Trump-Musk Fallout Recap, Circle IPO Post-Game with CEO | Dylan Patel, Jeremy Allaire, Kat Cole, Dana Settle, Anastasios Angelopoulos, Patrick Blumen, Blake Scholl

28:31The CEO of Cursor went on Ben Thompson and did an interview that was fantastic. And Ben spends half the time just marveling at how young the founder is and trying to kind of relate. Because it's so crazy. Think about it felt like six months ago we were talking and people were saying Sam Altman should be Gen Z. Right? Like where's our sort of Gen Z? I said that. Where's our Gen Z Decacorn CEO? I mean, Alex Wang. But there just aren't as many. And it just did feel like Gen Z was taking a while to find its footing in the private markets. Find their footing, a.k.a. get to a billion-dollar revenue run rate.

29:12Let's go. That's the expectation in this economy. And the other data point, the two other data points, they're already being used by more than half of the Fortune 500, including NVIDIA, Uber, and Adobe. And they say this scale will help us push the frontier of AI coding research. I mean, I want to know what's going on at the other half. Are the other half on Windsurf or are the other half just raw-dogging their IDEs? I mean, they could be using, I mean, there's so many different options. GitHub Copilot. But they better be using some. Somebody, Swix was saying earlier this week that GitHub Copilot is also at a$500 million.

29:50That's crazy. That's a public company. If it was out in the market, it's pretty remarkable. But yeah, Ben Thompson was talking to the Cursor CEO, and they're talking about how he got into software engineering. He's like, oh, well, I was a kid, and I wanted to build video games, specifically mobile games. And Ben Thompson's like, this is a huge mark for how young you are because there's tons of entrepreneurs that got into programming through video games. Oh, I was playing. This was me. I played Counter-Strike. Yeah, I played Counter-Strike. I built a PC, and then I would modify the levels and modify the skins and the spray paints and edit some config files and then eventually scale up from there.

30:31No, Cursor CEO started with mobile games. And he actually built one that was really, really viral and really successful amongst software engineers. So he's had a storied career. He's got to check the early Stripe account. Yeah, killer. What was the activity in their Stripe account when they were in high school? Yeah. That's the investment thesis. I mean, yeah, it would have been a good one. Who was the first investor in Cursor, I wonder? We've got to figure that out. I'll say NIO was early. Oh, yeah, NIO. That's right. But Thrive, Excel, Andreessen Horowitz, and DST are all in this Series C. They raised$900 million.

31:06David Tisch was also super early. David Tisch, let's hear it. Let's give it up for David Tisch. Let's go. Round of applause. I'd love to hear it. He's not going to post about being early as much as the rest, but we're going to talk about it. Absolutely. And, of course, Sam Bankman-Fried, also early into Cursor. I thought it was Windsor, but it's Cursor? No, it's Cursor. Oh, my God. Wow. SPF, what a picker. What a stock picker. I really wonder. So, yeah, FTX had a stake. Had invested, I guess, 200K. Oh, yeah, we heard about this. It had, like, insane valuation. Right? Right. Or insanely low valuation.

31:48But it was kind of rumored and the deal might have changed hands a few times or like the stake might have changed hands. They invested$200 ,000 and they sold the position for$200 ,000. Yeah. I wonder how well those FTX... Which is probably going to get re-litigated or something. That probably would have been like a 5X fund, right? If you just looked at the FTX private markets like startup investments as a VC fund, they probably did pretty well. What was the other one? Oh, Anthropic. Yeah, they had a huge stake in Anthropic. They were also in WorldCoin. Oh, my God. Wow. Wow, Sam should have just raised a fund, not had to deal with any of the dealing with crypto.

32:23Should have just gotten out of crypto, pivoted to crypto to AI. Yeah. Should have done it. Yeah, or told FTX users, do you want to be in part of Sam's slush fund? Yeah, that was rough. Instead of. Anyway, I think the story with him is not over. We'll see as it evolves. Apparently, he did get in a ton of trouble for that podcast he did from jail. Apparently, you can't just podcast from jail. yeah that that was wild you did that anyway um did you see this post says sales tax on autopilot spend less than five minutes per month on sales tax compliance numeral hq the golden retriever gets on numeral that's true uh you want to get a will or yeah i just thought this was hilarious will uh obviously of open ai fame says this was the motivation i needed thanks sfo airport It's just like, billboard, win the AGI race, dominate the leaderboard, win vibes.

33:18He's just standing seriously in front of it. Yes, this is great. Made me laugh. Win the AGI race. Don't care about the budget. Don't care about the pork. AGI does not concern itself with pork. The lion does not concern himself with deficit spending. Absolutely not. Absolutely not. We can skip this semi-analysis post because, of course, Dylan Patel from Semi-Analysis is coming on the show today. And we will ask him about Jensen's new math rules and have him break it down for us. This is a very fun post. But anyway, let's move on to Uber is evaluating stable coins potential for payments, CEO Dara Khosrowshar, he says.

33:55Yeah, I wanted to throw this in here mostly so we'd remember to ask Jeremy about that. Oh, yeah. Like all these different massive companies looking to adopt stables. Yeah, yeah. Apparently, payroll companies might be adopting stable coins for all of these different wire transfers that they do, both to speed up payroll, which is already pretty fast. but there's also a whole bunch of stable coins that, if you put stable coins inside a payroll company, you should be able to earn yield on the, like all the dollars that you're holding. Now, you don't actually hold that many dollars just from running payroll because it comes out of the company's bank account, goes for like one day into the payroll company's account, and then it goes out to the employees, right?

34:47It just gets split up. So there's not actually that much float there. But for tax withholdings, there actually is a ton of float because every paycheck you're taking out$1 ,000 and you're going to send it to the IRS. But you send it to the IRS nine months down the road or six months down the road on average, right? Because you're waiting. You only pay taxes once a year. So then, you know, half the year you're like on average. Sometimes you're holding it for 12 months. Sometimes you're holding it for one month. But on average, that's like six months of cash and it's a lot. And so right now that's held in the counts.

35:19But with stablecoins, you could potentially get more direct exposure to the underlying treasuries. And so there could be some sort of boon for stablecoin or for both stablecoin providers and payroll companies. So B2B SaaS undefeated, another beneficiary of modern technology. Just wait until quantum computing hits. Once quantum computing hits, everything changes in payroll. Payroll companies, especially global payroll companies are, you know, are really at the forefront of quantum. Oh, yeah. This is another thing. So what does the Elon Trump dust-up tell us about Elon's quantum computing timelines?

35:57Because as you probably are familiar, quantum computing can break encryption. So guess what happens when quantum computers are ubiquitous? Oh, there's Epstein files? Everyone's going to have access to them. And so if he really believed it. Quantum stocks should be down tremendously today. Exactly. this is the trade stay out of tesla stay out of stay out of rivian and ford got to look at the second or you got to look at the second and third order effects what does this tell us about quantum timelines probably that they're further out than we think because as soon as they crack those boom who's the what's the first what's the first file you're decrypting the epstein files obviously of course of course of course uh i like i like mike's a lot of this post about that he's like i honestly can't believe that we didn't know that already it just felt like internet lore that like of course trump was in there because there's like pictures but i guess like it's a different thing to be in the files than be in a picture but the picture seems more damning when i got at home yesterday yeah um sarah was like hey how was your day i was like uh did you see the news and she was like no i've just been with the kids like all afternoon just been bouncing around swim lessons music class all that stuff and i was like oh well like elon you know accused trump of like being in the epstein list and she was like wait like i thought like there's a bunch of everybody's the pictures of like them hanging out like having you know yeah there's video there's video of them like a party together and it was just kind of like assumed i thought but you know anyway anything to get the to get the the internet riled up elon's master master poster he owns it anyway Regardless of your view on quantum computing, you can express your view on it on public.com.

37:44Investing for those to take it seriously. They got multi-asset investing, industry-leading yields. They're trusted by millions. Go to public.com. Let's go to Naval. He says, the majority of voters are retired either by age, disability, or make work jobs. Paying for mass retirement requires crushing taxes, mass immigration, or rampant inflation. The only way out of a risky bet is on tech-driven growth. AGI will save us. But yes, the make work jobs is interesting. The AGI-pilled lion doesn't concern himself with deficit spending. Or unemployment. True. I mean, this is one of my questions for most of the truly AGI-pilled people.

38:27I try and formulate it as U.S. unemployment over or under 10 % by 2030. because the U.S. has rarely had higher than 10 % unemployment, but AGI seems like something that could do a lot of jobs. And so even in the Tyler Cowen model where, yes, medicine is resistant to AGI. Yes, farming and land and plumbing and all these different things are maybe resistant to just LLM-based agents. But even with that scenario, you should experience some pretty significant dislocation. And this is what that latest Dwarkesh debate with Shalto was all about. How quickly can white collar work be automated? Some people have very aggressive timelines, some people don't.

39:12But I always try and formulate it in like, let's put it in concrete, something that's measurable 10 % by 2030. And a lot of the AGI-pilled people, when they really unpack it, they start thinking about like, well, wait, we're really good at creating fake jobs. Like we could just have 10 more TSA type programs and all of a sudden people are, you know, still employed. And it's like, yeah, my job is pretty chill. I mostly just like watch YouTube and like oversee an AGI that does my job. But I still get my paycheck that way. You did a speed run to leadership in a census here in college, right? This is some incredible Cougan lore.

39:50Yeah, this was, I mean, the government's fantastic at this. So during the 2008 recession, there was a desire to pump money into the economy. More in the modern way we do that is basically with UBI and just sending checks to everyone. But it wasn't always that way. We used to have to come up with an excuse. Schemes. Schemes. So this scheme was basically counting. Counting. It was a counting. Counting scheme. It wasn't an accounting scheme. It was just a counting. It was a government-backed counting operation. Yes. And so the census in 2000, the economy was ripping. It was the dot-com boom. And so all the census workers were basically volunteers.

40:28My mom actually volunteered, went around, asked people in the neighborhood. It's actually kind of a nice thing to go meet your neighbors and ask them, like, hey, fill out the census form. Like, we'll be able to get more votes for the upcoming election. And it's very nice to use people, and people are happy to do it. But in 2008, the economy had just crashed. 2010 the economy was in the sham in in shambles after the after the recession and so they decided to pay everyone like twenty dollars an hour to do a job that people were previously doing for free and throw like way more resources at it to get like better census data but of course you you you can only spend so much and so I wound up getting like a management job there and getting paid a ton of money to do like basically nothing you were managing hundreds of people quite hundreds but dozens and it was it was actually my 21st birthday and And I remember going out and being like, yeah, guys, I can't really fully take advantage of this and drink that much tonight because I have to go to work tomorrow.

41:24You've got to look after your team. And do an orientation for 40 people that range in age from five years older to me than 50 years older than me. There was truly no one even close to me. It's still in school. But that's the way the government works. I would pay a lot of money to have a video of you up in front of all of them talking. It was a wild time. Wild time. Learned a lot about the government. Anyway, we have Patrick Blumenthal in the studio. Let's do one more ad read. Let's talk about out-of-home advertising. Adquick.com. Out-of-home advertising made easy and measurable. Say goodbye to the headaches of out-of-home advertising.

42:00Only Adquick combines technology, out-of-home expertise, and data to enable efficiency buying, seamless ad buying across the globe. Wrong screen share. Wrong screen share. We can talk about Eight Sleep 2. Get a Pod 5 Ultra. pot five has all this i had a brutal night's sleep it's been chaos i think uh i think the two i think the two one-year-olds took the elon trump breakup really really hard yeah i think the four-year-old as well very very difficult like it clearly like he he can't really articulate if he's a trump guy or an elon guy but uh he didn't let that on but it it was it was a dark moment for him well they're all america guys exactly exactly so it was kind of like in many ways it was an embarrassing time one yeah china one the twins first two words china one yeah yeah yeah they're disappointed dada china one yeah but rough night of sleep i got like a 64 he crushed me i mean i didn't do much better uh if you want to get on the official matchers of tbpn go to eightsleep.com slash tbpn They got a five-year warranty, 30-night risk-free trial, free returns, free shipping.

43:12Let's bring in our first guest, Patrick Blumenthal. I've been on a podcast with him before. We did PirateWires together. Good to have you here in the studio. Welcome to the show. Long overdue. How are you doing? Thank you so much. Break us down. Actually, first, let's kick it off with an intro. What are you doing? Where are you? Are you in America finally? Last time I was on a podcast with you, I think you were halfway across the world. I am in New York City right now for the illustrious decadent, incredible New York City Tech Week Fun fact I created New York Tech Week Co-founder right here I passed it off to Andreessen after the first year but yeah, the very first New York Tech Week was by yours truly One of the biggest still on the domain Yeah, break it down for us What are the best events been?

44:05Is it mostly VCs, entrepreneurs? There are a lot of pitching. How are the ads? Are there good sponsorships? There's a lot of ads. And I think the attendee to event ratio has shifted massively compared to previous years. I only went to two events, by the way. So everything that I say, take it with a grain of salt. I'm actually just in New York. And then it happens to be during Tech Week. There's a lot of events, a lot of sponsors, virtually for every single industry. Like there's a million NatSec events. There's a million. Interesting. At tech events. It's like virtually every industry, every type of person.

44:36What's the reaction? You seem a little tired, Patrick. Were you up a little late last night? I've been up late a lot this week and the previous week and the week before that, to be honest. New York is an exhausting city. It's not quite the same as San Francisco. Do you sleep more when you're in Ukraine or New York? Dude, I actually sleep really well in Ukraine. I find the food to be so much better, too. So you can like somehow smoke a bunch of cigarettes and drink a lot and still sleep really well. It's a true meme. The meme is real. Nice. Interesting. What's the reaction been at these cocktail parties to the Trump-Elon news?

45:11That's a good question. It came up a lot last night. It wasn't actually a Tech Week party, but it was a couple of dissident right events, we might call it that. I think people don't care as much as Twitter, to be honest. No one really had any strong takes. No one's taking sides, I guess. Yeah. There's just not that much to go on. It's still like crazy, but then in and out in a few minutes. But certainly something to track. But let's move on to the reason we wanted to have you on. Can you give us your reaction and breakdown to the Ukrainian operation that I believe happened last weekend? We covered a little bit on the show, but I'd love to get your take and maybe just introduce exactly what happened.

45:57And then we can go into some of the history and implications. Yeah. So at a high level, what happened? Obviously, everyone heard about it. But, you know, specifically, Russia had a very, very big part of its nuclear triad get degraded. So they lost like 33 percent of their kind of air launch cruise missile capacity. It's a bunch of strategic bombers that they manufacture during the Soviet Union, completely irreplaceable. And so one of the kind of big asymmetric advantages of Russia during the war has been that they can just launch a massive amount of stuff from really far away, like way beyond the range of Ukrainian fighters, Ukrainian air defense systems.

46:36They use cruise missiles, which obviously have a propulsion on the munition. So it goes much further than it otherwise would. And they also have glide bombs that they've been using for the last, you know, three years of the conflict, which is basically just taking dumb bombs and making them fly much further than they otherwise would. And so really all Ukraine has been able to do is destroy the platforms. Like there's just no way that they're going to shoot down every single cruise missile. They're not going to shoot down every single head. And so this is essentially the best thing they can do.

47:05It's just like go to their airfields and blow things up when they're still on the ground. You mentioned the nuclear triad. So these bombers that were destroyed were actually capable of carrying and dropping nukes, even though that's not directly relevant in this conflict. Yeah, they're all nuclear capable or a lot of them. Yeah. And so, you know, you have ballistic missile submarines, you have strategic bombers, you have ground launch ICBMs. Yeah. And ideally you want all three because the idea is essentially that you have too many nuclear assets in too many different places that, you know, you want essentially deterrence.

47:39So even if the US were able to first strike with, you know, hypersonic missiles, every single ground silo that the Russian Federation has, they'd still have a bunch of strategic bombers on station, a bunch of submarines on station. So we would still be screwed, basically. It's crazy to think of Russian nuclear submarines like just trolling around the ocean. I feel like they've been like so pushed back, so contained. But of course, that's obviously. Well, that's the point of nuclear submarines. means you can go and just hang out for quite a long time. What do you think the, is this type of attack a one and done?

48:18They operated it on cellular networks, which was a surprise. Do you think that Russia will make changes that make that a lot harder in the future? What's your kind of reaction there? I think the reality is that Russia is a very porous country. Like it's exceptionally easy to get things in and out to their advantage and to their disadvantage Like people don't realize how many Western components are in like every single Russian drone Like they're still heavily, you know using Western companies and components for virtually everything They get in through a wide variety of ways, but to our advantage Obviously the Ukrainians can do the same thing as they've demonstrated in this case I think, you know, the future attacks are going to rhyme.

49:01They're not going to be exactly the same Like what Russia is probably going to do is improve their jamming operations at airfields that Traditionally have not been subjected to Ukrainian drone strikes They're going to change their procedures around daytime operations. So generally one-way attack munitions both on the Russian and Ukrainian side happen at night It's kind of random that like on a Sunday morning, basically You know there were a couple hundred Ukrainian drones hitting all these airfields. It's like quite uncommon and we don't know where the failure was on the Russian side like were the jammers online and the Ukrainians just knew what frequencies the jammers were working on and so they just operated outside of that was the Russian jamming guy who just sits there with the big red button just not at his station at the time because he wasn't expecting it there's kind of a lot of open questions around that and I'm sure there's going to be like procedural changes but when you when When you kind of step back from the specific attack of like big trucks with garden sheds that are containing all these drones near air bases, and you just look at the basic kind of formula there, which is civilian infrastructure near military assets, the attack surface is just really wide.

50:13So like it could be a rail car, it could be a truck, it could be a van, it could be a ship in a port, right? The military facility doesn't have to be an air base, it could be a factory, it could be the Kremlin, could be any logistics depot, could be a base. And so when you multiply all that complexity, basically, across the 11 time zones that Russia has territory, with hundreds of critical facilities, it's kind of unstoppable, to be honest. Yeah, that's interesting. How are you thinking of investing in drones as a category? You've been on the ground in Ukraine, you've seen, I'm sure, a lot of the stuff that they're doing, which is evolving constantly.

