Thompson had a great summary of the problems with Apple's AI strategy, and he sums it up in these bullet points, which are really rough. One, Apple's head of software engineering didn't believe in AI. Apple's head of AI was skeptical about LLMs and chatbots. Apple's former CFO refused to commit sufficient funds for GPUs. Apple's commitment to privacy limited the company to purchased data sets instead of using their own, where Meta and Google and YouTube are all highly valuable training sources.

40:13Apple's AI team is understaffed, and the relative talent of an AI staffer still at Apple is uncomfortable to speculate about, but given the opportunities elsewhere, relevant. What should Apple do now? The baseline advice from Ben Thompson is make Apple's on-device models available to developers as an API without restriction. Make Apple's cloud interface, cloud inference available to developers as an API at cost. Allow Siri to be replaced by other AIs, perhaps for subscribers to those AIs who would revenue share with Apple. So you can bring ChatGPT into it. I bet for the access to the Siri button, Apple could get like 50 % up from their typical.

40:57Yeah. I mean, just hold an auction. Just hold an auction and be like, hey, yeah, like we're just not going to hit this out of the park. So we're just going to auction it off. It'd be great. They have precedent of selling the search bar. Yep. Right. Yeah. Why wouldn't they do? Anyway, the other news is people are now speculating about the exact device that Sam Altman and Johnny Ive will be building together at OpenAI. with the IO acquisition for$6.5 billion. Swix, who's been on the show, says they literally copied Avi Schiffman. $6.5 billion for what Avi Schiffman built in the cave with a box of scraps, quoting Iron Man, of course.

41:30And there's an old video from Avi pitching the first friend. And I mean, what a throwback. I remember this was 2023. I mean, this is what I was saying, I think on Wednesday or Tuesday. I was like, this could end up being extremely validating for Avi and Friends. if he ends up nailing. I mean, to be clear, he's also, I think, focused on a much different use case. I agree, I agree. Like friend.com. I agree. He's building a companion. To be fair, the bear case for Friend is that they wind up in the same position as Pebble, which is that they do the discovery to find out that smartwatches are a good thing.

42:07And then Apple comes out with the Apple Watch. They just leverage integration and it's just a much better product. And then Google didn't need to buy Pebble. they just created the Samsung Galaxy Gear or whatever, the Samsung watch. And so the existing hardware manufacturers, they haven't had to buy companies. And so they've wound up putting like a lot of the, there's not a winner in independent smartwatches. And so if there is a pioneering new format and OpenAI has a serious hardware product in the space and then Apple copies it quickly and then Samsung copies it quickly for the Android ecosystem, you're kind of left out to dry.

42:46So obviously I'm pulling for him and I hope he, I hope he can wind up navigating it and counter positioning in a way that he creates something that's truly disruptive, but it is, it is serious business right now. But Augustus sums it up, sums it up. Well, he says, I have no stake in SF AI hardware, but this seems right. And it's the giant going up against Avi Schiffman and friend.com. So good luck to him. But yeah, the current prototype is slightly larger than the AI pin with a form factor as compact and elegant as an iPod shuffle. one of the intended use cases really was a magic product yeah um were you too were you kind of too mature for a shuffle i don't know if i ever had one but i think it was more like a poverty problem i don't think i had the money for it um but it was it was very much like an accessory inspired a grind yeah i should have been grinding harder but honestly no you know what it was i was too much of a nerd so i had i had something called like an eye river which was like yeah that's what very unique thing that like it was it was technically more advanced than the ipod at the time like i had one that had a color screen that you could watch movies on it was like a two inch screen and it was well before the ipod video came out uh by like a year and so the release cycle was high it was was was faster and you get newer tech but it was like borderline unusable because there was no focus on user experience and so the company wound up kind of just not not not ripping um what else?

44:11Oh, Sam Altman. There's an exclusive in the Wall Street Journal by Berber Gin, a quote in here about what he's planning here. And we just got to read this one quote. Altman suggested that the$6.5 billion acquisition has the potential to add$1 trillion in value to OpenAI, which is such an insane thing to say. It's so crazy. But at the same time, how much is Apple worth? If you get a play in that space, like, yes, a trillion dollars of market cap is totally up for grabs. And so I think it's like, it sounds ridiculous. It sounds like over the top, but that really is the expected value. And so when you think about it as like, you know, what is that?

44:54A 0.6 % or yeah, 0.65 % chance at a trillion, probably worth 6.5 billion, right? In equity. And so that's the deal that they made was that, hey, we're going to get this team that's going to run really hard at this. If they hit it, it's low probability, probably, you know, a couple percent chance that they pull it off. But if they do, it's a trillion dollars. So it's totally worth the high number. At least that's probably like, that's one way to do the discounted cashflow and like the valuation. And we've sliced and diced the 6.5 billion a bunch of different ways. And I think we both came away with this idea that like, it's not as crazy as it sounds.

45:33It's 2 % of the company. You know, you're getting a great executive, you're getting a whole team. And like, it seems like a big number. They haven't shipped anything yet. That's crazy to buy something that doesn't even have a product for 6 billion. But when you think about the market, the opportunity, things start piecing together and it could wind up looking very good in hindsight, but who knows? Anyway, I don't think there's anything else we need to go into before we bring in our first guest. We got Jory from Linear. Let's bring him in. Let's bring him in. How you doing? Hey guys. Welcome to the show.

46:06We're great. Good. Having fun, reading through the information, looking at the new Sam Altman IO acquisition. Do you have a take on the new device what do you want to see in the world of ai hardware i don't know to be honest let's see when like things start shipping and like how things like evolve uh yeah have you had a chance to play like i've again like you know a driver's seat and like on the on the news like doing uh doing interviews and so on so i think like it's so like i had a had a break from for like whatever like four years or like so it's it's been a while it's more like john called out it was like Very strategic to go have him do a big interview at Stripe Sessions and then immediately the next week come in with this massive announcement.

46:52But I'm excited to see. I think every technology. Call that a coincidence in my parts. Yeah, yeah. But I think every technology enthusiast will benefit from him taking a real crack at AI. What else in the news this week? It's been kind of an informal AI week over here at TBPN between Microsoft Build, Google I.O., OpenAI I.O. Linear Agents. Linear Agents and Anthropic Cloud 4 dropping. What's been most interesting to you? What are you most excited to leverage? Pretty bad week for me for following the news. It's finished up moving, so I've actually been relatively off news. and like i think like uh on the docket for like long weekend to like start catching up more on the on like all the keynotes and so on but i don't know it's it's it's good stuff like having more agents or like it's really coding agents and like different like players like coming coming to the space so yeah uh i don't like it's it's it'll be like interesting to see like over the next couple of weeks kind of like what like performs like how well and also like uh what kind of like in user experiences are going to be there.

48:03Because I think that's kind of like the bit that we're at linear now following much more. We're also trying to work with different players and integrate stuff into linear. It's like, what are the patterns of usage that will happen with this? Yeah, what's more important? Just vague definitions around user experience patterns for agents versus standards like MCP and more technical advances. Yeah, I mean, like, those are, like, very two different things. I feel like, you know, like, MCP and, like, open standards, like, that's, like, personally, I'm, like, very, very, like, pro, of course, as an engineer.

48:42But it's been something, like, I feel like we kind of, like, lost in, like, last decade of, like, the big platforms owning more and more on their, like, own, like, trying to, like, hoard everything. But, like, now things are pretty much, like, being pushed out in the open. uh and people are like companies are like almost like forced to cooperate in the like the new world there's just so much demand uh and it's been interesting to see from our side too as like large companies like jumping very early and like so like a whole space is kind of pushing all the companies to like act fast uh let's see how much is like you know hot air like what like will come out of it but i think like overall it's like push for openness is good uh but then how have you so yeah so for for linear's agents product how do you evaluate how how open is the product if if somebody's building an agent can can they build an integration with with linear by themselves are you guys really kind of uh doing your own internal testing to kind of qualify potential partners there yeah we're we're kind of looking more on the promotional side i would say like we believe in like open apis and like trying to like have people access but they're not like that uh when it comes to like security and these kind of things it comes more on like what do you like promote sure yeah yeah i mean at the same time like if i get a really amazing agent that just can puppeteer my mouse and keyboard you can't really do anything about that maybe you throw up a captcha but agents are going to be able to solve those pretty quickly.

50:16And so at a certain point, you have to think like, like you're, you know, there's a reason why you have an API. There's a reason why you have an MCP server. There's a reason why you have maybe an agent integration, but are these like temporary steps in your mind? Or do you think that there's real durable value to an agent, like a standardization, something that you build in house, some sort of protocol or some or standardizing against things versus just mcp i kept coming back to like can't the llms if they're really so smart can't they just use websites but what's your take on that trend yeah it probably comes to like down to like the user experience that you want to like offer so you know like uh controlling a computer or like vm or whatever like that's the kind of ultimate like band-aid to like everything like it allows to do everything but it's still like it's not purpose built but and like mcp to a degree it's like a little bit like similar it's just like a way to like do like do uh request response like calls and get the information in and like share that but then it's more about i think like in our mind it's like what is the experience like inside the product that like you're gonna offer that maybe is like specific to that product uh and how that does how does that work like today like our implementation is like you know somewhat like rudimentary that we would expect like agents look like regular users but it's not very exciting to see the um the progress of the agent when they're like posting new comments in the thread that's a little bit of a hack that you kind of have to live without while we still see what's actually required and what can we build to better support the agent developers, what can experience give them.

52:13Of course, we're looking at this from the lens of linear, of course, running the company and like building the product. But it's also like, it's interesting to see how the, like where does the invocation of like agents happen? Like, is it like your CLI? Is it like a text editor? Is there like some kind of like web tool or like a desktop app? Of course, like for us, it's like, it's pretty natural like being like close to like where you work. So like integrating into that workflow. And I don't like that's why a lot of like these like smaller companies, especially like are pretty excited to like build in linear because they're already using linear themselves what what agent uh gets the heaviest usage internally at linear today i think it's been like a couple of the coding agents but those were like the first ones to you know to the market no i i mean i mean not even i mean specifically like your team at linear yeah like coding agents or or well you know, I, I kind of like mentioning names because like, of course we're, we don't have a horse in the race.

53:24Where's the interface for this? I love, I love horse races and I love having horses in races. Yeah. Pick a horse. But it's changing like every week. I feel like you said, like there's like a lot of like new, new agents that like came out this week from Google and like, uh, open AI and so on. So like, it's kind of like, it'll be like interesting to see like how those like play out and we're like looking for like working with like all of them of course but i think beyond i think like currently the focus is a lot on just coding agents um like it's it's natural because there's a lot of like value to be created over there but i think like we'll in the next couple of months we'll start seeing much more like non-coding agents like augmenting like in alongside with the coding agent.

54:10So you might have a task where you have a feature flagging agent, coding agent, and I don't know, who knows, a marketing agent helping you out to write the changelog blog posts and so on. Yeah, that makes sense. Are there any other AI features that you're implementing that feel like Windsor for cursor for project management, like something that lives in linear alongside, but it's not that asynchronous because in coding, we're seeing a lot of like, there's synchronous AI and there's asynchronous AI and these two different patterns. And it feels like the market's bifurcating, but no investors want to admit that because they want to say it's going to be winner take all.

54:53But in fact, we're seeing two different paradigms kind of emerge or maybe even three different paradigms emerge. What are you seeing on the project management side? Yeah, that's, yeah, I don't like the agents are roughly, especially in our case, are tied to issues or tasks. And that's kind of like the interface for them. Outside that we're of course, like linear is not only like issue tracking tool, it's like for project management, like organizing like your work at the company level, not only like an individual level. So and that's where we're going to have a separate work stream of we're building our own like AI tooling around that.

55:32how can we augment the project managers and like the product leaders to like do their work better and that you could like imagine looking a little bit more like a clod or cursor like alongside your linear data and but with that it's going to be interesting to see like how we can start like introducing like the external like agents or like mcps as like part of that um that of course like the issues is that's like the first first step because it's like much more natural what's your team's approach to testing new for example coding agents right it's uh in many ways like the linear approach from my perspective is you know really thoughtful ideally long-term planning around building you know beautiful products and and taking a calm approach to doing this and in that way, you know, enabling other people to maybe have more calm, uh, uh, effective, uh, product development.

56:35Uh, but at the same time, like testing new tools, they can be super effective, but testing new tools can also be super distracting. You know, we don't do a ton of engineering on TBPN. We actually do a surprising amount, um, kind of like back office stuff to kind of automate production. But, um, I, I can imagine in an environment where you have, you know, call it 50 engineers and on any given day, one of them can send a message into Slack. Hey guys, you got to check this out. It's like amazing, blah, blah, blah. And maybe they just got like a couple of good results and it's actually not worth directing everyone's energy to, but at the same time you want to stay at the edge.

57:12But do you have a philosophy or kind of an internal ethos around testing new tools? Not really. I think like mostly comes from people themselves. We're not like enforcing like certain tools. We're of course encouraging a couple of tools where we can maintain security and those kind of things. We're a mature company at this point, so we need to look after our customers' information. And a lot of it also lives inside linear, so we need to look after stuff. But when it comes to new tools, the team is looking at the news the same way we are, and like trying out stuff. And I think it's a little bit more, so you put out like linear ways, a little bit more calm.

57:59And I think like that also shows on the tool adoption and so on. Like you're like excited like to try out stuff, but like, you know, you take it with a grain of salt instead of like going on Twitter and like blasting, like we replaced like all of our team, like with AI. I mean, that's one way to do it, but that's definitely not us. Yeah. Nothing against that either, but yeah the Klarna method not naming any I said I said it I said it I respect it it's good marketing it's good marketing but it's not for everyone yeah I don't know like that that's that's been like overall like when it comes to developing AI tools like we started like when like everyone else started like a couple of years back when like the first GPTs like came out and I tried to do stuff and like you know tried to build a chat bot and so on like but i don't like very early on like we realized like well you get to like whatever 60 70 percent but then it's like falls apart uh the experience and like it's not and then it's really like hard to figure out like what's actually have to get to like the 95 percent where you want this is the apple problem you you linear and apple have similar sort of like desires for perfection right and and generative ai is in many ways completely imperfect right and unreliable so that's an interesting it's an interesting challenge yeah but we built we built a few things we shipped a few things we didn't make like a maybe like the biggest fuss about it we're not the ai powered like issue tracking software raising like gazillion like dollars.

59:43But now... Raising a lot from customers though. Yeah. Yeah. And, you know, I mean, like in the end, like they want to get their work done. And like, do they want to like buy hype or do they want to like buy product? Yeah. Maybe today, like you want to like buy more like AI and that's where we're seeing like a lot of like demand for AI solutions. And like now we're heavily investing in that. Yeah. And that kind of like our, we did a course correction over like the last uh like roughly six months ago when i think like the tools has got so much better like for me personally i was like on parental leave i leave and like doing the deep seek like came out and like just trying like the thinking mode was just like the light bulb moment like i think interesting yeah a little bit background on that i think uh just because of linear philosophy is try to build really snappy, purposeful tools.

1:00:38We always chase the millisecond and try to get something really fast. And when it comes to LLMs, you have inherent latency to it. So it's a spell, how do we bind this weight into the product that's instant? and I think like now with the new like thinking modes and so on, like that has changed the user, what people expect, like that paradigm. And now we can like build towards that and like people expect it to take a little bit like more time and also like more UI patterns to support that. But like you get a lot of value out of it. Totally. well yeah check out some of google's launches from earlier this week what was it they were doing 3 000 tokens in like half a second with the diffusion models yeah very cool um anyways this was awesome congratulations uh on the launch this week and uh come back on again soon yeah hopefully yeah cheers later jory talk to you soon um we got zach weinberg coming to the studio and we also have Doug Bernhauer from Radiant joining at 2.30 to talk about the nuclear radio which I'm very excited about welcome to the stream Zach how you doing I'm good welcome to the stream bull soundboard now we're getting better constant iteration what other product updates have there been since I was last here oh I mean our chyrons stop asking hard questions about our updates I noticed a sponsor whoever makes the yellow drink I assume to be honest they're not sponsoring us to be honest I did grow up in the town that this company is from we have no proper affiliation we should just wrap these in ramp yellow we should good to see you Yeah, good to see you guys.

1:02:46Thanks for having me. I mean, we wanted to have you on initially to talk about the most favored nation drug pricing EO. It's been a week or two since then. There was kind of like, I think it dropped Sunday. It felt like years ago. But I mean, it felt like Sunday, like, oh, this is going to be Black Monday for biotech. It didn't happen. The biotech stocks did fine. They also, this was coinciding with the pullback on the tariffs broadly. So the market rose overall. Can you give us how you process the information, how you're thinking about it now? Does this thing matter? Should we even be talking about it?

1:03:22Should we just move on? Yeah. Maybe let's start with like. why it's probably just like a misguided idea in the first place or at least the trade that I think Americans would need to understand that you're making and as it relates to drug pricing and what I think we may have talked about this last time or I've talked about this with a lot of people like the idea that America pays a lot for new therapeutics right and and we do we do by the way like this idea that we pay materially more even on like a GDP adjusted basis than other countries and in particular other wealthy countries. So you take Europe is probably like the best example where the amount that we spend on new therapies as a percentage of our, like it was called GDP adjusted price.

1:04:12Sure. If you will, is still on the upper bound. So like America pays for the innovation, uh, for the rest of the world. But don't we get the innovations like five to 10 years faster than the rest of the world? We definitely get them faster. We are usually the first place all these new drugs launch. There is also a set of therapies that we have that you just cannot get anywhere else. Like the other countries refuse to pay for them. It's socialized medicine, right? So it's like a single budget. And then there's also the factor of the rebates where the headline price that you see quoted in America is not what Americans typically pay, even if they don't have insurance.

1:04:53So you might see that some like Alzheimer's drug is 10 times or 100 times the headline price in America versus Japan. But realistically, Americans are paying three times more, five times more, which is still a lot. But it's not it's not the scary number that people often quote. Personally, I like paying more for drugs because it inspires me to grind harder. Yeah. Right? I think many Americans feel the same way. Wait, which drugs are we talking about here? Caffeine, nicotine, creatine, protein, testosterone mostly. Protein. Yeah. Full just TRT juicing before the show. Yeah. No. So like we do pay more, which is where I was going.

1:05:35On a net basis, like when you add in all the discounts that are given out, it's not as bad as it makes it seem but obviously like when you read about it in the headlines and they're like oh this drug is like 65 000 a year 100 that is the list price and so you know the media doesn't understand net pricing versus list pricing and so the headlines are much more sensational than they uh actually are in reality but yes the delta still does yeah exist um and we are we do pay more Now, part of the reason why we pay more is because if we don't do it, the drugs don't exist. Yeah. And I just think that little piece of this is really hard for most people to understand, which is like biotech.

