I see a large IPO on the horizon. You're surrounded by journalists. Watch your position.
Overnight success. The laser 3. It's 10:37 on Wall Street. Money never sleeps. Use your theater, right?
There's misinformation. Clearing order inbound. Let's just roll.
We are surrounded by journalists. Hold your position. Come, get up. Trust the experts here. We are experts. Founder Mal.
I see multiple journalists on the horizon. Stand by. UAV online. Blaze.
Double blaze. Triple blades. Double kill. Michael is 10:37 on Wall Street. Money never sleeps. You should. We are experts. Triple blade. Let's just roll. Right.
Clearing order inbound. You're surrounded by journalists. Hold your position. Strike one. Strike two. Activate golden retriever mode.
Mark clearing order inbound. 5. I see multiple journalists on the horizon. And founder.
You're watching TBPN. Today's Wednesday, July 8th, 2026, and we are live from the TBNN Ultradom Template Technology Forces Finance, the capital of capital. Let me tell you about Ramp. Time is money. Save both. Easy to use corporate cards, bill payments, accounting, and a whole lot more all in one place.
That's right. It's great to be back.
Yes. The Invest Like the Best hype song really, it really has, I don't know, it just burned its way. Yeah. Puts it in an incredible mental state.
100%, 100%. You know who else is in a good mental state? Jeff Bezos, because he's getting a markup on the 25 billion he's put in.
He's marking himself up alongside CO2.
Yeah. The news is that Blue Origin is raising outside capital for the first time. He's participating. Fund CO2, who he has also invested in through his family office, is also participating. But there will be an opportunity for some other true outsiders to get their first slice of Blue Origin. Andrew Ross Sorkin had the scoop in the New York Times today. You have to wonder if this is gossip coming out of Sun Valley. The company is raising 10 billion dollars. This is the first outside fund raise. It's hilarious to think about this 25-year-old company as being previously bootstrapped, but that's technically what it is, basically. One founder's capital just chipping away at a really hard problem for a quarter century. And now we are here. Bezos has been funding this basically solo on his quest to visit the stars. He's doubling down.
People can't really complain about the valuation because he's a big part of setting the price.
Yeah. Yeah. So he's doubling down. He's putting in $2 billion directly and CO2 is getting twice as much allocation, 4 billion. There's another interesting CO2 tidbit I alluded to earlier from the New York Times piece. Bezos's family office is a major investor in CO2's innovative strategies fund, which is focused on emerging technology startups. So I'm not sure if the money is coming out of that fund specifically, but there is a connection there. He's been building a relationship with CO2 for a long time and they are optimistic about Blue Origin's prospects. The valuation is expected to be $130 billion, big for a private company. I was listening to Gavin Baker and Travis Kalanick talk about how crazy it was when Gavin was at Fidelity and he wanted to value Uber at 14 billion and everyone thought he was crazy. The valuation turned into an auction, came in at 17, and everyone was like, this is unprecedented. We've never seen a private company this big. And now 17 billion, even 60 billion in the private markets looks quaint by modern standards in an era where we have several trillion-dollar private companies, many up in the hundred billion-plus range. It's a wild, wild time.
Brian wants us to warm up the gong. We got to warm up the gong. Warm up the gong for Bezos. Give it a couple warm-up hits just to get it warm. Just to get it warm. Here we go. Yeah. Fantastic. Nice and warm.
Yes. So, the Elon bulls are saying, 'Oh, SpaceX is underpriced now.' Because you have to imagine, I don't know if there's an exact revenue figure that's leaked for Blue Origin, but based on the size of that business, it has to be small. And so, this has to be a very high multiple, but you're not really valuing this company on cash flow or even revenue. You're valuing it on the capability, which is it's the second company in the world, I believe, to bring a rocket to orbit, land it successfully, you know, prove reusability. They got there before China, which has been trying to copy SpaceX for years, and so lots of really solid progress. Of course, there was that setback with the launchpad that exploded, but they figured out a solution to that and things are chugging along just fine. The company is estimated to burn five billion this year. So this looks like a pretty standard 12 to 18-month fund raise at 10 billion. And it's burned something like 27 billion to date, which, you know, a billion a year on average over 25 years. Not too bad for such an ambitious project. Still a lot of money, but a big opportunity. And you have to imagine there are a number of other comps like AST SpaceMobile, as you like to call them, AST SpaceMobile, Space Mobile.
AST, $27, $28 billion, $30 billion company, and I don't believe that they've gotten much to orbit yet. They've done some experiments, some tests. They're still early, but it's a huge market and we've seen how fast concepts of mass to orbit. Yeah. And we've seen how fast Rocket Lab and various... it's a big, really, it's a really, really big market. And so when you're going after a big opportunity and you've been working at it a long time and you got Jeff Bezos funding it, lots of reasons to be optimistic. Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it.
Another comp investors would certainly look at would be Rocket Lab, sitting at just under $50 billion, and there'll be multiple players in launch. That is for sure.
Well, we have some heartbreaking news. Heartbreaking news. Apple has scrapped plans for a cheaper Apple Vision Pro display and is winding down work at Samsung. This is from MacRumors. Really hits me and the other three Apple Vision Pro fans particularly hard. Brutal. I did... Who did I talk to? I talked to somebody who said that they actually are a daily user of the Apple Vision Pro. I thought I was in the top 1%.
Was it the founder that's making... Yes, that's right. He says he watches it, he uses it every night or something like that. That's a lot of Apple Vision Pro use. Obviously, the product never really found product-market fit. Although we have one in the studio, we've been having fun with it. I still think it's an incredible piece of technology, but a little bit of a chicken-and-egg problem. The developers never showed up and it was very expensive, very heavy, very cumbersome. It requires a lot to put that thing on to watch Lawrence of Arabia. But when you do, it's worth it. It's worth it. What did I watch most recently? I watched Dunkirk in it and I watched Godfather Part Two and Master and Commander. Master and Commander, that's a great movie, especially in Vision Pro. Really, you want to be in the cinematic feel, feel like you're on a warship. So Apple has reportedly stopped development on a cheaper display for a lower-cost Vision Pro. This feels like what the industry needs to move forward. The actual display was fantastic. It just needed to be cheaper and lighter so more people could get them. That at least was my thesis, but Apple disagrees with me. What do I know? So the supplier Samsung Display expected to formally end the project by September. And if you remember, there were rumors of the crazy hoops that Apple had to jump through to get that incredible display because they really leapfrogged Meta, who had been working with the acquisition of Oculus and they'd been improving the display fidelity. They got rid of the screen-door effect. They were increasing the resolution per eye. Everything was getting better about that display, but on a very linear trend. And allegedly what Apple did was they went to Samsung and said, 'We know that you are working on next year's display. You're also working on displays two years out, three years out, four years out. Now, what's the problem with the one that's four years out?' Well, it's amazing, but it's not ready for manufacturing at scale. It's not on an automated line. It needs a lot of manual work to get it to actually produce, and there's crazy low yield. So in chips, in displays, yield rules everything. If you run the machine a bunch of times, you want to get 99.9% yield, 90%. In the early days, when you're prototyping a new technology, you might be getting just a few percent yield. And Apple said, 'That's fine. We'll just jack up the price and pay a ton for these cutting-edge displays. Jump to the front of the category with the best display technology. We'll charge a fortune for it, but we'll have the best product.' And that will be our market entry strategy. And so I was waiting for that display technology to commoditize because it's been a few years. I thought it would just naturally get cheaper. It probably has, but it has not solved the problem of virtual reality, augmented reality. We haven't seen someone pick up.
Do you think they're just bearish on the near term of people watching movies?
Maybe. Yeah. In VR, because I don't... they're certainly not stopping R&D on their specs and their sort of everyday glasses product.
And so it might just be that they studied you and found that you're built different and you're one of one, the only person on earth.
Yeah. No, no. I mean truly, like sitting down and watching a full movie is a rare experience today. People aren't reading. They aren't watching movies. They're just scrolling and people seem to be satisfied with their phone. And also the nature of a doom scroll is somewhat interactive, somewhat multiplayer, although obviously people see it as brain-rotting and very isolating. There is a game to be played by actively scrolling, which is something that doesn't require virtual reality, certainly not enhanced by virtual reality. But then also sometimes you want to read the comments, sometimes you want to send it to a friend, sometimes you want to fact-check it and ask AI about it or spin off and do something else and go to someone's profile. So I think that the number of taps and user interaction experiences in a doom-scrolling session are obviously much, much higher than watching a film, which you just put on and hopefully you don't pause it at all because you're engrossed the entire time from start to finish. And so if that becomes as scrolling means more and more, maybe the VR experience is less and less relevant. So the cheaper display on the Apple Vision Pro, or maybe the Apple Vision Slim or Mini or something. I don't know. They could have used a different name. Pro always left the door open to just the regular Apple Vision at a more approachable price point, but the cheaper display that they were working on would have cut the headset's pixel density roughly in half. That's not what they need to do at all. They need to keep that pixel density because that's the best part of the experience. The move likely reflects Apple's broader shift towards smart glasses. Bloomberg's Mark Gurman previously reported that Apple paused work on a lighter, cheaper Vision Air. Yeah, that's the term. In October 2025 to fast-track its Ray-Ban-style glasses effort. This does not mean the Vision Pro itself has been cancelled. Apple refreshed the headset in October 2025 with an M5 chip, which does very little, slightly better tracking. I think you can run some applications at slightly higher fidelity, but it is by all accounts essentially an identical device, but they are long overdue for a proper refresh if they were taking this seriously and thinking about this as a true growth driver. I think they have recognized that this was an experiment that needs to be put on the back burner for at least a little bit. And so this report only suggests the cheaper display project is no longer moving forward. So the Vision Pro fans have hope. It's not fully cancelled. Maybe there's something else going on. Maybe there's more in the works. But anyway, it's interesting. With all the acceleration in AI, I was hoping that we would get some advancements in this since it is an adjacent technology, but we've been few and far between on the updates. We got to get an update from the Bigscreen VR team because they have done a great job creating high-fidelity VR that's also lightweight, but it needs to be tethered to a PC. It's a little more prosumer, a little more dev-kit styling. VR really attracts these visionary chads that are willing to go that... that seem... No, no, I'm actually... I could make that joke, but I won't. But just people that are willing to just continually look silly for a long time to ride through the hype cycles and just keep building. I think Bigscreen is an example of that. The founder that we had on that has a very cool glasses demo product. He's, from my understanding, probably using AR and VR more than anyone else on the planet. Like so committed to just like winning. Yeah. And so I really appreciate that about the category.
Yeah. Let's see how Apple's doing on the news. Are they trading up or trading down? I push the button. Will it come up? We'll see next time. Anyway, let me tell you about Figma. Figma agents. Agents meet the canvas. Your AI agents can now create and modify your Figma files with design system context. Production swearing at the team chat.