50:49I'm sure you've met a lot of the American drone startups. And obviously, you know, there's a huge range of players from Neros to, you know, Anderol to, you know, counter drone startups like ACS. But how are you looking at the category and where do you think it's investable? So I think Ukraine is interesting specifically. And then I'll answer your question. Ukraine's interesting specifically because you have a lot of like consolidation around the same exact ideas. So like I've seen a lot of drone factories spread out around Ukraine. They're all making generally the same stuff within FPVs, at least.

51:28And like you have different components like, oh, this FPV, we're going to use fiber cable for it because this area is jammed. But at a high level, every FPV in Ukraine is like exceptionally attributable. It's like 3D printed, very cheap components from China are now sometimes domestically produced, but like very, very cheap. Those are not really what I think of as venture backable. You also have obviously all the added unique risks of their factory might get blown up by a cruise missile. The founders in many cases are also active duty soldiers. So you have to kind of underwrite the additional risk of them getting killed on the front.

52:01Not to be so morbid, but it's true. In the U.S. ecosystem, more like a traditional startup, obviously, like there are no founders who are active duty in the U.S. There isn't exactly the same set of risks. and they're much more focused on selling to kind of the creme de la creme customer, the DOD. So the capabilities are just much more exquisite. And so there's a lot of stuff to bite off because the U.S., similar to Russia operating in 11 time zones, the U.S. operates in every time zone because we have bases literally all over the world. And so every single climate, every single potential adversary, Iran, China, North Korea, Russia, etc., they demand different solutions in each theater.

52:42They demand different solutions for different operations during different times of the day, different weather, like all these different things. And so the reality is that there's actually a lot to invest in. Specifically with drones and counter-drones stuff, I think the future of like the integrated air defense system, like, you know, we think of as like the Iron Dome, it's going to be much more heterogeneous, right? So like Israel now has like five or six different types of missiles they use. They're testing lasers. They're testing a kind of other couple stuff. But it's like a much more stationary and static type of system.

53:15The integrated air defense system of the future is going to be like three or 400 different things like this is going to jam this one thing, this is going to intercept this one thing, this is just hitting an FPV so it can be a small gun like in the case of ACS. And then other things, you need ballistic missiles to destroy other ballistic missiles. So you basically are going to have this system like a command and control system that senses all these different things and then essentially triages. What is Russia doing on the FPV drone side? We've heard that Ukraine is producing, I think, millions a year now, it seems like really, really mega scale.

53:53Has Russia been able to match that at all? Or do they take a different tact? I don't know where the official numbers are, to be honest, but I would be surprised if Russia was producing less, to be honest. I mean, I think certainly within the same order of magnitude, Obviously, there's very poor fidelity on the Russian side. But one thing that you can just look at, and this isn't just FPV drones, this is just drones broadly. One of the things that you can look at is like how many Shaheds are attacking Ukraine every single night. It's in the hundreds. Like they're literally just lobbing hundreds of Shahed drones at Ukraine every single night of the week.

54:27So I don't know if you guys saw, but they did their quote unquote retaliation last night. It was basically like any other night. They just lobbed a bunch of drones at Ukraine. you know the Russian economy now is like literally entirely a war economy like I don't know what it would look like as you know defense as a percentage of GDP like that's that's one of the big issues with peace at least on the Russian side is what happens to the economy if you no longer need to be producing you know huge volumes of tanks and drones and missiles and all these other things It's so depressing. So many great mathematicians there.

55:02They should just be building hyperscale data centers and racing for AGI. They have the talent, and they're focused on kinetic warfare, fighting the last war. Well, that's probably got to end the war so that we can do joint AGI development with Russia. Exactly, exactly. And the war first. Yeah, or maybe just more brain drain. I wonder if there are more Russian scientists that are leaving because of how crazy the war effort is and if that'll make it easier to recruit, you know, brilliant scientists. There's a lot of people just hiding to mostly on the Ukrainian side because Ukraine obviously has the draft.

55:38But a lot of Russian scientists left in the first few days of the war through Georgia and Armenia. And now they're just like refugees. And then a lot of Ukrainians within AI and tech, obviously a tremendous amount of them have joined the military and they're now developing drones. But there's still a lot of Ukrainians are just like hiding in their houses, basically. And their wives just do all their errands for them. Like they're not allowed to leave the house because they're going to get conscripted. So their wife gets groceries and everything. Can you talk to me about what is actually involved in visiting Ukraine right now?

56:09Is this something where you need to fly to a different country and take a train in? Is the airspace dangerous? Do you need a visa? What does the State Department think about visiting Ukraine right now, if anyone in the audience is considering going? What's the vibe on the ground? Where do you stay? How do you actually get connected? Can you walk me through your experience visiting Ukraine? Yeah, it just kind of depends on what you want to do. So I don't know how specific I should be about my next trip, but the more east you go, the more difficult things become, obviously, both in terms of getting approvals, but also just like transportation, and then also your risk tolerance would have to go way up.

56:45So like, you know, hypothetically, if I go to Kharkiv, which is very close to the Russian border. It's basically on the Russian border. The risk considerations are just very different, both in terms of what to expect every night from Russian strikes, what to expect in the context of the war. Like there is always some kind of an offensive going on around Kharkiv. And so you do have to be careful. You know, what I would say is that I obviously have a high risk tolerance. What I would say is it's not as dangerous as people think statistically. So like, you know, the likelihood that you're going to get killed even in a drone strike, when like, if you're in, if you're in Kiev, let's say, or Kharkiv, and they're striking it hundreds of times per night, these are big cities, you know, so the likelihood that you actually get hit by anything is pretty low.

57:30In terms of getting it in the country, you have to fly in through Poland, basically. So you go into Poland, and then you drive into Ukraine. And it's a big country. So it's like 14 hours from Poland to Kiev. Yeah, I mean, we saw during the global war on terror, there were plenty of journalists that made it in and out. Obviously, it's extremely dangerous, but it is doable if you're careful enough. Can you talk about the public relations strategy around the Ukraine attack on Russia? I found it interesting that we were getting high fidelity video, really detailed analysis. Yeah, first party video.

58:05Exactly. And the sentiment in the early war was very much like support Ukraine. Then there was recently some pushback about, are we spending too much on this? Is this even our war to be fighting? And that was a kind of political football that went back and forth, both on the left and the right in some ways. This makes, this feels like, it feels like the takeaway is that the Ukrainians are innovative and they're adapting and they're not letting down. And so maybe it's not time to back down in supporting of them. Is that deliberate? where do you think this goes from here in terms of support from America?

58:41I think there's like different groups of people that the Ukrainians want to kind of attract or get attention from these kinds of attacks. And the reason why they're so public about it, including posting footage, is that they're trying to basically convince the rest of the world that they're worth supporting in a very pragmatic way. It's like, forget about the fact that Russia started the war, that Ukrainians are being killed. Any of that's kind of irrelevant. it. The main calculus is, hey, yes, we are, you know, one fifth the population of Russia. We have a much smaller economy. You know, we have manpower issues.

59:12All these things are true. You know, Russia, historically, you don't want to fight a land war against over an extended period of time. We do these operations, the Ukrainians do these operations because it's a way of showing that they've essentially found asymmetric ways to offset it. And that if you send them money, if you send them weapons, if you provide intelligence support, they're going to keep finding new ways to do it. So it's just them basically showing every six months, really, we're going to find a new way to reinvent ourselves and keep Russians on their feet, basically. Very cool. Well, good luck on your next trip.

59:44Stay safe out there and enjoy the rest of New York Tech Week. Appreciate it. I will. Thank you guys so much. Be safe. We'll talk to you soon. Great talking to you. Bye. Next up, we have Jeremy from Circle. I'm so excited for this. I'm still learning about quiet periods and when folks can talk, but we nailed this one. We're so excited to have Jeremy on the show. Welcome to TBPN. How are you doing? How are you feeling? Congratulations. Great to have you. Congratulations. Thank you. Thank you. I'm feeling good. It's been a really intense couple of weeks and obviously a really exhilarating last couple days.

1:00:23You got any sleep last night? I did. I did. I managed to... My family was able to join me for the ceremonies here in New York and got dinner with my family and got some rest. We talked to Brian Armstrong about the initial rollout of the Circle partnership years ago. This has been a classic overnight success, very clearly. Probably an overnight success. Yes, yes. Overnight success. We love an overnight success on this show. But can you give us a little bit of the history and the story you told about the company when you were taking Circle Public? Yeah, absolutely. I mean, look, you know, my background had, you know, prior to this, been building internet software platforms and infrastructure, from like programming language, developer and software from like programming language.

1:01:26I got deeply into the technology behind crypto in 2013, really formulated this idea that it would become possible to build a protocol for dollars on the internet, where you could issue what I thought of as a full reserve dollar digital currency that could be made accessible in a protocol like HTTP. So you could have a protocol for dollars on the internet. And I believe that we could do that on these blockchain networks, which were super nascent at the time and there's no legal framework for any of this and it was like you know totally crazy but we believe we could do it and you know I think the belief was that over time these networks would become scalable the infrastructure would work we could come up with laws and regulations and that once we achieve that that basically the storage and transmission of value would become like a commodity free service on the internet and money velocity would explode and then in particular I was really excited at the founding of this idea of programmable money and smart contracts which it was like ideas on napkins at the time and that ultimately not only would you just sort of have this explosive way to and low-cost way to move value around but actually the utility of money could increase dramatically and that you could actually rebuild the financial system into a internet financial system which is how we founded the company at Circle Internet Financial originally.

1:02:49And so we've just been pursuing that. And now it's actually happening. The infrastructure works. There's laws and regulations. We have these full reserve digital dollars that have central banks and prudential regulators looking at them. And the technology is sort of starting to kind of go into the background. It's no longer crypto. It's stable coins and it's digital currency. And so, and we've, you know, we've built, I think right now, a really interesting business to bring that. And so that's sort of the origin of it. And so kind of it's been, you know, it's been a long road for sure. Congratulations.

1:03:24I have like 12 more questions. I know we don't have a ton of time. You know, I want to get into, you know, some of kind of two angles, how you're thinking about partnerships on a go forward basis with institutions and partnerships with, you know, other large, you know, public companies. It seems we saw Uber in the last 24 hours in the headline talking about, you know, exploring stable coins. We were talking about payroll companies potentially. 100%. And I feel like this gives every, you know, my last company was a fintech company. We enabled stable coin investing with USDC. And so I've been a big believer for a long time.

1:04:02But, you know, you guys being public now, the market reaction to it, it feels like it gives every single institution in the world permission to say, hey, let's adopt this technology in a meaningful way. And so it truly feels like you guys are on day, this is like day zero. Yeah. I mean, look, I think this is like a mainstream moment for the adoption of this technology. We're classically crossing the chasm, right? From early adopters to like mainstream acceptance, the technology usability, the laws and regulations. And, you know, having a publicly traded company like this, I think it really does matter to trust, transparency, compliance, governance, things that really matter when other major companies want to figure out who to build with.

1:04:40But at the core of our business, our whole model is, we're a market neutral infrastructure company. We're like a platform that people can build on top of. We have banks and neobanks and payments companies and fintechs and exchanges, custodians, every type of institution we work with already. And they face us as a market neutral platform. And that's really, really important. We're not competing for end users. We're not competing for businesses. We want to let other people build on top of both the kind of public utility that we put out on the internet, like our public protocols, and like the kind of market infrastructure of the money issuance side of it.

1:05:18And so it's now, it's definitely getting to a point where, yeah, I think major public companies, commerce firms, internet firms, financial institutions of all types, really can, you know, build on this now. It works. It's liquid. You can connect to it around the world. And we have the ability through innovations like CPN, a recent product that we launched, to abstract away from an end user perspective, a lot of the things that are, you know, tend to be barriers to entry and make it work compliantly as well. So it is a really, I think, significant moment. And I do think that, you know, lots of mainstream companies can now really feel like they can do this.

1:06:02And there's a great platform company to work with. I'm wondering about other real world assets on change programmable money. A lot of the conversation around I want something that's low cost, high speed, programmable and stable supply. That makes me think gold, stable gold coins, basically. I'm sure you've gotten this question before. But why didn't the market evolve that way? or even that original like 1999 Peter Thiel talk about, you're gonna be able to transfer$1 worth of S &P 500 to someone else. And there seems to be some demand for that type of stuff, but the market didn't just break that way.

1:06:39And so what's actually going on? Well, I mean, look, I mean, the dollar itself has extraordinary network effects. 65 % of trade is settled there. Sure. And frankly, like a huge driver in our business right now is demand for digital dollars around the world. people, households, firms want to hold these safe, full-reserve digital dollar instruments that have the utility of the internet. It's like when people woke up and said, hey, I can use WhatsApp and I can communicate with others. I don't have to pay my carrier, these SMS fees, and it's over-the-top internet money. And so people have figured that out, and people want dollars, even though there's a Sell America trade or whatever you want to talk about, or there's a fiscal cliff or whatever it is, like all these things that are there, people still want a unit of account and they want the power of the internet as a medium of exchange.

1:07:26And relative to almost every other fiat currency, the dollar still remains super preferred. And so I think that's certainly going to be the case for a while. And so I think that's there. But the programmability piece is the really exciting next step, right? How can you intermediate financial arrangements, commercial arrangements, labor arrangements, arrangements between AI agents, like all of these things that can be machine mediated with provable code, you know, pamper resistant provable code on the internet, which is what smart contracts give us. We're just starting to see the unlocks there.

1:08:02And it's going to be an entrepreneur and developer driven phenomenon for sure. That's amazing. You mentioned AI agents, Ben Thompson was reacting to the MCP standard and the lack of a payments layer in there, at least in this version of the specification. Do we need a V2? Well, you know, it's a great question. We're working with all kinds of teams on agentic payments. And I'll tell you, we worked on a new standard. Actually, it's reusing an old standard. We worked on it with Coinbase and a couple of others, which is X402, which is essentially it was a protocol extension of HTTP, which was originally a protocol extension to support payments native to HTTP.

1:08:47and no one ever did anything with it. And so it is essentially the payment request feature of HTTP. So that's been reincarnated and where Stablecoin is the payload. And so now you actually have built into HTTP an extension, which is a Stablecoin payload. And so it's literally designed for AI agents that need to basically handle these. And it kind of abstracts away what chain you'd use underneath. And so it's pretty cool. and we're pushing it into our dev products, so is Coinbase and others. And so I think actually like that was part of like, you know, early on Marc Andreassen complained. I was about to say.

1:09:26Yeah, I mean, that's the famous thing. Hey, we had this thing there, no one did anything with it. Well, now we have something to put in there. And now we have the demand, which is like machines that talk to each other. And so it's interesting how a lot of these things kind of come together at the same time. Yeah. Do you have a view on that idea of micropayments on the web? It just feels like we have micro transactions that happen across APIs. They just get bundled and reconciled at the end of the month. There's still a ton of value in stable coins. I'm not saying that like, oh, the narrative is bust because that might not be the end game.

1:09:59But we saw this with Bitcoin where people were saying, oh, there's four different ways that this could break privacy, blah, blah, blah. It wound up in having a very strong narrative. It kind of narrowed down the focus, but it's still a wildly successful technology. Well, I mean, look, at a high level, right, we think about the TAM here as legal electronic money. And then narrowing that down a little bit, you have cash and non-interest bearing demand deposits, which globally is like$60 trillion of the money supply. And so if you can have a digital cash instrument that has the superpowers of the internet and the like that can basically move and transact at almost no cost, there's an enormous amount to grow into.

1:10:39Now, the other part of the TAM is all of the different utilities that we have for storing it, moving it around, and those are being reconstituted on chain. And so you start to think about like, okay, you know, cross-border payments is one that we talk about a lot and we're seeing a lot happen there. The kind of point I like to make is like, when's the last time you sent a cross-border email? And so it's like, well, that doesn't make any sense. We're like, we're entering that era with money, right? Where basically like stablecoin money. And that could be a business that's got a$10 ,000 payment to buy something from someone in Asia to get product in Latin America.

1:11:16It could be an investment, right? I want to make a$10 ,000 investment in a startup in an emerging market or in the US. Or it could be, and as we see today, like the biggest electronic trading firms in the world are moving hundreds of millions of dollar transactions over blockchains using USDC to settle bilateral trades, the exact same transaction is being used to buy a 25 % digital object in a Web3 game. And it's the same, right? Just like if I send you an email with a picture of my breakfast and you send me an email with a CAA dossier, it's the same payload. It's whatever it is, 25 byte or a byte of data or whatever.

1:11:56It's the same thing. And so that's the cool thing about this very, very scalable. And so you don't need like streaming payments or micropayments as like a killer use case. You just need to be able to support all these different kinds of value exchange just in an internet native architecture. Can you talk a little bit about the geopolitical environments internationally that are most receptive to stablecoin adoption? You could see demand for crypto products track or correlate with authoritarianism, But then you have the pushback of a lot of authoritarian countries might want to ban or restrict or make something illegal.

1:12:34So so will we see more count almost counterintuitively or like as a second order effect? Will we see more stable coin adoption adoption in more democratic countries first? Yeah. Well, what's interesting is the demand for stable coin money is very global. And we do see it, for example, in countries where there's weak regimes or the banking system isn't as adequate or there's not as effective access to dollar banking. Sure. And so that's a valuable thing for the people and businesses in those markets. But it is sort of this, again, this over-the-top phenomenon. Like, you know, WhatsApp came over the top and, like, all of a sudden everyone was a Facebook customer.

1:13:14And then you're dealing with Facebook for whatever issues in your country. or the Arab Springs were an example of like, hey, everybody could share media with each other and like, whoa, we just lost control of things. And so I do think there are geopolitical dimensions to this. I think we're in the very early - But then you guys have partnerships in Japan that allow you to launch USDC there, which is more direct. We do. And that's our philosophy, right, is if a government is going to have rules around stable coins in their society, we want to make sure that those rules are fair, obviously. And we want stablecoins that are allowed to be used to be held to really high standards, right?