1:06:20There's this really amazing analysis that was done by RA Capital, which is like a giant biotech venture fund. Yeah, we had one of their principals on, I think, the Monday after. Super sharp group. Top five. You know, mostly publics. but like, this is like a, you know, expert, expert, expert group in therapeutics. Yeah, we had Tess Cameron on. Yeah, I know they have like a few hundred people. Their founder, Peter, is super sharp. We respect them a lot. They actually were investors in one of our company's next rounds. Anyway, good group. They did a very beautiful analysis of the expected value of a biotech investment at various rounds.

1:06:55Seed round, A round, B round, so on and so forth. Like, what is the EV of a dollar that you put in? And I will tell you that at both seed and series a, the EV is negative, negative at our current prices. And the reason for this is like drug discovery is the single hardest fucking problem that you will ever encounter in your life. because you are trying to make a drug and using all of these tools that are not actually the human being until the absolute end of it. So you have to take all this insane amount of risk and spend tons and tons of money. And the data that you have to make your decisions are like in a Petri dish or in a mouse or in a monkey or in a dog.

1:07:35And like, yes, that's useful, but it's not a human. And so like the amount of companies where, because you don't see this in tech, right? You don't see companies that have raised 300 million dollars do one product launch and then go bankrupt yeah and like that happens all the time in biotech that's actually like the norm is basically you know you've spent a few hundred million dollars you think this thing is possibly going to work it works in a dog it works in a monkey whatever the drug-like properties seem all really great and then you run a human clinical trial for 150 million dollars and it doesn't work and the company's bankrupt and everybody loses their money and like and it's not like 20 million it's like a hundreds of so So, you know, biotech is an expected value negative investment business, even at current prices.

1:08:21And so what that means is - Really quickly, is that just a power law thing where the best biotech venture funds are doing great at seed in series A? And just like, you know, there's plenty of funds that I could point to in consumer SaaS or enterprise SaaS that are terrible and negative EV, right? I don't think you could point to the asset class being negative EV. True, true, true. Right? Like, yes, there's always a skew in the results, but here we're talking about the entire asset class is negative EV. And so if you're on the front end of it, great. It's rough. You stick it in the middle, you lose money.

1:08:56In tech, I don't think you get the median fund in tech and you lose money. No, no, no. The median fund might be like one and a half X or something. Exactly. So it's not shifted this way. The whole category. Why is it shifted this way? There's a bunch of reasons. the biggest one in my opinion has to do with this concept of called like better than the beetles which is basically like it's a very simple idea when it clicks for you when i try to launch a new drug the thing that i have to beat in terms of its efficacy like how good of a drug is it is the current standard of care meaning what a patient would get if this drug didn't exist so it's not placebo in most cases it's not like a water you know sugar pill right it's like the current thing.

1:09:38And so as we get better at treating the current thing, the bar gets higher. So it actually like the better we do a drug discovery, it gets harder, not easier. We should do that with payroll software, unless you're better than the best payroll software, you shouldn't be allowed to promote your product. Well, and even more like, here's the crazy, you shouldn't be allowed to launch. Here's the analogy. You wouldn't be allowed to launch the product okay the government would say no no no no you cannot launch until you are better than the current your corporate card isn't better than ramp so you can't launch it's just ridiculous could you imagine that world where you're like oh my new you know someone would launch like i don't know yeah of course chat gpt search engine and then like the federal government's like nope google's better you can't do this until you're better yeah yeah yeah that is how biotech works that is how biotech works and by the way is that the right is that the right way for it to work does that contribute does it i i you have to imagine it contributes to prices being high because it's somewhat anti-competitive right if i okay for in the payroll in the payroll market right like a payroll platform gets super bloated some founders like and it costs you know a hundred dollars an employee some founder comes in launches an okay version also there's just less iteration on drugs, I would imagine, than on software, right?

1:11:03No, no, you can't like iterate through it. There's no like learning. You got to like, it either is better or not. Here's the problem. Are you going to take the drug that's 30 % worse, but 50 % cheaper, even though you don't pay for it? Absolutely not. Exactly. Yeah. Yeah. Exactly. Now, if you were on the hook, or maybe you saw some of the rebates and, you know, went into your pocket, maybe you'd be like, all right, you know, like my psoriasis isn't like that bad. I'll take the like 30 % less effective drug and I'll like clip some, you know, coupons and, but that we don't do that because, you know, in, in America, we view healthcare as like, it's supposed to be free for everybody.

1:11:41Uh, and so kind of do it with, with over the counter stuff. Like if you get like the mucinex extra strength max, like it might be$3 more than just like generic stuff. Try telling, try telling the cancer patient that you're going to get very different for that absolutely like you're just not going to do it and so you know the fda is kind of like well what's the bar is current standard of care that's the standard we've put in place yep look if we wanted to create a system where you could launch slightly worse drugs cheaper like that would be an interesting system but the problem is the people using the drugs aren't the one paying for the drugs at least not directly and so cheaper doesn't win right like you can't shift because the patient at the end of the day does not want the worst drug.

1:12:24Like I can't even, there's so many situations where you're just like, absolutely not, especially in serious disease, chronic disease, cancer, all the stuff that really matters. So going all the way back to pricing, one of these like weird, weird issues in biotech is simply like as the industry gets better, it gets harder, not easier, which is, there's almost like no other industry where that is true. Yeah. And so - It's the opposite of learning curve pricing. It's like the opposite of fabs, the opposite of software, the opposite of internet. Yeah, when you see all these stats of like innovation in biotech has slowed and all, you know, a lot of it is because of this and they call it better than the Beatles, which is this idea of like, imagine if you wanted to release music and the only way it would work is if you were better than the Beatles, because that's the best.

1:13:06And so like, how much music would you really have? And that's, that's the I like the phrase because it's whatever, catchy, like the Beatles. uh anyway so it does the one thing i'll say is it has an interesting effect of pushing for true excellence right yeah yeah i mean that does not happen that certainly doesn't happen in sass there's a lot of people that see a good sass product and they're like i can make an okay version of that and make yeah a couple million dollars a year there's less cynicism and there's less like we don't have a slop problem in farming well there's no and there's no like move fast and break things because you can't tweak your way to the front right you've got to build the best drug before you put it into people right there is no tweaking so i i say all of this which is to say like yes drugs are expensive in the united states um but like if you wanted cheaper drugs today you are basically not going to get any in the future and i know everyone's like oh that's not true but like i am on the front lines of this every single day i mean we are one of now one of the larger biotech venture funds at seed meaning like we take the most we're in that negative ev category uh different risk on yeah yeah and so um if the price on the other side is not big enough for the risk that you take it's not that you like tone it down it's that you just don't do it you can't i mean just to be very clear like you don't need to do this right like you could you could walk away from the game right and yeah like there's that and by the way it's not just me it's it's our investors our lps like if if my fund and every other biotech in early stage investor can't make money our job will cease to exist because our investors won't let us exist right like there's this the oh money will always be there no absolutely not and so you know you need a reward on the other side of it um and i think that trade is really hard for people to think about because you're like well it's a lot now and you're like yeah it's a lot now but if it's if it weren't that means 10 years from now you're going to have the same drugs available in 10 years as we have today your kids are not going to be any better off from medicine perspective so the reward does have to be really big but at the same time like do the europeans kind of like preload off of us yes and so this kind of like rough idea of like all right this is what they're trying to do is say we're going to index the u.s price to like a basket of of comparable international countries basically like wealthy first world countries mostly european countries and that we can't be like materially higher than they are on a gdp adjusted basis uh At least that's the idea behind the executive order.

1:15:53Maybe two thoughts. One, this will not take a drug, US drug prices down. It will not happen. What will happen, this is not a good thing, is it may force, first of all, okay. This is likely not enforceable. I don't believe, legally, I don't think the EL works. And that's part of why you didn't see the biotech market react, is every expert - No one took it seriously. Like it's not, it'll fail in the court a thousand times. It'll never get implemented. that doesn't mean some future version of this there isn't some like you know loophole or whatever that maybe the government slowly figures out but like this eo seems as if it's basically never gonna happen yeah what about what about the is is biotech being squeezed on both sides in some way right now given that there's kind of an attack on the fundamental foundational research as well and so like if you take that basically that's like investment that's sort of like you know non-effectively uh the biotech industry gets the benefit of like the negative ev bets are already being subsidized to be by the closer to zero but not this administration is doing absolutely everything it can to destroy the future therapeutics that you want every single decision they've made basically so far has been negative to future medicines existing right like you cut biology research so we don't really understand what drives disease you you cut skilled immigration do you know who is like half of biotech if you go to like any biotech company's website you go to our usually somewhere between 30 and 50 percent of the people on staff are foreign born yeah yeah because you know who like doesn't want to do the 17 years of school to become a chemist and like do all the training americans and so you know we already have a science a skilled scientist shortage and that's even with current immigration and so like kicking these people out of the country not letting them come here in the first place to train, like, we're just going to have a massive, you know, skills gap.

1:17:47This isn't like bring manufacturing back. This is like, you need to be a PhD in chemistry plus 15 years. Like these people don't exist. And so most of our senior team at Curie is foreign born. Yeah. And I imagine that, I imagine that, uh, the biotech industry could very quickly absorb like twice the amount of PhDs and just do more research. Right. And, and it's not, it wouldn't be like job displacement if that happened. So the solution and like biology and chemistry PhDs and proteins, like specific types of PhDs, you know, do we do we need like marine science PhDs? But, you know, so I don't want to put words in your mouth, but are you saying the solution is that we need a million Brian Johnson's?

1:18:29We're looking for scientists, not crazy people. So, you know, no, we need we need a million, you know, David Luz at MIT. and uh what about uh what about that guy in china uh hood john kui have you been following this guy he is the one like christopher baby yeah yeah exactly and he posts these like very vague posts where i thought it was an ai i thought it was an ai like meme account for months and he keeps posting like you know uh biotechnology should never be used for to create super soldiers and it's just like a selfie of him and it's like it sounds like you're making a super soldier man And it's unclear how much he's in on the joke, but recently he got married and his wife can't get into the country.

1:19:12Have you been tracking that story at all? No, because it's like fringe. It's like the New York Post of science. What about - All right, I want to go back to the yo. Because I just like, this is the class, this is this administration being like, I'm angry about something and then I make a decision and nobody thinks about then what happens. And it's just like over and over again. You're like, these fucking people are so stupid and I hope they're listening. like you are all so stupid and like you're just that's fire you're just ruining the thing that has made america excellent in science by killing all these things in you know science funding immigration blah blah it's like it is it is idiotic um but even even if you say like we're going to index the u.s price to international prices remember that the united states is a giant fucking country if 350 million people were the wealthiest in the world like all 50 percent of pharma revenue comes from the united states and then you're like japan is kind of second okay so now put yourself in the shoes of a pharma company right you're going to index you're going to index our price to the european price and let's say the europeans are like 15 of your revenue and the americans are 50.

1:20:20are you going to lower your price absolutely not you're just going to raise the european price yeah right and just to bring it up because you can't lose the u.s revenue yep so like you'll never give this revenue up so the price the u.s price will not change this does not work all it does is it may even if it were illegal it may force pharma to bring its european price up and you know what happens when that happens because this is like and then what the europeans will go okay we can't pay for it we're not going to buy it and all it does is it keeps the u.s price the same and it shrinks pharma like the net net is the united states will pay exactly the same that we paid before because pharma is not going to drop price in its largest category by far.

1:20:58It's the only place you can make money in pharma is in the U S and that price not coming down. If you have to bring the price externally up, it shrinks the revenue in that market, which shrinks pharma, which shrinks biotech. Because if you, if pharma is a smaller, you know, dollar value, if the, if, if the profitability of pharma comes down, you know, what they stopped doing buying biotech companies, right? All of us, the same effect, which is like, well, if this is less profit, it's just a different way of making it less profitable, right it's the same thing of like if it's less profitable i can't take as much risk and if i can't take as much risk i stop doing risky drug discovery and therefore the whole industry just kind of like like this and that's not a win how do how do uh how do companies get valued if they have one drug that's like working and maybe it's the best right so it's actually able to be in the market, yet there's a sort of sense that there are very real competitors that might come in with a drug that is substantially better.

1:22:00How is a pharma company that's exploring M &A going to value that? Is it still the same perspective? This is what they do all day. What you just described is exactly what these people are spending teams and teams of people on, which is like, okay, what is the value of the current drug that we've got? Which is usually if it's at the place where it's approved, you're looking at, you know, say what a peak sales and how long am I going to be able, how long is my patent? Right. What do we, what do we expect generic competition to come to, to, to look like? And then you're looking at the pipeline of competitors and you're looking at the data that they shared.

1:22:35And sometimes there's some data and sometimes there isn't. And then you're trying to predict the future, which is like, okay, just cause, you know, company X is working on a competitor doesn't mean it's going to work. so do we think it's going to work why do we think it's going to work if it does work what are the so you do like scenario planning like okay you know 10 chance this beats us 50 chance it doesn't like should we be concerned and all of this math goes into the spreadsheet of like what do we think this drug is worth uh and what's the risk appetite so like this is this is day-to-day you know biotech pharma commercial analysis there are very very smart people who do this uh you know this is what public hedge funds is what the ra guys will do you know and it's it's deep science um but all of it is just like at least the drug pricing stuff is just it's a trade that is hard for people to understand but it's kind of like if you don't put a reward on a very risky thing the risky thing doesn't happen and you just don't feel it until 10 years you won't even really know because it's the drug that never got started in the first place and then eventually what you're doing is you're outsourcing all this to China because the Chinese will pick it up and do we have a longer conversation about China so I all of it is just kind of like stupid um the other thing I would say I don't have to jump but like this is my defense of of medicine and uh therapeutics if you will as like a great investment they go generic yeah like nine to 12 years let's say roughly these are generic which means they are basically free for you in 10 years I know your kids in your kids and your future kids.

1:24:13Like it is an unbelievable investment. You pay a lot upfront, you get a lot of innovation, a lot of people taking risks. You have this like window in time, which is like, yeah, you know, you got to eat it. You're going to eat, you're going to pay a bunch for it. It's not going to feel great. And then everybody gets it free. There's no other industry in the world where that happens. And even within healthcare, you know, doesn't go generic doctors, hospitals surgeries like why why why is there so much focus on the therapeutic which is only also only about 10 of medical spending is drugs which then eventually goes generic versus the 90 of it which is basically just interesting and those physicians they're not well it's because the margins right what what uh strikes you as something that like the pill only costs a sent to make and so exactly just emotionally you think about high margin things as wasteful what's something positive that could come from the eo we we had a uh infamous uh health care exec actually actually no we had an infamous health care exec message us while we're live said ask him if he reads that the eo is maybe trying to solve the consumer pricing problem gross to net rather than actually making pricing comparable in the end given that there was a specific call out for consumer d2c pricing and which might have been why consumer d like healthcare is about the new drugs that are really expensive not like consumer d2c pricing of like you know whatever consumer medicine like i'm talking about real healthcare very sick chronic disease people with neurological conditions with with cancer with heart disease like the serious stuff which is what i just saw a company that they'll give me one pill that has uh function medicine and hair loss medicine in one pill is that not real medicine what's going on those are generic meds sounds like amazing it's amazing and you know why they're so cheap yeah because they're generic because 15 20 years ago we invented this shit we you know pfizer made like a crap ton of money selling viagra yeah that's right everybody now you get to get your six in one dick pill for dollars and like that's a pretty cool deal right like it's cheap and all this like shores there's some like gross to net whatever that's not what health care is that's not i'm sorry like hair loss is not health care interesting uh what about crispr on demand uh there was this this infant child that was saved with a custom gene therapy.

1:26:46It seemed like a very inspiring story. There was inspiring video all about it. It seems like an amazing testament to something we've been hearing about for a long time, which is CRISPR, maybe a decade. We've been talking about this. Is that an opportunity for new companies? Is that, is there already a company that's like ripping on the back of this? Or is that just kind of like an outgrowth of what we'd expect with the technology, given the maturity? I believe the company that did this was Beam therapeutics is that them who did it yeah um yeah it's cool uh this is i mean this is a perfect example right which is like that i think it's being i gotta remember i can't remember offhand which one it is um they that company that that successfully developed this uh this therapeutic is trading for less than money raised whoa wow okay break silence it's terrible right and why Why?

1:27:39Well, because like these gene editing technologies where you're going after essentially like disease that is very clearly defined by some mutation in your DNA. And that's a small subset of total disease. It's like very, very rare conditions that this applies to. So like the market, the way I think about it is like the patient market for this technology is very small. These are horrible diseases that in theory can be treated through gene editing. There's not that many of them, but you can, you can do some really amazing things if, if you can, if you can make these edits, which is the first time this has ever happened.

1:28:13But like, if you add up the patient population of like all of these conditions where you could possibly do this, it's not that many. It's probably, I don't know the math offhand. I bet you it's probably like tens of thousands at most. Okay. And so like, plus the variable cost of manufacturing this stuff on a per patient basis can be hundreds of thousands of dollars. Aren't we going to get some blockbuster gene editing for hair loss and then we're back in business no uh because hair loss is not you know there's no like one base edit to you know fix hair loss but yet um yet for i think i think i like this techno pessimist on my stream zach well one of the things you learn as you go into science is that the dna is not actually the whole story it's not even close to the whole story and so like it's not even you wouldn't you're gonna say there's is vibes a part of it is that It's actually astrology.

1:29:01It matters. Taking account the DNA, but also the astrological sign. Infamous astrologist. I mean, vibes are kind of like it in a way that like your behavior as you're alive does affect your epigenetics. Your epigenetics matter, right? Like you sit in, in, in a smoke all day. Like you're going to have some problems. Anyway, I like this as an example because this company that just made this like unbelievable breakthrough is, this has never been done before in humans. They're curing diseases. is now trading for basically less than cash raised. I got to go find the actual numbers. What's the name again?

1:29:34I swear it's - So there's, I just got an answer. So this was a bespoke CRISPR-based editing drug. It was built and tested by physician scientists at the Children's Hospital of Philadelphia and Penn Medicine with translational help from Innovative Genomics Institute at UC Berkeley. Reagents and GMP grade components were donated by several suppliers, including Acutas Therapeutics and integrated DNA technologies and Aldveron, both Danaher companies for guide RNA. Yes, on that one, there was another story. Yes, there was another story and I have to find it. Well, we will have to, we have to, but we'd love to have you back and kind of dig into some of these other biotech companies and take us through like the wins and losses.

