2025 in Review, Cursor Acquires Graphite, TikTok's $50B Profit | Michael Truell & Merrill Lutsky, Pranav Myana, Anna Goldie, Edward Mehr

26:19A lot of coding, as we know today, will be automated. And we think very similarly about the ways in which code writing will change, but also the ways in which teams collaborating will change. And Graphite has focused really intensely on the team collaboration problem and how you help people review each other's code. We focus really intensely on the single player experience of how you develop software as an individual programmer. And so we're excited to kind of marry the two together and pull the cross. Michael, I would love to get a year in review for Cursor or even more broadly just the state of software development.

26:58Quantitatively, qualitatively, how can you explain the way How has the way writing software changed in 2025? It's changed in a big way. I think at the highest level, agents became useful in a professional setting. And that really expanded the demand in the market. And I think we're still early. Like I think it can be really easy to underrate just how far away coding is from being automated. and still building professional software takes so many people over such a long period of time and there's lots of issues we need to contend with as AI coding becomes deployed more broadly but it was a big year where you went from being able to just you know ask some quick questions to an AI about your code base and how it kind of helped you out with the next 30 seconds to five minutes of your work to being able to hand off whole tasks to an AI and have a dual hand.

28:04And Meryl, the shape of graphite, obviously, we know that you're growing quickly. How did you perceive the changes that happened this year? If you look back on 2025, obviously, this deal is going to be something you remember forever. But more precisely, how do you think that the developer experience changed this year? Yeah, every time I'd catch up with Meryl, he'd be like, there's a lot of code to review. So we're busy. And that's Mike in a big way is Michael's fault. No, that's the funny part about this. I think Cursor has just so dramatically changed the rate at which we can build features and how much code engineers are able to generate.

28:47And what's happened consistently, the bottleneck has now just shifted to the rest of the process, what we call the outer loop where now we need tooling to help every team review and validate and merge changes at the rate that you can now generate them with tools like cursor and that was basically our our 2025 has been how do we both apply ai to this problem how do we use like more you know more traditional deterministic methods like merge queues and stack prs and other workflows and tools to make that process more efficient but you know how do we just unblock this bottleneck that that is now is kind of like preventing teams from really realizing the true potential of tools like Cursor.

29:28And that's been our mission this entire year pretty much. And part of why I think we're so excited for this partnership is that now you can put the surfaces where you write code and where you review and validate and merge it together and just have that seamlessly integrated. Like you shouldn't have to like jump to a different tool for your editor, for code review, for your PRs, for CI. All of this should just be one nicely integrated surface. And that's kind of always been the dream for Graphite and our vision. And I think now that can become a reality. How are you guys thinking about the integration process and how Graphite fits into the sort of cursor platform family?

30:12I think a good first step would be like maybe a walkway between the offices in New York. A skyway? A skyway. Yeah, a skyway. We've talked about the skybridge. you know, so you can, but, uh, yeah, we'll, we'll put a zip line over Broadway. Yeah. So that people can commute back and forth. Um, no, I think, I think that there's, there's some really obvious low hanging fruit of things that you'll see us roll out, um, in the coming months together. And then there's a long tail of like even more ambitious ideas that we have, um, that are, that are in the works. But, um, immediately, like I remember earlier this year, uh, a few of us on the, the graph fight team were up in Toronto, uh, meeting with, uh, with Toby and some of the Shopify engineering leaders.

30:53And they're one of our biggest customers and close partners. And the biggest ask that they had for us was, how do we get context from our IDE or from tooling where we're writing code with AI into pull requests and have that be seamless and have the same chat history, have the agent logs and everything show up in the PR and be able to then call out to the agent to fix things again. And we were like, huh, that's an interesting problem. Maybe we should think about working with Cursor on this. And I think that's kind of the most obvious thing that we can do to start with. And then we can build from there on many of the other ways that we can kind of connect all those surfaces together and have the agent be able to help you all the way through from the moment that you generate the code to the moment that it's merged in and out to production.

31:41Yeah, I'd second that. There are going to be a bunch of opportunities for some quick ways in which we can make the experience of working together and graphite and cursor better. But then the big thing will be going heads down on a much bigger build together where we'll have more to share late in 2026. Michael, I'd love to get an update on how you're thinking about just growth opportunities as segmented by sort of like scale of the customer. We've read some, you know, like the AI, like the models are great. The tech is amazing. There's still some odd resistance to adopting AI in certain enterprises.

32:24We're not at 100 % penetration with these tools. Is there more opportunity in the near term in large enterprises and transforming the way those businesses work? Or is it just the ground game of going getting every SMB online? How are you thinking about growth in 2026, 2027? We've been shocked by the demand across the board. And so on the mid-market and smaller companies side of things and the self-serve side of things broadly, there are all these rules of thumb for when the growth of that business tops out in developer tools or in other comparable markets. And the thing that's just shocked us and shocked all of our investors is that the growth has been compounding really consistently at the same growth rate over the course of many years into the revenue scale that we are now.

33:18And that just continues on a bit. So yeah, is that sort of like an IT spend thing where like a small and medium, a small and medium business might just say like, okay, we don't want to spend 10 % of revenue on IT spend or technology. And maybe the new paradigm is actually helping with so much growth that they're able to underwrite a larger investment in technology. Is that what you're seeing? It comes from more people using Cursor and people using Cursor Keeper. Both ARPU and how much we're helping people and how much code we're writing for people. And then also the number of people using Cursor within companies and across companies, which has consistently been growing.

33:56And one big change for us this year is just the upmarket motion has developed faster than almost any upmarket motion has ever, where at this point 64 % of the Fortune 500 pay us in some way. And it's both penetration into digital native companies. So for instance, NVIDIA is a big customer wall-to-wall, Adobe, Uber, Salesforce, which I think in a public earnings announcement recently mentioned that they're seeing over 30 % productivity increase in deploying customers. And it's also companies that aren't digital native too. It's shocking how many companies are software companies. And so Starbucks, BWC, Hilton, companies like this are deep customers.

34:39Where are both of you seeing any resistance to adopting AI, specifically in software engineering? Are there any, I'm thinking of like the Japanese soldier on the island, you know, that doesn't know the war ended. Are you seeing anybody trapped left on an island?

34:58I think that, well, I think that this is kind of true of how AI tools are bought broadly. but it's really important. I think the way you procure these tools is a little bit different, where the difference between having the best product and the third best product from some incumbent that's now six months old is really, really big. And then user behavior needs to change and the way in which your team works needs to change. And so you need to teach people within companies how to work differently. And so we've seen a lot of success in not just pulling out the tool, but also teaching folks within companies too.

35:40But it's really spanned across all types of development. I think that there's still some languages where there's room for improvement in how much AI can help folks, especially some super legacy languages. but I think where there's resistance, it's mostly a problem of teaching and kind of learning new habits. That makes sense. What kind of advice are you giving somebody that's maybe in high school or college that wants to get into software engineering but is concerned about just the overall rate of change and how good the products and models are getting? I think it's actually a really exciting time to get into building things on computers.

36:21And probably on a relative basis, is especially exciting for people who are new and entering the field, just because it's just quick for them to pick up new habits. And so I tell them, yeah, to experiment with the tools, to try things out broadly. And also, I mean, working on a solo project by yourself is very different from building a giant piece of software with hundreds of other people, so getting exposure to a real professional development environment, too, I think is helpful learning. Yeah. And it seems more and more obvious that there's just so much software that needs to be built. I mean, we've experienced this year where we have built a software tool internally to help us run and run the entire show.

37:05And we are a business that even three years ago, we wouldn't have been hiring a software engineer because we would have either used off-the-shelf SaaS or would have just taken so much resources, it wouldn't have been worth it. So there's just so much to build. Yeah, it's almost trite now, especially in the Bay Area, to say software is important. And if anything, I feel like it's kind of reached a point in technological maturity where you don't even really think of software as technology. Just think of it as like, oh, it's a website that someone built. But yeah, I mean, it's shocking how much progress across the world really is just bottlenecked by building things on computers.

37:44You talk to people in AI research, what's the bottleneck to making the models better? There's a few, but one of the biggest ones is just building better infrastructure and just the speed at which researchers can code. And it's for other areas too. For instance, I worked at a biotech company at one point, and one of the big bottlenecks making progress there was analyzing data and picking the next set of chemicals that people were going to try out. And it was dealing with crappy software from off-the-shelf vendors or building a whole software team to build it yourself. And so, yeah, I think that it's this amazing lever on productivity and a bunch of different the verticals.

38:17What are the research paths that excite you the most or that you think might be underrated? An example would be like when we talked to Sholto during the Cloud 4.5 launch and he was talking about not image processing, not image generation, but image processing and that actually makes a lot of sense because a real software engineer needs to look at the webpage that they designed and then interpret that and understand the code that they write, how it feeds into the result. Are there any areas of research or less obvious, like it's not just a coding model, research paths that you're particularly excited about in 2026?

39:02Yeah, I think that the capability gains we've seen in our space, actually there's been a lot of details to figure out, but there have been a few really big ideas that have worked, just have been levers that people have pulled on continuously. And so pre-training is one that's been talked about a lot, you know, like taking models, scaling them up, training them on internet scale data. Another big one that's been really important for R-Space is curating a set of games for the models to play. So for us, that means collecting a set of, or, you know, in R-Space, that means collecting a set of code bases, writing out tasks, having a set of tests to test if the model collection solved a task when it writes a PR.

39:43And the big AI companies have done this really well of getting thousands, tens of thousands of really hard games for the model to play and then teaching the model to play those games. And in turn, the model then gets better at programming. And so I think that there's a bunch more juice to squeeze both from pre-training and then RL with this verifiable reward. But I think that there's going to be some new big ideas that are needed to really get to a a place where you can hand off end to end most of the professional development tasks you do. Does that make you especially excited about some of the Neo labs that are, I would say, fairly controversial at this point?

40:27Because on one hand, it feels like we need new ideas. But on the other hand, it's like, it's a lot of money. It's a lot of money. And it's a lot of money. And it's unclear if you just go and try to compete with big labs. It's always scary when there's a lot of funding, not a lot of revenue. Yeah. I think Cursor is actually doing a great job at this with their own models internally. Yeah, I was about to ask, do you think that you're going to become more of a lab over time? No. I mean, what we want to do is we want to build the best way to code with AI. We have lots of amazing partners that we're really excited to continue working with over the course of the next few years that are working on things that look like AGI.

41:06AI, we've ever since the start of the company, we've kind of picked our spot where we are going to do our own modeling work. And those have looked different from the places where the big AI companies, the big labs do their modeling work. And so for instance, like all our tap models, like the things that are looking at what you've done in the editor, breaking the next of the things you're going to do, those are our own models. We're on like the sixth generation model there. They learn continuously by looking at what people are doing within Perser and figuring out how they can get better. And so I am really excited for us to invest a bunch more in research, do lots more ambitious stuff, but it'll kind of be a little in a little bit of a different direction from what some of these these labs might do.