1:13:54High standards of trust and transparency and compliance and governance and things like that. So that's important. And what we're now seeing is in markets where stablecoins have exploded, Korea, Brazil, Turkey, South Africa, Kenya, like all of those governments are working on stablecoin rules right now. And, you know, so I think it's a space where that's intersecting. And so, yeah, we've dealt with this in the history of the Internet, right? It was when the Internet started, like you needed a broadcast license to have a radio show in Italy, right? It was like or there were national monopolies that were the only entities that could actually do radio shows in Italy.

1:14:32Well, now here we are. Right. So I think there are these kind of transformations and society ultimately votes on what they want to adopt. But it is still a political economic negotiation. And monetary sovereignty is not something that the Internet has really faced off against in a big way. But people talk about Bitcoin as doing that, like as this non-sovereign store value. But then when you have kind of exportable digital fiat, then it does become a slightly different issue. Can you talk quickly about the competitive dynamics on a go-forward basis? The IPO was massively successful. It seems like they're, right now, from my point of view, infinite excitement for Circle and USTC.

1:15:19That's obviously going to inspire a lot of different groups, whether it's early-stage startups to other institutions to compete. And I can imagine myself why you guys will continue to dominate, but I'd be curious to hear from you. I mean, look, I think the way we think about our business is that we've built a stablecoin network. And it's an internet scale platform and network utility. And we take very much like an internet platform mentality. We want to build open public infrastructure. We want developers to build on it. And developers build applications that integrate to it, which adds utility, which then expands interoperability.

1:15:57And so you create these really great classic developer flywheel effects. And that's what we've done. And we've done that with liquidity as well. Like it's got to be available, the digital dollars, et cetera. So this is a, you know, it's a business that has network effects. And those are, I think, strong network effects. And then it's a business that there are these very significant hurdles we've had to work on for 10 years to become licensed, become regulated, to become supervised, to become trusted, to integrate into these different government regimes and the like. And that's a lot of work. And it's definitely work other people can do.

1:16:32So I have no doubt that given the TAMs that we're talking about, that there are going to be a lot of great companies. And what I've sort of said is I think this is a winner-take-most, not a winner-take-all market. And you sort of see that in the market structure a little bit today. And I expect that market structure to persist. And I think three to five years from now, within that set, that market structure, there's probably a couple of players that don't exist yet but will exist and will be double-digit players. So it's just hard to know exactly who that's going to be. But I think at the end of the day, we want to build things that are useful for end users, for developers, for businesses, and provide that and keep innovating.

1:17:13We're a product-led company, a technology-led company, so we want to just keep doing that. And as long as we do that, I think we can continue to grow and thrive. Yeah, you guys are in an incredible position personally. John bought a car this week. and were the dealer able to accept stables, it would probably be getting delivered today. But instead, it'll be Monday. I've actually paid people internationally with stable coins years ago. It was a fantastic experience. It was in Argentina. I have one last question. Once you've had stables, you'll never go back, right? It's like, oh, wait a minute.

1:17:46I can't believe this. Yeah, I have one last question, and then we'll let you go. Sorry for going too long. But I want to understand the shape of your business. I'll give you some examples. There's the Silicon Valley muscle. There's the Washington, D.C. muscle. And there's the Wall Street muscle. And when I think about the Bitcoin project originally, it was almost entirely technology, no regulation, no financial work. The grayscale Bitcoin ETF was very little technology, almost all lobbying and regulation and Wall Street financial. Are you a perfect blend of all three? Is there some focus? Has it shifted over time?

1:18:23Are you out of the woods on the DC stuff and now it's just finance from here on out? Well, I mean, look, I mean, you know, the best way to look at a company is to look at its OPEX and within its OPEX, you know, the largest piece of it is labor. And so when you look at the labor OPEX of Circle, like approximately 45 % is product and engineering. So we are very technology driven and that is leadership from amazing, you know, engineering and product leaders mostly out of silicon valley mostly based in silicon valley and um and so that's that's by far the biggest because we're operating this infrastructure and platforms and software and the like and then the second biggest which is you know maybe 20 is what i'll broadly call like our control functions and so that's like risk governance compliance you know uh you know financial controls because we're like this digital currency you know market infrastructure right so So the control piece is really, really key.

1:19:20And then you have like, yes, we have a legal policy and regulatory team that is much larger than a tech company and is expanding because we're now needing to work with and collaborate with governments all over the world. Yeah. So that part doesn't slow down. I think as this scales, the work that we need to do with governments only increases. the control infrastructure, the product engineering should kind of chug along. But, yeah, that's kind of it. I think about the compounding advantage of, yeah, you can compete with us. Just hire a team that can interact with every government on Earth. It's perfectly regulated and perfectly tested.

1:20:04You know what, though? I say this, which is like, you know, I've done the yeoman's work of legal bills for like a decade. and so like you know everyone gets to ride on top of all the legal and yes Satoshi had it so easy she had no legal bills well thank you so much for coming on I'm so excited for you and the team yeah classic tenure overnight success love to see it thank you have a great rest of your day we'll talk to you later awesome cheers Jeremy come back on soon we have Anastasios from LM arena we're going to get to the bottom of what the best LM is what the best LLM is welcome to the stream.

1:20:41Anastasios, how are you doing? Pretty good. How are you? I'm great. Congratulations. You have news. Can you first introduce the company, the news, and everything that's going on in your world? Sure. So, Ella Marina is an open platform for evaluating AI via human preference. What it is, is you can go to the website, ellamarina.ai, and you can type in your prompt, Often what you'll see is responses from two anonymous LLMs. And the LLMs are some of the best. Gemini from Google and ChatGPT from OpenAI and Cloud from Anthropic, and they're sort of randomly sampled from a pool. And not only do you see the responses, but you can also battle them against one another to see which ones sort of win for your vote.

1:21:29And then if you choose, you can vote for the one that won, and then we use that data in order to construct leaderboards. So we have a leaderboard of the performance overall, but also in subcategories like math, coding, instruction following, creative writing, and so on. And we sort of see it as like the people's voice on AI progress, right? Because you can go there, you can express your opinion. A lot of the tasks AI is doing is subjective. And by doing so, you can sort of guide the field. Yeah. So you raised money recently, is that correct? Yeah, we raised$100 million. Wow. Congratulations. Yeah, thank you so much.

1:22:00It just speaks to the strength of our community. and our duty to continue to improve the product for them. Yeah, so what is the business model? Yeah, what is the business model? Because I think people might not be so familiar. There's a lot of benchmarks out there, but you're actually building a huge business around this. Absolutely. The business model is to help, you know, one of the biggest problems that we see with AI-driven software today is people just don't know how to build it reliably. They don't know how to pick the right model to use for them. They don't know how to build like an agentic system with multiple subcomponents in such a way that they're gonna be able to ensemble it together into a piece of software that's like reliable and performant and It's hard even to define what these words mean because a lot of the job of AI is to interact with humans And so when you're interacting with humans, you know a lot of the responses people give are subjective So how do you make sure that you are using your models in a way that like ends up satisfying your users?

1:22:55Ends up making your product better and better you know, we're hoping to be able to address that problem because we have such a big community. Millions of people come to the site and giving their preferences, and we have a whole competition landscape of all the different models against each other on all sorts of diverse tasks, not just in tech, but also in, you know, medicine and real estate and school and so on. Yeah. So talk to me about the death of traditional benchmarking and the rise of big model smell. Have you been able to quantify the unquantifiable yet? Sure. So I wouldn't really describe it as the death of traditional benchmarking, I would just say that benchmarking is entering a new age where there are different types of benchmarks that are needed.

1:23:35So what does a traditional benchmark do, just zooming out? What you do is you say, hey, here's like a particular vertical that I want to evaluate. Let's say I want to evaluate image classification. Well, then I'm going to collect a bunch of images, I'm going to classify them, and then I'm going to basically give models a test. I'm going to grade them based on how well they do on like a held-out set of classified images. So what is the problem with this in the current day and age? Well, these days the way people are using AI is like so broad that you could never like annotate all of it with data sets.

1:24:07Like you could collect a data set, it's all it's gonna evaluate that thing but not everything. And you could collect 20 and that's gonna you know evaluate 20 things but really there's like a million different things that you should be evaluating. And you know arguably even every individual has their own point of view on like you know all these subjective tasks. And so that's why things like LMRNA come in as sort of a different orthogonal signal. Yeah, maybe it's still useful to say how well is my model doing on image classification or multiple choice question answering, but maybe I also want to know, you know, whether people like this model, why they like this model, whether they're using it in the ways that I intend it to be used, which, in which parts of the space is it good and bad?

1:24:45And our perspective is, those are the questions that we can answer by gathering a massive data set of human preferences and then going backward, basically inverting the problem and saying, let's mine that data for all the analytics that we can get so that we can tell you which model is best for you and for your use case. And that's where I think the power is in this sort of new age of benchmarks. What about humor? Can we expect any type of benchmarking from LMArena around just making the user laugh? That's a great idea. We should have a leaderboard for that. I think that that's a potential trap that humanity can fall into is LLMs get so good that you just refresh the button and laugh over and over and over and over.

1:25:28And then all productivity is lost and it's just the laugh button. If you're making people happy, making people laugh, that's not a bad thing. Sometimes I also scroll YouTube looking at my favorite comedians. And I think they provide me value. Totally. Yeah, I have a post here I want your reaction to. I said, now that AI can beat every intelligence test, we need new evals for the other human traits. Specifically, I'd like to quantify courage, fortitude, gallantry, mirth, equanimity, yearning, stout-heartedness, gravitas, panache. This would help me pick what LLM to use. How close are we to being able to quantify courage in an LLM?

1:26:11I love that so much. We need your vision. Yeah, I don't know how close we are to courage. Maybe we need to up the stakes. We don't have to compete in the real arena where they have some risk. We have some chips on the table, you know? Yeah, yeah. Our intern said, you could probably do this using control vectors. And I said, I don't want to put the courage mask over the Shogoth. I want to understand the base level of courageousness for the Shogoth as a whole. I love it. So we went back and forth on that. We're getting biblical here. I love it. Yeah, it's great. Where are you guys going to be? You obviously raised$100 million.

1:26:45Where are you going to be investing? What are kind of some things coming down the pipeline that you're excited about? Well, the majority of our effort these days is going towards taking the platform that we already have and making it a lot better for our community. Basically, inviting more people in. Because as you can imagine, how does our platform work? Like, people come to the platform. At this point, millions, but we're hoping tens of millions or more of people will come to the platform and vote for the ads that they like best. And then we take those votes and turn them into a number. Right?

1:27:11So what's the most important ingredient here is getting people into the platform and loving the platform to play with all these different AIs and like ranking them and toying with them and so on and so forth. And we want to build great product features for them so that they continue to love it and come. So just investing in Elamarina.ai, that's where we're really targeting a lot of our energy. We have Dylan Patel coming on next. He recently dropped the AI Mandative Heaven tier list. He's got OpenAI up at the top, followed by Anthropic, DeepSeek, Google, XAI down at B tier. Meta and Apple are struggling a little bit lower down in D tier and L tier.

1:27:49Does that track with you? What have you been most excited about in the recent developments out of the foundation model labs? What do you think is underrated right now amongst the various foundation models? Well, I think in general, the thing that excites me most is that there's a lot of competition these days. And there's also a lot of new modalities that are coming out. So like, you know, all the time we're seeing that the models are improving, improving, improving, and the pace is not slowing down. These things are getting way, way better. And there's a lot of the most brilliant people in the world that are just here, like in the crucible, developing like the best AI models.

1:28:31And then we're going beyond text, It's going from text to people are now doing audio, and people are doing video. And I don't know, there was a big video model release last week with VO3. That was huge. Are you already doing tests against the different video models? Because right now, I mean, I put the same prompt into Sora and VO3, and it wasn't even close at this point. Obviously, the next iteration will be maybe more competitive. But it feels like we're kind of too early to do those types of evaluations. where people would just immediately identify a VO3 video, but then there are some other labs out there, like Runway, that are doing cool stuff in video, and maybe it's harder to tell than people think.

1:29:10So we have a sort of initial cut of a video arena that we're gonna continue improving that was built by some awesome graduate students at Berkeley, like, yeah. So we have that, and we're gonna continue building it. But I think one benefit of some of these strategies, like arenas, is that they can actually provide kind of a touchstone for the field to come together and say, hey, here's what we're going to try to improve. Let's go. And then maybe it would motivate some labs to actually continue to build better and better video models because they want to be winning the game.

1:29:49What has been your take on the Lama journey? There was some accusations of benchmark hacking. A lot of people have been saying, well, it's like never bet against Zuck. He has a capital cannon and a huge incentive to stay in the game. And so we could be seeing some exciting things out of meta in the coming weeks or even coming months. I mean, all this stuff moves so fast. The leaderboards are constantly shifting. What's your reaction been there? So I would just say that we should all be gunning for meta. We should all be rooting for them. because they're releasing open models. And so I think open models are really great for the ecosystem.

1:30:33So I think I and everyone else should want them to succeed. Now, you know, there was a little bit of weirdness in the last release. I think it's okay. You know, they have a big group of really talented people. They're going to move on. And, you know, the models they released, they're good. They're getting better. And I hope the trend continues for them. Yeah. How has the AI safety debate shifted recently? It feels like Meta almost could have had a little bit of an out in years past when a model wasn't performing up to standard. You could never really tell if the foundation labs were really worried about safety or they just didn't want to release the product yet.

1:31:11It was an easy kind of out. We didn't see them take that this time. We didn't see them say, hey, we're delaying because of safety concerns. Has that kind of melted away from the discourse in your community? or do you think that there's a world where what you're building could be used to evaluate safety? Because there's so many different dimensions of safety now. I completely agree. And we definitely want to make safety a priority in evaluations. We have a red team arena now that has been out for a little while. And it's still a bit of a prototype, but we're going to keep working on it. And I think it's really critical.

1:31:46I don't know about the initial question about whether or not that's used as an excuse or not. That's kind of, I don't know, that might be above my pay grade. I don't hear a lot of those internal conversations. But I certainly think that, like, subjective evaluations are going to be important for safety. Jordy, anything else? No, this is great. I'm excited. Yeah, thanks so much for coming on. This is fantastic. Congratulations on the milestone. Great to meet the two of you and keep in touch. Yeah, come back on again soon when you have news. See you guys later. Bye. Cheers. Really quickly while we have our next— I feel like humanity has a duty to—everybody should be required.

1:32:21It should be forced. You got to go on Elam Arena daily, five minutes a day, every person on earth. What do you got, John? We got Bezel. Getbezel.com. Your Bezel concierge is available now to source you any watch on the planet. Seriously, any watch. And our next guest is here, Dylan Patel from Semi Analysis. We'll bring him in. Welcome to the show, Dylan. How are you doing? Boom. What's going on? Dude, this live audience is fantastic. I mean. We actually do have a bit of a live audience. We have four people in the studio today, but we also have a soundboard going. A giant gong. Let's go. If you want to drop a big number.

1:32:57Yeah, if you have any big numbers to throw at, we'll hit this gong. Why don't you just hit it, John? Dylan's been on a generational run. Yeah, yeah. Just hit it for semi-analysis. There we go. You got the gong cam. The gong cam. It's so loud. I'm sorry. What's happening? How are you? Great to meet you. It's fantastic. I'm having a fantastic day. That's awesome. I really enjoyed your China Talk episode, the update to the AI lab tier list. I wanted to, I asked some folks about that. I have some follow-up questions, so maybe that would be a good place to start. OpenAI, I mean, it sounds like they're not backing down from bigger and bigger training runs, but can you give me a little bit more insight into what it means to do a big training run now because they're still pre-training, like the RL stuff is scaling up.

1:33:50Like what is the shape of the next big run if that's even the right term to use in this era? Yeah. So there's efficiency gains on multiple layers, right? So pre-training, you're not simply just going to scale it larger and larger as your only avenue of spending resources, but you are constantly getting efficiency gains, right? Something that lets your model be 10 % more efficient to run or train, but involves changing the algorithm. So people are still doing pre-training constantly, but they're not, you know, OpenAI's Orion or GPT 4.5 was a run so large that they literally can't scale it up any larger until like end of this year when Stargate starts being operational, right?

1:34:32They literally just don't have a cluster that's bigger. So it's primarily them evolving their model architecture, continuing to get efficiency gains. So for example, GPT 4.1 was an efficiency gain over 4.0. And they're going to do another one of these sort of efficiency gains, step up in model size somewhat, but not as far as 4.5. So that's on the pre-training side. But all the eyes are on the reinforcement learning side. That's where they're doing all these crazy things or cool things. Whether it be O3 specifically for just regular reasoning and O4 is coming out soon enough, or it's deep research or it's multi-agent systems, there's tons of work they're doing spending training compute on training larger and larger models in terms of more compute spent on RL, not necessarily model size.

1:35:24Because your model size is baked in at pre-training. Yeah. Yeah. With Stargate coming online, is there really no other way around that? I feel like AWS is really big. There's other big clusters out there. Can they just not get access to those, or do they truly not exist? And will Stargate be first of its kind? I think it's funny that my little autistic obsession became the most political thing in the world. And so you have obviously US, China, now Middle East, there's all these geopolitics here. But of course, like the politics are between the companies too, right? So there's the whole like OpenAI Microsoft breakup, right?

1:36:02There's a reason Stargate is not in Wisconsin, right? Microsoft had a huge data center they're building in Wisconsin. That's where it was going to be. But now Stargate is in, you know, Texas and it's not being done by Microsoft because OpenAI and Microsoft had all these politics, right? A lot of which were really well reported, but a lot of which have been sort of silent. But, you know, and it's like, okay, well, Is Amazon going to give OpenAI Compute? Absolutely not, right? Is Google going to give OpenAI Compute? No way, right? It's like there's companies getting bought by OpenAI. The deal hasn't even closed and they're cutting off access, right?