Well, here is a crazy story. I mean, it's such a minor story, but it certainly grinds my gears. I think we're going to dig into it and see what you think, what I think once we dig into it. But Getty Images, we know and love them from their trove of Silicon Valley titans photographed at Sun Valley, at Sun Valley, which is happening now. If you don't have a Getty Images subscription, sign up now because you're going to be first to know when the new Josh Kushner paparazzi photo drops. But Getty Images was planning to merge with Shutterstock. These are two essentially stock photo sites. In theory, they should be absolutely decimated by AI image generation. Meta put out a new one. Muse image, Nano Banana Pro has been fantastic. Images 2 from OpenAI has been really good. And like they're honestly getting to the point where a lot of the example images are just like, okay, that's just a photo. I saw Adam Mosseri posted some on Instagram and I was like, is this AI or did you just take this photo? Like it's just you. And then he did a couple with filters and crazy hair and those were cool and that breeds creativity. So I think there's a lot of value in the technology. But in terms of just like I'm trying to illustrate, I'm trying to make a point visually. I need a picture of a forest. AI image generation is very, very competitive with the stock image industry. And this should be a business that's sort of affected like the way, you know, yeah, a lot of these businesses, their net new revenue over the last couple years is just working with labs that need licensing, yeah, all sorts of things. So certainly in the face of growing competition, consolidation of an industry is very normal. That makes a lot of sense. This doesn't seem like it should raise any regulatory flags and yet it did in the United Kingdom. Getty Images has called off the Shutterstock deal after the UK created some barriers around it. So the $3.7 billion deal would have created a visual content company able to offer a more expansive library to users, says the Wall Street Journal. So on Tuesday, Getty delivered a written notice to Shutterstock terminating their planned tie-up, which was first announced in January of 2025 and received clearance from the Justice Department in America in April. So pretty quick turnaround in the United States. The $3.7 billion deal would have created a visual content company able to deliver a more expansive library to users, helping meet booming demand for licensed images and videos as artificial intelligence disrupts the business of content creation. Getty stock slipped 1.7% in pre-market trading. Shares of Shutterstock ticked up 1%. So maybe the Shutterstock shareholders are happy about remaining an independent company. The termination came after Getty last week said that its board voted to not proceed with the deal if it meant selling Shutterstock's editorial business, a condition required by the UK Competition and Markets Authority. So the decision by Getty to abandon its merger with Shutterstock was ultimately a commercial choice. So the chair of the independent inquiry group that led the CMA's investigation of the deal, the inquiry group found that a loss of competition between the two businesses would reduce choice for UK media outlets and could lead to higher prices. And I would assume that this would give them some sort of pricing power in the face of, you know, essentially unlimited free editorial images in the form of AI-generated images. If you want to go with the real thing, you have to pay a little bit more, but that sort of offsets all the pressure of the customers that are leaving and no longer paying because they don't need to pay because they're generating them with generative AI. So this felt like a very logical story. And I'm a little bit uncertain that this is a good decision for the UK to help these businesses. It feels like it's just going to wind up hurting both of them.
Yeah. Are UK media outlets not allowed to use AI content?
I don't think that there's any broad rule against it. No. So it would be a deliberate choice and I'm sure there's plenty. I imagine that the Financial Times does not use AI images and that's by choice and they make that very clear and there's probably a disclosure. And of course they will use an AI image if they're talking about the AI image industry, but in terms of what you see on the front page of the Financial Times, it's almost always editorial. Yeah, that image, this is from Reuters, and that is an editorial distribution and something that is licensed. And occasionally they need a photo to illustrate something and they might use AI.
Okay. So I was trying to figure out why the headline number here was 3.7 billion. Yeah. Because it was based on the January 6, 2025 share prices, not today's prices, which are far, far lower. Yeah, Shutterstock is sitting at $329 million market cap. Getty is at $340 million market cap. So these businesses have been really, really beat up over the period since they decided, hey, we should tie up. Interesting. So is there a potential? So the UK, I think regulators are actually trying to just kick these companies while they're down. Is that what's happening or is it a situation where Getty... Oh, I mean that is possible. I mean it could just be like a complete mistake. I mean as you go back to Meta and you think about the UK, the US regulators blocked Meta from buying a VR fitness company that they said would create a monopoly in the VR fitness industry. And of course like the VR fitness industry like never happened. Like it's not an industry worth monopolizing. And so if you think about the business model of Quest, like would it have been advantageous? Maybe. But it's worth letting Meta take a shot at that and see if having a VR fitness company in the portfolio will actually get that product to scale. Tyler, what do you have to say?
Another interesting anecdote is the UK blocked Meta from buying Giphy in 2022. Yes. So to prevent a monopoly in GIFs, I presume this could be pretty dangerous. Yes. Which is also odd. What does Giphy power? The GIFs that you see in iMessage? You know what I'm talking about? When you go to, there's Genmoji, there's stickers, and then there's also #images and you can find images and you can send, you know, different pictures and GIFs. I wonder if these are from Giphy or if it's powered by Bing, not powered by Bing. Okay. So, there's competition in this market. Yeah. I don't know. Manda says Canva doesn't label AI in their stock images. That's so annoying. So, yeah. I didn't even know Canva was a player here, but it would make sense to like, you're in Canva and you want to, you know, have a picture of a forest on your greeting card and you can pick from a library. Some of them will be generated, some of them won't. And they probably also have a generative AI tool where you can generate a new image. But if you're just looking for a generic stock image that's already been described, you can probably pull from whatever library is out there. Anyway, yeah, Canva partners with Getty Images, but then they're also featuring generated images.
Anyway, let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. There is another interesting article in the Wall Street Journal on the cover, the front page of the main section. What do they actually call the main section? There's the business and finance section. Is this just world news? I don't know. Anyway, on the cover of the Wall Street Journal, it's the world's creepiest office and people won't stay away. This is the New York complex vacated by IBM has become a magnet for urban explorers. We have a video of this place. And if you're a scrappy startup looking to expand, this might be the key to success. The ultimate lock-in factory in sort of rural New York, I think. We'll get into it. So, this is in Somers, New York. I don't know exactly how far away that is from Manhattan.
Yeah. So, if you were like a NeoLab and your name was like Ominous Intelligence. Yeah. This could be a great sort of HQ potentially, you know, your first little micro data center.
Very James Bond villain sort of vibe. And yeah, looks like you could make it, you know, really, really homey.
How far is it from Manhattan? Let's see. Okay, it is 1 hour. You can be in the city in an hour. That's... Imagine walking these halls late at night after a long day at work. Yes. It wouldn't be creepy at all.
Yes. Robert Carlton was getting the mail one day in April when he saw a group of teenage boys sprinting away from nearby woods. He cut them off at the road and raised his arms. 'I just said, Stop. It's over.' The 62-year-old retired engineer recalled. Soon, he said, three pursuing New York State police troopers emerged from the tree line and arrested seven juveniles. The criminal charge: trespassing on the former IBM campus here. Disrespecting International Business Machines. Don't do it, folks. The long-vacant site has become a magnet for so-called urban explorers who prowl abandoned malls, hospitals, power plants, amusement parks, factories, and any other unused, disused structure they can breach.
Enough of the graffiti. Yeah. Do you think... put in here? We got to figure out how much dare they write all over the former home of International Business. Exactly. What was the IBM AI? Watson? Watson the machine was probably still at the frontier. Still at the frontier. It was super intelligent at Jeopardy. The global, which is short for urban exploration phenomenon, urbex, isn't new, but it's been turbocharged by artsy videos on Instagram and TikTok that spur others to create their own posts, luring more curiosity seekers. Police in Livingston, New Jersey, recently warned people to stay out of the closed Livingston Mall property. Jacksonville Beach, Florida, police issued a similar caution in April about the shuttered Adventure Landing amusement park. Going into an abandoned amusement park. That is thrill-seeking at its best. We are aware of a TikTok challenge to explore the property. Who's doing these TikTok challenges? And should they be allowed to post those? This is dangerous for some people. I mean, seriously, you go in there, there's a lot of broken glass and, you know, who knows about the structural integrity of a building that hasn't been inhabited in years or decades. Could be very... Is it on the market? Well, we have some news about that. I don't know if it's actually on the market, but I imagine if you make them an offer they can't refuse, this property could be yours. So Sebastian Capital, the Manhattan company that manages the 723-acre site, that's a lot of size, says it has beefed up security and appreciates help from police.
So, I'm shocked by the YouTube comments, yes, that are wholly coming to the defense of this building. They're saying, 'I hate there's a class of people who see something like this and feel compelled to start breaking things.' Why do people vandalize these kinds of places? Interesting. I can't believe these are all the top comments. I can't believe people go in there busting up the place. That's crazy. I would have sort of expected people to be on the side of the urban explorers, on the side of the TikTok challenge challengers. Andrew Proto, a defense lawyer, said a 15-second clip isn't worth a criminal record. Don't do it. If you've already been arrested, you're not the first call we've taken this month, and you won't be the last. Proto says he has represented or advised several minors arrested on the campus. The Somers town court clerk said some defendants received a six-month adjournment in contemplation of dismissal, meaning charges will be dropped and their arrest will be sealed if they avoid trouble. Some explorers who have posted about the IBM site say they follow an observe-and-preserve ethos and reject vandalism. They say they're driven by curiosity, the thrill of roaming forbidden spaces, and a zeal to document discoveries, and that they're careful and know their limits. Quote, 'It actually gives me hope when I hear that kids are out there getting into trouble,' said Bradley Garrett, a cultural photographer and author of the book Explore Everything: Place-Hacking the City about his own urbex adventures. He sees urban exploration as a gateway drug in a good way, sometimes into intellectual curiosity about history and culture. So you go, you break into this abandoned building and then next thing you know, you're an IBM consultant driving enterprise value for S&P 500 companies. Traveling businessman could happen. Who knows?
Could happen. Reg wants to turn it into an Aman. Aman would be good. Data center seems obvious depending on power. What about the first Aman with a data center with some local AI? Both? So, quickly, you asked about the price. In 2016, IBM sold the property for $31.75 million and it's owned by an LLC that shares a Fifth Avenue address with Sebastian Capital. So, everyone suspects Sebastian's in on the deal. A plan to convert it to a private school foundered during COVID and Sebastian said it is considered... Yeah, foundering is like failing, but yes, foundered. So floundered would be, you know, another appropriate term for what happened. But it was not successfully converted into a private school, obviously. But if you want to take a wild swing, go pick it up. Make Sebastian Capital an offer. Let's see. I wonder what it would trade for today. Lots of work to be done.