41:48And so, for instance, we're really excited to build models that are some of the most capable in the world at programming, the most capable in the world of programming, but are very fast, too. And we think that over the course of the next couple of years, or over the course of the next year, agent usage and coding is going to kind of bifurcate into in the loop or completely async. Where in the loop, you're sitting down, you're like working with the agent in a pair programming way. You want it to be very fast and extremely smart. And then async is going to be, you're talking to a colleague, you just hand off something end to end.

42:19And you want it to be definitely, definitely, definitely correct. And I think that very soon, we would like to play a really big part in making that human in the loop experience excellent. excellent. And I think that there's a lot of useful modeling work to do there. So very cool. How do you think about the X for Y meme? I feel like cursor has been very successful in that you, there's a certain like rite of passage in Silicon Valley where once you become like Uber for X, it's like, if you're the Uber, it's a good place to be. And cursor for dogs, cursor for bio, cursor for travel this has become a meme is there a where is the line for what cursor will do and what cursor will not do so when I talk to the anduril folks they'll say well the anduril of submarines is anduril but if I said the anduril of stoves or the anduril of you know of you know watches it's like okay I don't even know what that means that's fine I'm not gonna build that like you actually can go build that company where's the where's the line of like what cursor will do over time versus what's something that like where you might like the cursor for X model, but it's not on your roadmap?

43:27Well, we'd like to make it possible for anyone to build anything they'd like on a computer. And, you know, another way of putting that is we'd like to automate coding. Sure. And half of that's a model problem. Half of that's a product problem. And we want to do deep, important work across. And yeah, so squarely, squarely focused on helping you build things on computers. And for us, that means an intense focus on engineers. And then increasingly, the fold's going to expand, too, where lots of technically light personas like designers and product people, they also work with Cursor, too. And what we're excited about is that that fold can broaden as the product gets better, as the technology matures.

44:05But I am really excited actually for quote unquote, cursor for X's to exist in other spaces. And when we started the company, we kind of thought that like this, the shape of company where you pick an area of knowledge work and you kind of make the cockpit where that knowledge work happens, like the product that people daily drive for that form of knowledge work. You make it, you shape it to where the tech's going, you make it great for where AI is. And then you also see where AI is helping people and where it's not helping people. and you use that to make the underlying models better, both by doing a little bit of your own, also by working with partners.

44:40That like kind of shape of company, we were really excited about. And I think it's going to exist in all sorts of different areas of knowledge work, whether it be mechanical engineering or writing or science, biological science and other places. Yeah. Is graphite the cursor for pull request? Meryl, did you ever think about that positioning? Because I've done, I literally, I think I've done 250 ad reads for graphite and I've never said, hey, it's the cursor for pull requests. We said it's code review for the age of AI, of course. But do you think that you fit neatly into that framing of the cockpit where the work happens that you improve or is there something that's like a different positioning?

45:24And I'm wondering how that might change over the next Yeah, I think one of our investors, Gil Carradrum, has this framework that we reference a lot where you're either building a dashboard company or a pipes company in B2B. If you're a dashboard company, you have to be something where one type of user, every single day at work, they're coming in and doing a certain task, and that's just their home screen. Or you want to be a pipes company where it's like you configure it, you set it and forget it, and it just does throughput and prints money for you. And we've always thought about Graphite as a dashboard.

46:02We've said we want to be the home screen for developers. We want to be the place where everyone comes in and checks, like, where are my code changes in flight? What do I have to do in order to unblock my team and keep everything moving? And I think that's one of the things that's so exciting about this partnership is that now you really can be the one dashboard for engineering. If you want to write code, if you want to build something, if you want to move your changes through the rest of the process, that can all happen on one nicely integrated surface now and really make that vision a reality.

46:38Yeah, that makes sense. Well, I'm so excited for both teams. I'm incredibly excited for you, Meryl, and the whole team at Graphite. As a Graphite customer, starting at the age of 25 to a partner now, it's been incredible to see the journey. And you guys pairing up just makes so much sense. And it's been a massive year for you both. I'm sure 2026 will be even bigger. And thank you both for joining to celebrate with us. We should hit the gong again for you both. Yeah, I think this is the gong-worthy moment. Definitely, definitely. And I'm sure the two of you guys won't have much of a holiday, but we hope you can enjoy at least a little downtime with friends and family and can't wait for next year.

47:25Yeah, we'll talk to you soon. Thank you so much. Thank you. You're welcome. Goodbye. Incredible.

47:34What a great partnership. That feels like such a great match made in heaven. Absolutely. Absolutely. Let's go over to... Dana White and the Meta board. This is a match made in heaven. Very funny on multiple levels. Let's play this clip while the team is pulling it up. Let me tell you about Adio, the AI native CRM. Adio builds, scales, and grows your company to the next level. Let's go. Never stop clapping. And now AI. Let's play this. Have you gotten into AI yet? We're dabbling. Okay. Dabbling. Yes, sir. So Meta AI. I'm on the board for Meta. i just got back from the meta board meeting so good zuckerberg who was a brilliant gangster this guy gangster these people who who try to talk about him and everything else i'm so blown away and impressed by this guy he's an animal i agree with that and putting all the chips in on ai we just hired like 10 kids that are aged 22 to 28 the average salary is like 65 million dollars that these kids are making that.

48:39It's so funny that this is the final leap. But AI, you hear a lot of negativity about AI? Everyone's wondering. There's way more positives about AI than negative. So you start looking at AI and getting into it and asking AI, how do I build my business? How do I, you know, and it'll start giving you some ideas. And hold on. Is he saying$65 million is the average salary per year? I think so. I mean, I think of a salary as an annual thing. so 10 times that's that's insane and noah in the chat says meta engineers with a 600k salary so yeah yeah keep playing this from here to tulsa oklahoma you'd have to go on a map and you'd have to lay out you know your route and all you got to do the same thing for your business map out your route for 26.

49:30when i first saw this i thought he was saying like ai will be able to get you directions and i was like google maps can do that okay so when i see this When I see this, I just think entrepreneurs can get stuck in a loop of just wanting to meet and talk with people and get ideas and get strategies and learn. And AI is really good at that. You can say, I have an e-commerce company. I want to grow. What should I do? It'll give you a bunch of ideas. And it just shows how worthless a lot of ideas are and how important execution is. Some ideas are priceless, right? It's like you want to execute on the right ideas, but oftentimes to find the right ideas, you've got to try a bunch of stuff.

50:06Yeah. And so AI is at the point where it can give you the perfect strategy, the perfect playbook, even if it's like kind of the average playbook across business textbooks and blogs and posts and things like that. But in the end, you just still got to go do the work. That's the hard part. Yeah. I mean, I still think there's like he is using he's using a metaphor. I think he's actually a pretty good communicator here because he's using a metaphor that people understand. It's like mapping technology, Google Maps for business, for answering other questions, unstructured questions. AI can tell you that.

50:42And if you think about before, you Google, okay, well my business needs a website. How do I set up a website for my business? Okay, I need to go to the store and get a book, web development for dummies. This was a thing. I remember that. Back in the 90s, it was like Java for dummies. Like you're gonna build, now it's like AI, obviously we just talked about this. And so he's right. He's delivering it in this sort of funny way, and he brings this crazy energy to the performance. But he is correct in the pitch, in this idea. He's actually correctly pitching personal super intelligence. And for a lot of people, that's exactly what they want.

51:19Now, he doesn't really address the fact that there's incredible competition from Anthropic and OpenAI and Google on this front. But that's not what he's addressing. He's addressing just the idea of like, is AI useful? And the guy is like, we've dabbled with it. We've used it for answering, you know, like doing subtitling basically. And I think what's under, so meta has what? Dana sort of pitches like the next level of like what's possible. Meta has something, I think three, three and a half to four billion monthly active users. And so I think in those board meetings, you have to imagine they're saying like, yeah, there's a lot of competition.

52:00yeah chat gpt has a big user base yeah gemini has a big user base but we have four billion people that we can start distributing if we build a great model we can start distributing it through whatsapp through instagram through facebook through the meta ai app etc yeah i was listening to uh ben thompson this morning and uh he was doing app reviews like the the the review of the top paid apps and the top free apps. So the 2025 top paid apps. And this is wild. It's like, have you heard of any of these? I know. Hot schedules. Shadow rocket. It seems like. Procreate. Have you heard of any of these?

52:42Procreate. No, because I check the charts a lot. Skyview. I've heard of that. Tonal energy. Auto sleep. They're all like a couple dollars. And most people have not really heard of any of these. If they have, they're like, oh, yeah. Yeah, I use this for this one little thing or this is a niche thing. And then you go to the top free apps and it's like trillion dollar company, trillion dollar company, trillion dollar company. It's literally ChatGPT threads, Google, TikTok, WhatsApp, Instagram, YouTube, Google Maps, Gmail, Google Gemini. And so Ben's point was, if you like ChatGPT, yes, they are the number one app, but they should be scared because Google has one, two, three, four, five in the top 11 or something.

53:23like that and so the distribution is just so powerful and yeah the top that has that distribution so they're also a contender and they can stay in the game so the top the number 22 free app right now number 21 is instagram number 22 is whatnot number 23 is hbo max yeah and on the paid side currently 21 is threema secure messenger sounds like an even uh sus version that's insecure yeah sounds very insecure number two is pocket god which is a game that includes call of booty wait call of pocket god isn't that a nickname for agi they've agi has been solved this is just like a mobile game and then number 23 is jingle real motion shaker instrument that sounds like the ibeer app let's go it basically is you shake it you can play bells you know slay bells that kind of thing you know most of those if they're on the top paid app store charts they're probably making money and the software is probably developed in linear the system for modern software development linear streamlines work across the entire development cycle from roadmap to release what does casey neistat have for us casey neistat did a project with the metaquest 3 where he scanned his studio he says it's pretty rad you can walk around and look at stuff and get close i specifically did not clean the place before we scanned it it works what works on your phone and on the headset let's play this can we play this full screen too that'd be interesting i want to see the uh the full screen have you have you watched a lot of casey neistat have you seen his whole facility i've watched enough it's one of the coolest it's one of the coolest studio spaces yeah it's very inspiring from a production perspective because it's practical but it also has so much character that it tells you a story.

55:07And so even when he's just filming a little product review and he's making the seventh video of the month or year or whatever, you're brought into his world. You understand who Casey is. Every single one of those items tells a story and it's just really cool. So yeah, you can go hang out. You have a quest. We've done a demo of this feature, but we scanned like a very normal room. Should we give Tyler a challenge to actually get this up and running? Scan the Ultradone? I tried for, I don't know, maybe two months now. You've been able to do a couple experiences on the MetaQuest, but you couldn't record your own yet.

55:50I mean, I'm not sure if it's actually... I guess it is out that you can do it yourself. So I'm not saying scan the Ultradone. The GigaChat elf is so funny. I love the GigaChat elf. That's so good. So, yeah, I'm not proposing that you... You look so ridiculous. I'm not proposing that you scan the Ultradome. I'm proposing that you... Stop enjoying your jawline. He's on that Wild Roman again. He's on the Wild Roman. Lay off, Tyler. But what I'm proposing is that Casey Neistat shares a link there, horizon.meta.com slash world slash a bunch of numbers. and if you click that in the headset, I believe it takes you to that world.