1:36:39Sort of with the whole windsurf thing. So it's like, you know, it's like these things are so political and little nerdy autistic boys don't know what to do with politics. Yeah. How much of that dynamic is driven by just like true AGI pilling at the top level of the hyperscalers versus just this is potentially a multi-trillion dollar. Yeah. This is potentially multi-trillion dollar outcome. And like we need to have a player. We need to cut off a competitor. Is there a dynamic there? I mean, I don't think those things are mutually exclusive, right? Whether you're truly AGI-pilled, open-air anthropic type people where you want it for the benefit of humanity or you're like Andy Jassy or like Sacha Nadella or whoever it is, whoever you are at the hyperscalers, at the end of the day, profit is AGI.

1:37:30If you have AGI, you hope you can profit off of it. The people at the labs would be like, well, I don't know if AGI would let us profit off of it. But in the meantime, as we're building towards AGI, that is a trillion dollar, 10 trillion dollar, multi trillion dollar thing that you have under your control. Yeah. Do you think we'll be able to tax the profits of AGI even if we can't reap them? Because I was looking at the Trump-Elon breakup and I was wondering, what does this say about their AGI timelines? Because if you really... Yeah, the AGI-pilled lion does not concern himself with deficits.

1:38:02Yeah, if you believe AGI 2027 or AI 2027, the budget deficit doesn't matter, right? We're going to be printing 20 % GDP growth. We'll be able to pay back trillions of interest in just a couple of years. I think it's quite funny. You know, this Elon and Trump sort of drama, you know who's sitting there like, yes, yes. I know who. Sam Altman is like, yes, let's go. Because he's like, you know, like for a long time, Elon was like, I'm not going to allow OpenAI to become a non-profit or become a for-profit, right? And he's suing them, right? And part of his like, I'm sure part of his calculus for like going all in on Trump and, you know, saying he loves him more than a man can love any other man.

1:38:47Like, you know, four months ago was like him thinking he could, you know, convince Trump and the IRS to not let them convert, right? Now, like, it's like very obvious. Like, what's the impediment? Like they're going to get going to be able to convert. Well, at the same time, Trump was always a Stargate guy. He's a big Stargate guy. He announced the project, right? Trump is a big number guy, right? You get the hundred billion and he's like, hell yes. Hell yeah. Just because this is what it's about, right? Yeah. Before we move on from the open AI, the next big run, how real is the narrative of like they're going to pull ideas from DeepSeek?

1:39:25I think it was like FP8 was one of the innovations. There were a bunch of other innovations. Some of those were specific to the restrictions of chips, but it felt like some of them might be generalizable to just more efficient training runs. Are there real learnings for the American labs that aren't GPU poor or GPU restricted from DeepSeq? So I think there's certain things that DeepSeq did that the rest of the industry had already done in terms of like the OpenAI's, right? FP8 training was one of them, right? OpenAI has been doing FPA training since at least 2023, perhaps earlier. Now, on the flip side, there are certain things that DeepSeq did that were a bit beyond what the labs had done.

1:40:04And so one of those things is how sparse a model is. You have a mixture of experts model. You have so many experts in the model, but you don't activate all of them for every forward pass of the user. And so GPT-4 was 16 experts, two of them active. So that's a one to eight ratio. GPT-4-0 ended up being like a 1 to 32 ratio, but DeepSeq actually went even further. They were 1 to 64. I wouldn't say that that's something that like the labs weren't planning to do, but definitely DeepSeq did leapfrog a little bit, right? Went a little bit further as far as like publicly available models. And I know Google's models aren't as sparse as open AIs, right?

1:40:45And so everyone is headed that direction, but DeepSeq definitely went a little bit further. So I think it's like – I think some of the more interesting stuff that DeepSeq did was around sort of like making it open what the reinforcement learning verifiable reward paradigm is. Making it open how to do really efficient inference systems, right? These are the sort of things that I think DeepSeq did that a lot of people can learn from. And also there are special attention mechanisms. There's some interesting work there. even as far back as mid last year when DeepSeek released a paper, everyone was like, wow, this DeepSeek company has really good research, right?

1:41:20So I think there's things people can learn, but also there's a lot that like the labs had already done. They just weren't like, maybe they didn't go as far or they did it, but it wasn't like they were passing on all the cost savings to the user, right? Because they like to have margins. Yeah. How much of the DeepSeek narrative around like it happened because China banned high frequency trading is real? and should we ban high fructose trading in America if we want to pull AGI timelines forward? This is great. I think some of the most cracked people I know at Anthropic and OpenAI are ex-quats, right?

1:41:58Anthropic's best performance engineer who improves the performance of their, whether it's Tranium or TPU or GPUs, he's from Jane Street, right? They're missionaries. There's still mercenaries over in the high-frequency trading bucket. With a few laws, we could move those mercenaries over. I'm not saying we should do it. I'm just saying I want to know your take. You there? Yeah, I think that the – you know, with the mercenaries, right, it's like, well, the labs can just pay more money, right? Sure. And they do, right? Sure. They're luring over the people they want with these crazy salaries. Mm-hmm.

1:42:40I think we're kind of going back and forth on the... Can you hear us okay? We good? Yeah, I can hear you. Okay, perfect. Maybe let's move on to Google. What was your reaction to the Google I.O. news? And I mean, you put them in A tier. They have this incredible VO3 model. how much of that is due to the cornered resource of YouTube? Is that a durable advantage? We kind of saw the open web get scraped by every single LLM foundation model company. We saw all the code in GitHub kind of get exfiltrated one way or another. Is that something that is going to be a sustainable advantage for them, at least in video generation?

1:43:31So to some extent, it's like everyone trains on YouTube, right? Like every company takes YouTube and makes it into transcripts, trains on it, or even takes the videos and trains on it, right? Every video model company is ripping stuff from YouTube. This is like very well understood and known. But you know, there's limitations on how much you can steal from YouTube, right? I think my favorite sort of clip of all time, one of my favorite clips of all time is like when someone's like, hey, Mira, did OpenAI train on YouTube? And she's just like, you know the beep, you know the beep. There's definitely some data that got out, but there's the scale of data.

1:44:06Like with YouTube, it feels like just if you were to tokenize it and then compare it to GitHub's tokens, like you're looking at orders of magnitude and that seems important in terms of just general data. Right, right. I don't think you can just rip all of YouTube like that, right? Like they have protections against that, right? Sure, you can engineer ways around like pulling some videos, but like YouTube gets 500 hours of video uploaded every minute. It's like every second or something. It's insane. Yeah. And I mean, just the storage costs alone, like you would be able to see the storage center for the copied YouTube like from space.

1:44:38Well, speaking of scraping, I'm curious. So Reddit and Anthropic are in a lawsuit right now around Anthropic, scraping, Reddit. I was surprised that they didn't have a deal in place. Do you think there was, you know, meanwhile, Gemini has a deal. OpenAI has a deal with Reddit. But do you have any insight into why Anthropic would not even make something happen when clearly it's sort of a valuable and important platform to train on? I mean, like the data is the vote, right? And if you can get away with getting the data without paying for it, then you're going to do that, right? And I think like, you know, I don't know about the legalese, but like, isn't OpenAI's deal with Reddit like$100 million a year or something insane?

1:45:24Yeah, Gemini is like 60. Yeah, they're both around 100 million a year, and that makes up about 20 % of Reddit's overall revenue. But of course, that revenue is super high margin. 70 million a year. Yeah, from both. Google is 60. Yeah, okay. Right, so if that's the case, what does that lend you? $70 million a year is a ton of either, you know, it's like 70 researchers, right? Or it's like thousands of GPUs. So it's like, would you rather YOLO and just not pay for the data? And now Anthropic can probably just fight it out in court for a couple years before they finally pay Reddit. And then are they going to have to pay as much as Google and OpenAI?

1:46:11Well, that would seem unfair. They're kind of going to get away with having not spent the money, so maybe that's a good sort of view. Obviously, there's also the view of like, well, it's not actually like stealing data because it's like training on it, you know, whatever. Yeah, yeah, yeah. Totally. That's interesting. On XAI, you put them in B tier. There's a world where Elon being on the outs with Trump draws more scrutiny to some of his companies like Tesla and SpaceX. XAI feels like completely off of the political agenda in Washington. so maybe this is a bull case for XAI because Elon has more time to focus on it if he's not spending less time in Washington.

1:46:50What's your view on XAI right now? I have a fun anecdote, right? So I have a buddy who works in manufacturing optimization at Tesla of specifically how they press the steel parts and make the body, right? And he's like, dude, I haven't had a meeting with Elon in like nine months. I'm fucked, right? I'm just like, what am I going to do? He's going to come in. So the thing is, Elon's amazing because he comes in and he's like, why are you doing it this way? Why? Why? Why? Do it this way. And then you have to either do it that way or come up with a really freaking good reason why not. And he's just such a first principles thinker.

1:47:24But for like nine months, it was like, well, I'm going to continue to drive down the path that I think is best. Yeah, ideally you want that rapid iteration, right? It's like you want him to come and just get angry at you every day instead of once every nine months. It's actually better. I mean, it depends, right? Maybe it's not every day, maybe it's not every week, but there is a level of he comes in, he resets all your priors, and you reset your roadmap, kind of. And the flip side is for the employees, it's like, well, I might now have to go back to working 100 hours a week, whereas when he was gone, I could work 40 to 60, and it was great.

1:48:00So I know Tesla definitely had a lot of neglect. I know Neuralink had a little bit. XAI some, but XAI still had quite a bit of interaction from Elon, but obviously that's just going to grow more. And so I think it's probably a good thing that these companies have Elon coming back and devoting all his time and effort. The flip side is that XAI does have a government-facing sales, as does Anthropic, as does OpenAI. and now is it like XAI is going to lose out on those sort of government sales? But at the end of the day, government sales are like pretty small relative to private industry. Where do you expect most of XAI's revenue to come from a year from today?

1:48:47That's a great question. I don't think it's consumer, right? Because probably most of their revenue today is like the Twitter subscription, right? I guess because XAI is X now, right? So the categories that matter, it's like consumer, developers, and then traditional enterprise. Right? And is there like... People don't seem to be plugging Grok into Cursor or Windsurf. And we haven't seen people then Grok. But it's just so early that maybe they could win in certain... You need anti-woke receipt processing or something. Yeah. So it's just at some point, there's going to need to be a lot of revenue on the XAI side.

1:49:26X, the everything app, can do a lot of work. Maybe they add payments and things like that. I'm curious if you think that the merger is actually net good for X, the social media platform, in the long run. Because my concern is that it's actually X, the social side of the product that will really suffer over the next few years. But I'm curious what your read is. I think it was clear before the acquisition that X had a lot of debt and they didn't generate enough profit to pay it off so their EBIT was bad obviously before interest and taxes it was fine but they just had so much interest payments and so that's why I think XAI acquired it it is beneficial, XAI does get access to a data source that no one else has and the fastest data source out there, right?

1:50:24X is way faster than any other data source out there. So I think that's hugely beneficial. But as we think about what is the highest value ways of organizing the world's information and acting upon them, and it's like today everyone's highest revenue thing is code, right? Cursor just hit 500 million ARR, right? And all these other, Windsurf just got bought for this crazy amount of money, right? Anthropic is growing revenue. Even Copilot at Microsoft and GitHub, apparently is doing$500 million a year in run rate. Yeah, but code is where all the money is being made today. And the question is, I think all of the labs, at least OpenAI, Anthropic, especially Anthropic, believe that code agents, software agents, is where the most money and most progress is going to be made over the next six months.

1:51:12I think XAI also believes this to a large extent, but they also believe some other things. right? But, you know, I think automating businesses and helping them become more efficient is going to be where most of the money is made, not consumer. So, yeah, I mean, I guess you kind of, I can maybe guess your answer, but do you think Anthropic needs a dance partner or any dance partners? Because you could see them going out and getting Pinterest or Snapchat. I don't know if that would be valuable, but they're kind of the one LLM that doesn't have a huge consumer front door to their product. And Maybe that doesn't matter on the revenue side, but maybe it matters on the data side.

1:51:50Yes, I think they're just too AGI-pilled to think consumer matters at all. They don't care about voice modality. They just don't think it matters. And it's like, voice matters a lot, at least for consumer applications. They don't care about image generation. So it's like, do they need a dance partner? There's a really good Mexican song called Asia Baila Solo. and so like I don't know if you know but it's got a freaking trumpet and everything it's such a good song and it's like no they're gonna dance alone they're gonna dance alone staying in the Elonverse are you worried that cheap robotic arms will take the jobs of humanoid robots so I think like colloquially like everyone just talks a lot about humanoids humanoid robotics but it's like most of the applications that I think are automatable in the next two three years do not require legs, right?

1:52:45Like if I go to a warehouse, if I go to a data center, if I go to a factory, the floor is a flat slab of concrete, right? I do not need to spend all this weight and energy and power on legs when I could just throw wheels on it, dude. Like, come on. And so like, that's like one side of it. And the other side is like human manipulation. It's actually funny. Like AIs are going to be way better at us than like software, than they are going to be at like just manipulating, right? Like picking things up, right? Like it's like, it's like hands are freaking incredible. And no, we aren't even close to, you know, human level hands on a physical basis, let alone the AI that we have in our brain that's, you know, developed over thousands and millions of years versus, you know, the sort of language capabilities, which have only evolved over the last 5 ,000, right?

1:53:35So I think it's going to be hard for humanoid robots to be like, take off immediately but there's tons of applications where um you know humanoid robots will still be able to do basic things things that are um capable like low risk of getting it wrong so you can try again uh and then the other aspect is you know what about what about like i cut the cost of a humanoid robot in one tenth because i have wheels and so i cut weight down dramatically and it's way more efficient and instead of having fully formed hands i just have like three gripper fingers right and that's all i need for like you know data center automation i just got to plug in things and take them out, right?

1:54:10Unscrewing things. Maybe my arm is just a screw and then the other one is like a three finger, right? And it's like, this is way, way cheaper and specialized to the task than a humanoid. Not that I'm like against humanoids, but as far as like, you know, robot arms, yes, that's part of it, right? Like a lot of tasks, it could be a stationary robot arm. But the important thing to consider here is that literally all of this will be made in China and none of it will be made in America. Why is that? The supply chain for Chinese robotics is... Soundboard probe. We're very pro-America here. Yeah, the audience is very America-pilled.

1:54:48It's sad, but it's like, you know, the supply chain here is like, China's, China makes more robots. You know, if you go back five years ago, China made as many robots as roughly Germany, South Korea, Japan, and the U.S., right? Now China makes more robots than all of them combined, right? And the cost that they do that at is way lower. And like, well, the Japanese companies argued that the Chinese robots are shit. Yeah, but like you speed run up like quality faster than you speed run down on cost. And like the same applies to like Germany, although Germany sold some of their best robotics companies to China and all the tech got transferred.

1:55:21But like, you know, it's like we don't have the manufacturing supply chains to turn, you know, raw goods into motors or into actuators. And if we can, you know, or when we do, it's like it costs 10x as much. And so it's just like, it's impossible, at least without some huge industrial policy, for the U.S. to compete here. So, I mean, it seems like there's almost an analogy there to the semiconductor industry where China has been lagging there for decades, but has done a lot to at least catch up to the lagging edge. Are there any policies that, or investments or industrial policy that China did to at least stay in the game to the degree that they did that we should be stealing from them and doing here for robots and manufacturing.

1:56:1010 five-year plans back to back. We need 10 five-year plans. I think the funny thing is that when you look at industrial policy, the way China does it sometimes is more capitalistic than the way we do it. Now, obviously, giving a bunch of money to an industry is not capitalistic, but the exact policy mechanism of the CHIPS Act versus what China does is so much better. So China does like, oh, you just get X subsidy for manufacturing a certain amount, rather than like, oh, it's a company-by-company deal and this company gets this money for this plant. Or like, hey, we're just going to subsidize the land in general for this.

1:56:47Or, hey, if you build a fab that is 28 nanometer, you don't have taxes for 10 years. And so stuff like this is what China's done and that's had them like dramatically increase their semiconductor spend, you have things like the National Railway Company is now making power semiconductors, right? Because they make profit and they're like, well, if I make fabs, then I can blow all my profits from the railway in that. And then all my fab profits will just be profits without tax, right? And so now their largest railway company is like the third largest power chip company in China and like top 10 in the world.

1:57:23And over the next few years, they're going to be top five period, right, across the whole world. And I think like things like this are like interesting industrial policy. And probably we can get like the car companies, you know, and various, you know, like John Deere and like all these like automation, like these mechanized things to get them to start pushing into robotics, right? Because like BYD is pushing into robotics. And like obviously everyone knows about Unitree. But like, you know, Xiaomi is making cars and getting into robotics, right? All these companies are getting into robotics in China because of industrial policy.

1:57:58And they did the same with autos, EVs. They went into EVs because of industrial policy. Or they went into semiconductors because of industrial policy. America doesn't have that. We have the startups who are mostly just buying Chinese equipment and repackaging and pretending like they make it here, like Figur. or, you know, like we also we have we have American VCs who put American flags in their offices. That's got to count for something. I mean, I'm not saying it's not noble. It's just I'm joking. On Unitree, what do you think the perception of Unitree is in China? Because if you're if you're a believer in the wheels and the different actuators, it feels like they could be more like China's Boston Dynamics where it's like, oh, a bunch of cool videos, but not a lot of actual applications.

1:58:50But the reaction in America to Unitree videos is, oh my God, there's this crazy wave coming. But maybe that's a misperception. I love that every video of them is just like doing war things. Like you really want to scare the West? Totally. Why don't you just have it like petting bunnies? Like, you know, trust us, the Chinese robots are not dangerous. They should be doing that. Yeah, I was wondering, like, I fully expect that there will be one of those, like, North Korea-style military marches with 100 ,000 humanoids, and it will just be terrifying, even if the capabilities aren't actually there for real warfighting, just the demonstration.

1:59:27Did you see the humanoid marathon in China? Yeah, yeah, yeah. Yeah, yeah, yeah. But anyway, so to answer your question, right? Sorry. Unitree, yes, we see the humanoids, the robot dogs, but they've actually released a robot dog but with wheels. And so it's just like sticks and it has wheels. And they're releasing a humanoid with wheels, actually. They just showed it off at this conference in Singapore that one of the guys on my team was at. And so they're coming with the mobile manipulators, But also you have to recognize like for the cost that they make in ten thousand dollars Western companies make in like seventy five thousand or a hundred thousand dollars, right?