1:30:21I think it'd be very interesting to see what's going on. Imagine if that becomes a drug and we go, oh we can only pay$20 ,000 for it do you know what happens to that drug it sits on the shelf it's on the shelf no one ever does it no one ever makes it and there's your drug pricing in in reality which is like well it's got to be a reward so we're pro higher drug prices I guess you build us on that we want that we want the highest prices possible is what I'm hearing you want to you'll get a lot of shit you you put real high prices like people will come and and and take risks you know i'm sure there's some balance we gotta have you and screlly on together we need like here's my favorite stat yeah my favorite set the the i i i'll pull these for you so i don't want to sound like an asshole just like spitting numbers that aren't true uh i believe on a per american basis like you take our drug spend and you divide it essentially over like you know americans we spend drugs on it's something like a thousand dollars a year or so per person if you will uh we spend more on cosmetics than we do on medicine wow more on cosmetics let's hear it for america we're looking good what about fast food what about we might be dropping dead at 45 but we're gonna look good doing it uh thanks so much for stopping by this is fantastic we'll talk always pleasure cheers next up we have lee marie braswell from kline perkins coming in former uh colleague of mine at founders fund briefly uh excited to talk to her about artificial intelligence ai agents That's all the news of the day.

1:31:48And we'll bring her in. And we got some good chyrons for her coming up. We'll see if we can get them up on the stream. What are we talking for her? The term she tyrant herself. Lee Marie, are you there? Can you hear me? How you doing? Oh, what's up, Kugan? How's it going? It's good. What's new with you? I'd love to get your initial take on AI week. We got Microsoft Build, Google I.O., OpenAI I.O. We got Claude Four from Anthropic. It's been a massive week. What has stuck out to you as like the most interesting, the most underrated, the most worth digging into of the news of the week? Well, I guess just to start off, and it's so good to see you again.

1:32:33It's been a while. It's been too long. Yeah, thanks for having me on the show. Yeah, there's never a dull moment, really, you know? Yeah, you know, I feel like every day there's something really significant that you need to be paying attention to if you're going to try to make money investing in AI companies. Yeah. And so, yeah, I mean, I think the big thing that's – actually, there's so much this week. I mean, starting with some of the stuff you said. It's the best time to be a technology podcaster. Yeah. Investor is a close second, but podcaster is really where you want to be. That's where the action is because every day there's new news.

1:33:05Anyway, yeah, what is sticking out to you? Hey, Jordy. Hey. So there's, I mean, the IO acquisition in particular, that news, what an incredible video, what an incredible team. I mean, just really showcasing OpenAI's ambition to really, you know, get into a lot of stuff. And I mean, obviously, there's a lot of speculation about what the device will look like. I was actually looking this up. Sorry to interrupt. My favorite part was John and I are, I guess, ex-hall monitors. We're in the community notes program. Oh, yeah. So somebody was trying to community note the open AI launch video and the suggested community note was this video uses AI.

1:33:42These two scenes aren't real. If you look at this section, it's like you're trying to gotcha an AI company for using AI in a marketing video. Like, what is this? That's very silly. What do you expect, you know? That's kind of the point. Yeah. Yeah. Yeah. So, I mean, yeah, I mean, super excited about what potentially that could that could imply for just, you know, opening eyes ambitions, especially as it relates to consumer consumer hardware. I was looking it up on on O3, but O3 kept then kind of giving me some different answers. But it might be. And to be fair, I don't, you know, kind of know the current state of things inside of IO, Johnny's company, but it might be one of the largest, if not the largest sort of pre-product acquisition of all time.

1:34:23So, you know, I mean, six point five billion dollars is a lot of money. Yeah, we've been noodling on it. It's a couple points. It's a couple points for a key exec. Exactly. It's 2 % of the company, which when you think about bringing in an executive at that level, that comp for that team, it's not that crazy to get to that number. And then today in the Wall Street Journal, Sam Allman is quoted saying that if they can nail it and get a major consumer device, that's a trillion dollar opportunity. And that sounds crazy. But you look at Apple's market cap, you look at Android, Samsung, like, yeah, there probably is a trillion dollars in market cap.

1:34:53Well, there was also the tinfoil hat saying like, oh, he's trying to dilute the nonprofit. It's like, you know that they are issuing billions of dollars of stock to employees all the time. That's happening a lot. This is not some like, you know, crazy conspiracy. I do want to get your take on the actual device. We've seen some pre-renders, some AI images around like a pendant with a camera on it, the Her model. Maybe it's an earpiece. I imagine you get pitched a lot of AI devices. We've seen the rabbit, the humane, the friend. Are there any other ideas that you've seen that could potentially be just AI devices, even not from open AI, but just in general?

1:35:34Because Apple's got it pretty well covered. They got your ears. They got your eyes with the Vision Pro. They got your phone, the watch, the iPad, the laptop, the desktop. Like Apple's, there's not a lot of white space in the Apple catalog, but it'll be interesting to see if somebody can come up with something that's new. Have you seen anything that's interesting? Oh, man. I really wish I could tell you about it. But yeah, we actually have a company in stealth. I can't say much, but there's certainly different, you know. I'll be on the show when they're ready. Yeah. I'll be super excited, too. But people are being really creative.

1:36:04And then also, I mean, just generally, and I'm sure you're seeing this, too, just given like the pace of how much is happening and then also the competitive nature of these markets. Like it's clearly there's this big prize, maybe bigger than ever before. Yep. And so if you've got a really good idea, like, and if you don't have trouble raising funding, getting customers or hiring people, like, you know, why I want to talk about it too soon. So, yeah, I'll be excited to share more on that. There's also, I feel like people have just been generally willing to fund hardware, but also kind of categorically bearish, right?

1:36:34People are just like, oh, like hardware is really hard. Like I'm going to make this bet, but like it's really hard. The meme of hardware is hard used to be about supply chain. It used to be about all the things that can go wrong and the working capital. We're at a point where like Aura and Whoop are like big businesses that, you know, I don't know, like ultra specifics on either. But I bet that Aura will be a really big company in 10 years. From my perspective, it's almost harder. It's not, it is hard to make the device and manufacture it and deal with the fact that 90 % of your revenue comes in Christmas.

1:37:05this, but like the real hard part is actually changing consumer behavior and getting someone to take off their Apple watch or add a whoop band to the other wrist, which is, I think, the predominant pattern for most whoop users. But what are you looking for in hardware startups that you're looking at? What are the pitfalls that you want to see founding teams overcome before potentially backing them? Yeah, so I'll say I don't have a massive background in hardware. I mean, I will say I was at scale, I was relatively early there and was an engineer there for four years and then a PM. And we worked with a lot of hardware companies there.

1:37:44So, I mean, the main customer of scale back in the very early days was self-driving car companies and robotics companies. And so I have seen just through to working with them as customers, some potential pitfalls. And I mean, the quip is that hardware is hard. You've got all of the challenges of building a company. And then you've also got to just manage supply chains. And you have to be a lot more careful about the economics. Software is usually relatively high margin or it's easier to make it high margin. Hardware, obviously much harder when you've got to figure out where all these materials are coming from and ramping up.

1:38:17So even if you have demand, like ramping up in a sort of scalable way, and then also implementation can generally just take a really long time. And it distributes at the speed of the internet. Like you can go viral and get 100 million people on a website in a day. You just can't ship 100 million things anywhere in any reasonable amount of time. It just takes time. Like maybe one day. But yeah, it's like interesting to remember like Waymo was founded in 2009. Yeah. Right. So like just as that, like it took, you know, 20 years to now become this overnight success. But yeah. You mentioned robotics.

1:38:49I'm interested in the pools of training data that are out there. We've been joking about scale AI. Are they going to put everyone in motion capture suits eventually to get to the humanoid data sets? There's also a lot of work being done in simulated environments. Is there a data wall that we're going to run into with humanoid robotics? How are you thinking about the data challenges in robotics generally, whether that's in the humanoid side or elsewhere? Because it feels like the humanoid narrative is taking hold. People are funding companies. It's getting big. But if it's a 16-year journey like it was with Waymo, we could be in for a long one here.

1:39:25Definitely. Definitely. It's been interesting. I mean, like, I do think there are a lot of really good arguments for why we'll see robotics, like progress accelerate. And that's really, you know, obviously the data is more challenging to get. You know, it's not like you can just scrape the Internet like we did with a lot of these these LLMs. And the data is kind of like there and ready for the taking for at least like some of the early transformer based models. With an asterisk, like as we're now moving into agents, I actually think you have to be more intentional about creating different types of data.

1:39:56But then now with robotics too, like, you know, you've got to figure out ways to get real world data, whether that's, you know, internally or using somebody like a scale, or, you know, there's a lot of really exciting work going on in world modeling and simulation. And so, you know, I think hopefully a combination of being really intentional about the data, both on the sort of like human collection side, and then also some of this work that we're seeing in simulation and world models will hopefully mean that, you know, we don't all hit this data wall and we'll be able to sort of scale these systems more effectively.

1:40:25However, I mean, with robotics, it's always a little bit challenging to tell what that exact timeline is. Yeah, interesting. We were talking earlier about the agentic, the coding market, the AI coding market. I was kind of breaking it down to you. Is it three markets? Is it two markets? Is it synchronous work, asynchronous work? Is there a consumer, prosumer, bottom-up enterprise, top-down enterprise, sales motion? how are you thinking about just AI coding generally as a market, as the businesses evolve? Just give me kind of where you're at, and then we can kind of tug on that thread in a bunch of different ways.

1:41:05For sure. My favorite topic of conversation, definitely one of my favorites, is AI Cogent, near and dear to my heart. But yeah, as a disclaimer, so I'm on the board of this company, Windsurf. So agentic IDE, AI powered developer tooling platform. Yeah, it's been, it has been just utterly sort of shocking to me. Like, you know, when you invest in a company, you expect like, okay, you know, hopefully things will go well. Like hopefully the market will be there. And I just don't think anybody, I don't think anybody really saw this coming just in terms of like how widespread tools like Windsurf would become.

1:41:44And so right now, you know, you're certainly seeing like there's this sort of segmentation specialization you know you have tools like windsurf and cursor um that are you know prosumer so you've got people even with basically no coding or definitely no coding experience that are able to get up and running very quickly with these tools and really sort of like talk to it and it acts like this sort of programmer that will you know like respond to what you want to do maybe it comes back and becomes back with error but it can maybe automatically fix that error once you kind of give it some more guidance and you know it's just been incredible seeing like you know even even though i talk about windsurf all the time my father-in-law who i never mentioned it to not a programmer just like once he was on we were just talking all talking he's like you gotta go you're a vc you gotta go check out this company windsurf i'm like using it to like you know help me out at work and create apps and i'm just like wow this is incredible just how widespread he's like you outside of silicon valley all of that so so for those types of tools i mean you don't have to speak about windsurf specifically but just how much of it is just a ton of people on small consumer plans versus bottom-up enterprise adoption that gets rolled into a much larger consumption-based plan or even like a seats-based plan like how is how is the market for ai tooling in the enterprise evolving in terms of pricing because we've seen salesforce is talking about we want to price things not based on seats, but based on resolutions of customer service tickets.

1:43:12You could imagine charging someone for lines of code that get approved in pull requests or seat-based or usage-based or just pure inference-based. So how is that side of the market just evolving broadly? Yeah, I mean, so what we're seeing is, you know, there's certainly the prosumer segment for these tools, but it very quickly goes up, you know, bottoms up into the enterprise. And in particular with Windsurf, I mean, from day one, we've been quite focused on the enterprise segment. And so it's, you know, we're, we both have a large enterprise sales team and then these tailwinds from now all these prosumers and then, you know, sort of like professional developers that pick it up on their own and bring it in into their company.

1:43:52I think pricing, it's always interesting. It's something that we constantly think about. It's a very competitive market. So, I mean, you know, you also have to kind of pay attention to what other people are doing. I mean, per seat, easy to understand. But then, you know, obviously, a lot of these models cost, you know, cost money to run. And so you've got to figure out some sort of either credits or usage system. I don't think, you know, encoding where you have to the point where it's, you know, let's price based on outcome, like, oh, well, if we get this done, then then you pay. Like maybe, you know, I've seen sometimes in like customer support and things like that.

1:44:24But yeah, certainly having like a usage element, I think is pretty, pretty, Yeah. What's your reaction when an entrepreneur pitches you and they're like, well, we're building agents to go after this labor, this like end labor market. And the TAM is, you know, a trillion dollars because they're just adding up like global payroll for like that type of work. Like what's your reaction to that? Is that the wrong way to think about market sizing? If ultimately these are software products that can be delivered very inexpensively, which will ultimately create competition and drive maybe pricing down for that end service.

1:45:05Yeah, I mean, definitely, definitely some good points. I mean, generally, generally, I'm pretty optimistic. I think now just like kind of living through the cogen market where, you know, just to use this as an anecdote, you know, GitHub Copilot came out, my first kind of pessimistic reaction was just like, Oh, well, this is like so useful already. And GitHub's got all these advantages. Okay, that's like markets done. And they will be like, Okay, every developer like uses it, and then it's kind of over. And it just the opposite of this happened. Like it turned out that there is a lot of white space in the market for teams to innovate and to create really great product experiences, whether it's a startup, whether it's a foundation model company.

1:45:43And then also just like I think the market blew my mind. What was the secret? What was the secret to either the challengers to GitHub Copilot? I haven't really used them. is it more driven by flexibility around which model you're plugging into and GitHub Copilot kind of locking you into their model and not moving quickly enough? Or is it something more on the user experience side? So I forgot the exact date. GitHub Copilot started, I think, OpenAI only given the sort of relationship, but it relatively quickly moved to an option. So you can put in Cloud 3.7 Sonnet, which everyone loves. Yes. And I think actually recently, you know, I think I was reading this this morning, but, you know, GitHub announced like, OK, for these particular features, we're going to go with Anthropic only, like particular back end features.

1:46:31So it's clear like, you know, their GitHub is not locked into one model provider. I think, you know, what Cursor, Windsurf and others have done is just create, you know, different product experiences. And obviously, Getagoballet is still doing very well. But I think what this all has told me, and kind of back to your original question, was these markets are actually much bigger than we give them credit for. And so when I'm sort of being pitched maybe a ridiculous TAM, sure, it's probably good to verify that and really drill in on that. But I mean, we are in the world where now software can do work.

1:47:05And so you do have to kind of, you don't just get the sort of TAM being, okay, it's the incumbent software spend. And I think you have to also realize once the product gets that much better, people will want to buy more software and then also think about it as a work replacement. So you kind of get all these kind of bonus markets that, you know, end up being just like this huge thing. How do you process the tension between model providers kind of figuring out if they're wanting to end their own consumer or be a platform? I mean, the most obvious example was WindSurf didn't get immediate access to Anthropics' new models, which seems, you know, I'm sure you can't comment on that specifically.

1:47:49Or maybe you want to come in and put them in the hot seat. But yeah, it just feels like a very real tension that's going to have to get resolved quickly. because these are all big companies now, or at least the main players are big companies. There's billions of dollars of revenue on the line and everybody's in a constant race to be at the edge. You're right. I probably shouldn't say too much. I mean, what I will say, yeah, it's something that I think is going to have to be addressed in the next year or two. And I think it's just very clear that Cogent in particular is this market that everybody kind of wants to get into.

1:48:36You know, I do ultimately think that, you know, I'd be shocked if in a year or two, we don't see co-gen products that are both, you know, synchronous. So like assistant, you know, I can talk to it. And from the same interface, I can delegate tasks. I just think that that is something that a lot of people are building towards. It's like, there isn't a great experience today that kind of has both, at least, you know, sort of a generalized both. but I mean ultimately like all these companies are I think gonna in the in the sort of like even medium term come up against each other and ultimately what you want is sort of this distribution and I think you know all these companies too are realizing you also want the model and you want to be able to you know give yourself really favorable favorable economics so yeah it's going to be a super exciting even you know probably probably six to twelve months I imagine.

1:49:28So legal, the sort of legal AI space has been heating up kind of a interesting comp. In many ways, it felt like cursor had like a ton of mindshare and then windsurf just kind of exploded. People realized how quickly it was growing. At least that was my perception. And then Harvey similarly has been more in the sort of cursor bucket and that like tons of attention, bunch of rounds back to back and then lagora uh came out of feels like it came out of nowhere and they raised a big series b this week is there are you do you have a a kind of prediction or a take on what the next category to see these sort of battles play out i'm sure you guys already have some bets in whatever category that is she's already on the board of four companies that are going to dominate it the next four categories actually i can jump to the board meeting exactly what you just said.

1:50:23Yeah. Yeah. You know, I'm very lucky in that, you know, I think, I mean, even before I joined, so I've only been at Kleiner Perkins for two years. And even before that, a lot of my partners made really great app investments. So for example, my partner, Ramoon Incubated Glean, which is now, you know, kind of the enterprise knowledge discovery platform in this kind of new AI era. So certainly like we've got a lot of apps in the portfolio. I mean, Harvey, we did their Series B. Ambience, which is a medical scribe, also in the news recently, we did their seed in Series B. And then we have some investments in sort of sales AI as well.

1:51:01I work with this company, Nooks, which has started with a power dialer that people really like and is now expanding into other parts of assisting salespeople with AI. I'd say like those categories plus maybe add customer support, which we don't yet have kind of a generalized play in. those are kind of like the the battleground spaces these are where it's going down right now like it's clear the models work for the use cases it's clear there's a giant prize at the end so you've got you know inevitably a lot of companies competing for it in general especially with software you know i think it's basically impossible to be the only company in your can you can you fund a company that will just pick up the phone when i get like a robo ai call that's like trying to do some social engineering hack or whatever and just pick up the phone and waste their time really inexpensively.

1:51:47That's, I feel like that's what I need. There's definitely some cybersecurity companies out there. I think I've actually been pitched with a woman. I would have to find the name, but yeah. What do you think about competition from the hyperscalers? Obviously, Microsoft Build and Google I.O. happened this week. And there's one narrative that's like, oh, so many startups are now just bullet points in a Google I.O. keynote. But Ben Thompson wrote shortly after Google I.O. that Google only cares about search, and they only win when it's directly impactful to search. And so you might, as a startup, not need to worry if you're doing something that's more in the labs or experimentation phase for Google, and Google is more just almost doing it as a demo of their platform.

1:52:31And then they really would be fine with another company just buying a bunch of TPUs on GCP and building a real company on top of their innovation. But how are you thinking about startups competing in these new AI markets with the legacy incumbents? It's a really good question. It's something that I think about a lot. And obviously, somebody like Google has giant advantages, distribution, money. I mean, Google literally created the first Transformers paper. Um, so, I mean, certainly, certainly it's something that whenever you're investing in, in an app company in particular, you know, you, you need to, I think, at least think very deeply about what's, what's the roadmap of, of all these companies, open AI, Anthropic, et cetera.