Car and Driver, quickly. Tyler had something on the IBM complex. Yeah. So, okay. You planning to go? 1.1 million square feet. Historically, you'd see like I think something like around 10 watts per square foot. Okay. Maybe it's a little bit higher tech, right? This is IBM 80s. An estimate I think is reasonable is around 20 megawatts. 20 megawatts. Yeah. So, you're looking at a small data center. Is there any natural gas in the ground? Can we frack? Can we put solar panels on the roof? Can we build a nuclear reactor there? What are our options if we want to get this thing churning out tokens? We want to turn it into a token machine. Interesting. We should go check it out. Anyway, let's move on to the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. Just do it. And Jordi, Car and Driver says 2026 Lucid Gravity available with no interest due to no interest. What? Several months after launching the 2027 Gravity, Lucid still has plenty of 2026 models, so it's offering customers an interest-free loan incentive. Okay. I wonder how long there's no interest on the loan. That seems like a good deal. The Lucid is a very well-reviewed car. Very luxurious interior, very polished product. For some of the higher, more performance-oriented trims, incredible 0 to 60 times, incredible range. I think Larry David has one. They're well-reviewed cars, but it's an electric vehicle which has high depreciation and then you're signing up to work with a company that a lot of people think might not be around forever. So, what does the maintenance and repair and flow of products and new parts look like in a decade? And so, that is a risk factor for a lot of people. But the Lucid Air Gravity, I think it looks sort of cool. It's sort of like it wants to be a minivan. It's trying to do the job of a minivan, but they didn't just lean into the sliding doors. At least the Model X realized that what makes the minivan great is the accessibility of the doors, and they made them cool doors by making them go wing or butterfly doors.
Yeah. The problem here is that when I looked up Lucid Gravity for sale, I'm seeing a 2026 with just 3,000 miles that's selling for $125,000. A lot of money. Dollars. That's a lot of dollars. And regardless, even if you're paying no interest, it's probably losing, you know, it's depreciating thousands of dollars a quarter and so it's still going to be a rough ownership experience.
Yeah. Yeah. It will be interesting to see where they end up. I know they already have some big investors and potentially some corporate partners. It could eventually be rolled into another organization as sort of the EV strategy of a larger automotive group. Who knows? Did you want to look at End of Stay? Did you see this? We got to pull up this video because we were talking about exactly this. The 964 Cabriolet Jr. is a Porsche 964 that has been shrunk. Actually shrunk. Exactly what you said. Look at this video.
This is what you asked for. Fully customizable and look at this. It looks normal size, but wait until this guy gets in it. This is amazing. So, it is a go-kart that has the bodywork of a Porsche. And so, you can get one of these. They go like 60 miles an hour or something like that. Like they're pretty fast. Yes. So, I think if you put a governor on this vehicle, you'd have what you want, which is the vehicle for the kid that is fun, engaging, and highly stylized and made with precision. The group behind this is called End of Stay. Stay has periods after it. So, I imagine it's some sort of acronym. But what a wild... They did it. Yes, they did it. They did it. I'm interested.
You can see the YouTube video here, the preview of the 964 Turbo Jr. high-performance version. It's an absolutely crazy, crazy video that I got organically served to me and I was like, 'This is what we've been looking for for sure. This is exactly what we were talking about. I need one.' I think you do. You need one.
Anyway, GPT-5 Live, a new generation of voice models for natural human-AI interaction launched today from OpenAI, rolling out to Chat starting today. Taking shots at that guy on Instagram who's unsatisfied with legacy voice models. It's a lot smarter.
I have a feeling he's going to find a way to keep using the old models.
Oh, maybe. Or he'll just become, who's the Ply the Liberator? He's sort of the Ply the Liberator of voice models in that I know he's taking shots and he's like very unhappy with the quality of the voice models, but at the same time he does a great job of actually exposing and finding limitations, not necessarily jailbreaks, but by watching those videos you can very clearly understand, okay, I understand what's going on here. It doesn't have tool use so it can't count, it can't set a timer, it doesn't have, you know, a long-running context window that can stay with you for a long time so it gets lost. They can't talk to each other. There's a whole bunch of different ways. And once you have all of those documented by just a... he's basically like a white hat hacker. He's... He found a bunch of benchmarks. Oh, wait. What? He found a bunch of like good... Yeah. Yeah. Yeah. Basically benchmarks. Like if you can satisfy his videos and run those experiments again in a satisfying way, you probably have a better product at the end of the day. So, I don't know. I think it's cool. I'm excited.
Very, yeah, super excited for this one. I wonder if the real-time nature unlocks almost like a new kind of like pod, like real-time podcast format where you can just actually have a conversation about a topic. Yeah. Right. Yeah. We should have it sitting right here. It has a full-duplex architecture which means that it can listen and speak at the same time. So it can jump in appropriately and it can also route like certain questions to smarter models. Yes. To run in the background, you can still have a conversation. Very cool.
Which is great because my default flow, like if I'm driving, which I don't think I should be doing, but I will tap the record button, not the voice mode, and I will dictate a really long prompt. And then I will click send, and then I will wait for it to cook, and then I will have it read that back to me and that's all just like a couple extra clicks, a couple extra steps to get what I really want, which is a conversation. I was interested about the idea of where this goes. Like is there a world where you are talking to a model and you can have sort of multiple threads of conversations going within one conversation. So you can say, 'Hey, I need you to go and do a deep research report on, you know, IBM Watson.' And it goes and cooks on that. And then while it's doing that, you're still talking to it and you're still having a conversation. But then you also might say, 'Well, catch me up on the latest news.' And it has that at the fingertips. So it can talk to you about the latest news. And then you can go and it can say, 'Hey, by the way, I'm done with the report or I'm done with the code that I wrote,' but you're talking to me about other problems in your organization, other decisions, other research projects that you have. And it's all done in one continuous conversation like, you know, me interacting with you. Like, you know, we might, you might delegate something and then continue the conversation and then move on. Oh, I got to text this person. Okay, I'll text them. They'll do that. They'll come back to me in 20 minutes with the revision to the thing that we're working on. But in the meantime, let's talk about whatever else we got going on today. So interesting to see where this goes. In other OpenAI news, GPT-5.6 Soul along with Terra and Luna will launch publicly this Thursday. Preview access is expanding globally. Lots of people are having fun with this. Some people got early access. Other people were joking about clearly not getting early access but wanting early access. But a lot of people are having fun. Tyler, what did John Palmer say?
What did he say? He's in here now. He has one. He says, 'It seems I'm allowed to talk about 5.6 now. I wasn't one of the testers, but I heard it's really good. Again, I haven't used it yet, but this will probably be my default model from here on when I get access.' It's a good one. People are having a lot of fun. Kit Langton says, 'I can finally talk about GPT-5.6. I've been using it for many months, if not decades. And what is glaringly obvious to me now as I look back on my life is that you did not have access to it.' Oh, it's very fun. People are very happy with it. Interesting. There's a little bit of a fork around like how the vibe around 5.6 is different than Fable 5, how interactive it is versus somebody said like Fable is like light speed across the galaxy and Soul is like a Porsche at the Nürburgring, more engaging. I don't know. It seems like we're entering some fragmentation in like the right tool for the job.
Yeah. Ethan says, my big takeaway is that both Soul... Ethan had access. So he says both Soul and Fable represent jumps over previous models and have opened a large gap with the next best AIs. People will have preferences for one or the other, but if you're doing any work where better intelligence matters, those two models are your only choices. Dean says, I think for me the main takeaway with Soul and Fable is that I can't remember a time when the leading models were a, so decidedly ahead of everything else and b, so distinct from one another. So yeah, a lot of people are talking about how they're using the models together.
It's probably a function of just tighter distillation protocols. Like there's less pollution. Like also like not everyone's training on the exact same data anymore because it used to just be everyone just trains on the entire web corpus and now each company has different data sets, different data brokers, different usage patterns that they can train on and then different sort of research tastes. So go give it a try. I forgot to mention this, GPT-5.6 Soul is world-leading in computer use. It made me use it 100x more when we lost access to 5.6. It quickly started to go insane without it. So, yeah, people in the early access program obviously lost access for a couple weeks. Sure. Due to the DC action. But yeah, I saw some people saying like, 'What are some good uses for computer use?' And it's like anything on your computer. What do you use your computer for? They're not all going to be great uses, but yeah.
I'm curious. I still think it's always worth popularizing those genuinely useful patterns. I mean like deep research was that in many ways, just a very like, it gave you a 20-page PDF report on the topic that you were asking for and it did a lot of the work for you. And I still think the come up with a good prompt idea, find the right reason to use AI is a problem for a ton of people. There are plenty of people that adopt the latest AI tools, immediately find what the model's good at, how to use it, how to use it profitably, effectively, not waste tokens, not waste their time. But other people need ideas. They need to know that there's a new capability. You see this with the chat app. And you see this with the voice mode where people are like judging a model based on hallucinations that disappeared a year ago or judging a model based on some, oh, it has a cutoff window and then that hasn't been a thing for a year but it still burned its way into the memes of the online culture and thus people need to be updated on that. And so I'm in favor of coming up with good computer use demos. The real one is just, can it play Counter-Strike effectively? That's the real computer use test. Can it rank me up to global elite? Can it downrank me so I can smurf a little bit? This is the future. This is what I want. Anyway, let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more. Railway automatically takes care of scaling, monitoring, and security. I'm extremely excited to welcome our next guest to the show. We have Vincent Weisser from Prime Intellect. He's the co-founder and CEO, and he has some amazing news for us. Tell us what happened, then we'll dig into the news, the market, the story, the company. But first, give us the news because Jordi's already warming up over here.
Amazing. Yeah. So we've announced that we've raised $130 million at a billion with Prime.
To build the open superintelligence stack.
It's fantastic. Congratulations. Yeah, what we were talking about before the show is the ratio between dollars raised to run rate.
Absolutely. Incredible. Yeah. How much have you raised before this round?
Just $20 million total. So, I think we've actually got to this runway on like less than $20 million in spend.
Which is north of... which is north of 100.
Yeah, we doubled it, I think, since we actually closed the fundraise. Yeah. To 100. So basically, yeah, on track to grow much more from here on.
Okay. So reset me on the latest and greatest in the actual product offering. What customers are coming to you? Is it, I have used a frontier model to figure out that AI can in fact do some agentic workload within my organization. I want to cut cost and you're going to RL an open source model for me, fine-tune it. And then am I paying you for tokens? Are you getting me GPUs? What is the full suite of products that you offer customers these days?
Yes. So basically it is kind of like the...
Full stack to do training, deployment, and continuous improvement of models. I think customers basically come to us for anything ranging from compute to inference to the full RL and post-training stack to train models. And I would say it's like with my customers coming from different buckets and categories, a lot of them are like the AI natives who are basically starting out and oftentimes also like being multi-model, right? They use like the frontier models, they use open models for different use cases, like a good example is Ramp, for example, as one customer more in this AI native bucket, and a lot of the big AI native startups. But then we've also increasingly been able to attract a lot of the new AI labs, like a lot of the neo-labs. So a lot of those are using our full stack from like large-scale clusters to our pre-training and post-training and RL stack, but then also increasingly like traditional enterprises and even sovereigns coming to us who want to have their own end-to-end stack that they can run like on-prem.