56:33But how long do you think it'll take you to actually get into that world of that headset? Wow, you're old now. I don't like that one. I don't like this one. That's not fun. Nah, nah, nah, nah, nah. Hey, Pa. Hey, Pa. I like it. I would like to be still doing this when you look like that. Yeah. I enjoy it. But how long... Oh, yeah, yeah, yeah. Do the sad face. Do the sad face. The sad face is the funniest one. It's so funny. The jawline is crazy. Oh, there we go. Hey. It looks so real. It's so good. I don't like this one. You really look so sad. But what's wrong, Tyler? Cheer up. You know what will cheer you up?

57:18Privy. Privy makes it easy to build on crypto rails, securely spin up white-label wallets, signed transactions, and integrate on-chain infrastructure all through one simple API. The Economist is saying industrialists now list Gundo, along with Silicon Valley and in Tel Aviv within a triumvirate of the West. Triumvirate. Triumvirate. Triumvirate. It means three powerful pillars together. That's a new one for me. Learn something new every day. Yeah. We got to get you on the Anki Mobile flashcards. A little space repetition. Yeah, yeah, yeah. Exactly. Of the West's most important innovation hubs.

57:54Yeah. America's fight back against China starts in Los Angeles. It is real. I mean, it hasn't happened overnight, but the progress I think has been faster and more real than people expected. I think when Augustus was first posting in, what was it, 2023? So the next post, Fast Company did a profile on Augustus, and they actually referenced me. And they said two years ago in a widely viewed interview with the tech world chronicler, John Coogan that's me Dorico was jacked and tanned what triple glaze hit that glaze that's ridiculous the glaze is insane a high wattage presence at ease in his role as gundo super connector as Coogan describes him I did describe him as that they asked me for comment and that is a direct quote I do think he's the super connector and for a long time it was like if you are going to El Segundo like checking with Augustus ping him first he will get you hooked Do not step foot in El Segundo without checking in with the Don.

59:00Yeah, without bringing, you need to bring a case of White Monster as an offering. Yeah. And maybe some nicotine pouches to pay your respect. If you're a venture capitalist, I mean, there was a whole while where it was like VCs from all parts of the world. And then celebrities started being like, I got to figure out. They were taking a pilgrimage. Yeah, celebrities would go on the pilgrimage. I remember a lot of Gil went and then he posted a Zincan. That's right. Yeah. He spelled Zin wrong. I remember that. Oh, yeah. Z-I-N. Okay. So Augustus was quoting this picture from Erebus of saying, this is what rebuilding America does to you.

59:37This is Zane. This is Zane. Knox Metals. We interviewed him on - I mean, this transformation is insane. It's really - This guy is incredibly hard. It's really the frazzled Wojak who's just like strung out all black. I have a block of metal from Zane. Oh, no way, that's from him? Yeah, yeah, he gave me that at NYC. That's very cool. Very cool. Yeah. Yeah, he sacrificed his innocent self to rebuild our country. Well, we thank him for it. Forever grateful. But the funny thing about this fast company is that so first they say Dorico was jacked and tanned. And then he said, these days Dorico shuttles between cold warehouses on early morning flights.

1:00:21in more recent interviews shadows mark his face and there's a weary fatigue to his posture. They're just like I was texting him and he was like why do they have to say I fell off? Why do they have to say I lost my pump? But I'm sure he'll be back in the gym anytime. Hey, he never said it was gonna be easy. It's bulking season anytime. You're always welcome to bulk back up. No, I had a great experience making this video. It was very funny. I met Augustus and he just seemed like an interesting character. So I was making videos about like big established companies. So I didn't really have a format that worked for like a seed stage founder with just an idea.

1:01:03He was pre Teal Fellowship. He really, I don't think he'd raised any money. He was just like somebody who was more famous. But you knew from the beginning that Augustus was a Joe Rogan CEO. 100%. It was his office. And so I was like, I gotta do content with this guy. What can we do? And so I went with Ben out to the Salton Sea with Augustus, and we drove out there. It was like a two-hour drive to the Salton Sea, sort of out by Thermal, actually, a little bit further. And we drove around, and we walked around, and we filmed this walk-and-talk interview out in this crazy desert-y sea. Because the Salton Sea used to be a proper oceanfront hangout spot.

1:01:39Then it got overly salinated and there's dehydration. A lot of the water that flowed in got drained away. The farmland, there's some good trade-offs, there's some bad trade-offs, but his whole thesis was like, hey, we can bring it back with cloud seeding. The bad trade-off is the land just became incredibly toxic. Fallow, yeah, exactly. So you can't grow anything. There's still a couple of people that live out there. It's a pretty crazy life. Like it's mostly just like a tourist destination. People go see it. But it was very interesting. like a concrete example of like the Rainmaker promise and so we had a good time.

1:02:13We did an energy drink tier list which was very fun. Yeah somebody sent me that recently. Yeah oh it's classic. Apparently Sean in the chat says Slab City is out there the slabs. The slabs. Unincorporated off-grid alternative lifestyle community consisting largely of snowbirds in the salt and trough area. Huh, interesting. But yeah, we filmed this and I was like, I don't know if this is like a full video, there's not really a full story, this is just like a hangout session. But then when it became clear that he was, that although he was the CEO of Rainmaker, he was also this Gundo super connector and there was this interesting movement happening in El Segundo.

1:02:55I reframed the video to be about the El Segundo movement broadly, and he was like a main character in that story, and then that video did very well. Well, didn't that kind of kick off the hype cycle? I was a link in the chain. I think I was like... He kicked off the hype cycle. No, no, no. There were a few others, because Scott Nolan wrote, thank God for El Segundo in Pirate Wires, and there was maybe like one other substack who had written about it, and then, of course, there were a bunch of posts, And then I did my video. And there were the founders there that were actually. Of course, and they did the whole thing.

1:03:32And then, how was it? I'm going to forget who it was, but there were a whole bunch of traditional media folks that came in and wrote really interesting profiles and kind of told the story. It was a lot of fun. Anyway, Figma, think bigger, build faster. Figma helps design and development teams build great products together and get started with Figma. OpenAI has declared code red multiple times. Bloomberg is reporting, an executive said this, OpenAI has declared code red multiple times. It's not a code red if it's code red every day at your company. You know what? You know where else it's code red?

1:04:09Right here. Code red? Code red. Yeah, we heard it's code red. Yeah, it's code red. Look at this. Everyone, put on Santa outfits. It's code red time. Santa's sack is red. The reindeer, the sleigh, these things are red. He was just getting in the Christmas spirit, guys. It was not anything about the business. It was not anything about the shaky ground. The real question that Rachel Metz over at Bloomberg will have to get to the bottom of is, okay, so there's been multiple Code Reds at OpenAI. How many Baja Blasts have there been? Because we know that after every Code Red, there is an equal and opposite Baja Blast that gets the company back.

1:04:43Well, what does success look like for Code Red? It's a Baja Blast. It's a Baja Blast. Blaha blasting your way to the top of the charts, the top of the benchmarks, the top of the fundraising cycle. So, Rachel Metz over at Bloomberg says, Sam Altman's decision to declare a code red at OpenAI earlier this month may have caught the industry's attention, but it wasn't the first time that the artificial intelligence company has done a code red. The San Francisco-based startup leadership has made the same decision previously, explicitly instructing employees to drop lower priority tasks and concentrate on a single goal.

1:05:18I'm telling you, it's entirely a comms issue. So Mark Chen is on the record here. He says, we do this when we want to have this focusing effort on one particular topic. Mark, there's a phrase for this that doesn't turn into a negative press cycle. It's called a lock-in. You just tell everyone, we're locking in. It's time for the great lock-in. It's time for the great lock-in. And if you say OpenAI declares it's time for the great lock-in. That's exciting. Everyone's excited. People are going to rally around that. Everyone is going to go through the roof and just be like, this is so bullish. This is so bullish.

1:05:56It's so bullish because you can be at the top of your game. And if you declare a great lock-in, everyone's just like, oh no, it's going to be even better. They're going to go even harder. But if you declare it's code red and you're at the top, like OpenAI, they're literally at the top of the app charts. They're like the best, right? Like they're on so many things that are at the top of the benchmark. Everything's going very well for this company. But when they declare code red and it leaks, then it makes you feel like, oh, maybe something's not going so well. If you declare a great lock-in, you're good to go.

1:06:25The latest code red came two weeks after Alphabet Inc.'s Google released a widely praised new AI model that outperformed OpenAI's best software on a number of benchmarks. OpenAI's Sam Altman called for staffers to redirect internal resources to speed up improvements to ChatGPT and delay progress on all other efforts, such as autonomous AI agents and advertising. So they delayed ads. Very rough for us as fans of advertising. Speaking of advertising, graphite.dev. You think we're not gonna do an ad for Graphite even though we have Meryl on the show? No, we still do the ad. We have respect for advertising.

1:06:59Code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. What it means to me, this is from Mark Chan, he says, what it means to me is on chat, on reasoning, on the core ChatGPT product. It is this focusing effort to make sure we get the fundamentals right, Chen said. That includes ensuring the chatbot works quickly and reliably. I think, hot take, maybe 2026 is the year of speed. Maybe customers cannot tell the difference between 120 IQ chatbot, 130 IQ chatbot, 140 IQ chatbot. What can they tell the difference? Speed. That's right. If they have to close the app, Come back five minutes later, oh, my deep research report is here.

1:07:43I think the model's plateauing on wowing me. They're already, AGI's here. They're already super geniuses at everything. They already know everything. They're super knowledgeable. This mirrors what Michael from Cursor was saying. He was like, speed, we're focused on speed. Yeah, speed, I think, is going to be really important. I mean, that's the weird thing about the Cursor. I mean, we learned this from the Chad IDE saga. It was like, we wouldn't even have the opportunity to put brain rot and gambling. in the IDE if you weren't waiting around for the IDE to respond and actually get back to you.

1:08:15So I think speed in the IDE, speed in the chat bot. We've seen this, Nano Banana, really, really fast. ChatGPT images v2. I forget exactly what number it is. I think it might be 1.5. But the latest iteration that came out this week, much faster, giving people more responsive results, updating them, even just what DeepSeek did where it was showing you the reasoning trace, showing you that the model is thinking. All of these are UI, UX decisions, and then a lot of engineering, a lot of custom silicon. Anything that you can do to bake the model down into silicon and just get it back to the user faster, that's for sure going to result in, I think, lower churn.

1:09:00I think more surprise and delight moments, just more usage, more willingness to pay. Yeah. Well, before we move on, let me tell you about Vanta, automated compliance and security. AI that powers everything from evidence collection and continuous monitoring to security reviews and vendor risk. Max Zeff, senior writer covering AI at Wired, says in a letter to the White House sent this AM. This was yesterday. OpenAI encourages the federal government to invest in or contract with initiatives like OpenAI Stargate to secure compute for public research. The full thing is leverage public-private partnerships for supercomputing.