2:00:07So like the quality of hardware they can make so that's also like an important consideration. Yeah Said you said something earlier around all the you know, the next value in Models coming from automating businesses is the right framework for that agents is that how you how? how someone should conceptualize it, or is it an entirely new paradigm that's less about, you know, agent-human kind of collaboration and more so just like full autonomy? Yeah, I mean, the dream is always multi-agent systems, right? Many different agents working together and just doing the job. But today, you know, the time horizon for human model interaction has been growing rapidly, right?

2:00:49Initially, it was like seconds, right? It's getting to minutes, or if you use deep research, like 30 plus minutes, right? And this is going to start extending to many places like code. If you use cloud code, the time interaction is like minutes, not seconds. And so I think it's – and it can even extend to tens of minutes. I think these agent systems will – it won't be like a hard hit, right? Like everyone's expecting sort of like, oh, we have chatbots, now we have agents. But really, it's going to be like a blended curve of like, yeah, you know, like just like the amount of interaction that human has with the system becomes less and less and less over time.

2:01:30That's a good framework. Sorry to move back to China. But Alibaba, Quen, what, I mean, it seems like they're just pumping out like a million different open source models. What's the reaction been to the American research labs? Have we actually learned anything from that? We've had some folks on the show have been just generally excited to play with so many different research tools, but But are they trying to commercialize that and I guess my big question is like what is the actual chat GPT of China right now? Like what is the consumer front end? So on the on the latter point the consumer front end is Deep seek offered through the clouds right so some of the cloud companies like Baidu Tencent etc offer for DeepSeq, the model, but through their own platform rather than through DeepSeq.

2:02:22And some of them will offer it through DeepSeq themselves directly. But the interesting thing to note here is that we sort of have a very big dichotomy here. Part of the reason DeepSeq costs so much less on inference than US models is because the speed of DeepSeq's models, right? And there's this curve, right, of like how fast does the model respond to you versus how many users you can batch together, right? So if I have a system and I'm only serving one user, it's going to be screaming fast. But if I'm serving hundreds of users, each individual user is going to be slower. Now, the total collective number of tokens I'm generating is much higher.

2:02:57But this is interesting because when I use OpenAI, I get 100 tokens per second. I get 200 tokens per second. Same with Google, same with Anthropic, XAI, et cetera. You type the query and it goes maybe twice as fast as you can read so you can skim it. But if you do that to DeepSeek, it's like 20 tokens a second. And I can read faster than the model outputs. And that's a conscious decision because they're limited on compute, on, hey, I want to serve as many users as possible. And so the chatbot-style applications are actually very bad. And that's why Chinese people tend to use Open Router a lot to access open-ai anthropic models in China, even though they're not supposed to, or accessing other companies' models through other methods.

2:03:41So I wouldn't say that there's truly a one consumer front end. Rather, it's that as soon as DeepSeek came out, the government started pushing all the companies to offer DeepSeek. But there still isn't a good consumer experience in China. Moonshot is probably the most popular one, but their models aren't that good. So I don't think it's one out yet. Yeah, how much of an issue is owning the application layer if you're an independent country that doesn't want to rely on American or Chinese foundation models like what we've seen with Mistral and LeChat? There's this world where you spend all this money to build this data center in your country.

2:04:22You spend all this money to train a model. Maybe you're not at the frontier, but you're close. You're in the game. But everyone just by default in your country just goes to chat.com. And so unless you're willing to ban ChatGPT and you're not going that far, you haven't truly won in terms of deploying your model with your view on free speech into your populace. Yeah, so I think that's going to be challenging. There are sovereign AI efforts in all these countries, but like you mentioned, Le Chat is not that successful in France. and Mistral is the closest of any sovereign AI besides China and the US so you're going to get all these models but people are just going to use the consumer Google's models are so popular in India or perplexity is ridiculously popular in Indonesia for some reason I don't know why but these are just frontier models from American companies rather than the domestic model that could be made and so I don't expect any sovereign AI efforts to truly succeed on the model level.

2:05:29I do think that sovereign AI can mean many things. It could mean I'm making a bunch of data centers. It could mean I'm making a big NeoCloud company or it could mean I'm going to make the full stack with the model but at any point if it doesn't get economical you can always just rent out. Most of the costs of the GPU is you can just rent them out to someone else and then try and recuperate your cost. But you can't really do that with a foundation model that you've trained that has like mediocre weights, right? So that is kind of - Right, it's like you just, you burned all this money and flops on something that's effectively useless because Meta's open source model and DeepSeq's open source model and Alibaba's open source models crush you.

2:06:03Yeah, yeah. Speaking of Meta, I've been trying to work through what's going on there. There's obviously kind of the dust up around Llama 4 launches, but somebody asked me like, what is the bull case? And everyone always says like, oh, recruiting or, you know, some beneficial thing or benevolent thing. But I was wondering if there's a reasonable case to be made that if you try and just project out what Meta's LLM spend would have been on other vendors in five years, it would have been so big that even if they're constantly on the lagging edge, they're still recouping the investment in training LLAMA just because they will be able to serve it internally.

2:06:45Is that a reasonable thesis or is there something else going on there? What do you think of the motivations and problems with that project? They have internal llama models too that are trained on their internal code base that they can use internally. That sort of stuff is something they could just never get elsewhere. I think it's a multi-factor thing. The bull case is obviously a personalized AI assistant on my glasses. That's just always on my head, always serving me up slop. So truly waiting in the consumer use case. Right. That's the real more star. That seems to be what Meta is actually going for.

2:07:24But as far as like, you know, today they roll out WhatsApp, Instagram, and Facebook models, Meta Llama models to everyone. And they claim they have 500 million monthly active users, right? Which is a lot, right? Now imagine if they didn't offer that, right? Would people start using ChatGPT instead? and now they're out of that meta ecosystem and their screen time goes down and they get look at less ads or does like Apple intelligence become more powerful and people will go to Apple intelligence for their AI because that's the AI that's most accessible and now all of a sudden meta who already has been screwed by Apple multiple times on advertising is now getting screwed even harder because all of the like like more and more of the data keeps going to Apple and there's no way for meta to get access to it so it's like there's an existential threat here too right And so I think it's prudent that meta spends on these models.

2:08:15I'm also a lot more bullish on meta long term than most people are. People are looking at Llama 4 as a single point in time of like, wow, this really sucks. But there's two models there, right? There's Maverick and there's Scout. One of them is really bad and one of them is pretty good. What about Behemoth? Isn't that another? They only have a third? Rough name. Yeah, but they didn't release it. Rough name. They didn't release it. Yeah. Yeah, it is a weird name because it's very demonic and it's like you're unleashing this demon. Well, the funny thing is the architecture is super similar to GPT-4.

2:08:47Oh, really? Yeah, it's extremely similar. It was just a very funny naming scheme because the behemoth is the untameable beast and they couldn't tame it to get it out the door. And so the name really mimics what happened with that project. But so, yeah, I mean, are they gated by scale? Are there, I mean, the CapEx spend seems to be growing, but similarly to all the other hyperscalers, is there anything that's unique about meta over the next couple of years that could kind of give them a compounding advantage there and allow them to kind of come from behind? I do think they are still spending less on Gen.AI than other hyperscalers.

2:09:28note that 60-70 % of their GPU spend is on recommendation systems and standard business. Nowadays that means creating Gen.AI ads and those have more click-through rate. That's the holy grail for the next six months for them. So it's not like there's no Gen.AI in there, but a lot of it is on their classical business. But generally they are a player to be reckoned with. Open.AI has Stargate, Meta has their Louisiana project. I don't know if there's a fancy name for it. And everyone has their big projects, and so when we get to 26, 27, everyone's going to have these gigawatt-scale data centers with close to a million chips.

2:10:08And the game continues to fight on. I don't think that, like, this is a temporary setback for Meta, but they continue to recruit, they continue to have good talent, they'll be reorganized. Without pain, you don't get better. So this is a learning experience for them. So on Meta and Scale, they partnered with Constellation on this sort of basically agreeing to purchase nuclear energy from Constellation for the next 20 years. What's your updated take on nuclear? We obviously have these new EOs to help support it, yet it still feels pretty far out, at least for bringing new energy online. How are you thinking about the category?

2:10:51Yeah, so even in general, building new nuclear reactors is still going to take a really long time. And if you're a believer in AI-27 or AI-2032, whatever the number is, that's too long of a timescale to matter, right? This is still in the deficits can be largest possible timeframe. I think with regards to nuclear, that's the case for standard Gen 4 reactors. but when it comes to like SMRs they didn't ease the most important regulation which is that like the concrete containment zone this containment zone for the SMRs still has to be like monumentally big even though it's like a much smaller, much safer thing and so they didn't ease that regulation so I'm not that bullish on SMRs yet either and basically like yes you can sign these deals for nuclear power but that's nuclear power that was already being used by residential and now there's residential or like industrial, whatever it was, just turns to gas.

2:11:50Or on the flip side, right, if you look in like Louisiana, Meta's data center uses gas, right? Or if you look at Stargate, it uses gas. Or if you look at Memphis, Tennessee, XAI, it uses gas, right? Like, you know, you look at Amazon's Indiana facility, Project Rainier, it uses gas, right? Like everyone just uses gas and like that's - Are the ESG commitments that the hyperscalers made in kind of the prior era, like informing any of like, are they acting as like handcuffs or shackles or like leg weights going into this next energy build out? So the funny thing is some of them have backed off a little bit, like Meta and Microsoft have backed off some.

2:12:27But the other thing you can do is just like pretend like you're still green. And so what you do is you do this thing called a PPA, right, which is a power purchasing agreement. So you can set up like solar panels like 100 miles away and they'll generate power only during the day, but that power generated during the day is enough to cover the entire night of the data center as well but the actual electrons being consumed are not from the solar panels right the solar panels are just selling it to the grid but you purchased it and sold it to the grid whereas the power that you're buying from the grid is like coming from a gas plant and the actual electrons you're consuming are from there so then you're pretending to be green because you have this solar ppa backing it up or you have this wind ppa backing it up but in terms of actual electrons being consumed and new plants being built, it's gas.

2:13:11Yeah. And so I think this has, their regulatory ESG stuff is like malleable enough that this is allowed. They're working through it. Let's talk about some of the other sci-fi concepts for energy generation. People are starting to talk about data centers in space. People are talking about data centers in the middle of the ocean with tidal energy. Basically go where the energy is basically free or you get something for free. Maybe you get cooling for free or land for free. Is any of that relevant to what you've been focused on? So Microsoft's tried putting a data center in the water before for cooling, right?

2:13:45And that didn't work. And it's the same reason why space data centers won't work anytime soon. And as far as tidal energy, tidal energy is interesting. I think it could work, but it's a bit of ways. The main thing on why data centers don't work, whether you're sticking them underwater or you're sticking them on a barge in the middle of the ocean or you're sticking them in space is that GPUs are really, really unreliable. So we've done a bunch of testing of GPUs. We reviewed 40 different clouds, up to thousands of GPUs, in fact, in some cases, but hundreds usually. And many times, it was called ClusterMax, right, where we tested 40 different clouds, including Amazon, Google, et cetera.

2:14:28Everyone's GPUs are unreliable because they're inherently unreliable. And what happens when a GPU is unreliable? Sometimes it's as simple as, oh, turn off the server, turn it back on. Sometimes it's unplug an optical fiber, a transceiver, and plug it back in. Sometimes it's like, take the GPU out of the socket, put it back in. This is the vast majority of fixes. I don't know why you need to do this, but that's how tech works. You unplug it, you plug it back in. And you need people to do that. Yes, robots could potentially do it, but we're not there yet. the space of a data center, like the physical size of it isn't that large, really.

2:15:03Like they're big in an individual basis, but like they're not even covering like a basis point of American land, even in like a decade, right? It's like, they're very small. So, so like GPUs are really unreliable, which we've sort of proven with like ClusterMax. And then the other aspect of space data centers that's really dumb is like, how does heat get removed, right? Heat gets removed because like you have atoms contact the heat, they, you know, whatever's hot, they take the heat away and they're more excited as they leave, right? You're taking cold stuff, you're making it hotter and you're sending it away, whether it's water or air, right?

2:15:34But at the end of the day, you're radiating that heat out of an evaporation tower or chillers or whatever into the general air around you on land. But in space, there's no medium for you to like, like how many particles are hitting your spacecraft and carrying away heat? And also those particles that are hitting your spacecraft are generally quite excited anyways. but there's just like very few of them. And so like there's no way to radiate heat away and these things consume tons of power. So like they're unreliable, they consume tons of power. Power is cheaper in space, right? Solar panels, you can have them like constantly having sun but like it's like, you know, what about the fact that like you got to radiate the heat away?

2:16:15Like people try and put like tiny little computer chips in space and satellites and they have a hard time cooling those, right? So it's a problem that I don't think is like feasible. It's something that's feasible in the next sort of, within the deficit spending is okay timeframe. I like that. Going back to that. Are you bullish at all? That's a really good joke. Are you bullish at all on new consumer hardware, dedicated hardware for AI? Johnny Ive joining friend.com. You might be calling in from a rabbit R1 and we don't even know. are you excited at all about kind of a third device outside of the phone or the computer yeah I think like the Johnny I've Sam Altman video is so cinematic it's so good I'm excited for the product simply because the video is so good the quality of it but like like I don't know like there's I've seen I've seen that like this Apple device that has like a it's it's it's basically an airpod with a camera right um and like that that's pretty cool um or like the ar glasses right with the cameras on them like the meta ray bands um or you know all the other iterations of that i i think that a new device is definitely coming um with how good ai is getting it means that the form of human computer interface can change um and it can be more natural and more seamless right It's not going to be typing on a computer in front of a desk.

2:17:46You're still going to have that. You're still going to have your phone. But really, I think this hands-free compute device is coming. And it's enabled by AI. Now, will Meta win? Will Apple win? Will OpenAI and Johnny Ive win? I like Friend. The founder's really cool. Will they be a runner at all? It's a hard space. But I think there's three companies that seem to have a shot. it's like Apple, Meta, and maybe OpenAI Johnny Ive. Yeah. And Avi, of course. Yeah, and Avi coming from behind. The domain is just, we got to give it up for the domain. The domain is so good. Friend.com. Got to appreciate it.

2:18:23He spent a fortune. It's worth every penny. On Apple, you put them in L tier. If Warren Buffett can learn to love Apple at age 75, can we AGI pill Tim Cook at age 65 or however old he is? Is it possible? What needs to change for Apple to become dominant? They have so many advantages. They have tons of line time at TSMC. It's the AI that's made me laugh more than any other model. That's true. The tech summaries that go crazily wrong and hallucinate weird things, very entertaining product. Seriously underrated as a consumer product. But what's your experience? Break down what's happening at Apple, what changes you predict, what changes you would recommend?

2:19:02So I think they're clearly not that AGI-pilled. They haven't spent a ton on data centers and things like that. Now, they do have hundreds of megawatts going online in the next two years versus the hyperscalers have gigawatts, right? So it's like, okay, there's still an order magnitude off in how much they want to foom. And then the flip side is that Apple has started building an accelerator that hired some of the best people from Google. They already had a good chip team anyways because their chips for their phones and their Macs were good. But it's like getting the talent they didn't have. they're partnering with certain companies and they hired some of the best people from Google including their head of rack design rack architecture Apple's definitely keeping far enough behind that at any point they could leap into action but the challenge is always going to be how are you going to get people to join this company and so this is sort of my bull case or not my bull case but a possible scenario for thinking machines is that they just get acquired by Apple, right?

2:20:08Because at some point, Apple will realize, oh, God, we need talent. And it's like, there's no talent out there for us to hire. When we try and hire, they're like, no, screw this. And Tim barely makes 70. He doesn't even make 75 million. He makes 74.6 million, which is like a junior OpenAI researcher if they joined two years ago. So it's like, how can they compete? Dude, that's OpenAI's janitor, bro. Yeah, exactly. Is Lisa Sue cooking? she acquired bream earlier this week can you break down that acquisition uh was it exciting to you what's going on over there yeah so so they actually acquired uh two companies okay um this week they they acquired bream um which which is more like llvm compiler people um i think it's more of like an acquihire it's like 10 people or something like that i think uh it's not that crazy but it's it's it's it's more commitment that they care about acquiring software talent yeah and What's funny is when they acquire, acquihire a company, the people at that company are getting paid a fair salary, right?

2:21:09You know, like a very good salary. Whereas like sometimes their hiring process, they don't pay people enough, which is why they don't always get the best talent. Which is something that we've sort of mentioned. But yeah, yeah. So Brim's like an ML compiler and things like that. And then Untether is another one, which was like making an AI chip. And they acquihired them as well. Mostly for the staff on the hardware side, I imagine. Because they just need to move faster to be able to release chips as fast as NVIDIA's. Because NVIDIA's just like clockwork spinning out new chips. They're trying to go to that yearly cadence and such.

2:21:40So I think Lisa Su is waking up to how important this opportunity is and that she needs to spend a lot more. But note, they're still spending a ton of money on buybacks when they could be spending... I mean, so is NVIDIA, but NVIDIA's also making godly amounts of money. AMD's making good money, and then they could turn around and reinvest that, but they're not reinvesting as much as they should. I think these are good acquihires, both on Tether and Brim, which are both acquihires. They also bought Nod AI a while ago. Nod AI is where AMD's god of GPU AI software is. Anoush, he came from there. He's the king.

2:22:22And that acquisition was really good because they have, again, really good software talent that they were able to pick up. It's just we need them to move faster or if we want to take them super seriously. But the right direction. What was your read on the NVIDIA earnings and kind of the lack of guidance around the training to inference switchover? Do you have a better idea of what's going on? It seemed like they had quoted a 40 % number earlier. They didn't include a new number. But this seems like an important narrative or maybe it's not around the actual workload shift. And then I want to know if that's relevant to AMD at all.