1:53:15Um, I mean, certainly, certainly seems like sort of areas in search and sort of like social and, and prosumer that, you know, seemed like kind of the most under the microscope with these big companies, but you know, it, it wouldn't shock me if they tried, tried to compete in some of the other app categories we've talked about too. So in general though, startups, I like to bet on startups. Focus is important. And then, you know, I think if you move fast, if you're very strategic, there are a lot of good sort of like wedges you can sort of start with and then from there. What about M &A markets in particular?

1:53:50I mean, my reaction to the Windsurf acquisition was, and it's going to be kind of hard for you to answer this, but was basically that like, whoa, like maybe that whole meme of like, oh, don't build a chat GPT wrapper or whatever. Maybe that was overblown. Maybe that there's going to be a lot of app layer companies that gets scooped up. Obviously, OpenAI is moving very quickly. So they're kind of early to this. But you could imagine every hyperscaler needing application layer AI companies to slot in, especially if they've had distribution and they've actually scaled revenues and they have a really great team.

1:54:26And it feels like unlike in the previous era, like 2010-ish, the hyperscalers are more sclerotic than ever. It's been 15 extra years. They're less founder mode. And so I don't know what your read is on the M &A markets broadly, but how are you thinking about the hyperscalers approach to staying on the leading edge of these AI application layer companies. Yeah. I mean, I think it's important to remember and I have to remind myself that Chad GPT is less than three years old, I think. So it's like, this is happening so fast. And so when you think about like, I'm a hyperscaler and I see a market that is like, makes sense for me strategically, you know, is it really faster for me to go and try to build it internally or is it faster for me to find a really great startup, perhaps with expertise that I don't have and maybe some other advantage, whether it's the product or the distribution or something else.

1:55:26And then explore partnership conversations with them. And so, I mean, the one acquisition I can talk about is a company that I worked with at Founders Fund and still on the board of Neon. It's partnering with Databrack. Oh, yeah, cool. And it makes sense because it's like, you've got this team of database infrastructure experts and then you've got this company who clearly wants, their customers to be able to realize the full potential of AI agents. And so you do need a database and memory for those agents. And so it's certainly now I'm trying to think in my mind like, oh, well, what are the other categories that hyperscalers might be looking in and what makes the most sense for each one?

1:56:10The big one to me that I'm trying to figure out is just video generation, which we had, I wouldn't call it a Ghibli moment this week, but VO3 was close. Took a little bit. It just takes a lot more work and it's harder to get a consistent quality output. And there isn't a single meme where everyone, like the Ghibli thing was like, just take your profile picture, take your whatever's in your camera roll, two words, boom, one shots and you have something cool. With VO3, you have to be more creative with the prompt. Well, yeah, the interesting thing there is I can, it's so easy to imagine in the future where consumers are using video generation tools daily just for fun oh totally yeah there's there's some like consumers just like making memes on the internet then there's like kind of prosumer smb i'm gonna make like an advertisement yep and then there's the like hollywood studio that's like i'm gonna make a movie or independent filmmaker but where companies kind of fit into that i can see a world in which hyperscalers are kind of dominating like meta for example figures out a way to get really good at at at video generation because it fits into the meta ecosystem or YouTube with VO3.

1:57:21And I think that from my point of view, I can see some of these independent foundation labs that are focused on video having to really move up market to actually dominate because that feels like an area that Google's not going to necessarily build the thing that creates the next blockbuster video. yeah but if you're competing for the prosumer smb like feels like a really tough place to compete purely from a cost standpoint and with youtube or google's sort of youtube data advantage yeah yeah i don't know any any reactions to vo3 this week oh i mean super cool uh and we were playing around with it my husband's a founder and he was like trying to make this new video for something that they're developing internally and oh my god just insane uh so it's yeah really cool and i mean I kind of agree with some of the points that have been made, but then on the flip side, I will say like, especially in video, there's so many really great models and they excel at a variety of different things.

1:58:20And so if you're like kind of a third, you know, a startup that is not connected to, I mean, Google obviously is developing their own models. So if you're not connected to any one of the particular models and you can offer like kind of a best in class, like, okay, you can pick from, you know, many of them. Maybe, maybe that's an opportunity or if you're in sort of like very, very specialized so um yeah i mean generally still still optimistic just kind of given at this point there's not in my opinion a ton of platform risk you've got so many of the hyperscalers and labs like they're you know both closed and open source offering offering these types of models yeah very cool well thank you so much for stopping by uh we'd love to have you back this is fantastic this is great super fun all right see you later uh really quickly let me tell It has been on an absolute tear.

1:59:03Barely got to Neon. Another$1 billion acquisition. Fantastic. But really quickly, let me tell you about Figma. Think bigger, build faster. Figma helps design and development teams build great products together. Get started for free at Figma.com. Redesign how you design. Explore your ideas freely and iterate quickly. Let's also tell you about Vanta. Automate compliance, manage risk, prove trust continuously. Obviously, Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation, whether you're pursuing your first framework or managing a complex program.

1:59:41Go to Vanta.com to get started. Thank you, Vanta. Thank you, Vanta. Anyway, we have our next guest coming in, Stephen from Lambda Labs. We've had Lambda School on. We're waiting for Stephen. Should we hit some timeline? Yeah, let's hit some timeline. um what's uh did you see uh yo hey uh pocket is shutting down and so this uh uh we need a creator basically a board that says moment of silence yeah vibe coded a quick free open source prototype with some ai features called keep yard it still has issues but i challenged myself to push this in one night i thought this was very cool i've been joking about we need someone to vibe code a new Google reader since that was such a popular product.

2:00:24It should just be out there. And it's already starting to happen with replacements for Pocket. And it seems like Steven's in the studio. So let's bring him in and talk about Lambda. How are you doing? Welcome to the stream. Hi, guys. How are you doing? Welcome. Great. How are you? Thanks for having me on. It's good to see you. Looking fantastic, by the way. Thank you for wearing a suit. You know, naturally, I guess, when in Rome. But that said, this is how, you know, everyone in Silicon Valley should dress. I agree. I agree. I agree. You're just ahead of the curve. Are you in Silicon Valley right now?

2:00:57Can you give us a little introduction? Oh, drinking a monster. Drinking a monster. Let's go. This is so on brand. It's amazing. So I'm actually in my hometown of Charlotte, Vermont. That's where I grew up. Yeah. And it's where I live. and just got back from landed this morning. Cool. How would you describe the business these days? I know that - Well, first we got to talk about the AI scene in Vermont. Oh, yeah. Picking up? Yeah, I'd say that there's like one or two companies. There's a lot of, I'd say, independent-minded founders out here. but generally speaking, I'd say it's more of a farming.

2:01:47Yeah, my little brother who's not in tech went to college in Vermont and just stayed. I've tried to get him to come back forever. He's committed to the great state of Vermont. Well, let's introduce him to Masayoshi Son and maybe he gets started with building an AI company out there. It could happen. It could happen. It's not off the table. So yeah, brief introduction. How would you introduce the company? How are you describing the different offerings and kind of challenges right now? Yeah, so Lambda is an AI cloud. It's a essentially really massive GPU cluster that all kinds of startups and hyperscalers use.

2:02:33and we've both sort of taken, invested a ton of work. It's basically around a billion dollars worth of GPUs that we've deployed. Wow. And we've invested, at this point, probably around$100 million into the virtualization software that allows us to essentially take that massive cluster of GPUs. So you can imagine taking a cluster, let's say 16 ,000 GPUs, And you can sort of dynamically partition it and hand out little slices to people on as little as a 15-minute basis. And it's quite different from, let's say, a bare metal cloud where you need to sign a three-year contract. We've basically implemented all this virtualization software that lets you hotel on and off of it.

2:03:28Sure. while still maintaining high-speed Infinitan intercom. You did the meme. Another billion dollars for Jensen. Another billion dollars for Jensen. Is it actually primarily NVIDIA chips, or are there other GPU providers that are competitive at this point? I've been following George Hatz's journey with AMD. Is anything competitive with NVIDIA right now from your perspective? I don't think that there's anything that's competitive with NVIDIA right now. Yeah. What you just see in the customer demand is total and utter NVIDIA supremacy. Yeah. And I think that this is going to continue on for some time.

2:04:14The task I always like to say is you've got to get a customer. They've got to be able to download kind of any arbitrary model off of Hugging Face, run that model, train the model, fine tune the model. And then they have to buy from you and then buy again. Right. So that's the test, I think, for any chip manufacturer. And it's really hard to get a stable software stack. Do you like the term NeoCloud? Is that appropriate? You know, I think it's fine. I don't really have any opinion on it. We kind of look at ourselves as, we've always described Lambda as what would the cloud look like if it was reinvented from the ground up for AI.

2:05:06You know, training and inferencing large language models, training and inferencing large scale neural networks. Yeah. And I think that NeoCloud is an appropriate description for that concept. What is the balance between training and inference right now and how has that evolved over the last couple of years? Yeah, so, you know, we – for the overall business, we're just a bit over$500 million of top-line revenue. And I'd say that essentially – Congratulations. Thank you. Thank you. So essentially, historically, that's been majority training and then sort of minority interest, maybe 80-20 for training to inference.

2:06:09And now we've started to see that kind of switch over. I would generally say that when I see any sort of net new deal in the space for large-scale GPU capacity, it tends to be more inference-driven these days. Sure. It's great. We're actually using the models. It's not just training every bigger models with no – I mean, we saw that at Google I.O. Sundar Pichai pulled up the chart of tokens generated, and it's just completely up and to the right. it's completely up to the right. I mean, training is always going to be something where you do it once. And the whole point of training is that you want to amortize that over as many tokens as possible, right?

2:06:50Because that's just your fixed costs that you're wanting to spread across as many generations as possible. And so that's the point. The point of training is inference. We're there. I always kind of look at this as like a, I always like to sort AI companies by revenue and just go, okay, well, who's at the top? Probably like OpenAI today, 4 billion, 5 billion. It's a little bit hard to exactly guess. But then next one down, you kind of look, Anthropic, 800 million last published, mostly developer API, but probably fast growing. I would imagine they're probably in the billions now because Sonnet 3.7 was really good for code generation.

2:07:32um google doesn't seem to be charging um from what i can tell uh charging me 250 a month now just to it'll pop up to 500 too so they're getting my money what are they charging for so i am now on the gemini ultra plan okay and uh and i have 2.5 pro preview and then i also have access to vo3 but it's very limited like they can only generate two or three video clips per day It's heavily throttled, but it's a$250 a month plan right now, and it's going to jump up to$500 a month in a few months. So I imagine that across Gemini Pro subscriptions, they're probably going to grow that pretty quickly just because I'm seeing the number of five-star reviews.

2:08:15Zero to billion dollars in a couple of weeks. Absolutely. And so when you kind of do that breakdown, what do we see in terms of like total top-line revenue? It's all inference, right? I mean, all of that is inference demand, whether it's image or chat, GPT or video generation. And then mid-journeys, probably in the hundreds of millions of dollars revenue run rate. And so there's some substantial businesses that are being built right now. Yeah, what are some kind of narrative violations, things that you feel like the broader ecosystem is getting wrong right now, given you guys have unique insight into actual usage and activity yeah um okay so the the first one is just uh obviously i think if you look back there was that um that sequoia uh think boy piece um that was published on like uh where's the gp you know where's the ai revenue what's you know sure sure they sort of did that it there's a lot wrong with that analysis um so i don't need to go into that but i'd say like the general pessimism of like oh it's a bubble it's a bubble it's a bubble yeah i i just remember you know i started lambda in 2012 2013 was the year that like mark zuckerberg went to nerd and hired hired um yan lakun to to start facebook ai research at the time and ever I just remember everybody in the field kind of at the time looking around oh is it a bubble you know Google's just bought deep mind and uh you know Facebook's buying Jan LeCun so it must be a bubble because because Mark Zuckerberg just came to NeurIPS but um I think that I think that in general everybody underestimates how just what exponential growth looks like from, you know, it always looks the same in the same, you know, wherever you are, it always looks, looks essentially the same.

2:10:23And how good the code generation has gotten over the last 90 days, right? If you were to flip back before January, you know, state of the art for code generation was 4.0, and then 0.1 hadn't even, I think, been exactly released yet. With 0.1, 0.4, Claude Sonnet, 3.5, 3.7, all this stuff, the code generation is getting so good that I think I can say here, which is in a couple of years, you're just going to have a function that goes from cash into software. And that is going to completely change the way that businesses operate because you're going to spin up 500 different versions of the piece of software that you're searching for.

2:11:13You're going to be able to do this sort of like high throughput search through software space. It's going to spit out a bunch of things. It's going to have like a, maybe a tastemaker model that just rates it based off of the computer use compiled version of that software. It's like, well, these are all the source code bases for all 500 pieces of software. These are the top five. I'd recommend you launch this one. Go for it. And I think that's going to really change the way that the world operates in technology. So should you learn to code if you're a teenager? Yeah, absolutely. I think that you still need to learn how to code.

2:11:49I don't see... I think it's really hard to... In the intermediate period, you're still going to need to learn how to code. Sure. I think that you're just going to see just much higher leverage teams in the world. You might always need to think like a programmer and so learning to code is part of that process. That makes sense. What about the evolution of chips and semis? Before we dive into that, I feel like there's this general thesis that sort of, we've talked about this on the show, this idea of like uh we call it like gyro dreams of sushi software right this like really craft more art than science more yeah art and science like super intentionally built software uh that that is just you know super super thoughtful and you can see this in companies that like maybe go after a category like crm where it's like okay salesforce dominates crm but they come in and they build this sort of really beautiful, really thoughtful experience.

2:12:57And typically, if the teams are good enough, they'll end up doing well. I think there's this sense that that class of software is safe from completely AI-generated software. But in the scenario you laid out where you generate 500 different variations of a potential tool that you'd use, and it sort of automatically ranks it based on some taste-driven benchmark. Is it possible that that class of software is at extreme risk to disruption as well? Certainly, I think anything that's just software is at extreme risk. You look at a business like Salesforce, right? It's almost as if, I mean, how important is the software?

2:13:43How important is the brand? How important is the distribution model? And I think that when you look, the bigger the company gets, the more important the latter becomes. The distribution model, the brand of the company, how deeply embedded it is inside of the day-to-day life of the customer that's using it. And it's sort of like, well, what is the value of Salesforce? Is it the software? No, not at all. The value of Salesforce is the fact that every single company in the world, the first thing they do when they hire a VP of sales is make sure they have a Salesforce implementation. And then it sticks with the company until their S &P 500 component.

2:14:22And then all of the company's data is in Salesforce. Does that have anything to do with the software that Salesforce has written? Does the mode have anything to do with the replacement cost of developing the software? Crowd app. No. Right. so it's kind of interesting because like stuff like software will be solved but then stuff like how do you build moat and how do you build a business won't be solved so it's sort of like maybe just you know this is we're in the we're in the world for just uh the the business co-founder dominates the idea the era of the ideas guy the era of the ideas guy is upon us yeah that's amazing Yeah, that would be really exciting for everybody.

2:15:04I want to talk more specifically about the path to this future of like turning money into software and creating value that way. What is more important? Just bigger training runs, knocking down higher MMLU numbers, benchmarks, higher IQ points versus distilling models, faster inference, cheaper inference. baking some of these models down into silicon, what we're seeing with etched and putting the transformer architecture on a chip. They seem like two different vectors. Every time a new model comes out, my reaction is always like, well, this is good enough. I just want it to be faster. And so mid-journey V6 or whatever, I'm usually just like, yeah, I just love this to be instantaneous.

2:15:51As soon as I type the words, it's just generating in milliseconds. At the same time, the labs seem to be iterating towards bigger and better models and they have a mentality of like jobs not finished. But what's your take on the trade-offs there?

2:16:09Well, what we're seeing with the advent of reasoning models, as well as like the models that basically will do reasoning and then sort of like retrain the model to not do any reasoning, but sort of baking in the reasoning, um, is that this, the, the amount of compute that you, you know, the, the, the performance improves as a function of how much runtime compute you do. Yeah. And so if you can make a faster model, uh, and you can reason faster about it, you could, you can make an argument that, that, that actually, um, might perform, might perform better in some circumstances. I think that we're going to see all dimensions of that space explored by a variety of companies.

2:17:00There will be people out there at the edge building the biggest frontier models. There will be people quantizing and distilling those bottles down to something that runs locally on your phone. That was actually kind of one of the things I did before this iteration of where Lambda is. It was sort of trained conv nets that ran locally on the iPhone. Oh, wow. And this is like 2013.

2:17:26And you kind of see the same thing. There's use cases for that. It's super useful. You can have privacy preserving image recognition on your phone, but it's not going to be the same quality as like something that goes back to a data center. I think actually if there's one narrative violation, just going back to, you know, you said this like world of software generation. A lot of people are kind of stuck in this like, okay, AI is generating software. And I've got this entire, I think, thesis on where the future where we're going, you won't need any software at all. And that the neural network is going to completely replace all the software.

2:18:05And so let me walk you through this. The idea is that instead of generating a program, let's say a calculator or an Excel spreadsheet, just go to chat GPT and say, hi, please behave like a program. Please behave like this calculator or behave like this spreadsheet. Generate and ask a user interface for me. And I want you to essentially just respond, you know, implement the logic of that program in your mind. and that is what I call neural software. And it's really squished. You know, normal software is really brittle, right? If you make a typo, you leave a keyword out, you miss a semicolon, it's not going to compile.

2:18:48This type of neural software, it's not really possible to have a bug. It's really more just that you have a misunderstanding or you've misprompted it or something. And I think that that's where all of this is going. It's not code generation, but it's going to be your large language models are going to sort of take over more and more of the program space. And you will be largely interacting with these sort of transformer models or next token prediction models generally. And they will be the software that you interact with. What about diffusion models? We saw Google bring diffusion to text models.

2:19:34I was seeing something like 900 tokens per second. They generated, I saw a demo where someone generated a full calendar application, all the code for the calendar application in three seconds. It was 3 ,000 tokens or something like that. That feels like an algorithm from image generation that now we're seeing in the text world. simultaneously we're seeing images in chat gpt maybe do something more transformer or token based and so these lines are blurring can you give us any insight into what's happening there is that exciting or is this kind of just in the experimental phase well i mean this is like if it's whether it's exciting or not is going to kind of answer the question you had earlier which is how successful are baked in transformer ASICs going to be right because sort of like as the space the underlying space changes yep um then every one of those sort of ASICs now becomes a lot less valuable and you kind of have to go back to the the more general uh tensor processing uh that you see inside of tensor cores and you see inside of the architecture of like things like tpus yeah and away from you know really specific things that have to do with like um the kv cache and different transformer specific architectural things you might want to put into um an asic and so i think that it's it's interesting to see you've got diffusion models you've got things like mamba um and and where where where there's there's alternatives to the transformer that have what's called basically linear complexity in terms of the amount of memory that you need for the context length growing, which is better than the quadratic complexity you see inside of normal transformer models.