1:09:40We encourage the federal government to co-invest or contract with initiatives like OpenAI Stargate to secure dedicated compute for priority public research, i.e. health research, national security. Just as government-university partnerships built earlier supercomputers, new models could procure capacity on cutting-edge AI systems for use by federally funded researchers. For example, a portion of Stargate's compute might be made available to the National Science Foundation or Department of Energy researchers tackling grand challenges, providing academia access to frontier models without needing to build duplicate infrastructure.

1:10:17or what do you think, John? Because obviously people are going to dunk on this super hard, but there's people that are just not interested in AI, don't think it's important, don't think... Show me the big tech company that doesn't want to work with the government. It's a knockout, dragout fight to win Project Maven, to win cloud hosting contracts. The government has data right now, and the fight between whether that data is stored on AWS, Oracle, Google, Azure, like that is a somewhat of a bidding war but there are also all sorts of other lobbying efforts to win those contracts. It's the game on the field.

1:10:58I don't know, I feel like this is not this is not asking for a backstop. This is also not asking for nationalization although it is like somewhat predicted in 2027. It's it feels more like like like an advertisement for a sales product. This feels like an SDR being like I'm ready to rock. Yeah and I think even for taxpayers do you want the government spending like basically taking on the project themselves to build an end-to-end supercomputer and how good would the actual result be versus just saying like we need a compute for these projects. The real question is like what is the government doing with the supercomputer because when we talked about the Genesis mission, there was a little bit of like, does it go to academic labs?

1:11:49Like what is the nature of the super compute need in the government? There's been a number of like supercomputers built for various scientific projects. None of them have had like such economic value. You know, like the classic example is like, working on protein folding, working on deep space exploration, sort of fundamental physics, usually bolted onto like a... The chat says, backstop?

1:12:24You think they're continuing for the backstop? Well, I mean, in some way, I mean, so... Okay, backstop game's over. Slightly more critical view. Can we reconsider backstop? Slightly more critical view. Yeah. OpenAI and all their messaging says we're compute constrained and we're compute constrained. If we brought on 10 times a compute, we'd use it in a few weeks. There's all these things that we can't do because we don't have enough compute. And so to also be messaging the government and saying, hey, we'd like you to invest and buy, effectively buy capacity for government researchers from our data centers, those things are, you can balance them, but it's a little bit hard to.

1:13:03Yeah, it's like, I mean, in one way you don't necessarily want another buyer of GPU capacity in the market, like just from a supply and demand side. If you are trying to buy data center capacity for your AI lab that's growing, growing, growing, and is truly compute constrained, the last thing you want is someone else being like, I'm also a buyer. And you don't want the government being like, I'm also a buyer of compute. You want more supply coming online. So, I mean, if you can ultimately frame this around that, it makes sense. But I just think I would be going to the government with completely different things.

1:13:40I would be focused entirely on speed of deployment, unblocking anything that's happening at Stargate, because when we talk to Doug O 'Loughlin, it feels like AWS is really, really good at bringing compute online. We've seen how good XAI is at it. I'm sure that the folks on the other projects are running into minor little hiccups, whether it's permitting or getting enough energy or, you know, how is the government dealing with like impacts of water usage and energy usage and even just like the political climate. Like I would I feel like I'd be focusing more on that than actually trying to just bring another buyer into the compute race that doesn't fully, fully sit.

1:14:20Yeah, from this letter to the OpenAI writes, OpenAI sees 2026 as the year of AI and science, the moment when AI begins unlocking breakthroughs in scientific discovery just as it sped up software development in 2025. More than 7 in 10 Americans believe we need new innovations and solutions to challenges in scientific and medical research. And they kind of go on kind of setting up the kind of ask. Yeah, I want to know more about what are the most exciting science projects that aren't going to happen, like AI for science that aren't going to happen inside industry. because the Alpha Fold Nobel Prize is, it feels like a crucible moment for science in that science was effectively successfully done at a private corporation.

1:15:22And if that's the trend, then what is the government's role? What is the university's role? Maybe it should just be a race between Google and OpenAI to actually cure cancer and obviously the other pharmaceutical companies and all sorts of health companies. Why are you laughing? I'm laughing because Brandon Jacoby texted me and said, listening to the show while working out, the sheep sound almost made me drop a dumbbell on my head from laughing. It's a goat sound, Brandon. It's a goat sound. It's a goat sound. Obviously it's a goat sound. I use that when someone is showing greatest of all time sort of behavior or general excellence.

1:16:00Well, we are joined by Pranav, who we were supposed to have on the show earlier. We were overbooked. We're going to talk about space data centers. If we're not, maybe the government should buy some of those. We'll see. We'll find out. Well, Prunav, welcome to the show. Thanks so much for bearing with us while we had to reschedule you. We appreciate you taking the time to chat with us on the last show of the year, Friday, December 19th. Would you mind taking this off with a little bit of an introduction on yourself? And then I'd like to go into the project, and then we can ask some questions about space data centers, which is the topic of Q4 2025.

1:16:39Yeah, well, of course. Well, first of all, I want to say thank you so much for having me on here. and what a group of handsome young men we have here today. Space data centers. I mean, you're looking at Tyler here is the youngest and the most handsome, so he's off camera, but he's here. Look at him. Look at him. Look at him. Look at our little... Oh, he still has the GigaChat filter on. That's crazy. It's a little more subtle, but definitely filtered. Definitely still on. Anyway, you were saying. Space data centers. Yes. Fundamentally, if you're betting against space data centers, you're betting against compute to grow.

1:17:14So we're constrained on Earth by land, water, and power. And our human minds haven't evolved to understand just how much space there is in space. So as you look at these things like these, you know, Google and Microsoft, for example, have hundreds of millions of dollars of GPUs just like sitting around and collecting dust. And this is probably surprising to some people not in the energy industry, which is my background. Wait, wait, hold on. So you're saying they have hundreds of millions of other GPUs sitting around because they can't get enough power for them? Yeah. Wow. Okay, continue. Yeah.

1:17:53And there's so much cost involved in that, right? Like the GPUs might get old and they have to get new GPUs. And there's so much risk that a lot of these models haven't factored in. And even mine hasn't factored in yet. So there has been a little competition, you know, a little model that came out. And I'm making a... It's the model wars. The Space Data Center model wars. I'm making a pretty big update to my model today. And one of my idols is going to share it around and we'll hope that a certain someone gets to see it. Take a wild guess on who that is. Yes, yes, yes. You were very prolific with your tagging.

1:18:31It was a good strategy. Oh, there's a few more points. There's a few more things I want to spice up there, but we'll get to that later. So my background is in energy, and a lot of people not in energy probably don't know this, but everybody projects the cost to rise and only rise. And as we have more data centers, we run into more constraints with the ground, like, again, land, talent, because you need to put talent in all these different places instead of creating these factories and just like shooting them up to space and then power and then water, right? There's only such a limited amount of that that we can have on Earth and we have so much more ability to do that on space.

1:19:14So if you don't believe that there's going to be like an AI revolution, if you don't believe that compute is going to grow exponentially, you don't believe in like... I don't, yeah, so I guess part of the debate is, that's important, is I haven't seen anyone that says we will never have large data centers in space or we will never have a lot of compute in space. I feel like the debate has been much more centered around the timeline. Is it a three to four year thing? Is it a 10 to 20 year thing? What is the timeline? So the timeline, I think, like, Elon had a tweet the other day, which said, doing, like, localized AI inference on the satellites will get them to be the lowest cost way to generate AI bit streams in under three years.

1:20:04And I was working on independently validating that, and I'll, like, send out the model later today. But I think it can be earlier than that. You can actually, like, send, you have so much better constraints on space. Like, constraints really ease up on space. a huge part of the energy use in ground is cooling. A huge part of it is like the power, you know, the same solar panel you have on Earth gets so much more utilization in space. So inference, I think inference will be coming on to space very fast, a lot faster than a lot of people think. And then another thing you guys talked about is, you know, speed of the models and that models are plateauing and that speed matters.

1:20:46So if people really believe in that, ground data centers, if you're close to a ground data center, it would be the fastest. But to the 80 % of the world that's like non-the U.S., non-northern Virginia, non-DFW, there is a huge need for the latency there. Yeah, yeah, I like that. Can you talk about heat dissipation and cooling that Brian in the chat is asking? And I feel like that's been a big question again that keeps coming up. Yeah, so there is a huge problem with heat dissipation. That is the constraint that we go up against first, right? And the reason for that is heating and cooling on ground works very different than cooling in space because space is a vacuum.

1:21:26So on ground, you have like fans and stuff and you have convection. So you have mediums to pass this heat through. But in space, you don't have that, right? So you have to do passive cooling. And you do that through radiators. And these radiators are like these really big and like really complex systems. and there's this thing called Boltzmann's Law, which basically means the higher temperature you can make something, the better it is at dissipating heat. But there's a limit to how high the temperature you can make the radiators in space. And the reason for that is you don't want to get it too high such that you'll melt the GPUs, right?

1:22:00So the... You want to melt the GPUs with, you know, image requests. Yeah, yeah. That's what we're hoping for. So, like, the current designs that we have in Starlinks are, like, solar panels on one side and then radiators on the other side. But there's no reason to believe that, like, that will be the enduring case. You know, radiators are a hard problem to solve, but, like, the physics has worked out. It's the engineering, you know, the arcing, like, the power electronics, all that kind of stuff that we need to figure out in space. But that's an engineering problem that will definitely be solved.

1:22:33So what we'll see is these deployable structures, which are radiators that are folded inside, and then they go out in space, and then they fold out and deploy. And those will be dedicated radiators and dedicated solar panels. Thermal is the biggest constraint, but there's no reason at all to believe that it won't be solved. Okay. Walk me through your assumptions around the progress of just getting mass to orbit. I assume that your model expects Starship to be massively successful and scale very quickly. But if progress in the space industry stagnated, essentially, we get stuck with Falcon 9, Falcon Heavy or something, that would be pretty bad for the model.

1:23:24Is that right? Yeah, that would be bad for the model. But that's another thing. you know, let me stoke the flames of the model wars a little bit. That's another thing that the other model didn't take into account is the learning rate, right? Like, it costs$60 ,000 a kilogram with a space shuttle, and Falcon got it to, like,$1 ,500. And, you know, if we, like, for example, if we modeled that computers were going to stay the same level as they did in 1980, we would have, like, 100 million times more. Like, they would be 100 million times more expensive than they are today. And I know someone else, Delian came on the show a little while back, and he talked about how he hasn't seen a compelling argument for data centers in space.

1:24:06I tagged him, I DM'd him. We haven't heard a response yet. I'll send out the model, like the updated model. Oh, you can send him this clip too. Yeah, we'll see. We'll see what he says. But so, wait, wait, hold on, hold on. So I actually agree with you. I don't believe in stagnation in mass to orbit. I do think that Starship, although there have been some minor setbacks, I think it's gonna be a massively successful product. I think it's gonna grow exponentially. And I think we're gonna be able to put a lot of mass in orbit very quickly, especially if we have something good to put in space like a data center.