2:22:57Is there an opportunity there? Yeah, so I think that NVIDIA is going to have a hard time knowing how much compute is going to inference versus training. It's actually just hard for them to know. They don't get to know what workload their customer is running. They can guess. They have the most data to guess, but they can guess. Would people not want to tell NVIDIA? I feel like if you're buying 100 million H200s, you might just be like, yeah, sure, I'll tell you what I'm doing with them right now. No problem. I'll fill out your survey. Here's my NPS. I think these companies are very secretive. OpenAI won't tell NVIDIA what their model architecture is, even though if they just told them what their model architecture is, NVIDIA could optimize their next generation chip to it way better.

2:23:38And so NVIDIA just has to guess and hope that they know what OpenAI is doing, but they don't. They have a good idea. I think that's one aspect of it. The other aspect is that the lines between training and inference are blurring. In the days of pre-training, your cluster would be training, training, training, and then you deploy on inference clusters. But now with reinforcement learning with verifiable rewards, one interesting thing is that traffic is not usual, right? It's like it oscillates. And actually on the weekends, chat GPT is used very little relative to weekdays. During work and school hours, chat GPT usage is huge.

2:24:16During night, it's not really, right? Except for when like Ghibli slop is trending. Yeah, it's the only time you have like usage, you know, higher during non peak hours, which is like work day. Right. And so what's interesting is that these clusters have lower utilization. But now I can turn around and start using these clusters at night. For example, I can spin down a percentage of them and I can turn them to reinforcement learning to verify our reward. I can generate a bunch of data, verify it, see if it's good or not, whether it's math problems, coding, or some other reinforcement learning paradigm like computer use.

2:24:50I can do it, and then I can keep that data. Then if I ever need to switch over back to inference as usage is picking up, I'm just like, okay, stop, no more of that. Put the normal model on there and start doing inference. And that can take a minute or less. So it's like there is a blurring of lines of what is training versus inference. And then the other aspect is, can AMD get in there? What's interesting is that AMD has generally taken this approach of, hey, we're 80 % of the cost and we're the same performance. That's sort of where MI300 shakes out. Some scenarios it's higher, some scenarios it's lower, but we're 80 % of the cost.

2:25:32Now, when you look at this new generation of GPUs they're launching, I think next week, MI355, it's two-thirds of the cost of NVIDIA's Blackwell, but performance-wise, is it two-thirds or is it less, right? Or is it more, right? And so this is going to be a challenge. It's going to rely a lot on their software. Meta's really excited about it, but not too many other customers are. And so the question is, can AMD get in there? They're targeting inference mostly, but NVIDIA's moving the game really fast, right? They're coming up with these new software libraries that don't just like, it's not just CUDA, It's like they're building the entire inference engine.

2:26:10And AMD can't plug into that. It's called Dynamo. So what NVIDIA is doing around Dynamo is really locking AMD out from inference at the non-hyperscalers and labs because it just doesn't make sense for anyone else to try and replicate the software that NVIDIA is giving away for free on NVIDIA GPUs. What do you hope to see out of companies like Prime Intellect over the next one to two years? I believe you're an investor in Prime Intellect. Is that correct? but so I'm sure a little bit conflicted, but yeah. Yeah, I mean, I invested in the seed, but to be clear, I'm pretty impartial, right? We had this ClusterMax review and we didn't give them the best review even though I invested in them personally, right?

2:26:51So it's like, you know, this is, I think, but I think what they're doing I'm really excited about, right? AI is super centralized, right? And this is the only path I see forward mostly for central, is like that AI continues to be more centralized. There's fewer and fewer players. They have more and more control over our data, over automation of the entire world's economy. This is potentially quite scary for someone who likes decentralized power systems and capitalism and such, is that centralization begets tyranny. Not that I'm being that dogmatic, but their technology is mostly a technology reason.

2:27:26They had cool demos around decentralized training. They've trained models in a decentralized way. They've done reinforcement learning with verifiable rewards now in a decentralized way. I'm very excited about the research they're doing there and they're open sourcing everything, right? Which is the cool thing, right? You can just go look at what they open source. You don't have to like trust, you know, anything. It's like they have some pretty good software. I'm hopeful that there's a path to all AI and power, not just continually being centralized until we, you know, are like enslaved by the machine god.

2:27:57Except if semi-analysis were to centralize it, I feel like, you know, people should be able to trust you. benevolent sort of dictator world leader I can see that call me the Lee Kwan Yu of AI for sure I wanted to go more into those like studio Ghibli moments when people say quick fire off would you hire somebody with a studio Ghibli profile picture or is that a negative signal in our Slack someone has a studio Ghibli profile picture okay I like it And his public profiles are also Studio Ghibli. Okay. Wow. That's good. Wow. Whenever there's these big meme moments, everyone says the GPUs are on fire.

2:28:40Obviously, they're not literally on fire. But what is breaking? Because is it that there's actually more unplugging and plugging it back in in the data center? When we see Studio Ghibli go out and the queries are clearly failing, what do you think is happening internally? Is it just too much load? Because I feel like the chips are under incredible load during training, right? So they're used full times. But why are we seeing so much failure during these spikes? What's actually going on there? With training, we know what's happening, right? We're saying, hey, in this iteration, you're going to process this many tokens.

2:29:18You're going to update the weights. You're going to exchange them all. Okay, here's more data for you to train on, right? And you just keep iterating many, many times. Every 15 seconds or 30 seconds or 60 seconds, you update the weights. and you do it again. And you know how much data you're sending. With inference, you know, you have these clusters and it's like, it's that curve again, right? If I just have one user, it happens really fast, great user experience. If I have tons and tons of users, it happens really slow. And while I am serving more users and I'm getting more output, each individual's user experience is worse.

2:29:45But at some point it gets so bad that you either run out of memory or, hey, a GPU does break. Now all of a sudden you're dropping requests constantly, right? So not only does the user experience get bad because it goes from fast to super, super slow. Look at how fast Ghibli is processed now. It's crazy fast compared to what it was initially. But also when something bad does happen, you can't just be like, oh, take all these requests, route them to these other GPUs. It's, oh, all these queries fail. Oh, I got more requests than I'm able to handle. Well, these users, I'm just not going to start processing their requests.

2:30:21I'm going to say, sorry, it doesn't work. And so that's sort of when the GPUs are on fire, that's what happens, right? Now, also, I think it's mostly that dynamic of like you add that incremental user and things break. Or if something breaks, all those users now have their stuff broken. Well, thank you so much for coming on. Everyone in the audience should go to semianalysis.com if you're not already subscribed. And we would love to talk to you again. This was a fantastic conversation. Super fun. A million, a billion, a trillion, I want to call. A billion. Hey, hit the size. Go add a new plan on Semi Analysis,$5 million a year.

2:31:04And we'll ask some of our VC buddies that just have a bit too much management fees. We will shame them into... Oh, you're not on the pro plan yet? Yeah, yeah, yeah. The max tier? The presidential plan. You're on Semi Analysis Ultra Max? Wow. Wow. You must not be serious about artificial intelligence. Oh, you don't really invest in AI. Oh, you're not about the AI thing. You're more of a tourist. I get it. Yeah, more of a tourist. Yeah, yeah. You're a B2B SaaS guy. Yeah, it's cool. It's cool. It's great. I'm sure you make a good living. I'm sure you make a good living. But anyway, this is fantastic.

2:31:37We'll talk to you soon. Thank you guys for having me on. Super fun. We'll talk to you later. Talk soon. Bye. That was a lot of fun. All right. You're going to love this, John. I think you were too locked in. We're going to have Blake calling in at 2.15, 45 minutes to talk about the new EO. Oh, is it political? You know I can't pay attention to politics when I'm talking to an AI person. You know, I just get too locked in. But that is fantastic news. Very exciting. I see your post. Blake will be joining TBPN live at 2.15 p.m. PST to discuss the new supersonic EO. Very excited to talk to him. And we have Dana Settle from Graycroft coming in the studio.

2:32:15How are you doing, Dana? Good to see you. Hi. Great to see you too. What's going on? Thanks so much for joining. Very exciting times. The IPO window is squeaking open and it's open more every day. We talked to the founder of Circle today. Very good showing. We'd love to talk to you about that, what's going on in the markets, what you're investing in today. But would you mind introducing yourself for the audience members? For anybody that's been living under a rock. Yeah, living under a data center, as you like to say. Yeah. Thank you. Thank you. Yeah, Dana Settle, co-founder and managing partner at Graycroft.

2:32:49been a venture capitalist for a very long time. Started out in 1998 on Sand Hill Road. So seen a few cycles. Yeah. What was the first big deal that you kind of really, really tracked or invested in or followed? Like, were you tracking? Yeah. What deal allowed you the right to be a VC for more than maybe four years? Because it's, you know, the first four years, it's kind of like, you know, fair game. You can make some bets, but eventually, you know, there needs to be some results. Yeah, it's a good question. So, you know, when I first started in venture, we were investing in all kinds of things, but more, you know, it was like networking equipment and semiconductors and storage and, you know, sort of that was building the last, you know, really the first generation of the internet.

2:33:38And so some of the companies from that era were, are, no longer exist. they're actually bought by much larger networking equipment, you know, companies, etc. But a similar dynamic, you know, where we would invest. I didn't wasn't expecting this question. So basically, on the name of this company, but literally, it was like, you know, from Oh, Q Terra, God, optical equipment company. And it was, you know, in like 18 months, sold for$2 billion. Yeah. Wow, wow. There we go. That's amazing. Yeah, it is fascinating. Even though I was the number two on it. Of course, of course. Yeah, but that's great.

2:34:17It is fascinating how much the networking equipment space has kind of come back into focus with the AI boom. We talked to Kevin Wheel over at OpenAI. He's on the board of Cisco now, right? Yeah. And we've talked to another founder who is working on optical cabling in the data center because as we scale up these models. And so everything old is new again. Totally. But what is most new for you? What have you been focused on? and what's been in the news in the last couple weeks. Yeah, well, I mean, in terms of what's new and what we're investing in is certainly in all of the, you know, sort of AI, everything from infrastructure up through applications.

2:34:52So that's nothing terribly surprising there. But it is interesting to really look back to the history of the internet and see how much things are repeating now in terms of what's being built. The other thing that we do is we actually have a sustainability fund. And we've done that in partnership with some strategic partners. So Coca-Cola and eight of their largest bottlers globally. And what's been so interesting about that is we are there's such a huge intersection essentially between AI and sustainability. People don't really think about it. But, you know, if you look at energy consumption doubling by 2050, I mean, something's something's got to give.

2:35:32It's all oil and gas right now. But I think that's where we see a lot of opportunity, not just in renewables, but actually in like the software and hardware for sort of reimagining how all of that's consumed. Yeah, you imagine that all the hyperscalers are going to need to kind of reevaluate the ESG mandates. There's already a lot of companies in the space like Crusoe Energy started by flaring natural gas plants. There was wasted energy. And there's obviously a ton of news in solar and nuclear and all sorts of stuff. The energy landscape has been fantastic. Would you mind taking us through the latest IPO experience?

2:36:09How was that? I'd love to hear that story kind of from start to finish. because there's so many VCs that have been in the game for a long time and not actually been through that process because companies love staying private. Well, other reasons too. There's probably more and bigger reasons. For sure. But there's been a lot of pressure to stay private, right? Well, it's really interesting. I mean, I don't know. That's like a whole other tangent, and I'm kind of obsessed with it because it is really interesting if you look at what the public markets are today, which I think is really interesting.

2:36:40It's just really different. And it's almost going back to potentially what it what it used to be, you know, where it was like smaller IPOs and it was actually, you know, access to growth opportunities to more retail investors. So so anyway, but but our most recent IPO is a company called Mountain. And it's a pretty amazing story. Incredible founder, serial entrepreneur, technical background. And really, you know, he essentially started the business. I mean, we invested actually in 2011. So this gives you a little bit of a sense of success. We love an overnight success. Overnight success. We have a soundboard here for moments like that.

2:37:23It's every founder we talk to. I mean, we talked to the Circle founder and it was the same thing. He's like, yeah, I've been doing this for a decade. You know, today's the focus. Well, it's funny to see Jeremy Allaire, you know, talking about, I mean, I remember when And he had a video search startup in the last generation. And so it's really interesting to see these things sort of come full circle. Incredible entrepreneur. No pun intended. Yes. But anyway, back to the story of Mountain. Back to the story of Mountain. So Mountain, Mark Douglas, really, again, incredible founder. He had started one type of an ad tech business, essentially, back in sort of the display days.

2:38:01And when we first invested, just to give a sense, it was a$4 million Series B. Wow. No way. Just think about what that means today. Now it's like$4 billion. It's a series that's around for ants. It's around for ants. But it needed the capital and it worked. Yeah. And really, that market changed pretty dramatically when all of the regulation around cookies, etc. happened. And Mark had the foresight to take what was$100 million plus revenue business, super profitable, and essentially build an entirely new business, saw that connected television would be the next major ad platform. I mean, and again, seems very obvious now, but definitely wasn't consensus at the time.

2:38:44And he essentially just built that out of cash flow from the old business, sunset that business, and built this one up to now the company that went public two weeks. We're talking to the founder of AG1 next. I'm interested to hear your take on celebrity partnership. There's so many different configurations, whether it's just the celebrity investing, just like any other investor, and then maybe they've added some value to co-founding the company, incubations. What is your view on the roles of celebrities or people that bring differentiated value add in the startup ecosystem? Well, I think what you just said is the key, which is differentiated value add.

2:39:23And, you know, in the case of Mountain, we did partner with Ryan Reynolds and his marketing agency. And, you know, and that was sort of a fortuitous connection. But, you know, it really just I actually was introduced to Ryan and his team as they were sort of thinking about what to do with their agency. and it was just in this moment where I was like wow what you guys are trying to do in terms of really making ads that sort of you know move at the at this at the pace of culture like I have the perfect platform for you and you know really Mark and Ryan just sort of saw eye to eye on on really how to build this business um but again that was so really um authentic to Ryan and his team it's just like what they want to do they want to make really cool ads that really resonate with people not to just make ads but really you know to sort of make them at the pace of culture and mountain is the perfect platform for that and and that's why i think it's worked so well and we saw literally i mean the crazy thing is when you think about i think about celebrity or influence as being largely marketing right i mean it's and so if you sort of look for like a dollar for dollar replacement totally mountains inbound leads went from like two percent to over 60 percent wow that's crazy.

2:40:37Well, it's also interesting to think in a B2B business like Mountain, like somebody like Ryan Reynolds effectively acting as a sales rep and just like messaging a CEO and being like, hey, I'd love to set up a demo with you. And it's like that, just that impact alone in terms of bringing spend on the platform. I mean, we've been kind of joking about this. We talk about ourselves as like enterprise influencers because all of our sponsors are basically B2B companies for the most part. And it's just like, it's kind of a new era, but it's working. And Ryan's clearly carved out just a very unique position.

2:41:05He's not just somebody you just pay to show up for an ad. And there are celebrities that have built great businesses on that, but he's taking a very different tact. I'm curious, when you made the Mountain Investment, was there any, was there people maybe in the investment committee or other investors that were friends that sort of were writing off TV at the time digital was exploding? And now, you know, looking back, it's like TV advertising has exploded it's continuing to grow even in the u.s i think in 20 past sometime past 2030 it'll be well beyond you know 100 billion dollars in in uh spend just here in the united states so tv advertising is clearly not going away but i could imagine 10 years ago a lot of um you know you know a facebook pm that became an investor would probably have written a memo that said like this market is going to go away or at least shrink.

2:41:58Yeah. And I would clarify, I mean, a couple of things. I think 10 years ago for sure. I think TV advertising has changed. I mean, this is largely advertising on connected television, streaming television. So really the distribution has fundamentally changed. But yeah, I mean, we actually had investments in some companies that, you know, like 15 years ago that were actually, you know, improving the process for television ads. And we sold those businesses, gratefully, and made a profit on them. So when Mark proposed this, I think what was counterintuitive is all of those businesses were subscription.

2:42:37I mean, the idea was not to have ads in those streaming services, right? And so what's happened in the last five years is now all of those streaming models have moved to where the majority of subscribers that are coming on. I I mean, if you look at the public earnings from, you know, Netflix and, you know, Disney, etc., that the majority of new users are coming on to their ad platforms. Yeah. Yeah. Do you have, I mean, I'm sure, you know, investing and being in L.A. often, like, there's probably celebrities that come to you with like, hey, I want to get into venture. I want to have exposure one way or another.

2:43:13And there's so many different models. Do you have any recommendations for all the celebrities that are listening? for the best way to get exposure to tech or just patterns to avoid so that they don't get over their skis. Yeah, I mean, look, I think it's like anything, like whatever you're truly, really curious about and you really want to dig into, I mean, that's where you should spend your time, right? I mean, I think if you're just sort of dabbling, I mean, I guess if you want to have investments to sort of talk about because you think it's like just in the zeitgeist, then great, like invest in some funds, you know?

2:43:49Exactly, I was saying the same, more so it's just LP. Yeah. Either LP or like be deeply involved in the company. But yeah, exactly. Or if you just want to, you know, kind of learn about investing, like that's a great way to do it. I think if you want to actually be, you know, direct investing, angel investing, you know, I think actually, again, just sort of following like your true curiosity and the areas where, also where you can really make an impact. And that's why I think like with, you know, again, going back to Ryan and his team, I mean, they're amazing marketers. I mean, like best in class.

2:44:20And so I think that's where I would just sort of like encourage any celebrity to do what they're truly passionate about. And I think if you look at where things have really worked outside of Ryan, I mean, I think what Gwyneth has done with Goop. I mean, I think it's just these are just very natural to I mean, there are things in Goop Kitchen, which is another sort of spinoff from Goop. I mean, it's it's the food side of it. And it's again, it's just what's so natural to her. It's what people expect her to be sort of recommending an excellent ad, right? Yeah, yeah, yeah, totally. What are your thoughts around the evolution of LA's ecosystem on a personal level?

2:45:04I think my last five LA county investments were in the gundo and haven't done much outside of that. But I'm curious, are you disappointed? Are you accepting? Are you still excited? Do you see a lot more growth here? It seems like San Francisco has certainly sucked a lot of the oxygen out of LA. If you want to be. I had one portfolio company that was a consumer AI business that was in LA and just had to bail almost within two months just because of the talent. A$350 billion company popping up in San Francisco will do that. But yeah, what is your... Give us the temperature check on Los Angeles. Yeah, I mean, you know, look, I think there's no question.