2:21:32And I think there's going to continue to be a lot of innovation in the model architecture space. And that will probably benefit Nvidia a lot. Got it. Are there any other side projects in semiconductors that are exciting to you? Huge wafer scale computing like Cerebrus. We talked a little bit about baking things down onto ASICs. We saw that path play out with Bitcoin, with the FPGAs and then the ASIC kind of Bitcoin mining. There's other approaches and I'm sure Nvidia is not asleep at the wheel. Jensen Wong's in founder mode. He's aware of the boom. He's aware of the demand. I'm sure people are asking him, how can we run diffusion models faster?

2:22:15How can we run transformer models faster? What are you expecting on the semiconductor side over the next few years? Well, it's pretty clear that I'd say the front runners for competing with NVIDIA, which are all very far behind NVIDIA, but the ones that are sort of, I think the farthest along is probably, I would say today is Google and to some extent, Amazon with Tranium and Inforget. Yeah.

2:22:51that space is always evolving fast, but I think it's kind of a little bit telling that AMD hasn't, with all the resources, with all the sort of clarity on what you're supposed to build for the market, hasn't been able to kind of capture enough market share. Dylan Patel is going to turn it around. He's going to write just one more report and AMD is going to be back. I'm bullish. That's all it takes. All it takes. I love it. But in terms of, can you give us a little perspective on, obviously, if I'm doing a ton of inference on Lambda, I'm probably in the CUDA ecosystem, probably in NVIDIA land. If I want to take that over to Amazon or Google with TPUs, how much of a barrier really is that?

2:23:46can you kind of explain? Because at the same time, we have these incredible code generation models. It feels like putting an AI agent on rewrite this CUDA for TPU, that seems like the easiest thing to do. It seems like a problem perfectly tailor-made for AI agents to just sit there and write boring translation code. It's not even feature design, right? Wasn't that in the TL where someone was making a joke about the anthropic safety snitching on the user? And someone said, port this PyTorch code over to Jax. And then it says, you know, searching, calling FBI. Calling FBI. Do not do this. You're going to pull the rug on the entire economy if you do that.

2:24:33Preserving the underlying chips that it's running on. It's so important to America. Um, so, so I think that, I think that, okay, the, the, the, the, the stuff that this code generation is really good at today is still sort of what I would call within the realm of what, you know, one shotting a basic program, one shotting sort of a, a one page program where it can be much longer than one page, but it's sort of like single file. Yeah. Or a function or, or, you know, a class or something like that. not necessarily something maybe i'm just behind on it a little bit and i'm not you know doing what the kids are doing or something with with with uh sort of an ai ide but i just kind of like when i'm when i'm vibe coding i will just tell either chat gpt or gemini or claude i'll go do this in one page like have it be a one thing i want to copy and paste this into my my thing and run it um and it's it's quite good at that now when you talk about like going through an entire code base um fixing the compilation errors because there will be like subtle yeah uh subtle bugs that get introduced it's not quite there yet yeah it's just the problem is that i now am at like 100 confidence that it is going to be there uh in just a couple of years and that that's kind of why why i know that every sort of megawatt that that we build and every gpu that we deploy is just going to get met with demand on the other side because just two years out like if you just look two years ago 2023 code generation was primitive yeah primitive it didn't work now today it really works for more simple programs.

2:26:25I think two years from now, it's just going to, it's going to make you feel sick when you look at it. So what are the big bottlenecks you, uh, you foresee between like, uh, energy, water, land, uh, are you going to be building the data center in space? What do you think in the future looks like for you? Um, so I think that the, the bottleneck is definitely what I call like wrapped power. So the, I think there's plenty of generated power. Right now it's just not wrapped up in a data center shell. It's not in a powered shell. It's not in a facility that has direct to chip liquid cooling integrated into it.

2:27:07And so it's sort of like that wrapped up power. That's like ready for the current generation and next generation of chips. I think that there's definitely some regulatory bottlenecks, or I would just say regulatory hurdles that can be removed. And I think there's a lot of hope that this administration is going to start to remove those. that's like whether it's like looking at sort of the way that we run utilities where you kind of have to in some cases become an unregulated utility and put like let's say you know build a data center power plant which is to say behind the grid or not attached to the grid power generation station next to a data center and not every state's going to allow that and I think that there's there's probably a lot to do in the regulatory side to unleash the free market and let people build.

2:28:01And so I think that I think there's some hope there. And the, the other thing is just really, I think building a large contiguous spaces is like, is pretty clearly the, the answer in my opinion. Well, good luck with that. I'm glad I don't have to deal with it. it's a lot of logistics it's a real pain it sounds like a lot of work but you've been doing it for 14 13 years yeah classic overnight another decade in you at least so good luck to you yeah we'd love to get you back on this is a fantastic conversation thanks guys it's so many fun love what you love what you do long time fan this is i didn't i didn't clock you as a monster guy too we got to go deeper on energy drinks yeah there's so much to talk about lots to talk about but this is fantastic thank you so much for coming on great chatting uh quickly let me tell you about numeral sales tax on autopilot spend less than five minutes on sales tax compliance should we sing your sales tax ag sales tax ag uh go get uh go get numeral head over to numeral hq.com how'd you sleep last night john um oh terrible i woke up at 4 a.m it was a disaster i'm coming I know I'm going to get cooked, but you don't have to be cooked because you can go to eightsleep.com.

2:29:19Use code TBPN. I got a 76. It was brutal. 99. I can't get to it. 99. I'm back to back 99s. I don't know what I'm doing wrong. And then also we got to tell the folks about public.com investing for those who take it seriously. They got multi-asset investing, industry-leading yields. They're trusted by money. Let's give it up for multi-asset investing. And their partnership with Aston Martin, which we might have more information on soon, but stay tuned. I cannot wait. We got Sam Lesson coming in the Temple of Technology very quickly. Do you want to do some timeline while we wait for him? Is he here?

2:29:55You know I love timeline, John. I know Dwarkesh Patel dropped a banger on Claude Ford Day. He sat down with Trenton Bricken and Charlton Douglas talking about Claude Ford, how far reinforcement learning can scale. I haven't had a chance to listen to the whole thing. I listened to about half of it, kind of in and out while I was trying to sleep last night. But just a very cool vibe of like three people having a conversation right on the day when you want to know this stuff, talking all about this fascinating metaphor for AI safety because it's just been controversial. Of course, there's like some drama around some random anthropic news.

2:30:31But the cool thing that I liked was they said, basically, if you want to tell the AGI what to do and how to behave in the best interest of humanity. You could give it specific rules, but it might actually be better to say, imagine there's an envelope, AGI, and in that envelope is what I want you to do, all the rules I want you to follow. You can't access this envelope, but you have to behave in accordance to what you imagine to be in this envelope. And so the AGI is just like, I have to behave in the interest of humanity. It was a cool thought experiment almost. It was very cool. Anyway, Patrick O'Shaughnessy was shouting it out.

2:31:08He says one of his favorite genres of podcasts is recurrent expert guests on the same show. If you're interested in the nitty gritty of AI model progression, you'll enjoy this. So go check that show out this weekend. It's a great listen. And we have Sam Lesson from Slow Ventures in the studio. Welcome to the studio, Sam. Gentlemen. There he is. It's Jessica Lesson's husband. Put it on the camera. In the flesh. I saw her like literally minutes ago. That's how famous I am. Tell her to come by at least and say hi if she's not busy. Yeah, yeah. She's busy this moment, but maybe later. Yeah, well, we got 30 minutes, you know?

2:31:44We got the crossover of the century, TBPN, the information. It dropped today. It was great. I love that. Yeah, actually, I knew about the development of that story from dinner conversations. Oh, yeah. It was a lot of fun. Oh, I bet. I bet. Abe is the man. Yeah. There you go. It was fun hanging out with him. and um kind of the seymour you guys you guys are the cover of the weekend section is that right we are we are the big are you guys gonna do a print edition for us coveted space i always thought that they should actually do a print edition i think it's time everything goes in a cycle i got the i got the wall street journal here i need yeah it really i know i was in a once weekly uh saturday dropped off on the doorsteps of every hundred percent i would like hire the kids to deliver it with the, you know, throwing it out the window.

2:32:31Look, I was in an airport lounge in Germany two days ago, and they had all the print newspapers. And I'm like, look, there's a missing opportunity here. The Germans want to read about the information. They want to read about Technology Brothers. Yeah, of course. Let's get it to them. You guys are big in Germany, I'm sure. Well, somebody started translating our streams into Dutch, and they're doing other things. That doesn't seem like the first choice. Like, don't you want, like, Mandarin or something? I think it doesn't seem like your number one. We're trying to get big on Weibo. So yeah, that's next.

2:32:59Yeah, but we're expecting to have some pushback.

2:33:05Anyways, it's great to have you on as always. Congrats on your success. You guys are everywhere. This isn't the new studio yet, though. Well, you're a part of it. You're a part of you're a part of that. You have to come down next time you're in L.A. Come by in person. Be great. Yeah. Tuesday will be the first show. Yeah. It's a lot closer than Englewood Cliffs, New Jersey. You know? Yeah. Yeah. What's what's there? Oh, God. Is that the new Netflix? The NBC. You can't make this whole CNBC rival thing and not know where CNBC is. Everyone always says, Andrew Reid was like, oh, they're making Squawk Box for tech.

2:33:37And I was like, got to start watching Squawk Box.

2:33:43Netflix has a new billion dollar New Jersey studio. Maybe we're taking the fight to New Jersey in the future. We'll see. LA is fine. Where were you traveling this week? i i i went i i've had an insane travel schedule but i was uh monday i went i was new york uh frankfurt new york san francisco was my was my week what's going on in in germany germany was a personal thing i had to take care of but but new york was the solana acceleration uh what is it the salon big solana conference which they call ship or die very dramatic nice nice uh you're infamous there right the gm the gm guy do you still get comments on that you know i saw i saw raj backstage you know i was i was one of solana's first investors i like my my um jason calcanus i was the first investor in uber was like i like do that for solana right and so you know i've known those guys for a long time and i saw raj backstage we were reminiscing when i did the gm thing which became one of the first meme coins and blew up and got if you remember i got raj banned no I know it was about that yeah this was like Twitter still right so it was like somebody at Twitter was like didn't get the joke obviously didn't have context that like you guys are boys and just yeah the thing is I unfortunately have other people on Twitter legitimately threatened to kill me which that's not so good you got to get that taken down but the the Raj was a pretty funny one to for them to respond so so quickly to yeah yeah quick reaction uh there's so much so much to cover it's ai week it's ai week it's in every week ai week boys no but this week was particularly ai week because of i o microsoft google i o open i o anthropic uh a big uh are you are you taking a victory lap on the idea that this is a sustaining innovation that google's gonna win you guys actually i'll tell you a funny a funny end of You guys know my associate Jack Raines.

2:35:47You know this guy? Yeah. Jack's sitting. He's right over there. Jack, he can come wave for a second. Hey, Jack, come say hi. Jack sends me this long missive this week about how amazing Google is and how Google is going to win the AI war. What's up, dude? And I'm like, what value are you adding? I've been saying this for two years. I'm like, thank you. I'm like, I want to check out all the little Google. I was like, Jack, you're an associate at an early stage venture capital fund. Telling us to buy Google stock is not exactly... Hey, guys, I want to see slow as the next crossover. Yeah, crossover.

2:36:23Become an RIA. Put 30 % of the fund. We've actually looked into it. It's definitely not a thing we'll ever do. No, I mean, my favorite comparison, I forget who posted it, but they were like, do you want Google at 17 times earnings or Costco at 50 times earnings at this moment? at this AI inflection point. Look, here's the negative on I.O. The negative I.O. is that per usual, Google's actually terrible at marketing, right? And so they just launch everything. And so you're like, I have no idea even where to focus right now, right? Like it's not - My value comes in figuring out. Thanks, Jeff. Focus on generative AI videos, VO3, that's the thing.

2:37:05It's like so many things you're like, I don't know, but the net of it is like, they're just crushing it, right? like from a tech perspective. And I got to say, I'll tell you a funny story. I spent like 20 hours, quote unquote, vibe coding over the last week because I was on so many airplanes. And it turns out airplane Wi-Fi, vibe coding, perfect match of activity. So I just like was like messing with these things till my eyes bled. And you know what happened in hour 11? What? They stopped using ChatGPT and went all in on Gemini because it's actually better. 2.5 Pro is just like way better at a bunch of these coding tasks.

2:37:36It's better at structuring. It's like a bunch of stuff. So I think it was absolutely a Google week. But were you using Codex? No, I wasn't using Codex. Oh, there you go. This is a skill issue. Look, this is one of my other insights from the week, spending a week literally flying around the country, vibe coding in the air. Yeah. The whole what model to use thing is a complete fucking mess, right? And I think the analogy I now have in my head is it's kind of like, imagine walking into a crowded room where you don't know anyone and saying, who should I trust, right? And then by the way, every two seconds, a new person pops into the room.

2:38:12You have no, and like, people are like, oh, or you can trust this person 95 % of the time. You can't trust them with your life, you know, but totally, that's a totally separate chain of thought, which I think is important. I was talking to some entrepreneurs today about if you think of what AI is today, it's like having it a B minus employee or like an 80 percenter. And the question is, how do you get value out of 80 percenters is actually more difficult than it looks in a lot of ways, right? Because the reality is so much of work is about trust. But isn't it more their B minus because you're averaging their effectiveness on different things?

2:38:45In my experience, it's like for certain tasks, it's like 80, 90 % of the time, it's amazing. And then some percentage of the time, it's so bad that you would ask the employee, is everything okay? Kind of. But I think the thing I really think about guys is, you know, people are always like, oh, there are all these tasks that AI is more efficient at. And so clearly those jobs will just go away and it'll be it. And I think that's exactly the wrong model. Because the reality, what people don't understand is the real reason you hire people to do work is not to do the work. It's to have a throat to choke when the work goes badly, right?

2:39:21It's like, why do you have an accountant? You don't have an accountant to do your taxes. You have an accountant so that when the IRS comes and says, what's wrong with law? Why did you get this wrong? You're like, I didn't do it. That guy did it. Right? And so I think that's the interesting thing is like, there's all these models like, great, good news you get rid of the employee and now get a cue that's mostly right and you approve and you're like i don't want to approve that stuff like the whole point is to have someone else yeah i think our example is like we use a lot of the different deep research products and like we still would love to hire an amazing researcher yeah so that you can fire them when they're wrong right like you need like just the point is like there's this whole i think there's a lot of like well it's an agency thing too it's like if you could tell if you could tell deep research to like try itself and then try other products and then combine the outputs of all the different things and maybe that's a research agent that can there's there's all these things but i think the ultimate model in my mind is i don't even care what the average persons are using i don't care about some review online that's four stars for coding for this type of i don't care what i care about is one of the smartest people i know in the world using and like how do you make that transparent so i could just be like oh it's this type of task i'm gonna trust this model for that and like that needs to be up to date.

2:40:30It's much more like an identity problem or like a human trust problem, right? Than it is, you know, a technology problem in a lot of ways. And it's just changing really fast. It's been a fascinating week. But yeah, net, I mean, pulling up, pulling all the way up. Yes, I'm taking a victory lap on Google wins. And yes, I'm taking a victory lap on incumbents win because that's what's happening. So you're saying to Jack, like, thanks for reading my memo from two years ago yeah exactly well in fairness to jack he wasn't on the team two years ago yeah he was very excited to discover a notebook out but love you jack i gotta i gotta gotta get him out of the pool house yeah anyway uh i want to talk about some other stuff uh what what do you think about the uh johnny i've news and and hardware generally i mean where to begin i know I know.

2:41:21This is my favorite part of having you on the show. We don't have to prompt at all. We just like to give you three words, Johnny, I've hardware. Well, okay. A few things. First of all, I mean, my, my, my, my tweet on this one was happening was like, well, I guess all these hundred million dollar AI researchers are cheap, right? It like these, it's the designers that are going to get paid bank on aqua hires in this, in this era. Right. Look, I think it was marketed in a really interesting way, right? Like it's marketed as like, they tried to pick the biggest number possible. It's a 2%, they paid 2 % for this thing, right?

2:41:51For him. They market it as it's this big number. I think because that's, it's kind of the pissing match of AI, right? Is you're supposed to look really big and powerful with big number. And I'm sure Johnny doesn't mind being told that he's worth$6 billion. But like, it's kind of one of those interesting PR stories of like how you shape the narrative as much as anything else. The hardware thing is interesting. Like on one hand, I actually think it's like, if you're open AI, you do need a hardware play. You're going to get fucked otherwise, right? Like, you know, Meta discovered this years ago, which is like being beholden to those hardware is really tough.

2:42:24Right. In this case, it's only more so. Right. Because of kind of the way these chips are going to fall with Android and iOS. So I don't think they're wrong. And like Meta's got its glasses. Everyone's like it's not strategically wrong from first principles. I think the problem is that I have is I just think that like the likelihood they're going to successfully introduce a new piece of hardware is like zero. right um like and that's just you know it's you know you're gonna have a phone and they could try to compete on the phone but it's very hard for them to imagine building like a fully featured phone that you know whatever and then also having something that's so great well the one thing that does seem clear haven't they said this is a new type of device that will integrate with your phone and your computer and that screams to me okay well that is is app yeah yeah i said septum piercing where i saw you know my take was did you see e-cane so this was so when i left went in 2014 no but sorry sorry to finish my thought that the issue with that is if you're saying we're going to create a new device that integrates with your apple ecosystem that seems like a bit of a like a uphill battle look what happened to pebble right in the apple watch right in the end of the day it's like look you know you have the iphone is a hugely breakthrough device this is by the way not my line It's a friend's line.

2:43:42But like a hugely breakthrough device. You already had a phone, right? You know, it wasn't like a new thing you were putting on your body. The glasses work. That's a thing that was already on your body. The watch work. It's a thing that, you know, it's really hard to imagine you're going to get through with some new device. So my joke, but it's only half a joke, is look, you know, 12 years ago, I got obsessed with the idea of making an E-cane. So imagine like bringing back the cane, right? That's what you meant. I had no idea what you meant. I had no idea what you meant. I had no idea what you meant.