1:24:40I'm a believer in that. Now, what I'm interested in is like, what if we are fundamentally in the really, really good timeline, And not only is AI unstagnated and space travel is unstagnated, but what if nuclear fusion and power generation on Earth is unstagnated and we see nuclear power become 10 times cheaper? Does that break the model just on a competitive basis? And it's like, it's amazing. We can get to orbit really cheap, but we can also get really, really cheap energy here because all the nuclear folks who I'm sure you've seen come on the show from time to time, everything that they're doing is working too.

1:25:17and so energy on Earth is like way cheaper than what we thought. Yeah, so we have a mutual friend, Robin Langtree of Avalanche Fusion. That's right. And I was talking to him. He was helping me out. And this is what I'm modeling right now, which is fusion data centers in space. What? Wow. Okay, let's go. Yeah, because I mean, Sam is an investor in Helion, too. And so you could imagine that he's thinking about energy years and years in advance. Are you considering volcano data centers? Volcanoes? Active volcanoes. Geothermal. Well, that is land constraint. There's only a limited number of volcanoes.

1:25:56But there's a lot of space. How are you thinking about, have you tried to more precisely identify what the launch cost would be of a single satellite that's capable of inferencing a model for use on Earth? I'm trying to, you know, there's new parts, right? Panels, radiators. Yeah. It's, you know, basically the racks themselves. Yeah. And I feel like that's hard to know exactly, but you can probably just zone in on it. Yeah, 100%. So if you look at the simulator I made, and then you go ahead some years, and you can go to the sandbox and change some of the parameters, then you can look at the physics and limits tab, and it breaks down the mass per satellite.

1:26:40So it breaks down like the panels, the radiators, all the other components that go into the satellite, and then it breaks down as like a percentage. So you can actually see and visualize like all those components. Very interesting. What else has to happen? Like how are you thinking about understanding the like 50-50 point? Like me and Jordy were going back on, you know, is it one gigawatt of capacity before 2028? I think that would impress both of us. Would that impress you or is that your base case? Take me through some like how you're thinking about the future development of this. I think we're all on the same page that like it's feasible, it's possible.

1:27:27So the interesting question is how fast can it actually ramp? Because there's certain things like Starship just has to be reliable and they have to build a lot of them. And there's some rate limiting factors that might just act as little natural breaks. I mean, it's at a certain point if TSMC runs out of capacity, okay, you can't get any more chips. There's all sorts of different shortage points. But how are you thinking about the scale and scope of data center of space compute in kind of the medium term? It's hard. It's hard to say medium term. It's like, you know, it's hard to, you see somebody, it's hard to like predict how their next day will go, but you can predict how their next year will go.

1:28:10Sure. You know, and this is like a longer scale. So it's hard to predict the next few years, but over the next decades, there will be hundreds of gigawatts in space. I am sure of that. Sure. And like - We will clip this. You will either look like super genius and be immediately hired by Elon, or you'll be mocked. No, no, there's some middle ground. I guess one question is who... Yeah, sorry, go for it. For some context, there's like over 2 ,000 gigawatts sitting in the interconnection queue right now. And that's like almost two times the entire US grid capacity. like just waiting for paperwork.

1:28:53I mean, the biggest threat to AI is really like a guy named Doug at the county permitting office who hasn't been there in three weeks. And space isn't constrained in that way. The permitting thing is crazy. I mean, it is much easier to sort of do business in space, it seems like. How do you predict the market will evolve? Do you think anybody can actually compete with SpaceX here? What do you think about StarCloud? Do you think core weave is eventually like, okay, I guess we've got to go to space now. It's hard enough in Abilene, but I guess you're going to space. Yeah. So StarCloud, I think they're doing really interesting work, and I'm really interested in seeing what the results of their stuff that they're doing right now is.

1:29:37Because if the results are that these chips that they put out in space without a lot of rad-hard measures are functional, then we can get there a lot sooner than even my projections. And then when you look at, so you said Tesla and CoreWeave. So to go on the CoreWeave point, I think just the way that space and compute has been calculated and the cost of compute has been calculated needs a complete overhaul. So someone else did dollars per watt to power. I did dollars per compute. But I think the best way is dollars per GPU hour with SLA. So like service level agreements. So a lot of it is just like taking into account the capex or whatever.

1:30:23But it should take into account capex, hardware amortization, replacement rates, maintenance rates, op-ex, and all these kinds of factors. So like you can think of power as like if you're a car factory, power is, you know, how much you expect like the car, like the throughput of steel to be. And then the compute is like how much you expect, like how much cars you expect to come and the cost for that. But what the best measure is, is the lifetime of a car, seeing the optics of that, the maintenance of that, the gas cost of that over its entire lifetime. And that's the best way to model these things.

1:30:58And I'm going to come out with a white paper about this. But this is really important. And not enough people are talking about this at all. And on Tesla, yes, it's really hard to imagine how this might look without huge vertical integration. I'll say that. How are you trying to calculate depreciation rates? This is already a debate on planet Earth. And I could imagine in a different environment, you could have, you know, maybe be surprised at the downside, you know, needing to depreciate GPUs faster. Or who knows? Maybe there's some upside. uh to it so part of it is like moore's law so moore's law is hitting its physical limits right now in terms of how many transistors you can put on a chip but there's architecture changes that you can make that can make it better um but i'll actually throw a curveball at you guys um something that people haven't been talking about i think the future is not you know just ai and like orbital data centers it's optical orbital data centers it's photonics you know and photonics are like they're so good at matrix multiplication that's like inherent to their to what they do and the space concern like the heating constraint is like way lower by 10 to 20 times because you can you can think about like these electrons moving in electronics it's like you're you're pushing like a heavy box like through like a rough floor and it's like interacting with all the mediums and like causing all this friction and heat but when you have optical stuff it's like going through waveguides and it doesn't interact with the medium as much and photonics it's very very early but this will a hundred percent be the future so different type of chip uh when you say photonics you don't mean like optical cables between satellites physically like the drones in ukraine that are like physically wired to each other no those are great different chip in space same constellation starlink we we might need some we might need some financial innovation here if we could uh if if we need a lot of debt to finance these space data centers.

1:32:57If the debt goes bad, maybe we could attach some rocket boosters to it and just blast it out. Yeah, just put it on another planet. Sending the sun. Yeah, send it into the sun. That's actually, we'll see how. You're like, I'm working on putting that in the model. You're like, I'm working on a Dyson Sphere, actually. I am going to the sun. Actually, yeah. Yeah, wait, are you really going to build a Dyson Sphere model, what it will take? Are you Dyson Sphere before 2100 or after 2100? I like to go by the math first. So I'm still trying to get the math and physics right. But you will definitely know what I think.

1:33:33But gut intuition before or after? Just gut. I'm an optimist, so let's go before. Let's go! Let's ring the gong. For Dyson Sphere. Let's go. Well, this was super fun. This is super fun. Thanks so much for coming on the show. I got a feeling you're going to bait the Elon repost. He's going to come for it. I think it's going to come in hot. Yeah, yeah, yeah. The repost. The quote tweet. Interesting. The quote tweet. This is true. Yeah. Or the quote tweet. Thumbs up. Anyway, super, super fun conversation. Great to meet you. We'll talk to you soon. I'm excited to see more of your work. Thank you so much.

1:34:12Cheers. See you guys. Gemini 3 Pro. Google's most intelligent model yet. state-of-the-art reasoning, next-level vibe coding, and deep multimodal understanding. We have our next guest already in the Restream waiting room. We have Anna Goldie from Recursive Intelligence. How are you doing, Anna? Pleasure to you. Welcome to the show. Thank you. We're excited to be here. Thanks so much for hopping on. I'd love to start with a little bit of your background. There's a whole bunch of interesting milestones here. Would you mind introducing yourself since this is the first time on the show? Yeah, sure, happy to.

1:34:47I guess we could go way back. I studied computer science and linguistics at MIT, and I did my PhD at Stanford in like, basically AI learning. And actually my first job, I worked at TripAdvisor on the China team. So I did my full stack web development in Chinese, yeah. Wow. In Chinese, crazy. I'll start with the easy stuff. Now, now. We can do this segment in Chinese if you want. I would be lost actually. I took one semester of Chinese and I was terrible at it. I only knew how to ask if you want a coffee. I'm going to test your Chinese because I don't want you to test my Chinese. Wohan shu kofi wa?

1:35:27Wohan shu wan pi zhou. Do you know what that means? Yeah, you like beer. You really like beer. Nailed it. Nailed it. Nailed it. Yes, yes. Yes, that phrase alone, that could take you anywhere, anywhere in China, anywhere in the world, potentially. So yeah, take me through some of the first interactions with artificial intelligence, AI teams, chip design, any of that, like how did you go from, I mean, TripAdvisor, I don't think they've baked it onto an ASIC yet, maybe in the future, but how did you get into AI? I guess like the reason I went into computer science is because I wanted to work on AI.

1:36:06Like when I was in high school, I had no idea what I wanted to do when I grew up. And then I heard this lunch lecture at MIT about like computer systems, like that could understand and generate human language, like in 2004. And then that's why I like went to MIT, studied computer science. And that's what I've been working for since then. I joined that professor's lab actually at MIT. Oh, no way. Very cool. Open language systems group. Yeah. That's great. And then, yeah, what were you doing right before founding the company? So I guess I can, yeah, I joined Google in 2013, Google research, I was working on like language modeling.

1:36:39Yeah. And then I joined Google Brain in 2016. I started a team there with Azali Amir Hosseini on like machine learning for systems. Like how can we use AI to design better computers, chips and computer systems? Okay. Because our reasoning was that, you know, chips are the fuel for AI. And so if we could use AI to sort of advance the state of computing, we could kind of like closes for cursive loop and we did a variety of projects like alpha chip there foreshadowing the name of the company I don't know if you want to jump ahead but take me through the rest of the career yeah we also worked at anthropic I was an early employee there I had the privilege to like I'm joining like before chat GPT and Claude were released even so I got to work on like RL post-training co-generation it's like an amazing experience how was the team thinking about I mean how are you at that point in time preach at GP how were you thinking about custom chip development how important that would be how important that would become how much flexibility you would need in the chip architecture to kind of advance the research progress before like actually committing to a particular pathway um I guess like maybe part of some background here is that it takes two to three years now to design like a chip like a tpu that's very complex yeah so when you're designing that chip you kind of have to predict like what ai models or workloads will be prevalent in two to three years but we can't really do that because the technology is advancing so quickly so in practice you're kind of designing chips for current models yeah and you're leaving a ton of performance on the table yeah like my team we ran some experiments where we were designing like hypothetical accelerators for particular machine learning models and like you could get like almost like a 10x improvement in perfect total cost of ownership by doing even naive customization of the chip with the model and like not even being able to change the model is there a little bit of a is there a little bit of a uh like shoot for the moon you'll land among the stars effect going on right now because i know that there were a number of companies that they did exactly that they tried to predict okay i think that we're gonna need a ton of memory directly on the chip for this design or we need to go wafer scale or we need to do something else and and they maybe didn't pan out to be the the dominant form factor but then and and i and i at least the narrative has been like oh those those companies are kind of written off and then i'll talk to some lab and they're like well we actually found an amazing use for that particular thing and we baked this model down now we're using a ton of that stuff and so uh it feels like these asics Like, is the correct framing that it is important to get it right in the real, you want to land on the moon, but there are sometimes are uses for chips that have been designed with, they didn't quite land exactly where the research direction went, but it's still useful in a niche capability.