So how I see this is to some extent is like there's just increasingly, I think, a demand for having your own software and AI stack, right? It's like your own open end-to-end stack, but I think this is almost like complementing in many cases the other frontier models and stacks.
And you're using sovereign in not just the geopolitical sense, but also if you're a Fortune 500 company, you want to own your stack from start to finish, potentially.
Exactly. Yeah. I think like there was a lot of recent talk about this, obviously, like folks like Satya and Alex Karsh were also making those points and arguments over the last few weeks, and I think it's increasingly something where a lot of enterprises want to basically build this compounding data moat and flywheel, where ultimately I think the future that we are starting to see is a lot of these companies want to build basically self-improving agents. And the way to do this, I think, is generally you have an RL environment for your use case, you scale on it, and then you ultimately deploy that into production and then continuously improve with the user interacting with that agent. It's almost like the Tesla autonomy levels, but for knowledge worker agents. So I think like the Ramp example, I think, was one that people really resonated with for basically them being able to within a week and less than 50k of training spend outcompete a frontier model at automating spreadsheets and finance at a fraction of the cost of running the cheapest models, basically. Like they were able to outperform Opus at a fraction of the cost of Haiku and like 30% faster, like much more tailored to the use case. So I think in many ways like these specialized agentic use cases are a big unlock, and I think in many ways you still need the godlike big mega model to do the orchestration or planning, but then I think oftentimes execution can happen with these specialized step agents. So I think, yeah, those have been extremely fruitful and useful for a lot of the companies adopting and training their own agents.
Then I think the other big trend is just like an explosion in new entrants that are taking model training seriously. Obviously hundreds of neo-labs really got stood up in the last probably 12 to 24 months. So we started powering a lot of those with our full stack, including with also large compute clusters. So like NVIDIA also joined us as part of the round, and we've since day one basically operated large clusters, and this is in large part why we need so much money, right? It's like to basically be standing up more and more GPUs. Like right now we run like 15,000 GPUs and scaling to like over 30,000, and so this is like a lot of the capital needs even.
And walk me through when you're pressing the huge cluster button. Is that because there's broad demand for specific models that you need to inference at scale, and so there's just a lot of inference demand, or is it that for the level of training that you're doing, you need to be cluster scale to even kick off the training run?
Yeah. Yeah. So basically it's for all the needs from training to inference, you need obviously compute, and in a lot of ways like our customers are doing a lot of these large-scale training runs with us, leveraging our full stack including our compute, right? So it's like pre-training for billion-parameter models, like for example what we did with RC, like Q4 last year, or really large post-training runs and RL runs, right? So it's like increasingly basically we abstract all of that away so basically people can just hit a button to do a post-training run without having to lock into a GPU cluster and managing the bare-metal GPUs. So we basically abstract a lot of this GPU complexity away. But yeah, for that we're basically also ramping up our own clusters to basically just serve hundreds and thousands of customers in parallel on those.
Yeah. So what you're saying is like there might be a situation where I have a 50k budget, like you gave with the Ramp example. I have a 50k budget to do some post-training, and I need to deploy that 50k into like thousands of GPUs, but only for a short amount of time, and so the budget's actually going to be relatively small, even though if I were to buy all of that, it would be way more than 50k, millions and millions of dollars. So you sort of absorb some of that cost, orchestrate it, abstract it away, and then you can deliver for the customer just a slice of it.
Yeah, so basically I think this is also a big unlock, like also on the inference side, right, where customers don't need to frontload hundreds of millions in capex spend for large data center commitments.
So we basically also increasingly serve a lot of inference. So it really is the full stack, and I think a lot of the compute, like we have quite a clear visibility into the demand just because people come with us with large requests or large contracts to basically do larger training runs and deployments. So we basically just need to secure the compute to be able to fulfill all of those. And I think right now we would probably be at many orders of magnitude more compute, like compute but also revenue, if compute wouldn't be as constrained as it is right now. So it's like right now I think in the current environment, GPUs are basically sold out for the next few months and for almost the rest of the year, especially if you want to stand up data centers in the US. So I think this is actually the main bottleneck to scale right now for anyone in AI, and this is what we've been able to really scale through extremely well because we've basically partnered with almost every data center out there since over two years.
So basically able to stand up like data centers in different geographies and orchestrate them globally.
Any predictions, forecasts for American open source over the next six months?
Yes, no, really great question. We joined also like NVIDIA to help them train Nemotron. So they're doing the Nemotron Alliance and coalition, and so we were one of like I think 10 or so partners. So it's like I can't speak about all the things there. But I think in general, both on that front and other things we...
Anything just between us, just the three of us.
So basically, I think in some ways the leading American efforts actually are Nemotron and actually like RC with the Trinity model. So both of those we're very deeply involved in, and so I think those will continue to lead with American open source. Like we have some of our own models cooking, and then we help a ton of customers with theirs. So I think there will actually be a huge resurgence of frontier open models coming out of the US, also led by players like NVIDIA itself and a lot of our customers and partners. So yeah, I think in general, I think the interesting geopolitical question is actually what happens with Chinese open models, right? It's like obviously we've seen news that China might consider export restricting them, and there might be policy interventions on both the Chinese and the US side, right? So I think this is actually in some ways the motivation to build a stack and to also push American open source forward is to have less reliance on models that might get banned any day, be it more closed. Right, it's like I think this is something increasingly a lot of enterprises are starting to realize that there's actually huge dependence in a sense, mainly from the policy side. So this is something I think where we want to just enable everyone to have their own end-to-end open stack to be able to create their own models. And ultimately I think there's obviously a lot of now also neo-labs, and a lot of them are actually thinking about things in quite an open way. So I think we'll be surprised that that's maybe one of the hot takes is to see more neo-labs release open models. Like a good example that we also partnered with recently was at Poolside, for example. If people were expecting them to be closed, but they actually opened their models. So I think there will be more of those, especially given there are hundreds of well-funded neo-labs now. I think there will be a huge resurgence just of the age of research in terms of these neo-labs pursuing truly novel directions, right? It's like going where the big labs almost can't go necessarily, taking big bold bets that also might not all pay off, but if you have 100 shots on goal, some of them will probably pay off and result in interesting outcomes. So I think this is really also what we are empowering, right, is kind of helping these new neo-labs and big AI natives and enterprises to move much faster and not have to basically all build the same stack each by themselves in-house, because obviously a lot of them are like tiny teams of tens of people instead of thousands. So we're able to help them move much faster basically by adopting ours.
When you hear of a major American tech company, let's say someone in the Mag 7 that has extra capacity, do you reach out to them? Do you ping them? Are you at the scale yet that a deal would be interesting to them, or are they really searching for these, you know, $10 billion plus opportunities?
For sure. Like we already are sourcing pretty large clusters, right? It's like sometimes up to like 5 to 10,000 GPUs. So we're basically looking at every source and pocket of supply that we can find. Like we've literally turned every stone to find the last GPU available on Earth. And we've been quite good at that. Just like in some ways, I think that has been one of the core tenets almost since the beginning, is that we really wanted to orchestrate the global compute supply efficiently. So we did a lot of optimization for tolerance and distributed training even in the early days to train across distributed compute. So it's definitely, I think to this question, something that we're definitely considering. Like if there's anyone with some spare GPUs, you can always...
Back on China. Do you agree with my perception that if there is some sort of lockdown on Chinese open source export controls, that would not affect...
Are they mad that American companies are distilling on or post-training their open source models by design? I thought it was a goal of the project. But what I'm interested about is, would they try and claw back access to Kimi or GLM, any of those models that have already been deployed? Because there's an element of like once the weights have been downloaded, it's sort of just out there. Of course, you could go and try and sue every company, but that's very hard if you're in an American jurisdiction and you're a Chinese company. But how would you see that playing out? Would it just be going forward, no more open source, and that's the CCP telling Chinese companies, or some other shape of restriction?
Yeah, I think it's really hard to know. I think like a good example was like in some extent the most popular Chinese model already has been closed, for example, which is Qwen, right? Like in some ways they've actually changed their licensing, stopped open sourcing. I think there will be both companies changing their policies, right, but then also I think probably state intervention, right, in different shapes and forms. So I think basically one can't rely on the steady stream of frontier open releases out of China in many ways, right, from different perspectives. And I think similarly a lot of other sovereigns, like obviously including the US, also doesn't want to build on necessarily China. So it's like I think a lot of American enterprises obviously also like sovereign use cases need American models. But I think similar for other continents, right? It's like I think there's been a lot of work obviously, and we're collaborating with people also across Europe, across India, across other geos. So it's like I think we'll see a much more multipolar open ecosystem of open frontier models. So I think that's a healthy balancing away from just relying on Chinese open models, even though they've obviously been a great accelerator of open source progress. I think it's also unhealthy to rely just on an ecosystem, and yeah, I think there will be a bunch of quite unpredictable changes, similar to how the last few months have been quite unpredictable to everyone involved. So I think this will continue to be the case in the sense that it will be quite a dicey AI policy landscape for the next few years.
Mhm. Last question. Data center construction, new data center construction, is deeply unpopular in America domestically. Simultaneously, the workloads that are being done in AI data centers, latency is basically irrelevant. For training certainly, for most inference tasks, if you're waiting a couple minutes for a result, an extra 500 milliseconds to get across the world isn't going to matter. This feels like a recipe for global competition, for aggressive data center expansion in a country that does see it favorably, that does see an economic boom from it, or does have a lot of energy. What is the state of the non-domestic, non-US data center buildout?
Yes. So we're certainly building a lot of data centers and partnering with a lot of them outside of the US as well, but I think there's still a lot of demand actually for data centers in the US. I think like even from a data perspective, there's a lot of now the financial industries and highly regulated industries also adopting obviously more and more agents and compute. I do think inference still matters, like speed in a sense, like maybe not for long-running agents, but for a lot of use cases people still really care about having extremely snappy responses. But I think there's definitely a general boom in data center buildouts outside of the US as well. I think there's just a limited set of countries who are strong allies with ample amounts of geopolitical stability, but then also energy and green fields and enough space.
Yeah, US has unique access to natural gas, which is if you're trying to bring energy online super quickly, it's hard to beat.