2:44:11If you think about the cane, it's actually, you put a shitload of batteries in it. Yeah, yeah, yeah. Big antenna, right? And you know what the nice part is, like, the world is getting less safe, so you can beat people with it. Yeah, yeah, yeah. The world is, like, the world is aging, so, like, everyone needs canes. Everyone needs canes. You know, like, there's all these, like, macro trends in favor of bringing back the cane. And then, you know, could you drop, like, a GPU and a microphone and a camera in the top and, like, talk to your cane? Smart cane. Yeah, that's smart. Yeah, smart cane. smart cane no more dumb i don't want to be pitched anymore non-ai canes okay i've had enough like i'm just saying like if they actually if open ai came out like we were bringing back the cane i'd be like all right listen for all my shit well you know that johnny i've got his start at a design firm that was working on bathrooms and so maybe there's all like smart toilet you know yeah well there is someone just came out there's a there is like a new fitbit for point for your poop okay i know some friends that have funded that that's what i think i i funded that you funded that i know some people i thought it's called it's called i think it's called throne i think it's interesting uh for a bunch of reasons one is it's just like personally as like a quote unquote biohacker there's so much information on how your digestion what's happening with your digestion your hydration all these different things look in the end of the day you want to copy behaviors people are already doing and you know in college one of my roommates brian would send me pictures of his poop all the time so like you know you just gotta double down let's move on let's move on brutal uh hyperscaler capex give us anything to get us off of this hyperscaler capex here's my line on it i flew through jfk today they're spending 20 billion dollars renovating jfk yeah i'm like this capex is nothing spend more money on ai like if you're gonna spend 20 billion dollars on like renovating jfk all of a sudden spending 100 billion on ai doesn't seem so crazy what about say as much as you can about what meta you know met as playbook over the next kind of six to twelve months I know they've been having talent you kind of retention yeah llama behemoth delayed delayed look at the end of the day I again I as you know have huge respect for the team there and the leadership there I think the beauty of meta as I've said many times is it's heads you win tails you win right you know from from a from a investor perspective right which is if you think about it it's like the most obvious highest value use case in the world for everything that exists is making ads way more targeted and effective right and that's happening and it's an incredible trend to ride and they're going to crush it you know mark and that team is like deeply competitive they want to win win win right and so just from a pure competition perspective i think they're going to like go to the mat to win win win and like there'll be many chapters to that But the good news is they have a massively profitable business and an immediate use case, right, that is deeply aligned.

2:47:11And so I just, you know, I learned a long, long, long time ago to never bet against Mark, especially when he cares about it. Yeah, it feels like with Meta, there's huge value to having Llama be a free LLM for them that they can vend into every corner of the app, even in places the consumer doesn't feel or see. so uh profanity filtering and uh understanding what content goes where ad matching uh generative ai to clean up instagram photos all these things that will just be under the surface yeah as opposed to what google's doing which is like now we have 10 new apps notebook lm like you're we're not seeing that for meta and maybe that's the right strategy i look at the end of the day meta has the most natural beautiful use cases for ai which is what you're pointing out and therefore like they don't need to do a bunch of crazy shit right they just need to like win at like the core and And then basically, again, the open source thing is clear.

2:48:00It's like they actually, like think about ads. They don't really care if they're doing the AI or someone else's because it's all going to go to them either way, right? At the end of the day, it's just better if more people have access to it and the innovation will happen. So, look, one of the things I did in my VibeCon, I got really into like all the platforms, right? I spent a bunch of time. Remember DigitalOcean? Like DigitalOcean is doing a great job with AI, right? Like it turns out like their services are really good. And it was really easy to boot up some agents on top of a knowledge base, on top of an applet.

2:48:29That's always the worst part. You write some code and you want to actually get it on a server. It's always still a hassle. And it's like they've actually done a great job. I was incredibly impressed. But then you say, where is Llama showing up? It's like, well, it's right there. You just click a button and you're using Llama. It's like no bullshit. We've got to get a killer video generation model out of Meta. They're the only one that has another data source that's similar to YouTube. market how do you see that market evolving there's a bunch of foundation model labs doing video generation and i can see some of those getting to the point where they could crush it for big hollywood movie studio you know making very specific types of scenes or assets but then it feels like the low end of the market for you know if you're a random smb and you want to make an ad or something like that yeah maybe use some vertical sass but also i wonder how much it matters i mean like you know i had this conversation actually with my good friend dave morin on our little weekly you know bullshit session more or less more or less we're talking about dave also somehow cracked like the top of the angel there was like oh yeah he's number one list and he he was the top dog it's amazing so well let's take on the validity of that list we'll talk about yeah we could but i mean i just like to see my boy dave at the top look my boy kevin my boy my partner kevin was like number four on the list it was a very interesting let's go but the uh i wasn't on the list but you know that's okay the um the uh but uh what i was gonna say is the um the thing dave and i were talking about is like look you know if you think about what's going on like with memes right in this whole world you have this is like high road low road thing everyone's really excited about like super high-end video generation and hollywood and you know what's happened in the media landscape those things don't make any money no one cares about these movies anyway.

2:50:18They're consuming low-end crowdsourced memes on the internet. That's where attention is. That's where the energy is. That's where entertainment is. And so I think obviously the pure technologists and the highfalutin people are really excited about the narrative, storytelling, beautiful, da-da-da-da. But when it comes to what tools will actually be used, how things will be marketed, how humans will interact with each other and communicate in the store, I think actually it'll be much more budget-y tools that actually win. and the other stuff is kind of just you know intellectual you know masturbation a little bit it doesn't mean you shouldn't do it like people speaking of that like from a speaking of that why is speaking of uh masturbation uh why is only fans getting priced at eight times earnings you know i i so desperately want to buy only fans guys i wanted to buy it for years i had plans like There's been a few iterations of this.

2:51:10No, it's so good. It's so good. What do you have against OnlyFans? It's a mess. It's so good. It's a mess. Your reaction is part of the reason I want to buy it. Yeah, you love the controversy. The buyer pool is small, so the price is low. Well, that's part of it. The other thing is everyone, because of the buyers, it's actually the best AI short you could possibly think of, right? because part of the reason it's getting sold my understanding and like the whole story is like well we're gonna have all these ai girlfriends sure so like the creator economy is gonna get crushed on these things and i think that is like a very 101 simplistic take on what's the other the other thing is a lot of the a lot of the cost that goes into basically generating revenue on only fans is from messaging which is done by people in the philippines and ai so this is the thing I think everyone, OpenAI is like the greatest human AI platform in existence, right?

2:52:06From a position. Because you have, I think you still need real brands. I think that having human brands matters at the top end. Your ability to then like effectively automate, it's kind of like the Facebook story, which is your ability to slot in AI and bank it, right? On messaging, it's like off the charts with OnlyFans and improve the quality. So I think it's interesting. I think it's like this classic thing where it's being sold. I've always loved the asset. It's the number one creator platform in the world. it's so defensible because they actually have a very low take rate because they're so big that no one can no one can assault them right because of how they're kind of set up what's the take rate it's like it's really low i want to say it's like 10 maybe 15 but it's like much lower than you think and as a result like you have this incredible cash engine i mean youtube's like 45 percent and many other sites are 30 just for general content creation so just the upshot is like this incredible position it's it's i think there are huge product opportunities that always have been with it but then you have this double storm of like one people like you guys boo it because you patrician technology brothers right like and you won't like acknowledge human nature and people want to do this and then two there's all these fancy investors who like think it's dead because they read the front page of the wall street journal about ai and don't understand what's actually going on i'm like that's great this is my zone christ yeah i have no i you're gonna have to become an ria to do it probably.

2:53:26So good luck. Well, I got to figure out how to not do that. I also like for what it's worth, I'm a seed investor. I don't know how to raise the$2 billion I need, but call me if you got it. Well, speaking of seed investors, I don't know if you saw this chart from PitchBook. Emerging managers are on track to raise less money this year than they have in a decade. That sounds right. It's now under$20 billion. During 2021, it was up at$60 billion. It was really, really spiked. 2022 was around$50 billion. And now we're down in... Just one JFK renovation. That is a lot of money for emerging managers.

2:54:01I assume it's defined as first fund. We're going to put$20 billion as an emerging manager. How much did they get in 2021? You put it in$20 ,000,$50 million funds. That's the idea, at least. Or that's what happened. I've been so... I don't see any of this working out for anyone. Like, you know, I know it's one of the they'll always be the random winner. Yeah. But isn't that the nature of the game? It's like startups follow a power loss. So do funds. I just think we have to hit a size gong because the 64 billion that they raised in 2021 still earning 12 billion of fees. Yeah. And so let's just give it up.

2:54:39Congratulations to the emerging managers. Guys, I think you guys are cheering too much. If you do the math. No, I'm joking. to be clear. Doing the math on being a small fund and being locked in for 10 years as like a solo GP with your emerging manager fund and your kind of shitty subscale, it's actually a pretty bad business. What's wrong with being locked in? No, no, no. What you're saying is so real, which is imagine you have a startup and it's like, okay, you have to commit to this startup for a decade, even if by two and a half years in, it's not working. Is that real though? I feel like most emerging managers, They write the checks in the first year and a half, two years.

2:55:18And then, yeah, they get investor updates. They pass them along to LPs. But they could go and get a job at big tech and just become a CEO. Even if it's a couple hours a week for 10 years? Yeah, I guess that is kind of hassle. Look, it's also, it keeps you from doing other, it's just messy and complicated. And the reality is, again, if you're earning$2 in$20 ,000, you're never going to see the$20 ,000. Like the$2 ,000 ends up actually on a$50 million fund after you pay for actual things. You're doing fine. Like no one's crying. But like, it's not like some great business, right? Like, I think the only reason to raise a$50 million emerging manager fund is like to have enough card flips to prove that you can size up a bit.

2:55:55Now, I think the flip side is true, too, which is I don't know how anyone makes money on a seed fund that's over$200 million. I just know the math doesn't work, right? I think there's like a sweet spot of fund sizes that work. Too small, you know, the economics to make it work. Even if you have a banger, it's like not that important. And like too big, it's just like the law of big numbers. You're never going to produce an important fund, right? From a returns perspective. So, look, what we do, seed investing, we're, I think, in the sweet spot, obviously, because why would I not think that? You know, we've made money.

2:56:23I'm very proud of that. I care about DPI, whatever. But here's the thing. Sorry, sorry. Sorry we're laughing. We have to explain because you can't see it. But the chyron right now says seed investor. Don't become a seed investor. Stay the fuck out. But then the reality is, even from my perspective, it's not the best business. You do it because you love it, right? But if you just did the math, being an asset manager and earning 2 % and just getting enough DPI to justify more 2 % fees is for sure a more scalable business model, which is why most people do it. Do you think part of why every emerging manager jumps into seed is just like they can't do growth?

2:57:03Because I feel like the skill set of being a growth investor is in some ways more repeatable. You get people that come out of investment banks and they have all the skills to underwrite a company that's at a billion dollars. There's less risk. But at the same time, when you think about someone who's starting a new fund, if they're going out and raising a growth fund on fund one, that just feels harder to marshal that capital. question it's like who's gonna give you lp level dollars right and like by the way if you're if you're going a later stage fund you're trying to like model to like a guaranteed whatever it is yeah two and a half x or whatever why would you give it to the crazy kid right the crazy kid you give money for because they might be right yeah yeah yeah and so you want that seed bet and then also i feel like a big part of the emerging manager dynamic is just like essentially scout funds and and there's a lot of like horse trading around like yeah we will do we'll do the deal for We started that way like 10 plus years ago.

2:57:57And like that when the market was much looser and it was a different market. Yeah, yeah. And it worked for us. And we were able to scale up into being like a legit seed fund, you know, with legit dollars and legit institutional investors. I just think like most things, all paths get super crowded, right? And so once the path is super crowded, it's just really hard to differentiate. It's really hard to make a good business out of it. And I do think like almost everyone's going to go away, right? Don't become a seed investor. You heard it here first, folks. Last, I want your quick take. XAI's Portland Data Center Fire.

2:58:31You think that's just a little accident, or you think Portland is doing Portland stuff? That would be really funny. That would be really funny. God, have you guys ever read the science fiction about Parable of the Sower? No. Great book. Great book. I'm just finishing it now, but there's a whole slob plot where there's all these people who are addicted to this drug that makes them think that fire is the greatest thing on earth so they're just like setting fire to everything and like wow this is funny it's it's i just looked it up it's a post-apocalyptic story set in 2025 wow if you caught that when you're like they're basically about people like trying to walk north from la to like north to seattle to escape craziness but the um yeah so i don't know if you told me that that drug was invented in portland wouldn't blow my mind yeah i think we got to get a this is Fantastic having you on.

2:59:20Thanks so much. Always a pleasure, Sam. Happy Friday. Happy Friday. Talk soon. Cheers. Let me tell you about adquick.com. Out-of-home advertising made easy and measurable. Say goodbye to the headaches of out-of-home advertising. Only adquick combines technology, out-of-home expertise, and data to enable efficient, seamless ad buying across the globe. And we got Bobby. They are simply the best. Coming into the studio. Yep. I'm going to do some VO3 generations for all of our partners. get those in the mix pull those up i think it'll be fun getting a little creative uh we got bobby coming into the studio welcome to the stream how you doing thank you for having me where are you are you in miami right now i'm the last one standing in my yes let's give it up yeah i've seen some of my i love miami actually stuck around thank you very much yeah yeah i i love miami it's beautiful every time i get to go visit it's always struck me as uh like the it's like a classier version of Las Vegas for tech.

3:00:17And it actually worked really well where in every industry, there's always like the CES conference that happens in Vegas. And it's a little bit below tech's, you know, prestige. And so if tech could assemble in Miami for events, that's great. We gave it a real college try. It was happening for a while. You know, there's Hereticon. Yeah. I mean, Hereticon is the best conference I've ever been to. Exactly. And it was cool that you get people that come from New York and they're not just going to go home on the uh after the conference they're going to stay same thing with sf same thing with la so it was actually a feature that not everyone was there i feel like if you throw a conference in la or sf a lot half the people are just going to go home it was special for a moment i anyway i got a what's up i got a dox bobby oh yeah because he pulled up the hereticon in a ferrari let's go we love you remember i called it out we were walking it's actually just getting the buttons be done oh really we were just talking about this you have the button you have the button issue no it's seriously it's it's like the worst thing to be in this magnificent machine and then it's like i don't know if somebody yeah sticky like ate a hamburger you have a date and you you you okay great i'll pick up my ferrari and then she tries to adjust the seat and then she's like this is disgusting oh no it's terrible what were you doing in here this is terrible that's all fixed now.

3:01:42That's all fixed now. Okay, that's good. That's good. We're glad to hear it. Find us. But yeah, give us the news. Give us the update. Give us a little introduction on yourself for those who might not know. Yeah, so I'm so glad you guys had me on today. We had a very big announcement today. So this is something I've been thinking about for 15 years and we announced our new email client called Sunflower today. Cool.

3:02:08Thank you. Thank you. It is like basically the opposite of inbox zero. okay like i don't think you should have to have an email client that makes you work for it right in other words like you load up inbox zero i don't care how many keyboard shortcuts i don't care what what you give me but if i'm having to triage every single email that comes in you know keyboard shortcut this way that way yep from the applebee's email you know no yeah yeah You need an email client that actually works for you. Okay. So on our marketing website, we don't mention the word AI one time. Huh. Wow. But this thing is entirely, this product would not be possible five years ago.

3:02:56Okay.

3:02:59It's, you know, if you want to be, if you think you need to be a superhero to use your inbox, there's a product for you. We're building a product for actual humans. Bold. And it launches, you know, the wait list launches today, sunflower.me. Yeah. And the product is like. I should actually put my phone before this because literally we broke slack today. Like apparently there's a limit to how many notifications it can send you. It just keeps dinging, dinging, dinging. That's great. Amazing. And the product is like already like you guys, you guys are the product's like real. You guys are using it, right?

3:03:40This is not, you're not just launching a wait list. Like there's something under the hood. We've been working on this for a year and a half. I use it every single day myself. There's about a dozen people right now who have this. Yep. Give you guys early alpha access. Yeah. So talk to me about the actual workflow. I imagine I still have an email address. People can still email me, but I imagine an LLM is scanning every email that comes in. Am I seeing the emails? Am I, I mean, I've used Gmail's Gemini features. I, one time I randomly clicked it and it just typed a response and it was like kind of decent, but it's honestly just a lot of clutter in the UI right now.

3:04:17Um, how are you thinking about hot on the hunt here? I'll be honest. Yeah. Well, I mean, thinking about this, I've actually was thinking I should just get a, I should just get an LLM to summarize every email and just text it to me. And then I can just text back if I want to respond and I just never look at my inbox ever again. Apple tried and it didn't work. Current mail.app. It's it's, it's not good. Total dog. Yeah. It's just so funny that email, like the default state of email is like, here's stuff that's important. Here's stuff that's not important. And then it's only ranked by when it hit your inbox.

3:04:51Yeah. Which is kind of silly because status quo of email is a to-do list filled up by other people. Yep. Totally. And that's, that's it. Right. You should be beholden to that. Right. Just because someone sends me an email doesn't, I mean, I get a lot of emails. yeah uh anyways so yeah you were you were kind of close uh we're gonna reveal our hand more and more publicly but today was like the firing the starting gun like we're building in public now so you're gonna hear a lot more don't share your revenue if it gets too high people will copy you we don't like building in public based on the signups yeah today like we are cash if if everyone converts but how much uh how much of your time are you are you gonna be spending here versus investing going forward yeah um that's a good question you know i personally funded this company uh for the alpha like you know i i to be able to read people's emails you mean i'm kidding i can't by the way i can't read anyone's emails an exhaustive security review process with google right now yeah so no there's no reading you can sign up just google just sundar just sundar yes sundar can read your emails but not bobby yeah i bet i bet he can actually i bet i bet they're pretty locked down over there um but yeah i mean and i remember uh the mailbox days and that launched um uh and one of the things that i loved that i loved was that once he got off the incredibly long wait list it basically tricked you into just archiving everything, which I thought was a great paradigm because you kind of declare email bankruptcy.

3:06:33You're not deleting these emails. So they're still there. And just that idea of, Hey, you can, you can archive things. I had never archived an email before. And I think a lot of people are in that boat where they have unread emails and they've read emails and they have a number that's like 10 ,000 next to their email app. What do you think? Do you want to hear a funny story about that? Please. I was friends with Gentry when he made Mailbox. And I worked at Facebook at the time. And I went on this hike with my friend Bo. And literally, it's like the Kalalau Trail in Hawaii. It's like a two-day hike.