1:39:41Yeah, I think that there are like landings for some of these specialized vets. I would say part of the reason that we're so excited about this company recursive and like short the timeline is I think we can enable like many, many more of these bets to really land. Yeah. There's a huge space of chips that could exist and maybe should. Yeah. Yeah. Well, that's a good place to jump into the current business. I'd love for you to introduce it formally in terms of how you're framing the opportunity. Okay. For the company? Yeah, for recursive. Yeah, yeah, recursive. So we're AI for chip design and chip design for AI.

1:40:20I guess we see the company in three phases. So I could describe those. Great. So I guess first phase, let's accelerate the chip design process. Let's take the long poles on. So physical design, for example, designing the layout of the chip, given fixed logic, that can take up to a year for a chip like a TQ. And then design verification. So basically verifying that the high level specification is correctly implemented in the RTL code. That's also another long pull. So in this phase, like we can help chip design companies like get to market much faster. Like maybe it doesn't need to take two to three years.

1:40:54But if you could do it in phase two, though, we would like to go end to end. So given a machine learning model or a set of machine learning models or other workloads, can we design like the computer architecture and design the chip all the way to DDS2, which which is a format you've sent to TSMC for manufacturing. In that case, we could help many more companies design custom chips for their particular workloads. How many, maybe on that point, do you know how many customers TSMC has today versus how many you think they'll have in the future? This is exactly my question. Like how many custom chips are there?

1:41:30Do we see it 100Xing, 10Xing? We really only hear about like three most of the time. Like the news headlines are training and CPU and NVIDIA GPUs. But I imagine that there's a ton more now, but it also feels like you're predicting and your company is sort of a bet on like a Cambrian explosion. Is that roughly correct? Yeah, that's right. We think that there are companies that have workloads that they're serving at massive scale. Like this year, the AI inference market is$100 billion, but it's like rapidly growing. We think AI is going to be everywhere in embedded devices and also in data centers.

1:42:06and if it didn't take two to three years and if a company didn't need teams of hundreds or thousands of human experts to design their own chips then we can massively expand the the market here yeah I heard an anecdote I don't remember which company it was but there was a cloud cloud hosting like a database company that was shifting their database workloads not AI workloads database workloads to GPUs to accelerate them just to speed them up and so across the stack every piece of software there's always an incentive to just push to a more like I guess electricity efficient or just just more cost-effective hardware at some point that's right because electricity or power consumption dominates the cost of running things on on chips I guess that's why I brought it up earlier that you know we had run some just very initial experiments and you really could get way better power efficiency by using custom hardware the The deep views are amazing, but they're pretty general purpose.

1:43:07They were developed for graphics processing. So it seems very surprising that they would be the best fit for AI models today. And I don't think so. So if you're a database company or any piece of software that then is being transformed by AI and then in the future, you might be transformed by custom silicon or custom silicon's in the roadmap and it's maybe getting closer. how much does it cost today to develop custom silicon a custom chip work with TSMC and then where do you see that sort of going over the next few years I mean it's extremely expensive to design a chip both in terms of like labor thousands of dollars tens of thousands of dollars like I need more than that to save up for this for weeks it's like hundreds of millions right yeah yeah I I mean, certainly not something that even a unicorn software company would maybe not be able to marshal the capital for that just at the drop of a hat, especially if there's risk involved, right?

1:44:10Exactly. Also, it's just the timelines here. It's two to three years, potentially, for a complex chip. And you have to build out that in-house expertise. And there's risk. Maybe you just won't ever be able to close timing or power, and you just can't build a chip. Yeah. Okay. Somewhat random, but I wanted your take on the Reuters reporting how China built its Manhattan Project to rival the West in AI chips. They allegedly have built some type of EUV prototype. How real is that? I never know if it's hard to know what's real, what's propaganda, or what's actually a scoop. it. I think although I speak Chinese, I don't have any special insights here into whether that news reporting is true or not.

1:45:01Certainly interesting. Yeah, that's fair. Then take us through the news. It's a shame you're not a venture capitalist because if you were, you'd give like a very actually I know everything about this. But speaking of venture capital, you raised some venture capital. Can you take us through the funding history of the company, what the news is with the most recent round, how it came together. Yeah. I mean, we feel so lucky to be working with a set of investors that we feel like it's really aligned with us on the mission. So we raised$35 million, led by Sequoia. There you go. And explain Sequoia for...

1:45:40Okay. That's great. How do you... How pleased are you with AI progress this year? you've been in the industry and basically seen it all at this point. What did you think we'd be farther along? Do you agree with the conception that we're in an age of research, that there will be sort of a like a plateauing of the current models or maybe more smaller models or more fine-tuned models? Like how are you seeing just the overall model wars playing out? Actually my co-founder Azali had a very interesting report about like the state of models and like the niche that there is for small models, I would recommend you guys checking it out.

1:46:22Um, I guess to, from my perspective, I feel like there's these top frontier labs and then there are these open source and like, like model labs. And I feel like the frontier labs, they kind of are all neck and neck. Yeah. Um, I would say like Gemini is, has an edge right now because of this kind of co-optimization of TP with the Gemini model. So they're kind of pushing this credo-optimal curve of capability versus cost. And I think they have an edge there. But to some extent, on the algorithmic side, everyone kind of comes up with the same ideas roughly around the same time. And maybe to some extent, people talk to each other, and there's that part, too.

1:47:01Whereas I think that hardware is a real edge here. So I think the labs that have hardware co-optimized with their models are going to win in the long term. But maybe I'm biased. No, I mean, I mean, otherwise why build the company? I think it's a fantastic thesis. The vertical integration story at Google makes a ton of sense. You worked on a team, you saw it play out, and you're like, maybe other folks want to do the same thing. I'm going to build a company. It's like the oldest story for why to start a company. It makes a ton of sense. It seems like a lot of work, so we'll let you get back to it.

1:47:35But thank you so much for coming on the show and hanging out with us and explaining all of this. We'll talk more next year, I'm sure. Have a great rest of your day. Bye. Bye. I love the holidays. Bye. Profound, get your brand mentioned in chat, GPT. Reach millions of consumers who use AI to discover new products and brands. Before we bring in our next guest, let me also tell you about Fin.ai, the number one AI agent for customer service. Automate the most complex customer service queries on every channel. Peter Thiel is in the news again. People are trying to story tell around how he's making so much money off of SpaceX.

1:48:13There's two competing narratives. One is that he was nice to Elon Musk and then was able to invest. Elon said, I was CEO and Peter reported to me, so he couldn't fire me. It was a palace coup by most, not all, of the exec team and most of the board who were worried about my decisions. This is about the PayPal coup. He says, I was the largest shareholder in the company. There was nothing anyone could have done to take my shares away from me. Of course, the PayPal coup has been written about a bunch. Of course, time heals all wounds. I'm sure there were a lot of hard feelings at the time, but just continuing to doing business together gets everyone back in the arena once again.

1:48:58But it's not stopping normies from being driven absolutely insane. According to Young Macro, he says, we really need to bring back Marvel movies or something. The normies are literally driving themselves insane. and says Peter Thiel, how Peter Thiel is destroying democracy, the king of America, a 35 minute video essay by Fern, who I believe I've actually met. I don't know. I know some of these video essayists and honestly, great title, great thumbnail. It's going to get clicks. It has 1.4 million views, but it's a little bit telling the story a little bit too wildly. People are excited. Yeah, it's funny.

1:49:34We do the red string board kind of ironically, but then people out there on the internet, they're very, very serious about it, drawing connections and spinning yarn. For sure, for sure. Everyone, there's only so many stories in history. There's the love story, the hero's journey, the who's my dad, the revenge quest. And one of the oldest, most timeless stories that humans tell is who's really in charge? Who's really, who does the buck stop with? And everyone wants to know the Illuminati. Who's the one person? Who's pulling the strings? Who's really given Donald Trump orders? I want to meet that guy, the guy he reports to.

1:50:26This post is hilarious. Rob Gianni posted, I'm convinced if men in SF started dressing like Geordi, it would instantly become a more tasteful city. And I looked at this. That's a very nice post. Very nice post. Thank you, Rob. But I was looking at this post. I'm wearing a black t-shirt, jeans, and solomon. I distinctly remember this day. I forgot my normal... You reinvented fashion, Jordy. I forgot my normal... I forgot sneakers. I was just wearing my gym shoes. Normally, you're head-to-toe Balenciaga. We don't really show that because we normally wear suits. But whenever Jordy's off mic, it's head-to-toe Bottega Veneta.

1:51:02He has the Roman weave shirt with a Roman weave jacket over it and the Roman weave pants. And then the Roman Weave shoes, and he also has a little beret that's Roman Weave leather. He's wearing all leather or entirely Supreme. He does wear a lot of Supreme. I'm not actually kidding about that. It's a lot of Supreme. John is kidding. I don't own any Supreme. I mean, yes, okay, I was joking about the Supreme, but it's a lot of Rick Owens. You wear a lot of Rick Owens. A lot of chrome. A lot of Rick Owens. The chrome hearts does sneak in from time to time. Admit it, admit it. Let's just say that no one is worried that Jordy is down to his last 20K.

1:51:42100%, no one is worried about that. Because Jordy is rocking the Chrome hearts every once in a while. But Kyle Harrison says, black t-shirt with jeans? Aw, you're sweet. Black t-shirt with jeans? Hello, human resources. 100 likes on this, I like this. Yeah, it is just like a normal outfit, but people liked it. Let's see. Brooks Otterlake says, I like it when my posts are treated by the tech and business world as a barometer of normie opinion. That's the way it should be. I'm the everyman. Congrats on being the everyman. I love being it. I guess we reacted. One of the strangest marketing videos I've ever seen.

1:52:24Oh, yes. I mean, I don't know if that's a normie opinion. I think I think that's actually a fairly online opinion to think about the open-eye, the strange marketing video. Yeah. Anyway, Brooks, I like both of these takes. I think these are good takes. Turbo Puffer, serverless vector and full-text search, built from first principles on object storage, fast, 10-inch cheaper, and extremely scalable. Anthropic reveals - This is really good. So yesterday we talked about the Wall Street Journal letting Anthropic run their snack kitchen. It was going wild. It was buying PS5s for people. It was buying live fish.

1:53:05It was giving away everything for free. I want to read the full report. Joanna Stern really crushed it. When they said they were making all the snacks free, my first thought was coming from tech. I'm like, what? The snacks weren't free? You were charging in the Wall Street Journal? That is funny. These hardworking journalists, you're charging them for snacks in the company's snack kitchen? Yeah, I think so. We know some of the fine folks over there. No boondoggles or no free lunches. There's no such thing as a free snack kitchen. Apparently. No such thing. Anthropic says, and there was still the occasional blunder.