Yes, and I think we're seeing this across Europe even, where there's specific regions that are much more active in data center buildouts than others, and those are generally the ones that have their own money to invest in them and cheap energy and a lot of local support for it. So I think in general from what we're seeing, there is new entrants almost to the data center buildouts, where it's like even places like Australia, New Zealand, and smaller, much smaller places, right, like Armenia, Kazakhstan, and others who have a huge amount of energy and are making huge bets also on data centers, but while also being allies of the US. So I think those are the countries generally which will see the biggest boom in data centers. I think at the same time we'll also continue to see a boom in US data centers, but yeah, it's definitely harder to get data centers in the US than outside of the US, and so it's something where I think we'll see both continuously expand. And then I think there might also be obviously, like to your point, a policy updating with upcoming elections and midterms. I think this whole almost populism in the US around data centers and a lot of the kind of public sentiment, as potentially also sovereign influences and other things from other nation states who are trying to harm almost the US data center buildout. Yeah, so I think there's a lot of things at stake for the US, right, to not pull back on their data center buildout, but I think there will also be specific states leaning in much more heavily than others, right? Like I think this is already what we're seeing where like...
So it's like there's specific states in the US that are just much more well suited to really ramp up data centers from here. And yeah, I think we'll see them ramp up globally and hopefully also outside of, like beyond Earth. And I think...
Like I think this is the only way to really properly scale.
I love it. Well, thank you so much for coming on the show. Incredible progress. We love seeing you and the team win.
Fantastic. Have a great rest of your day. Say hello to Will for us, and we'll talk to you later.
Goodbye.
Let me tell you about Cisco. Critical infrastructure for the AI era. Unlock seamless real-time experiences and new value with Cisco. Our next guest is Ben Thompson, the founder of Stratechery. He's been on an absolute tear during a slow news cycle. Does it feel like a slow news cycle to you?
I don't know. And the summer is always slow. I, you know, I take a couple days or I slow down a little bit myself.
I feel like you've done a good job of providing really insightful and interesting pieces during a slow time where I didn't know that I wanted...
Yeah, you are. Is that what I'm hearing?
I didn't know I wanted you to adopt the voice of Mark Zuckerberg for 20 minutes and go on a tear, but it's not like that was driven by like a specific, okay, there's a Wall Street Journal article where like or Bloomberg has a scoop and like we need to know this. It's more just like resetting on the Meta narrative. I want to talk about the Meta narrative, but can we start with Xbox?
I want to know first, let's go back in time a little bit, and I want you to reality check me on the story I tell myself about how Microsoft wound up in the situation it wound up in with Xbox. And I think it stems with DirectX, the graphics layer that allowed games to be run on multiple graphics cards. So that was a Microsoft-licensed piece of property. So this is why we don't have the Chrome... You want to go really far back, because we didn't get an Amazon gaming console, we didn't get a Google gaming console, right? Like why...
It's interesting. So you want to go way back. I mean, we can go really far back here. I mean, back in the day, the main technology for doing 3D was coming along was OpenGL. And OpenGL had various pluses or minuses. It was, you know, we're way back in history, but this is the origin of the NVIDIA story, making cards to do 3D graphics. You had like the Riva, the 128, then you had the TNT, then you had the GeForce, and like we're talking back in the 90s when I was, you know, your guys' age or whatever it was in college. And you know, making the first, you know, Quake comes out, like the real first 3D game. You know, there's actually a really interesting Twitter back and forth with the old id people talking about Quake and how basically they obliterated the company by burning everyone out by making it. But super fun, super fun back and forth. But Microsoft comes out with Direct3D as an alternative to OpenGL that was, I think, and this is a little out of my area of expertise, but I think was a little easier to use, had more niceties, was more tied into the platform, but this is all sort of PC gaming. Yeah. The Xbox, the way I think about it, and a lot of gamers kind of get annoyed when I write about the gaming stuff because I don't really care about the game part of it. I'm more interested in the strategy and how that fits in the overall company.
What is your review? Can we get a GTA 6 review?
No. Well, I feel compelled. Number one, I feel Rockstar, I think, is charging way too little for this game. They're charging what, $80? Like, this is a game...
This is... You're going to get cancelled for this. This is a nuclear take.
No, they should be charging like $200 for this.
GTA 6 is like the last great game. Like you think about it, it was all made pre-AI. Like it is the pinnacle of this AAA craftsmanship. Years and years and years of blood, sweat, and tears to the extent where you have like the Twitter analysts like counting cigarette butts outside of the offices to see like how much crunch are they in right now.
I don't even think he's trading the stock. I don't even think he's trading the stock. I think he just wants to know what GTA... the love of the game. Literally the game, the GTA.
Gamers are insane, so I always tread in this water very, very carefully. But no, so I feel compelled to buy GTA 6 just in honor of it existing, even if I don't know if I'm ever going to play it. So I'm definitely going to do that, and I'd be happy to pay. Yes, $200, like $80 is ridiculous. They should charge more. Anyhow...
Yeah, we had one, the youngest person on our team was commenting on people online complaining about the price and said something like, you didn't have... you had 10 years, you couldn't save up 80 bucks?
Seriously, whoever you are... but like I know kids that do lemonade stands and they clear like a hundred bucks, you know?
Oh, absolutely. You know the old $20, I'll keep the change. Nothing will make a kid's day better than that. Although unfortunately with inflation, I think now it's sort of like expected. It's unbelievable how expensive things are these days. But...
So anyhow, I got GTA 6, but we're back. So the console era. So the way I think about Xbox is you go back to Microsoft, and Microsoft had this whole thing. This is pre-iPhone. This idea of their corporate strategy was three screens and the cloud. And the idea that we're going to... we already own the desk via the PC, we're going to own the pocket via the phone. And people always used to say Microsoft missed mobile. Microsoft did not miss mobile. They came up with Windows CE in like 1999. Like they were in mobile. They just had the totally wrong conceptual model because they were basing it around, you know, as an extension of the PC. But then the third screen was the living room. How do we get in the living room? Well, people have consoles plugged into their TVs. If we can get in and own the console, we can own sort of the gateway to the living room. And we need to earn permission to be in the living room. We will do that by creating the Xbox, the game console. And what's interesting about this is this is a strategy that totally failed. And it failed for a kind of an obvious reason in retrospect, which is first and foremost, people buy consoles to play games, not to have a portal to the internet in their living room. And so you're selling a multi-hundred-dollar console to people who want to play games, but people who don't want to play games aren't going to randomly start buying consoles. What actually ended up taking that over was things like the Fire Stick, or, you know, Apple TV. Betamax played this. I still think it's like probably the best solution, but it's way too expensive. Now, the Apple TV's price got jacked to the moon with this recent memory increase. But the Fire Stick is probably a big one. You had the Chromecast, you had Roku TV, you had, you know, these, which ended up being... it's not that... it's a classic Microsoft thing. They were actually right. There was this new front page to online services to be built. They just took totally the wrong approach with Xbox. And what's funny is that this approach actually became this uber act of self-mutilation because the Xbox didn't succeed in achieving this because it's a games console that's way too expensive to achieve this. But then also they killed the Xbox where Xbox 360 was pretty competitive, and that generation is actually the most interesting console generation. I'm happy to expound on that if you want to. But they come out after that with the Xbox One, I think, which was super expensive because it had the Kinect built in, that sort of wave-your-hands-around sort of thing. And it had all much more restrictive DRM things. It was going to be more digital. It was basically that console was designed around we're going to realize our vision and own the living room. And gamers hated it. So they rejected it. So their vision, which they didn't achieve, also killed their console. And so my... I wrote over a decade ago, and the whole Xbox saga is filled with regrets, I think, first and foremost for Microsoft, but also for me, because from the beginning I was like, they need to kill this program or they need to spin it out or get rid of it. The stated goals, which made sense from a corporate perspective, were not achieved, and it is continuing to exist because it exists. And I was actually sympathetic to that. I worked at Microsoft from 2011 to 2013 when Microsoft was in the dumps, and you know, they still had a great business, but their stock had been like $40 for like a decade. And the one thing the employees had was like, well, at least we have Xbox, all the cool kids over there. Like it was such an important thing to corporate culture to have Xbox, even though this was a totally failed division in terms of achieving corporate goals and also was sort of self-immolating because it was trying to achieve corporate goals that it was not going to achieve.
Yeah. What do you think the naming scheme says about that project? I always found it very odd that they went from Xbox to Xbox 360. That sounds like the third. It's actually the second. Then Xbox One is the third, but it's one. And then now they're on Series X, which feels like the 10th, but it's the fourth. And then they have a bunch of subnames. It feels like they just never leaned into just like Xbox 2, Xbox 3, Xbox One.
Yeah. Well, you're asking... I mean, maybe it shows I worked at Microsoft. You're asking the wrong guy about naming things, I think. But yeah, total mess. They called it the 360 because it was coming out the same generation as the PlayStation 3, and so they wanted to... well, they wanted to be on the same level. Like, we both have sort of a three in our name.
So what's interesting about the PlayStation 3 is the PlayStation 3 was Sony's big misstep in terms of the whole PlayStation era. They had this crazy processor called the Cell processor that had all these capabilities, and up to that point consoles were differentiated by hardware. And so Sony famously helped... one of the reasons they beat Nintendo was they had discs, so they could have much larger media and things along those lines, whereas Nintendo was sticking with cartridges. And there was like other... they were much more favorable to third-party developers just in general. They had sort of taken over in PlayStation 1 and 2. PlayStation 3 comes along, they produce this... Oh, they wanted to get Blu-ray off the ground. That's another example of consoles getting screwed up by trying to attempt other corporate priorities. It worked.
Didn't they win? And didn't Xbox go with HD DVD and they lost?
They did, but at the end of the day, what happened was the PlayStation 3 was way too expensive. I think it ran super hot, if I remember. And all these extra capabilities were not used. And the reason they weren't used is that was the first HD generation, and you had this shift to 3D games and high-def graphics. And what happened was it used to be games differentiated themselves by how well they wrote the game, like how did it perform on your Super Nintendo, on your Nintendo. Now we moved to a world of actually all the actual game was offloaded to engines, game engines. Like, you know, obviously Epic got super big with the Unreal Engine, and so you no longer did that. What a game company did was actually design the game and then asset creation. And asset creation at HD was way more difficult, way more expensive than it was for standard-definition TVs. And so the economic imperative for developers came, we can't just be on PlayStation. We need to be on everything. And so that was the ultimate generation. And almost every game was made to run everywhere. Run on PS3, run on Xbox 360, and run on the PC. And that was also the Xbox 360's best generation. That was the one they were the most competitive on because they had all the games, because all developers were making them sort of run everywhere. They had a very simple-to-develop-for architecture. The 360 was very PC-like, even more so than the one previously. And that was their best generation. And then they come back the next generation shoving Kinect down people's throats. We're going to own the TV. And meanwhile, Sony learned their lesson. They're like, okay, we can't differentiate on hardware. We need to create sort of generic hardware that is easy to develop for, and the way we will differentiate is through exclusives. They bought a bunch of small studios, in contrast like buying Activision, which we can get to a little bit. They bought a bunch of small studios and pushed all those studios to developing killer games for the upcoming PS4. And so you got things like the Spider-Man, that was a huge game, Last of Us, that type of stuff.