3:07:06You're trekking it. The entire time, he just talked about email the entire time. He talked about email and the fact that he was about to propose to his now wife. And I said, what are you spending on this engagement ring? Anyways, so I took a photo of him. It's a beautiful photo, right at the end of the trail. And if you remember that app, when you got to inbox zero, you saw this beautiful image. And he hated that app the whole time. It just launched. He was ranting to me the entire time about how much he hated mailbox. And I was like, it's actually a pretty good app. I'm friends with gentry etc etc but anyway so when i got back from the hike i sent gentry an email and i said hey would you would you guys put this this image into the app like as the you know you're done like inbox zero state yeah and so and then i took a screenshot and i said to him i said you're in the app now you're in the app that you kind of hate you despise that's amazing and he took it a good it was great yeah we're not about inbox zero i think inbox zero is a flawed concept i think like you know it's funny i want you to do this for text right right now i have 3873 unread texts and i have that because i don't feel an obligation like just because somebody had my number at some point and like i really try to respond to the stuff that's important and pressing but i'm not gonna like get home from work and be like oh i i gotta respond to this person instead of hanging out with my kids right like it's it's not no it's not my everyone feels entitled to your time yeah right you know look i'm an investor i get the same emails probably we all do right like about 20 a day and they're all like drip auto campaign follow-up so it's like they send you a cold pitch email and then you get you get the email then you get the email and then and then every single successive email like tries to guilt you into actually you know replying yeah like start using more and more emotionally manipulative tactics yeah and it's like no no no just because know my email address does not entitle you to my time yeah i think that's right what do you think about are you gonna are you gonna tag are you gonna are you gonna like identify like emails that are being dripped versus you know artisanal organic farm to table emails we're working on that um you know what we're also working on i again you know there's a funny steve jobs quote isn't it funny a ship at least from the top but here you go um that's great um you know you know like email trackers you know like obviously superhuman does this i'm like you know you see you've seen it yeah we both allow you to block inbound uh trackers and we'll also let you send your own so our thing is like we are just so 100 focused on the individual user um so in other words like if you want to block um these trackers perfect we'll let you do that and then you can send your you can send a tracker of your own i like having those trackers i like letting the sdr now i read this 10 times i'm still not responding yes yeah that's that's That's the feature you should build is a feature to open and close the email.

3:10:45That's a better feature. Yeah. Yeah. It just keeps paying the pixel. So the SDR is like, this person, they're reading it for hours every day. What? But they haven't responded. What's going on? DDoS, they're tracking the pixel. What is your take on Eric Mikofsky, founder of Pebble? He started this company, Beeper, eventually sold it to Automatic. and the whole idea was get into the iMessage world, create one unified inbox for all the different messages. I think he had a knockout drag out fight with Apple because he was reverse engineering some of the APIs. Obviously there's changes in big tech around antitrust and it maybe now is the time to put pressure on that.

3:11:28Is there a world where we could see text and email kind of unify at some point? What's your view on kind of the long-term here with big tech? I mean, look, we have a long-term strategy. I'm not going to come on your podcast and just spill all the beans. Spill, spill, spill, spill, spill, spill, spill, spill, spill, spill, spill, spill, spill. Activate golden retriever mode. What was that? It was activate. What did I sign up for? Golden retriever mode where you're hot, friendly, and dumb. Yeah, yeah. Golden retriever would just tell the entire roadmap. tell everyone no no no no you can you can come on again and you know you know spill every time one of your companies and i pass another one okay um anyway yeah future of text email so i'm mostly concerned about email i think email it's like what's old is new again right and so if i again if i if i if i were a ship that's spilled from the top here's what you have that here's what i have to say um just like apple built i message on top of sms i think there's a layer to be built on top of email that's all i'm gonna that's all i'm gonna tease me anyway yeah no i think i think i think that's an interesting yeah framework so i don't i don't want to own all communication like i don't want to you know put a pull it on one place i just want to clean up the mess that is email okay yeah it is an interesting time right because like ai is transforming everything but yet i feel pretty confident that i will have an email address that i have to use in 15 years and like that sounds kind of crazy and silly but at the same time it's not going away fax machines we talked with tj yesterday about how he was spending 30 of their dev resources or something like that on fax machines right at pill pass When I went to Facebook, we launched as a joke, just the TechCrunch network only.

3:13:31We launched a feature called fax this photo. And they covered it. Wow. No way. Yeah, yeah. It was literally launched. Just to troll them? Oh, yeah. You can Google this story, like TechCrunch, Facebook, fax this photo. My engineering buddy, Evan, we were launching all these things one day. Fax your photos. 2009. Wow. yeah there you go that's amazing okay so facebook punked us this isn't really going live for everyone they they tried to like pretend like they were in on the joke but no they were that's very funny you know the joke either and so this guy blake ross he had to run upstairs like you know you know in a frantic rush saying guys guys guys guys it's just a joke it's just a joke yeah full stack verticalized mog that's amazing well thank you so much for coming on bobby this looks the real thing bobby this this was super fun congrats on the launch uh i'm excited to try the product and come back on come back on as you have more uh more news to share amazing sunflower.me if i can get a plug in for sure let's do it go check it out we'll talk to you there now see you guys and if you're trying to visit miami you gotta head over to wander.com find your happy place find your happy place book a wonder with inspiring views hotel-grade amenities dreamy beds top-tier cleaning and 24-7 concierge service it's a vacation home but better folks get on wander and we got oran hoffman coming in the studio he is here so welcome to the stream oran how you doing what's going on great i happen to be here very excited great to great to finally have you on yeah i should have done this a long time ago we we had a brief phone call i think a a couple of years ago.

3:15:16Good to reconnect. What's the latest in your world? Well, things are great. We've got a couple of different things. We run a weird holding company where we own a bunch of data companies, start different data companies. And then we're also at Flex Capital where we do a lot of seed stage investing. We did 20 deals last quarter. We're also doing about 100 per year. That's amazing. Yeah, we were kind of struggling to find what your title should be. So we just put founder and investor because you wear a lot of hats yeah that's good i think you got like sam lesson on today i'm like an aspiring sam lesson okay okay that's great that's great how are you on the slope so you're a good skier no i'm not aspiring well then oh then you're not at that stage yeah i'm only aspiring on certain things okay okay 20 other dimensions i'll never catch up he plays a lot of tennis he yaps a lot but yeah Yeah.

3:16:07He runs like three minute miles. Yeah. Okay. I've been talking about running. On the data side of your business, was that an AI bet that you made a long time ago? Did you get lucky? Did you have incredible foresight? I think I had the wrong bet. And so I think on the data business, I think we had believed that as the that these companies that sell data like data as the ingredients not just like analysis on top of the data the ingredient we thought that these companies would become like way more valuable because we thought the number of buyers of data is would go up significantly so if you think like you know people this is 10 years ago people made oh there are uh maybe 50 hedge funds that buy data today significantly there's 11 000 hedge funds the market's going to go from 50 to 500 I think the market went from 50 to like 58 over the last 10 years.

3:17:06So it really didn't go up. Retailers or maybe 20 retailers that bought a lot of data 10 years ago. Out of your, this will be tough to remember specifics, but a lot of like seed stage companies will be like, oh, well, we're going to generate this data and like, we're going to sell it eventually and it's going to be valuable. And it's like, oh, and then you kind of press them on it and it's like, well, like hedge funds will buy it. And then it's like, okay, like, are you confident that those 58 hedge funds or have you seen any of your - They're a terrible market. It's a very small market. And really, most of them are, even like the biggest hedge funds in the world, like Bridgewater, like the amount of data they buy is minuscule.

3:17:47Compared to the AUM, it's like, it's basically zero. And is that just because they spend more on like lunch? Yeah. Are they getting data from other sources, like more public sources? Are they scraping the internet or are they just not using data-driven algorithms? I mean, they have some data. They have data that you could put in an Excel file and stuff like that. And they're not like, you know, smart people who are thinking about quote unquote data, but we actually think of like a data business that's selling data, proprietary data, different types of data. And usually data that's like larger than you could like put an Excel file.

3:18:22Most funds are not doing it. they're really really if you think of like real estate so when we were starting like the number of real estate investors that were using data was zero and today it's blackstone it's like one it's going from zero to one real estate is a pretty big asset class now they're using some data so people are using still some data but really like we're talking about like real alternative data really have a very sophisticated system to be able to ingest it and use it it's very, very, very small. Is that driven by the fact that maybe it's almost like, I mean, we've been thinking a lot about Google IO and VO3 and Google's advantage in video generation, particularly because they own YouTube and YouTube feels like an incredibly valuable data source.

3:19:10It's almost, but the whole like data is the new oil. It's almost like, no data is like a Rembrandt painting that there's only one of. And so Google doesn't want to sell that data. Like Google wants to keep that. And so like, yes, the data market would be bigger if all of these assets actually transacted and were fungible like oil, but they're not, they're super proprietary. And if you have a brand, your data is more valuable. So if you, if you've got something like Reddit and everyone knows Reddit, they understand what people are talking about on Reddit and they say, okay, we're having all these like interesting conversations.

3:19:46they sell that to Google for$60 million or something to train their LLMs. But if you had lots of people having interesting conversations, but you weren't called Reddit, my guess is instead of$60 million, you might get$10 million or$5 million or something. The brand is actually the 10x thing of why people want to buy it as well. Yeah. What about just creating data for the purposes of artificial intelligence, not pulling data from other sources and trying to go find a round peg in a square hole with selling into a hedge fund. More like there is demand for robotics data right now and we are going to go get a bunch of robots and generate a bunch of data.

3:20:27Yeah, I like that. Yeah, I mean, obviously like Scale AI is a great company, right? So they're doing that. They're actually going and building those types of data. We just, we're investing in a company doing what you're saying with the robots. Sure. I think those types of things are good, but it's not always just for the AI. like there's other types of buyers because if you just are selling to like AI core model companies you're you're talking maximum of a dozen buyers that's not a great market um and he went from 58 to you went from 58 to 12. yeah exactly so you just want to sell to open AI anthropic google and facebook like and maybe a couple of others like I don't know that doesn't seem like a great business long term either.

3:21:10Yeah. And if the deal, the first deal that you do is 100 million, what's the deal size on the next one? Correct. Yeah. Are you going to create 100 million dollars worth more data that you can sell to them? Is it even that valuable? Anyways, wanted to have you on specifically to talk about a post that you had yesterday that went pretty viral uh you said that any vc that makes you pay for their lawyers is not founder friendly founders barely affording rent should not be footing legal bills for vcs with private jets and yet this is standard uh you kind of extrapolate on this but you're writing you know uh 20 checks a quarter maybe more uh i'm sure that almost every single one of the companies you back ends up uh having to pay some VC's lawyer bill maybe in the next round.

3:22:01But I would love for you to kind of break down the post and what inspired it. Yeah, so for those people who don't know, in a standard in a term sheet, somewhere in the term sheet, it says usually it says the company must pay up to X dollars of the venture capitalist legal fees. And that X can be negotiated and it can be as low as 10 grand. And I've seen it as high as 300 grand for more complicated deals and stuff like that. And it's there. And it just it seems it seems odd for a whole bunch of reasons. First, it's like there may be many different investors, but you're only paying the lead investors types of things.

3:22:39Second is that, you know, why are they why are they actually having the company pay? It's like a little bit of backdoor way to get your fees paid. um and then there's like there's like a more complicated argument as well which is i don't think this is lp friendly a lot of a lot of vcs will say this is lp friendly um because now i don't have to like yeah wouldn't the person paying presume you you can pass it i mean that the vc could pass the fees on to their lps so correct so they could pass it on to their lps and this is a backdoor way of doing that without actually telling the LPs that you passed it on to the LPs um by taxing the company it's it's taxing the company but but it's a it's it's it's even the LP isn't aware of it whereas if you actually normally if you if you use Tegas or something and you pass it on to your LP your LP sees the bill and they're like why are you spending 50 grand on Tegas like you should be spending less and they can have a conversation or if you're like if you're having air travel they can understand that as well wait wait can you give me more clarity around like the idea of passing fees on so like the real the real fees so so vcs have like two and 20 right but then there's certain fees that you can pass back that it gets it closer to even three you could get it up to like two and a half three yeah four percent and that could even be like when a vc pays for lunch with a founder you can pass that on correct Interesting.

3:24:10Yeah. And every VC has a different model. Different structure. Because I always thought it would be like there's 2 % fees and then you got to do whatever you want with that. If you need a private jet, like that's coming out of the 2%. If you need to pay salaries, if you need to pay for Tegas, if you need to buy lunch, like it's all coming out of the 2%, but you're saying it can be passed on. The true loaded fees can be much higher than 2%. Interesting. Yeah. Like at Flex Capital, we try to almost do everything in the 2%. If we go to lunch with us, we take it out of our own management fee, which means we get less competition because most VCs want to protect that 2%.

3:24:42That goes directly to them, right? So we get less competition, but we think it's more LP friendly. And then there might be certain things like your fund audit might be done. So you have to decide, okay, what's actually done to the fund versus what's done to the management company. So we are more to the management company. Every VC will do it different on how they do it. but some can be like, but at least regardless, at least generally when they're passing fees, at least the LPs do see it. And then they can have a conversation like, hey, why are you spending all this money with lunches to me? Why am I paying for your lunch to go to Spago or something?

3:25:19So whereas with the legal fees, they don't know. The LP has no way of accounting that money. What about like general pressure to bring down the cost of startup legal fees at the Series A level? Justin Kahn was working on a project with Atrium to try and automate the Series A documents. Before you can even get there, I think Ahmad had a good response to your post from Mercury saying that like he just basically tells the lawyers, we're going to cap it at this and you agree up front. And then the lawyers are super motivated. But that's still. that's and that better that's the mod's amazing and that's that's definitely step one like you got to negotiate that cap maybe you can't negotiate all the way down to zero which is what i would like to see happen more but you've got to negotiate it my guess is also it also depends on how the the higher the cap the longer it will be to close your rent so you know if most people know you get shouldn't it be the other way around no the higher if it's a high cap the the lawyers are just going to be like yeah this is gonna take six weeks because it should be it should be i mean so i'll pay you 300k to close it this weekend or 50k and yeah you can take a month i can basically tell you like here's how you close your round it's gonna be 15 days yep plus one day for every two thousand dollars that's in that i can guarantee you that that will that will that will be exactly what it is for every single round that is in there so backwards shouldn't one to one defensive of startups eating the cost is that, you know, if you have an emerging manager with a$20 million fund and they're barely, you know, paying themselves a dollar in salary and if they want to lead around that ends up getting priced, it could get expensive.

3:27:13But your argument is still like, okay, well, LPs are still eating that cost. Yeah, I mean, look, it can be negotiated and maybe in some cases it could make sense. that's in there. But why are you doing something so complicated? I mean, YC created the safe. The safe doesn't have any legal costs, right? Really. You can actually use a pretty boilerplate thing. Most of these terms, you should be negotiating in the term sheet anyway. So going from term sheet to close, which is where all the legal costs come in from term sheet to close. Why is that expensive anyway? It's because probably you're not managing the lawyers.

3:27:52The lawyers have no incentive to keep the cost low because they know what the cap is. If the cap says 50K, I guarantee you, you're getting a bill for no less than 49K. Like in the history of the world, I doubt it is almost ever come in less than 49K for a cap of 50K. Like how do they automatically always hit the cap? It's perfect. They're filling the space. They're built different. the other thing as an lp what you don't want to pay the fees and not get the ownership is the other thing right it's like if you're just getting past the costs and you're like eating that but you're not getting like the the incremental kind of dollars deployed into the company yeah like the incremental ownership is it does it i mean i i guess like where i'd want to go with this is what do you have conviction or any excitement around tools that can be like AI tools specifically that can be introduced into the process like basically that term sheet to close period to just make that more efficient where we could say okay in most cases but first of all we just said we're just involved in a company that did term sheet to close in in six days like it's not that hard, especially in the early stages of a company.

3:29:12Like how much diligence do you need to do? You need to, and look, the company did a good job. They got all their stuff in a data room. They did all the things that they were supposed to do. Then as the venture capitalists, okay, well, you've got to, you've got to somehow read it, read through it. Maybe you can use AI to read through some of it. You got to go check it. You have to make sure that the bank account is legit and all these other types of things that isn't fraudulent, but it shouldn't take that much time. The main reason it takes a really long time is just the back and forth. You put a red line in, I send it to you, then you don't have time to look at it that day.

3:29:46It takes you three days, then you send it to me. Then I say, if we actually all got in a room, that's what they did. So they got all in a room on day six. And they basically put like a four hour block. And they said, we're going to just go through every single issue right here. And they're able to iterate on it just live. And then you would put it on mute. You would talk to your lawyer to go back. It just got done. it's so much easier. And obviously even M &A agreements, like you should be able to get those done so much faster. And actually those often do get done faster. Like sometimes you can do an entire M &A from term sheet to close faster than you can do a VC financing.

3:30:24Many M &As go faster than VC. Because everyone's just motivated to really work on it and put in the time to make it happen. Yeah. Well, thank you so much for stopping by. Come back on again soon. Sorry we only had 15 minutes. I love it. Invite me anytime. Yeah, we'll make it happen. Yeah. Great. Hang in. Talk to you soon. Cheers. Bye. Yeah, we have some breaking news. The first nuclear executive order has dropped. I'm going to read here from one call out that people are celebrating. To maximize the speed and scale of new nuclear capacity, the Department of Energy shall prioritize work with the nuclear energy industry to facilitate five gigawatts of power up rates to existing nuclear reactors and 10 new large reactors with complete designs under construction by 2030.

3:31:11This is exactly what we were asking for with let's copy paste Diablo Canyon. We have Doug Bernauer from Radiant Nuclear in the studio. We're going to bring him in and get his breakdown on it because he's obviously been working in this industry for years and should be able to give us an update on what's happening. Thank you so much for joining the show. Doug, been too long since I saw you. I think I saw you in DC last, but should be an exciting day. But can you give us your read and analysis on what's happening? Yeah, it's unbelievably exciting. There's a whole lot of text. I had these fact sheets that came out, I think went to the media on all these executive orders.

3:31:50It's unbelievably exciting, but it's all coming in like right now. I think the actual official documents just got posted only minutes ago. So I haven't read. So you're an expert. You're an expert is what you're saying. Well, we really put you on the spot. But I mean, can you talk to me about a little bit of the history? I mean, first introduce yourself, what the company does, and then maybe we can build up to some of the maybe problems or areas for improvement that the industry has been asking for. And then we'll kind of dig into what's happening today. Yeah, thanks. So I'm the CEO of Radiance, a nuclear portable microreactor company.

3:32:29We're focused on mass production of one megawatt reactors. We'd like to see 50 of those per year. And they'd come off the factory line and then go out to the customer wherever they are, run for five years, and then come back and be refueled. So it's a very unique sort of model. And my background before that, I worked at SpaceX for 12 years. I worked on the first rocket with legs, amongst other really cool stuff for Elon, but eventually focused on nuclear. And it's a really unique model that we've been trying to work. You know, a normal reactor is about 1 ,000 megawatts. The thing we're working on is one megawatt.