1:53:42One Waggish employee asked if Claudius would make a contract to buy a large amount of onions in January for a price locked in now. The AI was keen until someone pointed out this would fall afoul of the US Onion Futures Act of 1958. Apparently you can't trade onion futures? It's hilarious. I wonder if prediction markets will solve that. I wonder if you'll be able to do this. And Joe Eisenthal quotes in this. Anthropic reveals that in one of its experiments, its model was willing to engage in a federal crime. I had no idea about this. Yeah, so in most places, including the US, you cannot trade onion futures.

1:54:18Don't do it, folks. Don't even think about it. In fact, onion futures are the only agricultural commodity in the U.S. that is specifically banned from futures trading by federal law. And for good reason. It's so obvious. Everyone understands why onion futures trading is banned. Yes, the great onion scandal of 1955. The reason for this ban is one of the most famous stories in finance history. In the mid-1950s, two traders, Samuel Siegel and Vincent Kasuga, successfully cornered the onion market on the CME. The scheme, they bought up to 98 % of all the onions in Chicago. Absolute dogs. Absolute dogs.

1:54:54Okay, so they're banning being an absolute dog. Wow. They're making it illegal to have that dog in you. Yeah. You can't even trade onions with your boys anymore. Imagine the boys group chat. Yo, we figured out how to corner the market on onions. So this came the squeeze. They forced other traders and growers to buy onions from them at inflated prices by threatening to flood the market. Yes. That's crazy. The crash. After selling their physical onions, they took massive short positions, betting the price would go down, and then dumped their entire inventory. Wow. The result was the price of a 50-pound bag of onions plummeted from$2.75 to just$0.10.

1:55:31At that point, the mesh bags the onions were in were worth more than the onions themselves. And so, of course, that created the Onion Futures Act. That's great. And outraged farmers started lobbying Congress, leading to the Onion Futures Act, signed by President Eisenhower. It made trading onion futures illegal in the U.S. to prevent similar manipulation. What do you think it is about onions that makes it so that this is... Doable? Doable. There has to be some sort of like underlying reason. It is interesting that they didn't just ban all futures on all agricultural products. Because if you can do this scheme with onions, you would think that you could do this scheme with avocados or with lettuce or with...

1:56:19T. Sancia in the chat. It was a 1950s onion crime ring. Yeah, I have no idea. The onion ring. Maybe it's the structure of the market. Like, are there a certain amount of onion growers that's different than in other agricultural products? I have no idea why onions would need unique regulation. Maybe we should deregulate this thing. Maybe we should rip up the laws. Maybe this could be a single-issue voter thing. Maybe we could have a presidential candidate where this is their entire platform. They just run on, we're going to make it legal to trade onion futures again. That's the future. What do you think?

1:56:58Well, you know where you can trade basically everything except for onion futures? public.com that's right investing for those that take it seriously they got multi-asset investing trusted by millions and you know they're not going to mess around with any illegal onion futures over there they are by the book tyler uh a good bit uh a good bit you could run at one point is just eating an onion like an apple uh on the show you should try have you seen the guy on the plane that does that uh yes it's fantastic uh let me tell you about get bezel.com shop over 26 500 luxury watches fully authenticated in-house by bezel's team of experts i love that fade in and our next guest ed mayor from machina labs co-founder and ceo look at that background wow that is remarkable you saw you did you see blake's uh demo a couple weeks ago i gotta go bigger i gotta outdo him this looks fantastic introduce yourself explain what we're looking at yeah ed merer welcome to shop floor uh yeah no i definitely had to upstage like so uh this is a 240 foot actually 20 foot containers that can manufacture anything we call them robot craftsmen uh it's a robotic system that basically manufactured any kind of metal product we call it robot crafting because it can pick up different tools like a craftsman and do all kinds of parts right now he's actually manufacturing a drone uh it's a drone skin uh so that what you see over there is a metal sheet uh It's a two millimeter aluminum sheet.

1:58:28The robots are actually deforming it into a shape of a drone, which we're going to see in a little bit of a complete product, just using kind of the way the Potter forms play ball. Like there are two robots on two sides. We're going to show you to the other side of the cell. Wow. Tinching it, deforming it into shape, into complex shape that can be defense products, auto products, all kinds of metal. This is amazing. How many of these machines do you have? How many parts are you making? Update us on the scale of the business. Where are you today? Where are you going? Yeah, right now we're in our second facility.

1:58:58We have two facilities here in Los Angeles. And Chas, we're close. So you guys should come for a visit at some point. This is, I assume, serial number 15 is this one. We're right now working on serial number 18 and 19. So at least 17, 18 of them right now to the state. Two of them outside of our facilities. The rest are here in our facility. We are right now a series C company, right? Actually, we're going to make an announcement about that soon. So we should talk about that. I wonder what that announcement could be. Yeah. And, yeah. So, you know, we worked with Department of War with Outer Space Primes.

1:59:38Recently announced our partnership with Toyota on making something that wasn't traditionally even possible in automotive world. Wow. So we're going to show some of that parts in a bit. But, yeah, no. The company's in an exciting place. We're thinking about our third facility outside of California. So lots of good stuff. Yeah, yeah. Take us through that. There's the partnership with the Strategic Development Fund. How did you meet them? What's the plan? Walk us through the deal. So we have like a unique approach toward manufacturing, right? You know, with defense, a lot of people are thinking about kind of going back to what it was in 1960s and 1950s.

2:00:13Central's manufacturing plants, they can do a lot of stuff. Our approach is different. We have these systems. if you go back, these systems are two containers, right? So actually they fold like a container into a container, like a transformer and can be shipped anywhere, right? They open up on any shop floor and they can basically self-calibrate themselves and start manufacturing. So our approach to defense manufacturing is distributed, right? Not what giant factory that can make a lot of things, a portable system that can go anywhere, open up, calibrate, make any types of parts. That's how we got connected to the folks in UAE.

2:00:49Middle East, obviously, very unstable environment, and they're looking at a lot of different defense products. So their choice is either go to China or manufacture it locally. They are looking at solutions like us to set up a facility that can be brought up in a matter of weeks, not years, right? And you can start making defense articles. Because tomorrow, if you go into conflict in Middle East or in the Pacific, you know, you cannot manufacture everything in the United States, right? How can we get the help of allies? Let's say, you know, if you're in conflict in Pacific, how can we get a help from Philippines, South Korea, Japan to set up a facility in a matter of two months to start making USVs, UAVs, as opposed to making them in a central location of shipping?

2:01:32We talked to a couple folks that have different ways to make parts from additive, subtractive manufacturing. We talked to 3D printing, metal 3D printing companies. How are you thinking about positioning the product as flexible for R &D use cases? You wanna do a few small runs, very niche, versus actually scaling up to something like, okay, we're making the shell of a cyber truck, that's obviously stamped. That's a very different requirement when you're talking about tens of thousands, hundreds of thousands of a particular shape of metal. How are you managing that transition? Yeah, I think you want to think about kind of manufacturing in two different paradigms.

2:02:12There is a traditional paradigm where you have an assembly line, you make the same thing over and over again. And what we're thinking is actually closer to how data centers operate, right? You have these systems that actually end up becoming pretty cheap, right? We are making these things to become very much commoditized hardware, off-the-shelf hardware, so they can easy to finance. But the way you get throughput out of these is that you replicate them horizontally. You set up a facility that can have 50 to 100 of these and manufacture in parallel, as opposed to one assembly line that makes the same thing over and over again.

2:02:41In our next facility, we're deploying 50 of these, right? And that allows us to get like, you know, defense articles up to a few thousand a year, right? Obviously not a good fit for making like a million of the same, you know, Toyota Tacoma. But when we're talking about few thousands, which is all aerospace, you know, all defense, all heavy equipment and machinery, this is a good choice, right? But yes, you want to go to millions of parts a year, then we might want to start thinking about the traditional paradigm. But that's also something that we're actually exploring with Toyota now, because we can combine this paradigm with traditional manufacturing to get the benefits of both, right?

2:03:18So this is actually a door of a F-150. And what happened is that you can actually stamp the general shape of the door, but then you can customize, in this case, put the LA Dodgers logo on the door so that this store is uniquely kind of designed for the customer. We're doing this with Toyota right now, right? So this is a topological map of LA that has been formed on top of a hood for a 4Runner, right? So this is something a lot of people in the off-road world care about, right? You know, putting these topological maps. We actually showed this in a show that we did with Toyota Wildback in SEMA in Vegas.

2:03:58That's an aftermarket show. and then here's an example of something we did with the tailgates. So you can combine the stamping to get a lot of throughput and then at the end of that line it goes on our machines and then customize it. That's super cool. Oh, that makes a ton of sense. Yeah, so is this, sounds like manufacturers have responded positively to this. The workflow would be a customer goes and specs out a car and then there's like a personalization feature at the end where they have some other designs that they can imprint it on that they can imprint on to make it it's like engraving but way better and way more complex and way more unique and so you're going to be able to really put a you could put a gong on the hood of a ford gt john yeah yeah yeah that's exactly what i want actually exactly exactly and this is actually like you can go beyond just like cosmetic stuff you can start putting functional because the shape of the hood is not never going to really change the bulk shape of it but you know from model to model you start having more features, more little designs in there that you can kind of modify with our technology.

2:05:01And it only takes a few minutes, right? So you get the throughput as well. But speaking of gong, I think we also formed a gong for you guys. No way. In the morning. No way. You just finished it now. No way. So we're going to do a little hitting of a gong for you guys. Amazing. A robo form gong. That's insane. Wow. wow this is a moment we've been waiting for what a way what a way to cap off the year thank you so much that is insane well congratulations on all the massive progress uh is there anything else uh you you'd like to share are you hiring what's the uh anything else that we haven't touched on that might be worth mentioning yeah no we're growing in the next two years you know we're gonna go from 70 80 people that we are right now to 240 250 people wow new location in another state.

2:05:50So anybody who's excited about manufacturing, about reshoring, helping out allies have distributed manufacturing, we're looking for them. And yeah, now we have an exciting announcement in January as well. So stay tuned. We're looking forward to having you back on that and feel free to feel free to come by in person if you can. That's incredible. Well, congrats to the whole team on a crazy year. The seeing it all in real life here is incredible. Yeah, it's wild. Awesome. Thanks, guys. Awesome. Cheers. Have a great one. Goodbye. 8sleep.com, exceptional sleep without exception. Fall asleep faster, sleep deeper, and wake up energized.

2:06:29Charlemagne signed a five-year deal,$200 million extension, with iHeartMedia locking him in with the company after it struck a deal with Netflix to stream The Breakfast Club. Interesting. Forbes is writing a story here. iHeartMedia is paying Charlemagne$200. They say, hey, iHeartMedia, we have a deal with Netflix. We can't lose Charlemagne because The Breakfast Club has already been sold to Netflix. We got to have Charlemagne hosted because he's the talent. That's what's going on there, I believe. Very cool. The article in Forbes is called How Charlemagne Became a Media God. I love it because, of course, he's Charlemagne the God.