Exactly. I mean, and so the PS4 was an unbelievable generation, and a lot of the reason it was unbelievable was because it had these exclusive games, and they just crushed Microsoft because for all the cross-platform games, the Call of Dutys, the Maddens, things like that, those always ran everywhere. And so what you do is you limit the economic potential of your studios because you're only building for your platform, and you make it back by all the licensing fees you get off the EAs of the world and the Activisions of the world because 80% of gamers are on your console, not on the other guy's console. And so they just wiped out Microsoft in that generation. And that really just continued to the PS5. And this is where we sort of get to the Game Pass disaster.
Okay. Yes. So, yeah, I want to... Yeah, let's just jump forward to that. How was it? I didn't want to ramble too long. I was going to give you a chance.
We love it. I have a bunch of other questions not related to gaming, but I love this history.
I think the console history is fascinating. It's really interesting how sort of because it tracks technological change to such a large extent, and it totally shifted sort of strategy. So, and I think you can draw a lot of real parallels to even just model releases over the last couple years. How, you know...
Certain decisions around one model sets you back a certain number. You know, they're iterating a lot faster, but I'm seeing parallels.
Yeah. The sort of development of... Yeah, it's interesting. You could always sort of draw lessons. So you get to Microsoft at this point again. I think I wrote at the time, okay, time to give up. Like I wrote multiple times through the years, they should give up on Xbox. And at this point...
Was there ever an option after they bought Activision just to make Call of Duty a console exclusive, or would that have been an FTC problem?
Well, we'll get to that because that's part of the story here. So Microsoft decides, okay, what we're going to actually try a new business model. It's true, Xbox, sort of selling a console, selling, you know, licensing it. We kind of do that, but we're actually moving to being a services company. The stock's doing much better now. You know, like we have Azure, all these things. What we need to do, and the way we compete with Sony, is we need to have the exact opposite of the Sony strategy, and we need to lean into services, lean into subscriptions. We're going to do Xbox Game Pass. And the idea of Xbox Game Pass is instead of buying every individual game for $60, $70, $80 — gaming pricing, by the way, far longer than it should have been, particularly for these AAA games, but we already got to that with Rockstar — so they're like, you pay one price per month and you get access to all the games in Game Pass. Well, of course, no publisher wants to partake in this because they make all their money upfront by selling games to their biggest fans. You know, it's kind of like the movie model. You make a bunch of money upfront, and yet in the...
Were they looking at what happened to the music industry as an example of saying like, yeah, it's cool to make your product more accessible. You get more listeners by bundling.
That was the bet. You had to make it up in volume, right? And so this is why the Series S came about. There's the Series X and the Series S. The Series S was a more underpowered but much cheaper model. I think it launched at $300 versus the X was I think $600 or maybe $500. And that was also a problem because you're kind of neutering the X because developers build to the lowest common denominator, not sort of the highest one. So that was one problem. But the idea was we're going to expand the market, and they have these package deals where you get an S and you get Game Pass for a year, and it's this sort of thing, and we will make it up. And the problem is no independent developer was going to go for that. So they had some of their own games, but they went out and started buying a bunch of developers. So they bought Bethesda, which included I think at that point ZeniMax. Like they have a bunch of great ideas, but those developers just aren't releasing incredible titles annually, except for Activision, which does release...
Well, exactly. And so the issue with buying these companies is what you're paying for is you're paying for their existing business plus the premium. And their existing business was selling single copies of games to all platforms. And what they were saying is we're going to basically force all these studios to take a bath, to lose a ton of money by putting all their games on Game Pass. We'll still sell on PlayStation because we have to, because that's a lot of money. It would be massively economically destructive not to. But what we think is going to happen is gamers are going to realize I could be paying $70 a pop on PlayStation, or I could go to Xbox and I could pay $20 a month and I can get all these games, and that's it. And what a better deal. And by the way, we'll get new gamers. People are going to come in and realize, wow, I can get all these games for $20 a month. None of that happened. All they did was cannibalize their existing business. They didn't expand the market for console gamers. All they did was have gamers who would have paid way more to play these games pay way less to play these games. And Microsoft ate it. And I think they, you know, the recent reporting was they expected to have 75 million Game Pass subscribers by this year. They have like 30 million. And that number is actually decreasing. It's not going up. And so it's just been... and so that's sort of the context of what's happened here. Phil Spencer got retired, which if you retire at that level of a company, you're probably canned. Probably too late. The whole strategy disaster. They brought in new leadership. And now there's sort of like a reset going on. They're laying off a few thousand people. They say all games production are going to continue, but there's a real sort of critical decision here. What do we do? Because we ended up getting stuck in the middle, and we're just... this doesn't work.
What did you read into their approach of laying off 1,600 people now and keeping a 1,600-person RIF? I mean, it's a terrible idea because they know like everyone there for the rest of the year is like... like every VC would tell you to cut hard, cut deep, and then move forward with the team. If you're on the ship, we're going to Valhalla together.
I think I do have some sympathy because the reality is I appreciated the note because they are in big trouble, and there's an aspect of just stating it clearly, we're not sugarcoating it. And if they laid out how much trouble they were in and then only cut 1,600 people, then everyone kind of knows they're going to be cutting more people, right? So I think they're just in a bad spot. So of course it's going to destroy morale. It's going to destroy everything for the next year because everyone's waiting to get fired. But there's a bit of... they did, I think, tear the band-aid off to the extent they could. And now there's just a... now they can be very overt. They can go through and actually talk to employees about what are you working on? Why should you still be here? And it's a reflection of the employment climate. Like people in general are more... they're just not going to bail. Microsoft, you know, they're going to want to keep their job, and so they can probably hopefully, fingers crossed, do better cuts by virtue of doing them in the open. But I don't know. I mean, I don't envy Xbox's leadership at all. It's going to be a very tough thing, but I can totally understand how they are where they are.
Probability of a spinout.
I mean, I think it needs to happen. Like, but you know, this is again the regret is this was my take for years and years and years. I'm like, okay, fine, Xbox Game Pass, that seems kind of interesting. Let's go for it. I should have stuck to my guns. So like, no, stop. This is going to be a disaster. I think, yeah, I don't... I don't get it. I just don't... I didn't see it before. The thing that got me to say, okay, let's give it a chance, didn't work. I don't see it now. I mean, I think it does give me pause. They hired Matthew Ball, who I've had on Stratechery multiple times as their chief strategy officer. I have a lot of respect for his takes. So I'm curious what he's thinking about.
This gets to the question before. Is there a bit where they just write off these acquisitions and say actually all this stuff's going to be exclusive now, where like Call of Duty, buy an Xbox, like basically giving in to the FTC's worst fears, which... but those fears didn't make sense at the time in the context because it's... by writing off what's funny, we lost billions and billions of dollars.
Yeah. And it would be a good pricing test for the power of a AAA game to say, okay, you can only get Call of Duty on Xbox now. And you would see like how many incremental Xbox sales do you get, because that means the real value of the game is actually in the hundreds of dollars if people have a PlayStation and then they'll try something new.
Switching gears a little bit, we keep having this experience where there's a company, you know, an AI company, let's say an application layer company that is or lab that is, you know, hot, but then you don't hear very much for a few months and you write them off, and then they come out and they've like added hundreds of millions of dollars of revenue and are doing a crazy up round. And it's this weird thing where like even, you know, historically, you know, there's power laws everywhere, but I feel like in this moment we're seeing more and more examples where like the 10th best company in the category is still having the financial performance that a company that would have been the leading company in their category 5 years ago. What do you make of that?
This is just a thing in tech in general that has always sort of stuck out at me. I mean, I think I...
Has it been though? Like CRM, for example, like if you go back to like 2015, was like the best CRM company...
Actually no, the example I always go back to is I think it was 2014 or so when Apple bought Beats, and they bought Beats for $3 billion, and everyone's going, the tech press is going crazy. Everyone's writing thought pieces about it, and I'm like, you know, of course writing about it. I'm like, $3 billion? Like, Apple is a $500 billion company. What does $3 billion matter? And what's funny about that take is number one, it was right. Who cares about $3 billion? That's in our couch cushions. Like, people were spun up about something that didn't really matter. Number two, Apple was only a $500 billion company. Like, the astronomical increase in valuation and numbers around this is absolutely crazy. So to your point, yeah, a 2019 CRM company, number one, there probably wasn't a number 10 even then. But yeah, the numbers are not even remotely tied to like any... I just sort of let them go and be like, I write down the numbers and they don't really mean anything to me. But it's very hard to keep track of like the inflation in tech numbers specifically. You know, you think that... what was I complaining about pricing before? Oh yeah, lemonade stands are bad. The Silicon Valley lemonade stand inflation is absolutely the worst in the world.
On mergers and acquisitions, there's this news today about Getty Images calling off the Shutterstock deal after UK regulators said that they would have to sell the editorial business for the merger to be approved. And it seemed very odd to me that a merger would be blocked by a regulator when that industry that is consolidating and is maybe becoming more monopolistic is under such insane attack from generative AI. I can't imagine a better defensive move than putting Getty Images and Shutterstock together.
This happens. This happens every single time. Like the natural response to your overall category being threatened by an external force is consolidation. Like it's the only way to defend yourself, to sort of, you know, you get synergies, like all these...
Companies are pretty established, so it's actually fairly easy to price, you know, exactly what things are. It's a much... like if you have growing companies you want to merge them or acquire them, it's really unclear because how do you sort of price out the future? These companies are declining, so it's very easy, like you could do a discounted cash flow and understand exactly how much they're worth. It's a very straightforward negotiation. It makes total sense for them to consolidate.
But precisely because they're knowable and understandable, it's very easy for regulators to block them. And you see this happen again and again. You see like pushback on... you saw this with TV and like different networks getting together and the various pushbacks happen to that, or in telecoms. Like this happens, this is a natural pattern. The moment that a company, an industry wants to consolidate, needs to consolidate, is the moment regulators are most likely to stop them consolidating. And whereas the actual time when action would matter, like when it's growing and a company is making the right acquisitions to take over an industry, regulators don't do anything, and also rightly don't do anything because you're dealing with the future and it's unknown. You don't want to screw it up.
Um, it raises the question of what use do regulators sort of do anyway in broad, broad strokes. I used to be... I've sort of, you know, I think that they probably do more harm than good as far as a lot of these sort of blocking mergers and things go, but um, I don't know. I'm the one that just said maybe Microsoft should take Call of Duty exclusive. So, so who knows? Uh, maybe I'll...
moment because we were looking, it's a $3.7 billion deal, but I'm like, how is this a $3.7 billion deal? These are two 300 million-ish market cap companies, and it was because the deal was like priced back in January of 2025, and so both companies had declined so much combined market over the last 18 months, and so maybe they should do the deal anyway. I don't know.