3:32:59It's got to be road transportable. We're trying to make the nuclear power that people want, because the problem that we see with why we don't have all this nuclear technology today is really that there's not as many customers. But that is all completely turned around, I don't know, in the past maybe just two years, right, with all these announcements of the big data center players coming in to buy nuclear, to restart old plants, to build new ones. and we've actually been fighting for a lot of regulatory reforms that that we see as probably being necessary for our particular model yeah where you've got a really a factory that makes reactors where you have assembly and you don't have any construction and if you read the rules that are 50 years old you get you hear about construction a whole lot yeah and so it's not really focused on or talking about the right sort of thing so yeah um when we last spoke you were working on the helium loop, going to test at a dome in Idaho.

3:33:51How are things going? What have been the most recent updates for Radiant? Yeah, there's some unbelievable updates. Just a month ago, we were announced as one of just five companies who will get access to fuel from the Department of Energy, something we've been working on for a very long time. It's unbelievably exciting. When I started the company back in 2020, we said we would do a critical test with a reactor at full scale in 2026 we are on target to do that and really i mean because of that announcement and the support that we're getting from federal government um another change just recently you know we've been working through a process they have the national labs uh we're working with idaho national laboratory um and the fun context that are here they've tested actually 52 reactors in their history although the last of which was in 1977 so it's been a while and we have a kind of old process and they are aware of it anybody was there anybody that was a part of that last test that's still there or is it all just kind of lore at this point it is sort of lore it is sort of lore for the new reactors but there definitely are staff who are there you know there are uh the advanced test reactor the atr reactors where they test all the navy fuel that's been operational for a very long time you can go tour it today still operating so there are definitely uh staff there who are very familiar with reactor operations and refueling.

3:35:10They also have a transient pulse test reactor, which is pretty amazing. It actually has an original core from the 1950s still running and operating. I forget what the numbers are, but it pulls us into the several gigawatts range. So there's some expertise there, but certainly there are not people there from the early 1970s anymore. Okay, I have the Wall Street Journal article pulled up. I want to read some of the key points and just get your reaction. So the high level is that Trump signed executive orders to boost nuclear power industry, citing the need to overhaul regulations and fast track licenses.

3:35:44A lot of this is driven by tech companies like Amazon and Google who are driving demand for nuclear power to support AI systems and data centers. And there's an interesting wrinkle there where a lot of the hyperscalers made very aggressive ESG targets and climate pledges before they knew that they were going to need 50 gigawatts of capacity. and so now they're really, really pushing for clean energy. The executive order aims to shorten approval times for new reactors, but critics worry about the impact of safety. Now, I want to talk to you about that balance between speed and time to market and safety.

3:36:19I wouldn't trust myself to design a nuclear reactor and get it approved in a day, but it feels like we haven't had a new reactor design approved in way too long, and so we're probably on the too slow side of the curve, but where should we be? What is a reasonable amount of time to actually approve a new reactor design in your opinion? Yeah. I mean, probably on the shortest scale, it's gotta be something like six months. Like if I use that context I said earlier, 52 reactors in up till just 1977, nuclear technology only existing since the mid fifties. That means they were doing more than two per year.

3:36:55Yeah. Yeah. So let's get back to that standard, right? Um, walk us through what, But just as a citizen, not as an entrepreneur or founder or executive, as a citizen, what type of tests do you actually think are the most valuable? Is this, I need a physicist to run all the numbers in a spreadsheet and run simulations? Do we need to just fire this up in a place where even if it explodes, we've seen that it didn't explode. And so we've done real word testing. Is it a balance of those? what kind of testing do you think is best that can be executed at speed? Yeah, well, it's a great question. The best is immediate testing and incremental testing.

3:37:37And there's no reason that you necessarily need to go slower. You can regulate along with the speed with which you can do the design, the analysis, the procurement, the assembly of a reactor. If a company is willing to spend capital to take that financial risk, um you know the the goal is don't put fuel in unless you get approval but there should be really zero barriers up until that point because there's no safety risk yeah um but then when you get to that point of like okay now there is risk we are going to load fuel in you do want someone to go and check your numbers and you want them to dig deeply and thoroughly into what your reactor is and does and what the plan is but the fact of the matter is when you load fuel it's not anywhere near critical you can make a you can look at subcritical multiplication factor that will trigger little sensors that are effectively it's you know something that sounds like a geiger counter we use fancier things today we use vision chambers we use boron line proportional counters uh you can do the analysis to show that you will get many many counts per second even with a very small amount of fuel loaded like let's say you construct one sixth of your core you can already validate models um and yet you are ridiculously far away from being able to go critical interesting They haven't even assembled the whole system, right?

3:38:48Yeah. So the FDA has similar approval timelines for new pharmaceutical drugs. But the reason that it takes a long time to approve a new pharmaceutical drug in America, it makes more sense to me because if I'm going to take a drug, I kind of want to know, hey, is it going to have an effect on me in 30 years? So maybe I want to see a mouse model or a monkey model or a dog model or even human testing over a long period of time. And sure, there's some acceleration there. We don't need to wait a full lifetime to approve a new drug, but there is an element of time. Is there anything that's comparable to that in nuclear where the behavior of a system over a two-year time period is different than a behavior of a system over a two-hour period?

3:39:36Yeah, I think the most similar thing would be the effective neutron fluence inside of the core. So if you hit some parts parts of your system with neutrons, they will generally swell up. Metallic parts swell up and their thermal conductivity will usually go down, their density goes down, but the physical shape of them changes. Sure. So you might have to test that in some special way to know what a 10-year lifespan looks like, right? Yeah. But a lot of these materials have been tested and there are these giant data handbooks that it's very predictable. So there's like the Nuclear Systems Materials Handbook, which is like put together in the 1970s, actually just released finally through this lengthy process where we had a ton of help.

3:40:16We went and recovered this great national asset and it's now available for all reactor developers. That's amazing. Yeah. So they're pretty predictable effects. So there's not, I would say there's really other than the effects of neutron fluence, there's nothing really that challenging. We plan to test right next year in the dome. That is a hermetic structure, totally sealed with its own radionuclide detection in it. And so you can partially fuel a core and you can do it in a hermetically sealed chamber at a facility where it's surrounded by scientists, right? And experimental hot cells and other test facilities that would allow you to see what any problem is, solve any problem.

3:40:55But we are using all well-known standard materials. Got it. Makes sense. uh were you is it uh funny to you that the the news of taiwan decommissioning their reactors hitting the same week that we're saying that uh we need more uh or or is there something about taiwan's as a as an island i i'd read something about there being a bunch of um you know enough earthquake activity in the area that like there was some potential reasoning around that but then And we also had someone else on who said we have solutions for that at this point. It's not a risk that should require shutting it down entirely.

3:41:36I don't know that these things are linked, actually. I think I'm behind on the Taiwan reactor news. Those might have just been around for a long time. Might be coincidental. No, no, I'm not implying that they're linked. I was saying more that it's kind of ironic in the sense that we have one country saying we're shutting these things down meanwhile and they're less far less energy and germany kind of doing the same thing it feels like some kind like some countries are following the american nuclear playbook from two decades ago and maybe they should be adopting the playbook from today and they need well we just announced it today and yeah so hopefully yeah how much like how monumental is this for you were you like betting the company in some way on eventually getting an administration that would like introduce an EO and like, you know, try to bring about policy around this.

3:42:31I think, yeah, in a lot of ways. Absolutely. If you're working on something innovative, there's going to be regulatory gray areas. You have to know about those gray areas and manage the direction they're going in to make sure that the vision that you have is the right, the way that the administration or the regulatory agencies go so that you're not decoupled from reality, but you do have impact on that. And you actually, you know you can't write down rules ahead of time for something that's not invented yet right so you but you do always take some risk in doing that but i think it's actually pretty beautiful you lay out a vision go like imagine you have like a bunch of reactors they're all built identical to one another so they're super reliable you deploy a whole fleet of them they all stream data back continuously we're in the modern world we're in the information age why can't all that data just stream to the regulator why can't it be completely modernized um and we've been talking about that for a while.

3:43:23And I think that has led to now that along with other things people have said have led to these sort of events where the administration goes, okay, we're going to just, we're going to actually go and support this. Yeah. Last question from my side. Can you talk a little bit about how you see the market playing out? Obviously, there's some news about where we might be getting really big nuclear reactors at the one gigawatt scale you're building at the one megawatt scale, is there going to be a 10 gig, a 10 megawatt company, a hundred megawatt company? Is it a smooth curve? They're all the same company and you just, yeah.

3:43:57Or, or maybe they're, maybe they're bundled or broken down or scaled up or scaled down. Like where, like, how will this play out and, and, and who will do what, I guess. Yeah. Well, as a consumer, what you want is for there to be two or more companies at each scale yeah right and uh yes there will be something at the portable micro reactor scale maybe two two companies there at least and maybe that's one megawatt and smaller um if you are willing to put a reactor in the ground somewhere and pull the fuel out on that site and put it in a system you can make a bigger reactor i call it a micro reactor 10 megawatt scale so you want two companies there you want two at the 100 megawatt you want two or more at the gigawatt also um because they're different different areas uh don't need these amounts of power so if you think about and we never think about these places they're remote places in the world like in alaska you have over 150 disconnected micro reactor locations or micro grid sorry locations that could use reactors canada has over 200 of these sort of areas and these are coastal you know look at those land areas on the map they're they're massive they have a need for heat and for power.

3:45:07They're extremely isolated. There isn't an actual grid. So you need a small system, right, on those types of regions. And if you put in a one gigawatt reactor somewhere, you had better have a lot of big power lines to move that power to enough customers to use that, right? So on the two, like, extreme ends of the scale, there's, like, pretty obvious use cases. And then in the middle, it's, like, if you don't want to wait years to upgrade the grid or it would be very expensive, like a million dollar a mile or more up rates to power lines, then you might go, well, hey, we have this region where we're going to use 10 megawatts.

3:45:41We could put a 10 megawatt reactor. That's economical for us, right? Without overcomplicating it or having to add power lines extending across states. Awesome. Well, thank you so much for popping on. We'd love to have you back again and go even deeper. Congratulations on the overnight success. The job's not finished, but I like this for you guys. Yeah, this is great. it's great news yeah it's really exciting i mean one of the coolest things they uh they asked to reform the process that we're using at national laboratories to test reactors that's going to be extremely critical to us staying on our schedule we will achieve going critical next year because of a lot of things happening today that's great news thank you you're the man for joining doug we'll talk to you soon milestone cheers good luck talk to you soon uh i wanted to pull up we're We're going to stay on nuclear for a little bit.

3:46:31We're going long. We want to pull up Scott Nolan, my former colleague at Founders Fund and the CEO of General Matter, a nuclear fuel refinement company, was in the White House. Scott Nolan, CEO of General Matter. We're an American enrichment company trying to bring back the US's lead in producing nuclear fuel. So just like car engines need fuel, nuclear reactors need fuel. Right now, the US is completely dependent on other countries to make the key step of enrichment in this field. and these executive orders are going to pave the way for the U.S. to regain its lead. So we really appreciate it.

3:47:03Would you be doing the AI plans because we have a lot of them going up now or soon going up and they need tremendous electricity? Are you going to be involved in many of these plans? Yes. Scott Nolan, CEO of General... There we go. I'm here for Scott Nolan. ...in the White House. Color temperature, a little bit off on that video, but otherwise, fantastic. Yeah, well, we have Asaya from... valor yeah waiting room we're bringing him in let's bring him in welcome to stream has he been in he's been in the capital yeah i think he was over there today in dc let's bring him in uh and he is building at the 10 megawatt skin and so there everyone's got everyone's playing in the different market gotta play there he is thanks so much for calling in can you give us the breakdown how you doing doing really well thank you so much for having me on guys i'm uh i'm actually just coming back from going live on a a little upstart called bloomberg tv so you know it'll take a while for them to catch up to your stature yeah but yeah it's kind of like a feeder it's like kind of like a feeder system yeah can you go horizontal with your phone yeah yeah sure there you go yeah that's right a little a little feeder network for the technology brothers so yeah great to be in the temple technology today yeah it's good to have you Break it down for us.

3:48:23We read kind of the highlights, but what were the key pieces of the EO in your opinion? Yeah, listen, this is a total rewrite of the regulatory system in the United States for nuclear energy. You know, nuclear grew up in the U.S. with a total dominance of nuclear power in terms of U.S. strategy, right? So we had the only nuclear reactors in the world for a while. We piloted essentially every nuclear reactor architecture that there is. And we wanted to protect that edge and protect that advantage. And so what we did is we built these high regulatory walls around the nuclear environment, hoping that nuclear wouldn't spread around the world.

3:49:02We wanted to protect this advantage. This is a natural instinct. I think it was a good instinct at the time. But the problem is that over time, our competitors or geopolitical rivals built nuclear anyway. And where we are now is that these walls that we built to protect our advantage in the United States actually eroded our advantage. And we ended up building a large wall around an empire of dirt. And so today, this is about bringing those barriers back down, allowing American entrepreneurs to build nuclear energy again. There's a couple of really important things announced here. One is revitalizing the Department of Energy.

3:49:34So the Department of Energy was created to be a nuclear test agency. Not a lot of people realize this, but the DOE, we kind of imagine it to be this broad-based energy agency. And it is, but it was actually created to be a nuclear reactor test agency. And since it was created, it's only done that one time. And so this is about revitalizing that. And actually, President Trump has given a strong commission to the nuclear industry. He wants three test reactors critical on U.S. soil by July 4th, 2026, America's 250th birthday. And just a few minutes ago, we announced with the governor of Utah for the first time that we're partnering with the government to full charge, full steam at hitting that goal and turning on a reactor by July 4th, 2026 at the San Rafael Energy Test Center in Utah.

3:50:22So that's what we announced today. Congratulations. That's amazing news. Good luck. Thank you. Overnight success. So yeah, I mean, it seems like with any nuclear project, regulatory is always the main risk, the main rate limiting factor. With regulatory out of the picture, is this more of an engineering challenge now? Is it a manufacturing challenge? Like, how do you weight the different challenges ahead for you to hit that deadline? That's absolutely right. Listen, you know, nuclear is a technology that we've been exploring for 70 years now. We've done an enormous amount of physics around this.

3:50:59We've done an enormous amount of research. The thing that has stood in the way is one company deciding to be the SpaceX of nuclear, right? Like deciding to own nuclear energy, full verticalization from the design of reactors to construction to manufacturing to deployment to operation. That's what Valor Atomics is attempting to do. There hasn't been a Ford of nuclear energy yet. There hasn't been a Tesla of nuclear energy yet. There hasn't been a SpaceX. Valor Atomics is intending to be that company. And regulation has absolutely been the long pull in that tent. So this executive order is incredibly important.

3:51:30I think this is really going to set the tone for the next 100 years of energy dominance. If we're going to win on AI, on manufacturing, on supply chain, we're going to do it with a ton of cheap energy. Talk about the first product, 10 megawatts. Is that correct? 25 megawatt electric is our commercial model. Why'd you sell them there? Yeah, we've gone out and we've really hit an ambitious milestone. I founded this company about 17 months ago. We were intending to go out and build our first prototype reactor in 18 months. We finished it in 16 months. This is a test reactor sitting in our Los Angeles facility today.

3:52:05We've operated above temperature, above pressure, not with uranium, not splitting atoms yet, but with electrical simulation. And really what we proved is that it's possible to build a nuclear reactor in under 18 months. Now that we've done that, we go and actually build it again in Utah and another one in the Philippines. And we actually load them with uranium for the first time, turn them on. After that, we go and build our commercial model, which is this 25 megawatt unit, and build thousands of them around the world. very cool Jordy anything else sounds pretty simple just draw the rest of the owl boys yeah just draw the rest of the owl yeah we make we make thousands of nuclear reactors we make trillions of dollars I'm very excited it's a good day in energy yeah I mean it's fantastic I I uh you know we were just talking with uh Doug from from Radiant and in many ways like if you started a nuclear company at any point in history rough you were betting on activity like we're seeing today and action and so yeah super validating for for you guys that had to get looked at like you were crazy in a bunch of pitch meetings early uh because it feels like one of those things where like culturally i think in 2020 people started kind of waking up to like hey wait nuclear is pretty cool and probably good and we probably want more reactors not less yeah and then companies that you know have started picking up How important is it for you to vertically integrate into the actual productive use of the energy?

3:53:33It feels like there's so much demand for energy. You could just become a mass manufacturing for nuclear reactors, and that would be enough of a business. When we talked last time, about a year ago, you were talking about generating fuel out of the air with this energy. But it feels like you might just want to let the free market decide what the best and highest use of the energy is. Is there still a plan to do something with the energy that you own and control and decide? Or is it just like, let's just make as many nuclear reactors as possible right now? Yeah, you know, one of the core realizations of Valor is that there are some products that the market actually doesn't know how to buy, right?

3:54:16So I think this is one of the most important realizations that Elon had with SpaceX, which is that if you're going to make a rocket company that makes really good rockets, you also need to fly them. You can't just build a rocket. You have to fly him, too. And the reason is that the customer actually doesn't know how to fly the rocket. A small sat company, a telecom company doesn't know how to fly a Falcon 9. And so it turns out the best way to use a Falcon 9 is SpaceX is to actually own the entire operational cadence. And then you sell products like kilograms to orbit and Internet connectivity.

3:54:46So that's the right way to think about these really, really complex technologies. Nuclear reactors are like this as well. So yeah, when we think about building lots of reactors, we think about building them on these very large energy campuses, and then we sell energy, right? We sell energy, not nuclear reactors. That might mean energy to AI data centers. Absolutely, it means hydrogen, and it means eventually synthetic fuels, right? If we can make a synthetic fuel cheaper than diesel, cheaper than gasoline, we can actually become the energy company of the world, right? Because the majority of the energy today is in a hydrocarbon format.

3:55:19So that's kind of the core realization that we had when starting Valor. Very cool. Well, good luck. Thanks for calling in. Yeah, huge day. Huge day. Congratulations. Huge day. Big day. Thanks, guys. To you as a whole industry. Cheers. What a great show. Whole world tour. Four hours now. I guess we just hit four hours. Anything else we need to cover before we get out of here? I think breaking news, John. What's that? It's the weekend. It's the weekend. And as always, our board has been very explicit about this. leave us five stars on apple podcast or less if you don't feel like it but if you've listened this far you're four hours in i'm guessing we earned that five stars so just do it everybody have a great memorial day weekend we actually won't have a monday show yeah because we're moving into the new studio yep but we'll be back in full force today in the new studio in the temple on tuesday we're excited cheers have a great weekend have a good one bye