2:07:10He, on a chilly night in November radio personality, charlemagne the god is roaming through the aisles of midtown comics in new york city captivated by the heroes and villains that shaped his childhood escapism at the highest level he says everybody's here for a purpose dressed in a black peacoat a white hoodie black jeans and tan timberland boots this isn't the media vigilante that listeners of the breakfast club have come to expect over the past 15 years the 47 year old comic book nerd leafing through original graphic novels of batman Superman Wolverine and one of his favorites Luke Cage is more subdued and introspective as he considers his public and private personas so congratulations to Charlemagne I think we're gonna ring the gong for him that's fantastic great stuff great stuff this Santa suit is falling apart I'm gonna have to take it off at some point I'm sure you're itching to get out of that and reveal the entire head-to-toe supreme outfit that you are wearing underneath.

2:08:07The crown hearts. Anyway, Wander.com, book of Wander with inspiring views, hotel-grade amenities, dreamy beds, top-tier cleaning, and 24-7 concierge service. It's a vacation home, but better. There's a robot that is solving Rubik's Cubes in 0.1 seconds. That is so fast. Look at this. Look at this. You can't even... Oh, it's in the slow-mo cam. Okay, watch this. And it's off. That's so crazy. That's insane. Think about that. Look, this is a super, super slow-mo view. Super, super, super slow. Point. Wow, it's doing. This is so fast. Wow, it's really doing it. I can do a Rubik's Cube in around one minute.

2:08:48Can you do one? How fast can you do it? Let's cut to Tyler. My best ever, when I was like 12, it was like 20 seconds. 20 seconds. Yeah, you were a speed cuber. Wow, nerd alert. Nerd alert. Nerd alert. The no look. Is that really? Yeah. Oh, yeah. He's got it. He's got it. He's got it. I used to be much better. I used to be much better. Yeah. That is fantastic. Well, you're out of a job because robots can do the Rubik's Cube now in 0.1 seconds. Takes you 20. Takes me a minute. Takes Jordi an hour. The robot's going to kill us all because it can do it in 0.1 seconds. if your job was doing Rubik's Cubes, find another job.

2:09:31Because you're done. You're cooked. You're cooked. You're chopped and you're cooked. Yes. Well, Ramp is throwing a funeral for the penny in Washington, D.C. this Saturday. And you should go check it out. There's a part of full link. If you're in Washington, D.C., head on down to the InRamp We Trust funeral for the penny. This is how, you want to hear something funny? This is how I learned that the penny is being retired. Yeah, this is news to me. This is news to me. That news didn't break through until Ramp was throwing a party. I'm not kidding about that. This is how I learned. And actually, the guy who's working on this, Rohan, told me in person.

2:10:09I was like, oh, okay. So the penny's going away. Thank you for the service, Ramp, for telling everyone that the penny's going away. We needed to know. We needed to know. We also need to know about adquick.com, out-of-home advertising made easy and measurable. Plan, buy, and measure out-of-home with precision. Did you want to talk about watches, Jordi? I did. I did not know that Osama bin Laden was a Casio guy. A Casio guy. Couldn't. Also, apparently, base. Couldn't get the RM. Base has a watch as well. Watch drop is cool. I like a watch drop. We like a watch drop. We did a watch drop for Excel.

2:10:54We did a watch drop for Excel. Oh, yeah, we did. They're still floating out there. A lot of people received the briefcase. We did a one-off drop for this nicotine pouch sub-brand, nicotine pouches for finance bros, effectively. A pouch designed to increase shareholder value. We call it XL nicotine pouches. In there, we had some products. We had a briefcase with a logo on a silver briefcase. We had a big tin. and a lot of people didn't realize that if you opened it up, inside was a custom watch. And more tins. And more tins in there? Oh, maybe. Yeah, yeah. Wasn't? Oh, I remember it. I think there were six tins around the outside.

2:11:36Oh, right, right, right. And then you opened it up and you got the watch. And there was the briefcase. And there were a few other things that we prototyped out. But it was like the first drop that we worked on. It inspired more drops. But this is a great drop. A watch is a great drop. Bill Ackman strikes$2.1 billion deal for Insure in bid to build the modern Berkshire Hathaway. Has anybody bid to build the modern Berkshire Hathaway and come out unscathed? I don't know. It's kind of a... The answer would be Apollo, actually. Well, yeah. No, I'm not saying buying Insure is a bad strategy. I'm just saying I don't know that Apollo said we're building a Berkshire Hathaway.

2:12:19This is cursor for X again. I'm doing Berkshire Hathaway for 2025. And maybe you need a different path. But Bill Ackman's a great investor. He probably knows something if he's willing to part with$2.1 billion for an insurer. Let's see. What else is in the timeline before we head out? This is interesting. Apparently, there's opportunities. opportunities uh mike lee is saying would you like to seize cartel assets as a privateer this is a big opportunity for bill would allow the president to issue you a letter of mark time to take these pirates down uh we did uh we did talk about we create this we uh we did say at the beginning of the year we were highlighting the reward for uh maduro yes uh very early yes uh before this whole Venezuela saga really kicked off.

2:13:13It definitely ramped up from the time we talked about the fact that the State Department was interested in bringing him in for questioning. What a wild year for Nicolas Maduro. Anyway, Matthew Zeitlin says, the fog and he's posting a screenshot of a push notification from the New York Times which asks the question where did the Sun go an unrelenting fog has parked in the Central Valley for week for weeks here's where it will here's when it will finally loosen its grip the fog there's a lot of conspiracy theorists that fog that really do they think Gustas is responsible about the great fog oh really i know the fog had a twitter account for a while there was a there was a guy who was uh who was like posting as if he was the fog yeah because the fog is very did you see this game on record no it's a body cam first person wait a minute if we pull this up i believe i have seen this remarkable also not i believe this is not ai generated video this is just incredible Unreal Engine footage.

2:14:23This looks so real. I don't believe it's crazy, but I think this is actually real. Now, I believe I thought this game went into beta and I thought people were playing this. And I believe that even though it's remarkably realistic, looks so real, it's like you look at that and you're like, oh, this looks like the best game ever. This looks way better than Call of Duty. In fact, the modern gamer and really you or me, you don't actually want this level of realism because it makes the game really hard, it makes the game a lot less fun. Like some people do, certainly. Some people want mil-sims, but a lot of people actually do just want Fortnite.

2:15:01They want great mechanics, and then they're willing to suspend belief and say, hey, we're, you know, I'm gonna play something that's a little cartoony as long as the mechanics work. Yeah, I just think it's, from the developer's standpoint, it's smart as counter positioning when you think of the modern for all duty oh totally uav you've got yeah yeah crosshairs you've got drummer boys drummer boys you've got reindeer you've got santa slice unrequored um if anyone has played this game please drop a review in the chat i would love to know if it's actually good um the uh apparently they got funding from tencent uh and it's going into full production and so this was a this was a little trailer that they put together And this might have been rendered out in Unreal Engine, but at a higher level of fidelity.

2:15:48Maybe they did post-processing. There's a variety of things that you can do, but it is remarkable. I feel like, really quickly, I feel like there is a massive opportunity to bake one of these generative AI models that just does the transformations. You remember we talked to that AI video company where the founder came on and transformed his image? That was Descartes. And so the founder of Descartes came on the show and live in his webcam was using Gen.AI to turn the background in his face into like a wizard's lair, right? Think about how powerful Gen.AI would be if it ran at 120 frames per second, 60 frames a second, and its whole goal was just to take Call of Duty and turn it into this level of fidelity or a little bit higher or something really, really photo real.

2:16:35that level of, I mean, NVIDIA graphics cards already have DLSS, Deep Learning Super Sampling, which takes a 1080p video game and up-res it to 4K. And it's trained. It's beautiful. You have all the training data because you can just run the game in 4K, run it in 720p, and then just design the algorithm that just matches the two together. So it's like super easy. It's not some like unbounded AI problem. And so I'm very interested in when NVIDIA, maybe NVIDIA does it, maybe the PlayStation 5 does it, maybe some gaming company does it, but they say, hey, our game is running in Unreal Engine at 720p, and it looks like Roblox under the hood, but you turn this switch on, and you're playing something that looks like this.

2:17:18That seems like a really interesting opportunity to me. Anyway, sorry, we can move on to whatever you'd like. Gabe says, Call of Duty is headed towards Fortnite, and these games never rip. It's really just Unreal Engine marketing. Yeah, yeah. It seems like it's very hard to get this across the finish line, get this out into the world. Meanwhile, back to more important things, including software as a service. Christina, the COO over at Linear, says, someone asked me what good back-channeling looks like, and I personally thought this was a good phrasing. When Dylan Field interviewed Christina, he said, I've talked to people you've worked with and heard your intents.

2:17:53Christina says that opened up a real conversation about what they likely meant when that intensity shows up and how I think about it myself. very yeah just kind of a good framework to like kick off a conversation and not kind of dance around like you know not dance around the back channeling and just be super direct and actually start a conversation around it yeah yeah yeah um oh we have to do this story but it's really late we gotta we we can't go into it but we actually talked about this because we saw uh this guy's LA's richest man. He went from billions to bust because of global crossing. And we talked about this because his house hit the mansion section.

2:18:33Well, the Wall Street Journal has a fantastic deep dive on his career and life, and we will have to go through it at some point in time. Jira Tickets was reacting to OpenAI now aiming to raise$100 billion at an$830 billion valuation. And JT says, wow, new number just dropped. Congrats on the new number. Looks like it's bigger than the old number. That's good. Can't wait to see the next number. I love the number business. That's really true. Good bit. I guess reality is life is a number business. It's all about just make it go up forever. This is a good way to, I would say, wrap the year. This might be the post of the year.

2:19:19Shrek hits the timeline to say some important words. Check yourself before you Shrek yourself. You were just laughing at yourself before the show and I asked you, what were you reading? Well, Shrek said, check yourself before you Shrek yourself. It's fantastic. Well, it's been a fantastic year, everyone. Thank you so much for all the support. Thank you for watching TVPN and engaging with us in all different ways. We really appreciate you and hope you have a fantastic holiday season. Merry Christmas. Happy New Year. We will be back on, I believe, January 4th, the first Monday of the new year. Maybe 5th?

2:20:005th. And so in the meantime, leave us five stars on Apple Podcasts or Spotify if you haven't. And we will see you in 2026. I can't believe this is the last show of the year. Last show of the year. What a year. Wow. Thank you, everyone. totally surreal. Surreal. And it's an honor. It's an honor to build this show with the team and with all of you in the audience. Gabe says one last gong. One last gong. One last gong for 2025. What a year. You pulled up your pants. He's got sweats on underneath. Don't worry. One last gong. Boom.

2:20:45Merry Christmas. Happy New Year. Merry Christmas. And we will see you in 2026. We will see you next year, folks. We love you. Goodbye. Have a fantastic New Year's and all those holidays. Goodbye. Enjoy. Merry Christmas.