I mean it could be worse. You could be Netflix and have the stock market sort of like make a bid for Warner Brothers that the stock market hates, abandon the deal. But they already decided you did the deal because you're desperate and you're screwed anyway. So, um, yeah, and you don't even get what you bid for. So, uh, maybe that's what happened here. I don't know.
Oh, well. Well, thank you so much for taking the time to come chat with us. Have a, uh, have a great summer. Uh, enjoy the break and we'll see you soon.
Yeah. Good luck with the lemonade stands.
Yes. Try to make it. Don't go into debt.
You're going to run up on a kid that has Clarina.
Raise some money. Yeah, I know. Exactly. Yeah. Claron lemonade stand.
Clarimate stand. Five easy. Silicon Valley free. Interest free.
It might. Well, thank you so much and have a great rest of your week. We'll talk to you soon.
Have a good one.
Let me tell you about Shopify. Shopify is the commerce platform that grows with your business and lets you sell in seconds online, in store, on mobile, on social, on marketplaces, and now with AI agents. And our next guest is Rodrigo Liang from SambaNova. The co-founder and CEO is here with us in the TV realm. Rodrigo, how are you doing?
I'm doing good. How are you guys?
We're doing fantastically, but it sounds like you're doing better. Give us the news. What happened?
Well, you know, we're super excited. We did a big, big funding announcement today with a billion dollar round that we did a first close on at an 11 billion valuation. So, super excited about announcing that.
There we go. There we go. This is reminding me there was somebody posted like a couple weeks ago something to the effect of like every day there's a new billion dollar deal in chips from some company you've never heard of, and uh, I guess the trend, the trend continues.
Yeah, uh, yeah. Is this an overnight success? When did you start the company? How long has this been overdue?
It's an overnight nine-year success. You know, I've been building chips for 30 years, a high-performance processor, and we started this company in 2017. So no, but it's been a great journey. It's just, you know, again, it's so noisy. So there's so many companies out there doing also all sorts of awesome things, but look, then, you know, inference is now right in front of us. It's kind of broken everything open. It's what we do and it's been what we've done for a long time. And so it's the new training run. And so people that have kind of been thinking about building just for training, well now it's the time for inference, and this is where we shine, and we shine, you know, and investors are putting all their money behind that, and so we're excited about that.
How, walk me through the phases of... I mean, you have some of the biggest companies in the world you're working with. JP Morgan Chase selected you for secure on-prem AI inference. And I'm interested in the thought process that large Fortune 500 companies, Fortune 100 companies are going through right now. I imagine that there's a lot of explore with frontier models, very expensive, they're token maxing. Then they shift to, okay, this agentic workflow is working, let's use an open-source model for it that is, you know, either the open-source frontier has caught up and can do this particular task. Maybe there's a fine-tune or some RL on some private data that makes the system perform better. When is the conversation shifting to actually optimizing the inference stack, the hardware? What's the conversation like there? What are the savings that people are focused on?
Well, I think what people are starting to realize after they started using ChatGPT and the frontier models is, look, there's a bunch of things I can do there and it's great. You know, we'll use it and it's going to be part of their business for a long, long time, right? And hyperscalers are offering all sorts of different services with different TPUs and GPUs and training chips. They're offering different things. They're again on the cloud. What's happening now is this next phase of production is people starting to realize, okay, what happens to the stuff that I'm not comfortable releasing to these models, like my private data, my security, the stuff that I want your models to learn and then somehow become part of the global knowledge, right? And so what happens with that, right? So I don't know if I have, you know, my client's banking information, you know, some very regulated information, what do I do with that, right? And so people are starting to realize, okay, there is a whole another class of workloads, a whole another class of applications that you got to go and do private. And so with someone like JP Morgan, which is fantastic, you know, I mean they are the top of the top when it comes to enterprise and IT, right, and technology, and so fantastic partners. So they selected us as the ones that can come in and go into an environment like JP Morgan, do full on-prem, secure, private. They can bring their open-source models, they can fine-tune into it, they can deploy that privately, securely, and control the models and control it forever, right? So I think we're starting to see the landscape start to steady, right? There are some things you use the hyperscalers for, there's some things that you're going to use the hyperscale clouds for, and there's some things you're going to say, hey, let me run on-prem and let me control the outcome, and especially since I don't know how the regulatory environment is going to settle. Let me just be secure for that. Let me just be safe and just run those things on my own models in my own environment with my own security around it and then be done.
Yeah, you can think, you know, a pharma company being like, I would love for all the other large pharma companies to share their data with a frontier lab. Totally. Uh, but I'll pass. Mhm. What does a partnership with you look like for a large organization? If I have, you know, a large IT footprint already, I probably have data centers, but they might be more CPU-based workloads. Maybe I have some GPUs, but I'm not doing, you know, frontier inference. I'm not fully scaled up, but I have some. Are you consulting with them at the level of like where they should be buying land for their next data center? How they set up a powered shell, who they should partner with, like how deep in the stack will you go with a company that you're working with?
Well, look, you know, I mean, if you look at the AI ecosystems up to date, you know, we've been really focused on hyperscale clouds, you know, the neo-clouds, the frontier models. We're really focused on building those out, you know, gigawatts here, megawatts there. I mean, just, you know, just this large-scale deployment. And then what you're still thinking about is, well, how do I get as many tokens out per, you know, megawatt, per kilowatt, megawatt in Google that I have, right? So that's kind of we're very focused on that. Some of what we play into that are really, really efficient models. We run the big models really fast. That's really what we're known for. Huge models, right? You know, the battleground for power and data center and all that is about huge models. It's not about the little models. It's about the trillion, multi-trillion parameter models. That's what we focus on and we run that faster than anybody else in the world. So, but then you flip it over to the enterprise, and for them it's not as much about how many thousands of racks going to fit into an environment. It's about data privacy. It's about security. It's about having the capacity when I want it. It's about low latency. It's about these things that again you go back to traditional computing. It's what they've always wanted, right? They always wanted low latency. Oh, high performance. Oh, security and privacy. They always wanted that. And so what we know is that because our rack is a 10 kilowatt air-cooled rack, compared to what traditional GPUs are, 140 kilowatts per rack, right? We're a 10 kilowatt air-cooled rack producing state-of-the-art AI. Sure. Now you can go into your existing data centers. You don't need to go buy new data. You already have data center. Roll out the old gear. Bring in the new gear. Off we go, right? And you're running the latest and greatest models at great performance. So that's kind of what's exciting about it. And they're first and foremost our customer, you know, they're buying racks and deploying them. But then broadly, they're really also a co-developer, you know, when it comes to enterprise secure AI for enterprise, right? Because they are the most knowledgeable when it comes to deploying these types of applications and they're just so smart about this stuff. And so we can partner with them and figure out how to bring the enterprise on board, which again, you know, if we all think about it, there are three classes of AI clients. You got the model makers, right? You got the clouds, and people forget enterprise is just waking up. And historically they have been a massive amount of spend, massive, right? So they're just waking up and you see JP Morgan, so banks and some of the other retail start to come on board, and I think that's going to be a part of the ecosystem. I think it's going to be something that's going to be around.
Yeah. Last question for me. Jordan, you can take whatever after, but walk me through the pre-ChatGPT era for your company. I imagine that since 2022, 2023, it's been on fire and demand has been through the roof. But how did the company survive from 2017 to 2022? What were you doing? Who were the customers? Did you have revenue? Like, was it all R&D and venture funding? Like, what was the early history of the company? Because that feels like a very potentially rough period, but how did you get through that time?
Well, look, no, I mean, you know, being in chip design for a long time, you always know that this is a long-term investment. And so, I mean, back in 2017, we were still trying to recognize cats and dogs in the web, right? This is a cat or a dog, right? Kind of what we were doing, right? We were trying to figure out, you know, what do we do with this thing, right? So, and we're building recommended models, all of different things for people. But the reality is this: what OpenAI did, it just flushed out all of that noise pre-OpenAI and focused all the value around let's start here. It doesn't mean that images aren't important, doesn't mean voice, all these other things aren't important, but what it did is focused us as a global market to say let's start here. Let's just start with language, and it's a very good place to start. And it allowed us to actually come, all of us come together, say hardware, software, models, everybody start focusing in on language as the use case. And now you've got so much production traction, and now we can focus on other things, like we're doing great in voice and we're doing great in video, and these are things that are also coming. But the ability to sequence these things in production so that you can actually see value has been incredibly important.
On that note, going forward, how are you thinking about, you know, reacting to customer needs? Like this partnership with JPM and some of your other, you know, new customers feels like reacting to this like real-time need of like, okay, we understand the potential, we're willing to invest a lot here, but we have certain kind of criteria around how we want to roll these products out versus skating to where the puck is going and trying to make predictions around R&D that you do today that will pay dividends, you know, three, four, five years out.
Yeah, well, when you're chip design, as you guys know, right, it takes you two years to design and you got to go through TSMC, you got to build these wafers and you got to build these systems. I mean, there's a three or four year cycle, right? And so they're always, it's a blend between engineering and fortune-telling, right? And so you're kind of mixing those. But the reality is you're always doing that. But here's the difference when it comes to enterprise. And we not only have JP Morgan, but last month we announced a huge partnership with Vista Equity, which has 93 portfolio companies building all these applications. And so what is the biggest thing that we run out with AI is we don't have takes. We don't have a takes. So now what you're seeing is the JP Morgans of the world, the enterprise of the world, the Vistas of the world, all these applications all turning SaaS companies into AI-first companies, and you're seeing the demand pull through. And so what we're excited about this is not only do we have the chips and you got connected with the data center that's there, but the next thing is connected with the end user, the end off-taker that applies AI into something that's useful to your businesses. So we're super excited about the JP Morgan, the Vista, all of these, that picture. That's our mentality is what are people using it for? Solve that problem. Everything underneath follows.
Mhm. Thank you so much for coming on the show. Congratulations on the massive round.
9-year overnight success.
Can't wait to talk to you again soon. Have a great rest of your week.
We'll talk to you soon. Goodbye.
Let me tell you about public.com. Investing for those that take it seriously. They got stocks, options, bonds, crypto, treasuries, and more with great customer service. Our next guest is Alana Palmedo from Paradigm.
What's up? Welcome to the show.
Our last guest raised 1 billion. You had to one up them. You had to two up them with 1.2. Tell us about the new fund.
Oh man, thanks for having me, guys. Really great to be here. Um, we're super excited about it. We raised 1.2 billion to go after the new opportunity set.
Thanks for warming that up for me.
1.2 billion for investment.
All right, guys. We're going big. We're going after all frontier tech. We're really excited about it. Obviously, you guys know we got our start in crypto back in 2018. This is our fourth fund. Yeah. Uh, pretty excited about it.