2:45:55I think there's a sucking sound back to the Bay Area. And, you know, again, it's sort of like early days of any major technological shift. You know, I think that's naturally going to happen. It's Silicon Valley, you know. And so we actually opened an office up there for the first time last year. We've hired a new partner up there, team up there, and we're all spending a lot of time up there. I mean, for us, I mean, I love LA. I moved here. I did spend 10 years in Silicon Valley before moving here, but I moved here almost 20 years ago. I think LA is an incredible market, and I think having access to this market and really having a deep understanding of a lot of the industries and network here is super valuable.

2:46:34But the talent density from a, you know, just technology standpoint right now, Silicon Valley is just a super important place to have a strong network. And so we're, you know, we're spending time both places. We've also, I mean, we are spending time also in, you know, the gundo and in other, you know, on other things, which are certainly big growth areas in L.A. We also led an investment that we think is super exciting called Whatnot, which is LA-based company, fastest growing, largest live streaming shopping platform in the United States. So, I mean, we think that there will always be great companies here and continue to be, but there's definitely an AI talent density in the Bay Area.

2:47:18I think we were talking to Gary Vee and he told us to put it. Yeah, when Whatnot exits, it'll be like, LA's back. LA's back. And then people will be like, oh, I never lost faith. I never lost faith. That's hilarious. Well, we had Scopely also last year. Yeah, Scopely was fantastic. That's a pretty great story and continue to have a good theme here. Yeah. And Service Titan. Mountain. Space Gear. Yeah. SpaceX. Yeah, there's a couple. It's not the worst place in the country. We're doing okay. Yeah. But I think LA kind of like surprises people when you dig into the numbers as opposed to it's not as loud because Hollywood is the thing.

2:47:52It's the Hollywood town that happens to be above expectations in tech and hasn't made as much noise about it. Anyway, this has been fantastic. Thank you so much for hopping on. Yeah, super fun. Congratulations on the big win. We'll talk to you soon. Great to see you guys. Bye. Cheers. Have a good one. Any other breaking news that we need to cover in between our guests? No, let's bring it. No, we're good. No politics? Yes. We're so back. Let's talk about AG1. Let's talk about consumer packaged goods. Let's bring in Kat from AG1. How are you doing, Kat? Good to see you. Welcome to the show. Boom.

2:48:27Hi, hi. Jordi, I'm going to let you take the intro. Thank you so much for joining. What's happening? How are you? I'm great. How are you guys? We're doing great. It's Friday. We wish it was Monday. We like podcasting a lot here, so the weekends are always tough. But how are you busy? Sorry for your wind down. I know. Monday will be here soon enough. I know. I know. I'm sure you feel the same way. Why don't you give a quick intro, a brief history. We don't like to make this biography, but I would love to have you kind of give your background for the audience. Sure. I'm Kat. I'm the CEO of AG1. I have spent several decades in consumer.

2:49:09I started opening franchises around the world when I was a teenager and had incredible opportunities in restaurants, food and beverage, and franchising for the first 20 years of my career. By the time I was in my early 20s, I was in 18. So teenager, got to back up. What were you doing as a teenager? You just sort of sprinkled that in there, but I'm sure it's a good story. Yeah, I started working at Hooters restaurants as a hostess when I was 17, putting myself through school. I'm the oldest of three girls. We left my dad when I was nine, helped raise my sisters. And so restaurants are very commonplace for young folks who need to work to start to work.

2:49:53And I grew up in Florida. I used to pick my first real job after soccer refereeing was picking up cigarette butts at a nightclub in the morning. I would get there at like 6 a.m. and that was my job. So very familiar with the sort of restaurant grind. in and it certainly it allows you to bring it's easy to bring this sort of service culture mentality into startups when you're you know responding to like tickets online uh if you've got your start in restaurants yeah i mean that that that closeness right that proximity to the transaction to the customer experience if you're paying attention it never leaves your bones and your soul and so i stayed in that industry for many years i was with hooters for 15 years um helped grow that business around the world and then moved on to a Rourke Capital, a really big private equity firm and restaurants and franchising owned portfolio company called at the time called Focus Brands, took over Cinnabon when I was 30 years old as president to turn it around out of the Great Recession and then built that business into a billion dollar global revenue business, then built a licensing CPG division out of that parent company and eventually ran Focus Brands, the parent company, a couple billion, eight brands, 80 countries for four and a half years, so 10 years at Focus total.

2:51:13And all throughout that time, I was just becoming more passionate personally about nutrition and fitness. I had kids later in life. I'm 47. I have a five and seven-year-old. My mom had breast cancer when I had my second, my daughter. And so health became this personal obsession. And so while I was running these big mass market brands, some of them were sweet treats and fun for you, but some were better for you. Jamba, Moe's, Schlotzky's, McAllister's. I wanted more, like more bleeding edge of health and nutrition. So I started angel investing in better for you businesses quietly on the side. That eventually coincided with when I left Focus after a 10 year career running those global franchise businesses, I was introduced to the founder of then called Athletic Greens, AG1.

2:51:57I had been a customer for a few years and, you know, here we are. That was four years ago. Amazing. What was your personal process diligencing AG1, given that I'm sure you looked at hundreds of businesses, run these sort of massive organizations? I guess, what were the qualities? Because I think, if I remember correctly, the time you joined AG1 was when almost everybody was writing off direct-to-consumer as a category broadly. They just said it didn't work. The economics weren't there you were giving all of your you know uh you would eventually pivot to retail and then that was completely changed the business model and you couldn't really get to scale yeah there were a few things one i did similar diligence to some of our investors which is we were customers i was a customer for two years and when you have a personal experience with the product and real passion for me it was energy immunity gut health support it made a big difference for me as a as a busy professional mom.

2:52:54So I was a bit of a fangirl of the product itself. But that's just one layer. The second was I actually advised the founder for half a year before I joined. And so talk about diligence. I had the opportunity to get inside the business, get close, meet the team, understand product. And as every layer I listed, what I learned was even better than I could have expected. And I had a bit of an ego moment where I'm like, oh, this company is so much smaller, you know than the other ones that I've run is is this really what's right and best for me even though I love it and very quickly I got over that too this is actually a brilliant time to jump on a health and nutrition rocket ship that's better than anyone knows and has incredible upside so after advising I joined in November of 21 so what were the first challenges when you came in it sounds like a lot of things were going right but there's probably some area that you were focused on?

2:53:49Yeah. I mean, there was a ton going right on the outside. Everyone said, oh, you guys are crushing it. We did this fundraise, right? 120 million raised at a 1.3 billion post. It was still the end of the frothier era, like end of 21, beginning of 22. And the business had done 160 million that year in revenue and bootstrapped up to that point. So the founder It was an incredible capital allocator, incredible subscription business, 90 % subscription, growing at multiples, not percent. So that's what was good. That's what was great. Fresh round of capital, the only round of capital we ever raised, and a lot of tailwind in the category, in the industry, and a clear leadership position.

2:54:31What was wrong is that we had scaled so quickly, and we only had one manufacturing plant in New Zealand. So all of that green powder was being blended in New Zealand and most of our business had grown in the US. And so job one was scale the operations to support growth. I wanted to launch a few additional products sooner. I wanted to go into retail sooner. But we had no business doing anything new until we really built the infrastructure. And that was expanding manufacturing to the U.S., which we did immediately to support the U.S. business. So now New Zealand supports the rest of the world. Yeah.

2:55:16We own our source, source, highest quality ingredients globally, then blend in those countries and And certainly that's helped us in the tariff volatility moment. And so that was job one. The second was then it was very clear pretty early on that even though there were so much energy and tailwind and growth in the supplement category of wellness, and a lot of that was getting hyped up and promoted on podcasts and with influencers and creators. And we were certainly a part of early, very early in that trend. We were fortunate that some of our early customers became thought leaders and had famous podcasts.

2:55:54But it was very clear that science was going to be a critical part of marketing for this category where it historically had not been. So a few years in, first, shore up operations. Second, actually pull back marketing a bit and reinvest millions into double and sometimes triple blind, randomized placebo-controlled clinical trials, which take years to build a PhD team, to scope the trials, to conduct, to analyze, et cetera. And so those were the first two things. Operations, supply chain in particular, which includes like QA, maintain our NSF or sport third-party certifications. There's a lot of like high quality standard we needed to protect as we moved manufacturing and expanded it, which was a little scary as we put our product in the hands of other blending facilities, but the team did it beautifully and then make massive industry leading investments that would take an annoying amount of time to be completed and pay off.

2:56:47And then use those two things as the foundation. Once we got them in place to then innovate. So we wrapped up last year, 600 million in annual revenue, one product, one channel. Yes. John's going to hit the.

2:57:06Crazy. Congratulations. Yeah, the scale is just wild. Let the gong have its waves. Yes. Yes, we deserve it. Our customers deserve it. That's fantastic. But it is time, and it was time a year ago, to make some thoughtful, large, focused moves outside of that single product, single channel model. So just this last week, we announced a direct-to-national partnership with Costco. So we are now a multi-channel company. went straight from no retailers to all of the largest. I don't think people realize how significant winning in Costco is because I'm an investor in a company that expects Costco in the first, I think, 18 months to add$50 million in revenue to the business.

2:57:59It's extremely significant. It could easily be into the nine figures. Absolutely. it's great if you can just go from one channel to adding retail as this individual channel that you can really focus on. Yeah, if you're running a small CPG company, our recommendation is just go into Costco. And win in Costco. Just win in Costco. It's easy. It's that easy. For anyone out there with a$2 million CPG business, I'm curious if you think most of the issues in D2C and consumer brands broadly are just operational because like when I think of the structural versus structure when I think of yeah versus structural where it's like hey you have a good product maybe a good brand and you're just or or a brand that could be good yeah yeah and it's just a lack of of pure leadership abilities internally because like we're very close with the the the Ridge Wallet team and that's a business that like shouldn't be a multi hundred million dollar revenue business right it was the original product was a um you know in the early days the ltv was like the ltv was like the first the value of the first that's it the first purchase right and so you would think there's no way this could really scale or be massive and they have against against the odds um but i'm curious as you look at other d2c businesses how often are you thinking this could be great but it's just not like it's actually just like a operational uh um uh lack yeah actually you know it's interesting indeed to see is i i've seen both ends of the spectrum where something is actually demonstrating outsized growth but there's not a lot of there there underneath and it's a window of time before they hit their cack wall yep and they end up having a lot of unit economic challenges and it doesn't mean retail will solve those challenges in particular and then you see others that have beautiful products, incredible retention, but just haven't mastered the art of e-commerce marketing.

3:00:00And so for us, at least early on, I think people gave us way too much credit for being amazing D2C marketers. We were not. We had a phenomenal product that it didn't matter what type of marketing we use to have people learn about it. Any marketing is just awareness and trial. What keeps people ordering months and years, I've got over 500 customers that have been with us for a decade, a decade. They've been drinking AG1. That is not because a podcaster told them something. It's not because of a billboard or sponsoring a sports team. It's their experience. And so I don't think we can remove consistent quality and that compounding resulting both growth from those customers individually, but we call it the referral quotient.

3:00:49Our number one driver of customer acquisition, our post-purchase survey, how did you hear about us? How did you first hear about us? Friends and family. Number one, by a mile, bigger than podcasts, bigger than influencers, bigger than television. We're full funnel. We're pretty well distributed from a marketing strategy. I was just listening to you guys talk to Dana. She's amazing and their portfolio is phenomenal. And whether it's television or community and activations. It's all a part of our approach. But D2C really has to have that incredible experience to have its chance. I mean, we were rewarded for our focus, right?

3:01:24Pretty unusual to get to 600 million with a single product. And we weren't even on Amazon. This is just drinkag1.com. And that's unusual. And I do think that might be part of a challenge for some D2C businesses, is if they don't get traction or they're not seeing that retention or the LTV they want, then they go wide, like SKUs and offerings, and you get distracted. But if you can focus and really differentiate, then that traction is there, and the unit economics perform to the best that they can as the algorithms change and as the structure, to your point of structure, of digital marketing changes.

3:02:03On the other side, retail, yes, it's a revenue channel, but it's brutal, and your product is in the hands of someone else. And but it is marketing, right? It is eyeball. The shelf is a billboard. And so if you can do it well, I believe it is synergistic with D to C, not the maybe predicted or feared cannibalization that some might think going into retail. I'm glad we didn't go in early. This is our time. This is the right time for us. But we have an unusual business and that people stick with with AG1 for so long. Yeah. Do you have a reaction to the news this week or last week that Nike is going back on Amazon, sort of a pivot away from their direct consumer strategy?

3:02:49Yeah, I think Nike got a little too inspired by you guys. I think they might have been trying to copy you guys. They saw you putting up phenomenal numbers. They're like, AG1 was the last holdout. They're the last holdout. If they're there, it all bets are off. Yeah, exactly. Welcome back, Nike. Yeah, but it gets to. They seriously said, we're not going to do Amazon. We're going to focus on our own Nike.com and direct response marketing. And that's the other thing. You guys have blended. You guys do, I'm sure, the vast majority of your budget's around direct response. But when you came on board, suddenly the AG1 brand was aspirational.

3:03:23Totally. It was like, you know, cool. And it was leveled up. It felt like almost like fashion. And that's been a critical part of the Nike brand is putting the celebrities front and making the athletes the heroes. And that kind of fell by the wayside during the direct response era. Hopefully, they're getting back to it. People were talking about Caitlin Clark is signed with Nike, but most people don't even know it because the particular shoe that she wears is like a limited release. And so they've gone this like very odd strategy. So just general lessons from Nike or reactions to the relationship with Amazon there.

3:03:52I mean, look, it's similar to us. And may we be a Nike one day. What a phenomenal brand. Seems like you're on the way. But they've had their bumps as all scaled companies do. And we, several years ago, as part of that pullback in that lower funnel marketing and investment in human clinical trials to raise the standard in research, we also moved the marketing spend we protected to more upper funnel, right? Building brand, brought an incredible CMO recently, former CMO of Yeti. He's a creative genius and really deeply believes in community. He always says niche to be mass, like go to your people, be deep, be there for them.

3:04:28And we have a lot more ground game happening now. And so that investment in full funnel marketing, you get a greater return if you're in workplaces, including the world's largest e-commerce retailer being Amazon. So for us, and I know this was a part of Nike and even some other brands calculus and either going off or never being on. For us, there was so much tailwind in organic search and growth. We didn't need to be there to grow the way we grew, and we could barely keep up with the growth that we had. It got to a point where three things changed for us, and I can't speak for Nike, but I'm assuming this is a piece of the pie.

3:05:07One, there was a growth of counterfeits and resellers, and to protect our brand at scale, and this isn't a shoe. It's something people put in their bodies. That's a safety issue. That's a brand issue. So when part of our thinking of getting back on was being the programs that Amazon in particular, transparency and others that they put in place to take down either those fraudulent products, people infringing on IP or unauthorized resellers. That was critical for us. We would not have come back on in the way that we did as in as big of a way if those programs were not in place for this category in particular.

3:05:48It's about quality. We added, we had to add holograms to our box with randomized unique numbers on all of our packaging. We did this pre-Amazon because the, you know, it wasn't just Amazon, all the marketplaces. There were, I'm talking real counterfeits, like weird typos and Grinch green powder, dark green powder, getting caught by customs. And imagine getting a phone call from your general counsel. And she's like, we just got a call from the feds, customers. something that says nse for sport so you're being you're being a very you know like a politician about this but when i talk to portfolio companies in the supplement space about amazon it is just their number one uh one of their biggest pain points and and just a massive source of of frustration around counterfeiting around you know all the sort of issues that you're when we got that gone it was gone yeah gone and um now we're meaningfully scaled business right so um they've they really this is not being diplomatic i had a ton of risks to my brand and it is now gone on that platform that's those are the facts um the other issue was that because the resellers were there even if it was real product being resold i couldn't tell you how old it was the shelf life, the stability.

3:07:09So I still had a quality concern. But the as big of an issue for e commerce marketing was it was driving my ratings into the ground. So our brand ratings on Amazon were awful. And so if you would search AG one on Google, you'd get these great ratings from our site or other sources. And then Amazon, who starts three star? I mean, it's horrific. And so as soon as we got back on, resellers are gone, counterfeits are gone. We have a limited product assortment there so we can be there meaningfully, but keep D2C as the place for expansive SKUs, ratings went through the roof because customers are having a better experience.

3:07:44So it's search. It's about search and discovery as well for us. Where is your team getting the most leverage from AI? There was news this week from Meta talking about wanting to basically help advertisers actually generate ads. Dylan Patel was talking about that. And in general, you know, when you talk to brands, like the number one thing holding back their growth oftentimes and efficiency is creative. So it's very exciting. We're finally there with VO3 where it's at least a drop-in replacement for some B-roll or stock footage that you might pull off the shelf. So, yeah, that's a fascinating question.

3:08:22God, the technology is unbelievable. It's unreal. For us, the first place AI showed up in the business is our customer service and customer happiness. We partnered with Sierra. We were very early with them and have no affiliation with them otherwise being a happy customer. And they partnered with us to craft what we needed. And what we got out of efficiencies, we redeployed into human resources for concierge programs, for loyalty programs, which we were actually very late to develop as a subscription business. So I used – there was a financial benefit. There was a speed benefit for like we learned that 50 percent of our customer inquiries were things we already knew this.

3:09:03We just didn't have the solve that Sierra helped us with. 50 percent were just manage my account. You know, like that stuff can be odd. That's prime for AI. So for us, AI was first there. Then we have tools that our team or supply chain team have brought in on transparency and tracing for ingredient sourcing since we own our sourcing. And so it's really helped us track and monitor and enable all because we have pretty rare high standard specifications of our ingredients and have unusual levels of testing of ingredients. And so to manage that through the supply chain has been very manual. And so AI has helped us there.

3:09:41Then to your point, marketing. We have some interesting thoughts on AI. We are using it for, as you said, digital background, digital B-roll, anything that isn't supposed to be real product and real human. Because the supplement space has both good actors and not, we are holding ourselves to a standard of if we're putting a person on camera or if we're showing our product, right or wrong, it's going to be real. but everything around it, the enablement, the editing, we are bringing every AI tool we can to make that more creative and more efficient. Yeah, we were talking to a food photographer earlier today who was very nervous about artificial intelligence and AI image generation but was hoping that some of the rules around the way food is photographed from the FCC.

3:10:34Yeah, marketed because you've seen those famous pictures of like, here's the burger. I wasn't going to say the name of the company that does it, but here's the photo and here's the actual burger and they don't look anything alike. And you can imagine that being a big issue, even if AI can be used in the post process. There's probably still value to showing the customer exactly what they're actually getting. Last question from my side for now. Now, do you think that people should be significantly more concerned with the supplements that they choose to put in their bodies? There's this, obviously, in the Silicon Valley, the tech world, people will take any supplement if it's like, hey, this is going to make you 1 % more focused today.

3:11:17They're like, I will buy this from some sketchy website and take it today. And, you know, you keep coming back to this sort of like traceability testing, you know. We see this with the creatine gummies. Yeah, well, that's an issue too, but there's no creatine in it. But I mean, if you test supplements, it's like, yeah, you could be supplementing lead not knowing it, you know. And so I think that not every supplement company has the same standards. And I don't think people fully appreciate the potential consequences of that today. Yeah. Yeah, you know, the very fast answer is yes, people should be incredibly discerning while there are truly so many companies doing the good work.

3:11:57Third party certifications, publishing certificates of analysis like us on their site, all kinds of third party labs. there still are so many who, whether intentionally or not, in the mildest form of the issue, what is on the label is not what's in the product, right? So you think you're supplementing with X amount of this ingredient and whether it degrades because of the shelf life and they are blissfully unaware or their manufacturer pulled one over on them and they are unaware, I would say irresponsible in those cases if you're going to sell a product, that could be happening. And in that case, it's just you're not getting what you pay for.

3:12:37On the other side, to your point, there could be ingredients that you don't want or don't know are in there. And so what I would encourage people to do, whether they're in Silicon Valley or anywhere, is recognize this is a huge industry and where there is growth, there will be both good and bad actors. And the great news is there are ways to know who and what to trust. You got to look for third-party certifications and don't just take the words that someone types third-party certified what third party yeah nsf for sport is the gold standard they're incredibly expensive and a pain in our asses because if i change one thing in this formula it has to go back through their process through shelf life testing to prove what's on the label is what is in that bag yeah you actually have to go you have to really verify too i i helped start a company in the water filtration space and we went through the full nsf certification process it costs hundreds of thousands of dollars and then we have competitors that will put the nsf badge all over their website and they're like actively getting sued by the nsf or in legal battles but they just keep it up because it increases conversion rate but if you go that's right surface it's like there's just nothing that's right i think the other piece to for for people to think about too is this is unfortunate, another layer of confusion is you also then have people doing what you just described, starting their own labs.

3:13:57And then they are testing product with some unknown methodology and then creating viral content around whatever they find. And that just makes it harder for the customer to know. So I would just say this is not, and anything you put in your body should not just be a listen to someone and click and buy and put it in your body. Go to the website, look at the ingredients section, look at the research section. Do those supplement companies have any research on human beings at all? And I mean, you can imagine, you know, because you've got investments in the industry, having four double-blind randomized placebo-controlled clinical trials with over 100 people proving the same outcomes of supporting gut health, 10X gut bacteria, beneficial gut bacteria covering nutrient gaps in four different populations, that's not cheap.

3:14:48and um but we have to right we're the leader we don't make claims we can't substantiate i could have done two studies and that's enough to say clinically back we did four right we're just we we have to lead because there is so much to be questioned with those in the middle and i will offer if anyone is starting a supplement business and if they genuinely want to know how to do this right i will help them my team will help them it is bad for the industry for people who don't take this stuff seriously. That's amazing. Well, thank you so much for hopping on. This was fantastic. Congratulations on the massive revenue numbers, all the progress, classic overnight success.

3:15:26We'd love to see it. Yep. Come back on next time you have news. This was super fun. Love to. Love to talk to you soon. Thank you guys. Great to see you. See ya. We are shifting gears over to the supersonic world. We have Blake Scholl from Boom Supersonic. Is the speed limit lifted? Are we Post speed limit, I want to go 2 ,000 miles an hour tomorrow. There he is. Smiling. He's smiling, so something must be good happening. What happened today? We've been live for the last few hours, so. We got a text message. We got to have you on. Yeah, nothing big. We just actually broke the sound barrier today permanently because the band is reversed.

3:16:04No way. That's amazing. John's going to hit the gong. That's for you, Blake. Congratulations. We've been hoping for this for a long time. So the rules are going to change, and you can go supersonic as fast as you want, so long as the sound is a gong. Yep, that's great. I would happily let gongs ring upon the fields of America. From C to shining C. I'd love to see it. I mean, I know this news is breaking, but how did you find out about this? Was there expectation that this was going to happen? who are the folks who have been working on this and then break down what actually happened. Yeah. Well, I mean, in a certain sense, people have been working on this for decades, since 1973 when it first went into place.

3:16:53And we've been working on it very intensely since February 10th, which is when we announced we could fly a supersonic without an audible boom. I think it was that night I left LA where our test flights were and went to DC. And by the way, horrible flight would be much better supersonic it was like like left at 11 p.m got it at like 4 30 a.m it's awful it should totally be supersonic and then by the time i landed i had an invitation to the west wing so um the uh so we you know since then uh what we did what to me felt like a very long frustratingly slow campaign but what everybody tells me in dc is actually supersonic So a bipartisan bill dropped three weeks ago in the House and the Senate called the Supersonic Aviation Modernization Act, which, by the way, I think we should still pass because that locks in all of today's goodness.

3:17:51And, you know, and we just sort of continue to educate people. By the way, we don't have to have the boom anymore. By the way, not all booms are created equal. Not all are bad. And that's something a lot of people can get excited about. So, I mean, I know a lot of the technology is about flying at very specific speeds at specific altitudes, doing all the math to make sure that the boom bounces back into the atmosphere or something like that. What kind of speeds does that actually work at? What are we expecting? Because supersonic goes well past. It goes basically infinite, right? Right. Yeah, exactly.

3:18:26So there are two ways you can kind of tackle sonic boom. And one is what you're just describing. Make it go away entirely by making it make a U-turn and go up to space. And the great thing about that is there's nothing to argue about whether it's too loud. The less great thing is it only works up to about Mach 1.3. You can go about 50 % faster, but you can't go 100 % faster or 200 % faster. Put that in terms of like a commercial airliner. If I fly at like 500, 600 miles an hour, you're talking about like 700, 800 maybe? Yeah, let's talk about in terms of flight times. So with a 50 % speed up, which is what's achievable routinely, you can leave New York at 9 a.m.

3:19:07and be in San Francisco at 9.30 a.m. So it's a three and a half hour flight. Okay, wow. Which is like, that's a W, right? Yeah, that's massive. Now, if you want to go higher than that, faster than that, you have to convince yourself that some way or other the boom is not a problem. Yeah, okay. And I don't think – people don't agree about the answer to that. I want to get out there, fly the airplane. I want to hear some real booms. I've heard a bunch of booms in Mojave, by the way, we were doing our test flats. Booms are commonplace down in the Mojave Desert. Military makes them all the time. And there's some that kind of get your attention.

3:19:40You're like, wow, we probably shouldn't do that. At least not at nighttime. At most, that's a daytime thing. You know, but there are others that like sound like somebody like dropped a book on the floor. And, you know, and those are, you know, so think of it as like the wake of a boat. The bigger the boat, the bigger the wake. The closer the boat, the more the wake you feel. And so if you have a right size boat at a sufficiently far away distance, it really ought to be considered a non-issue. So I think we're still kind of taking in exactly what's the language of the EO, but my expectation is that this will be a one-two punch where if there's no boom, there's nothing to argue about.

3:20:18If there is a boom, there should be a way to sort of demonstrate that it was an acceptable one yeah yeah so uh specific flight corridors uh where people are opting in or something like that and yeah i mean that's it's just really the the you'll hear and not if you're in kind of boom full mode that the boom corridor is actually like 50 or 100 miles wide oh very wide i mean you know we can't build a railroad in this country yeah like you know the positive thing is like i imagine When you're traveling over oceans, we can have a little bit more flexibility. Yeah, there has never been a restriction over ocean.

3:20:57Fortunately, there's nothing to do there. So the way our first airplane, Overture, will do this is it'll fly boomless over land at like plus 50%. And then when we get to the coastline, you gun the throttles, and then you go to full 2x speed. Yeah, yeah, yeah. Yeah, and I'm sure that the economics change when you can do 1.3 Mach over land. because there's so much more over land, air, travel generally. So it's just a bigger market, so it justifies everything on the investment side. It does. And the economics are actually really interesting. So if you fly sort of lightly supersonic, it's worse for fuel burn.

3:21:30You can do so many more flights with the same airplane. You save a bunch of money on other things. And so on Overture, going right under the speed of sound and going boomless supersonic is actually the same dollars per seat mile. Right? Like the fuel slice gets a little bit bigger. Yeah, on the cost side. The cost side. But the revenue is going to be – I'd pay 10 times as much to get from New York to SF. Oh, yeah. Revenue is obviously better. I mean time is money. Save both. Ramp.com. Also, boom. Boom Supersonic. You can go to both. Time is money. Everyone's going to save you time. But if the boom comes down, reflects, bounces back up, and I'm in a plane there, is that going to break my plane's windows?

3:22:13Yeah. Okay. So a few people asked that question. It's a good question. The answer is Concorde did that for 27 years and nobody noticed. Oh, interesting. And one time they tried to test it. We know the pilots that were doing this. They sort of arranged for like a 747 to be here while Concorde is kind of passing overhead. And they had told everybody on the 747 that they might get to hear something. And they didn't hear anything and everybody was super disappointed. Okay. So it just doesn't really affect. Maybe you feel something that feels like turbulence. I don't know. It's like a nothing burger.

3:22:45But it's not going to blow the windows out, which is like the worry, of course. No. I mean, if that was going to happen, you know, Concorder knocked out every other airplane in the sky. Yeah, of course. Airplanes fly above. Yeah. Airplanes fly below. So Concorder is booming for 27 years over the Atlantic and like nobody got hurt. Yeah. Wild. Anything else? The way to look at this, by the way, is like the energy intensity and some of the physics of sonic boom are very similar to thunder. Right? You know, you could be like, oh my gosh, like, you know, could boom over the water, like, disturb the main tables of like, I don't know, the aquatic spotted owl or something.

3:23:21Sure. Well, anything that was going to be put out of commission by a sonic boom was long ago put out of commission by a thunderstorm. Sure. Yeah. Sure. Was there anything, was there anything to you that was surprising about the EO?

3:23:38What did you maybe not anticipate? I mean, I admit I haven't read it word for word yet. Yeah. I read the fact that he's been partied, and I need to calm down and read it. There's a lot of stuff in it that appears to not just say, great, you should be able to fly a supersonic, but advances R &D. I got to read all of it. I want to say what it says on Reddit. Yeah, yeah, yeah. I'm sure there's a ton of work to be done on the licensing and approval and everything that happens with certification. Some of those things might need to change or speed up. Some of them might need to stay in place if it's a safety issue, of course.

3:24:13But what's next for the company? What's the next milestone that you're tracking?

3:24:19Let's go. So we're building our first engine. Okay. And that thing is like 60 % the parts you're in in the manufacturing process. And we'll go run that bad boy later this year. The skeptics always say the startup can't build a jet engine from scratch. So we'll find out. and then we'll start building the first full-scale pre-production prototype Overture next year. It'll be roughly three years from where we're today to be back and there's a bird in the air. Last time we flew it looked like a fighter jet. Next time it's the Overture. It kind of looks like a Concorde and a 747 had a baby. It's got this little hump up at front.

3:25:01It's not a double-decker but it's a really unique airplane. Not yet. We have to get at the gold-plated 747. Those are really hot right now. Lots of people giving them to their friends. And so I could imagine handing these out to world leaders being pretty effective. I have to ask, did you get a chance to watch the rehearsal by Nathan Fielder? You're giving me a blank stare. He's probably too busy. So I'm assuming you're too locked in. This sounds like a pop culture thing of which I'm ignorant. Yeah. Well, the reason it's relevant is because this sort comedian Nathan Fielder has a show called The Rehearsal.

3:25:39He learns how to actually fly a commercial aircraft to demonstrate to the FAA some of the issues they have around safety. Specifically on pilot communication. He identifies that the vast majority of commercial air crashes that have happened over the last few decades have been the result of a captain making a mistake and the second in command not having the kind of energy or confidence to stand up to them because they're meeting for the first time. And so he dissects this in a bunch of hilarious ways. My main takeaway from your answer is that it's extremely bullish for Boom, that you have no idea.

3:26:18There's a certain class of hard tech founder. I always notice Scott Nolan, he'll repost their announcement like four days later because he's so in the trenches. that like, I'm like, Scott, like, like you, you know, you launched three days ago. Like you should be retweeting at least like a post because he's so offline. But anyway, I have one last question about being online. You've been learning from Elon about manufacturing online. You had an interesting back and forth where you were talking about manufacturing parts. I barely followed it. Can you explain what did you learn from Elon about manufacturing?

3:26:56And yeah, yeah. Break that interaction down. So Elon has this thing he talks about called the idiot index. Mm-hmm. It's a beautiful Elonism, right? Yeah. Where it's sort of a, you've got a final part and you divide it by the part cost. Like, it's a turbine blade. It costs, I don't know, a mil... A real-world example. We've got a set of turbine blades. It costs a million dollars for a set of turbine blades. What does the raw material cost? A thousand dollars. So what's the idiot index? A thousand X. Like the cost goes up a thousand fold in the manufacturing process. And so it's a sort of measure of like, what's the money efficiency of the manufacturing process?

3:27:33There's another thing that I think is actually even more important that we call the slacker index. And the slacker index is how long it takes to get something out of the supply chain divided by how long it actually takes to make it. Oh, so more time-based. Right? Like, so we were getting these 3D printed turbine blades for our first jet engine, And we go quote them in the stupid old aerospace supply chain. And the quote comes back and it's like, it's six months and a million dollars. And I was like, wow, okay, well, how long does it actually take to print the blades? Oh, like 24 hours. So what's going on the other 179 days?

3:28:09You're like waiting for, and I was like, oh, it must be that the machine is like really hard to get. And there are only a few. Well, no. Turns out the machines are off the shelf. They've got them in inventory. You can get one in two weeks. And by the way, what do they cost? Two million dollars. So for the price of two inches worth of blades, we were able to buy a 3D printer and print our own blades. And so what happened, it compounds from there. If engineers can only make a new blade every six months, they really work the blade design. Because they don't want to get it wrong. Because if they get it wrong, it takes six months to fix it.

3:28:41But if they print it the next day, then all of a sudden the rate of iteration goes up. And they're out of engineering hand-wringing and trying to get it right the first time goes away. You want low idiot index and you want low slacker index. I love it. I love it. Well, hopefully this EO unlocks a lot of that and we can bring down the index, bring down the, the, uh, the slacker index. Definitely had a high index. So I'm glad, I'm glad that's been a deep six. Well, we are one step closer to a supersonic podcast and I cannot wait. Yeah. I haven't forgotten what you're on. I know it's happening. It's happening.

3:29:18Well, congratulations enjoy the weekend hopefully you can advance the technology but also celebrate a little bit and we will talk to you soon Blake thanks so much for hopping on jumping on and congratulations bye let's give a little air horn for supersonic flight yeah I love it I can't wait to hop on a boom supersonic take it to an exotic locale stay at a wander find your happy place find your happy place book a wander with inspiring views hotel great amenities dreamy beds top tier cleaning and 24-7 concierge service. It's a vacation home, but better, folks. Any other posts we need to go through?

3:29:54We talked about Polymarket partnering with X. That is fantastic news. Shane Copeland has the news. He says, proud to announce Polymarket's partnership with X and XAI as their official prediction market partner. I'm very interested to see how this actually integrates. Obviously, Polymarket's gotten very good at sharing images with a very catchy image and then the market, not just the, the Polymarket screenshots have been kind of a staple of X since the election, but they've gotten better at that. I'm wondering if we're going to see the data actually hydrate live like community notes so you can attach a Polymarket to a post and it will stay updated.

3:30:31That would be very cool. So you could be tracking it on the post. It'll all be very interesting to see where that goes. What else should we talk about today before we leave? I think we covered a lot of it. Scott Belsky is coming on the show next week. Very excited for that. He has a post that I was just enjoying because it's about product leadership. He said, underrated characteristic of top product leaders, compulsiveness, obsessively capturing decisions, ideas that need to be taken, baked that would otherwise be fleeting. This is how I felt about reading this post. Incessantly iterating over options in mind, morning till sleep.

3:31:09This is what the golden retriever does. The golden retriever incessantly iterates over options. Should I get the ball? Should I get the snack? Should I roll over? Try and get a belly rub? High energy. High energy, trying to be efficient. Not always getting it right, but doing their best. World-class products are crafted via compulsion. There's no better example than the golden retriever. Golden retriever never says, nah, I don't want to get the ball today. Yeah, no, never. So go out, get that ball, enjoy. Be a golden retriever this weekend. Fantastic week timeline in turmoil lots of chaos in tech, but we're still here We had incredible IPO window is open.

3:31:49Yep We had some drama some drama on the timeline, but who knows it's a new week We might see completely new things and the and the relentless march of techno capitalism and artificial intelligence well, it marches on and And as always, we're Newsmaxing. Ben is standing behind the camera right now giving us a threatening look. He says, go leave us a five-star review. Yeah, he does. Spotify, Apple Podcasts. Yes, please. It does help the show a lot. Thank you. And we hope you all have a fantastic weekend. We'll see you Monday. See you Monday. Goodbye. Market clearing order inbound.