But I feel like, I don't know, this is being somewhat positioned as like an expansion of the vision or the thesis, but I feel like Paradigm's been investing in broader frontier themes for years now. Like, can you talk about some of the history of expanding beyond crypto and the key moments in the fund's history that sort of woke you up to broadening the thesis?
Yeah. Well, I think in 2018, crypto was the most interesting frontier to be investing into, and like that really was the heart of how we started building the firm. And it was the thesis behind being the sort of researchers, builders first, how do you get really close to the tech to be able to understand and underwrite these things from first principles. And then I think over time it's become pretty obvious that it's not the only frontier, and really hard to ignore what's happening in AI. But actually, like, I would say it's more than that. It's like the world itself has started to blend together, and drawing these concrete lines is like no longer applicable. I think that like even something that called itself a crypto project years ago today, like things that are right on the periphery, tech is just all touching each other. And when you think about trying to separate it, it just doesn't work that way.
How are you thinking about the early stage? There's been some chatter on the timeline about, oh, like every VC just wants to pile more money into the biggest private companies, the trillion-dollar private companies. Uh, why would I waste money or time talking to an early-stage founder? How are you thinking about opportunities at early stage, mid-stage, late stage? What even is mid-stage now? Is that a $10, a $50 billion company early stage at this point? Who knows?
Yeah, I think, well, I think stage is a hard thing to define because like what are we really talking about? Are we talking about the stage of the company, the valuation? Like, um, I think it's impossible to not be backing some of the earliest founders and folks who are, you know, coming out, um, with new great ideas. And so for us, it's like talent-led. I would say like you find great people. Oftentimes they're in their early 20s. Um, so I don't think you can ignore the early stage, but you also can't ignore the dynamic that you're describing, which is like the power law of what's happening with these larger later-stage companies and how much they're able to capture and the compound rate of growth. Uh, so I think it's like you have to do both.
Uh, from an actual product standpoint, are you breaking out the fund? Is there some amount, you know, segmented for the early stage, some for growth, some for incubation? And how should founders be thinking about the new vehicle?
It'll be truly stage-agnostic. So, we're going to invest in everything from super early, seed stage, even pre-seed in some cases, all the way into growth markets. We also, you know, we have a core flagship fund which is an open-ended vehicle, and so we invest into public markets too.
Oh, interesting. Very cool.
Uh, what is the effect of this year's IPOs? What have been the internal learnings from the height of the opening of the IPO window? It feels like it's sort of open but maybe not fully wide open. Uh, we've got a lot of exciting IPOs that have done well. Uh, but how is that affecting the fund and your feeling around the market right now?
That's a good question. Um, I think it's a hard thing to answer. One of the interesting dynamics happening in IPOs is like Hyperliquid and some of the trading activity we've seen in advance of the IPOs. It's been pretty fun to see like the 24/7, 365 markets really come alive, which has been a long-term thesis that the crypto markets have had. And I think Jeff Low at Hyperliquid has just done an incredible job pulling it off. And, um, and Jeff Low at trade.xyz and Jeff Yan at Hyperliquid. Yeah. Um, and so like the Sarah Bruss IPO, SpaceX IPO, like all of these now, like we had all of this trading activity going into those, and so there was a lot more transparency on pricing in advance of the IPO event itself. Um,
Yeah. How accurate was the like on-chain pricing relative to what these IPOs actually opened at? And do you feel like... surprisingly accurate?
Yeah, surprisingly like quite accurate actually.
Interesting. And have you heard of any institutions like starting to, you know, bigger like firms on Wall Street that you think will, if they're planning to invest in the IPO or trade it after it opens, actually start their activity on-chain? Is that something that you see happening in the future as we get sort of more evidence that on-chain trading tracks, you know, real life after these IPOs actually happen?
Yeah, it's a good question. And I mean, if I were them, I would be doing that. Um, I don't know what the rules are and what the playbook is for them today and like what the liquidity and depth of those books looks like. But I mean, I think that's the thesis for what 24/7, 365 markets look like in the future. Um, is you're going to have a lot more institutional activity. Um, certainly we're seeing that in perps as a derivative product themselves. Like I think they're just a much more efficient pure-play product for big hedge funds to come in. And, um, I would even say like the same thing can be true in prediction markets. Like what we're seeing is, you know, some people are relating to prediction markets as an institutional product for getting exposure to things and bets that they want to place that is much more precise than we've had in the future or, excuse me, in the past. Um, and so the future I think holds like still a lot of white space in terms of how these markets are going to evolve, but we continue to be really excited about it.
That's great. What do your internal meetings look like? Do you have like, you know, a meeting once a month where you guys pitch new frontiers that you're kind of kicking around? Like, what is forming a thesis look like? Is it all partner-driven or is it more collective? You guys have kind of a general idea of a category that you're excited about and then you work together. Mhm.
Well, I'd say top down, like thematically we're investing into areas that are structurally long AGI first and foremost. Um, so what are the things that are going to have a natural tailwind to come out of that? We don't run like necessarily a top-down process though down to an investment and say like let's go do a deep dive on a particular theme. Um, I think that oftentimes you try to gain conviction, um, and do sprints on individual areas in the market, um, that were maybe like there's a particular founder that we are excited about. Um, but the process, yeah, I would say it's more organic. Like each individual partner may have a different thing that they're excited about and digging in on. I do think like, you know, fundamental to our thesis is like we have to be close to the tech, and so we ourselves have to be able to go really deep and like understand these things down to, you know, lines of code. And a lot of folks on the team are computer scientists or physicists.
About down to the metal. You guys have a little data center in the office just humming away.
No data center in the office. Um, yeah. I mean, I like orbital data centers, are they the future? I don't know. What do you guys think?
I don't know. I think you guys should get on the next SpaceX rocket and run a little experiment.
I think it's all just... Yeah, I'm on the Elon timelines, which is take whatever he says and multiply it by three. And that's probably true, which is still going to happen, but just not next year. Yeah, I think Elon will will things into existence. I...
With enough time, you will get there. Yeah, almost always.
Uh, but it takes time. Uh, anyway, thank you for taking the time to come chat with us and congratulations on the new fund.
Yeah. And excited to meet all the new founders joining the portfolio.
Yeah, super pumped about it. We got some great folks on there. I love Zipline.
Evan Rogers. Keller is amazing at Zipline. Yeah.
Fantastic. Super excited.
Have a great rest of your day and we'll talk to you soon. Great to have you on, Alana. Cheers. Bye.
Let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish. Our next... next up is Byron Boots.
We got boots on the ground. Boots.
Let's bring in Byron Boots from Overland AI. Welcome to the show. How are you doing?
Oh, great. Thanks for having me on, guys. It's really exciting to be here.
Yes. Great to have you. We're very excited. Uh, can you reintroduce the product and then take us through the latest news and the contract that you just signed?
Yeah, sure. So, um, good news is I'm working here out of our factory in Seattle. So, you can actually see the product that we have, right? So, uh, we're building autonomous ground vehicles for defense. Basically, you think about that as like robots for the battlefield. Yeah. Um, so do you guys want to see it?
Absolutely. Walk through it a little. Give us a tour.
So, so here's one of them. This is our Ultra product. Um, you can kind of see the top part of it. Yeah. Uh, it's a large off-road ground vehicle. You've got sensors in the front. So, there's stereo cameras and LIDAR. Um, you've got these big wheels here and long-travel suspension. Um, down here is a payload deck, so you can think about putting, you know, big payloads on the vehicle. Um, right in here. There's onboard power. There's compute down below the deck. Um, comms in the back. So that gives you just a brief overview of one of the things that we're building. And then, you know, you asked about the contract and we're super excited. So, uh, we're the first ground autonomy provider to get a production contract with the US military, um, $20 million contract with the US Marine Corps.
Oh, it's fantastic. That I'm thinking that truck needs a gun. What does the payload look like on there? It seems very modular. It seems like you can put a lot of different equipment on that. Is that something that's delegated to the Marine Corps? They decide, they partner with other primes, or do you have like other relationships where you can bring the configuration to them?
So, we've put 30 different payloads on these vehicles. It's designed to be modular. You can see that there's L-track down here that makes it really easy to attach things. Um, so this is everything from kinetic payloads, like remote weapon stations, drones, EW. Um, so it depends on what the customer wants. We can help them integrate anything onto this vehicle.
Uh, dogs as a form factor. There's been a bunch of sort of creepy dogs, robot dogs, or actual dogs. Well, we should get into real dogs. Uh, but, uh, there's been a bunch of videos circulating recently. There was one maybe a week ago of a robotic dog and it seemed like they had solved like some of the recoil challenges. Do you think that's a... like how are you thinking about that form factor? Obviously you guys focus on something else, but I'm sure you understand the trade-offs of both. I can imagine like your product is much better for high-range, you know, situations where you need a lot of range. Uh, but how do you think about it?
Yeah. So, one of the reasons why we love this form factor is that it's first of all large enough to put a serious payload on it. It can handle 1 to 2,000 lbs depending on what the configuration is. Um, it's got wheels, so, you know, it has really good use of energy. You can drive it 100 miles. Um, so robot dogs are really cool too. So before I started Overland AI, I was a professor running a lab where we worked on machine learning and robotics, and we had a bunch of robotic dogs in the lab, and they're fun to work on. Uh, but I think wheeled vehicles are really the place where we can get, you know, the most...
You're saying the wheel is underrated and it doesn't need to be reinvented. That's what I'm hearing. This is crazy.
Yeah. I mean, so not only do you not need to reinvent a wheel, I mean, if you have squishy tires, you can do things like drive up and down stairs and do all sorts of stuff. So,
I never thought about how simple that is. You can just drive up the stairs if your tires are big enough. How big are the tires on that truck?
I think these are like 38 inch. Um, so pretty big.
38. That's fantastic. I love it. What does the actual manufacturing process look like? Obviously, you're a very well-funded company, $100 million round recently. At the same time, there's probably a lot of experimentation, a lot of putting different parts of the supply chain together, a lot of modularity. So, uh, how hand-built are these? What does the automation story over the future look like?
So, the Ultra vehicle like this one is actually based on a Polaris RZR chassis and drivetrain. Um, so we get that from Polaris and then we upgrade the suspension. You can actually see a great view of that here. Got it. Um, we pull out the seats, we just add this flat payload deck, um, and, you know, we do all the work to add the compute and the sensors and wiring, power, all of that kind of stuff. So we pull all that together in our factory here. We're also very field-forward. So we test these out in the field all the time. You know, we're in the field every single day working on the, you know, the software, the autonomy. And I think one thing which is really, really important about these systems is that this is a completely autonomous robotic system, right? Like it has sensors that allow it to perceive the environment, plan through the environment, control it so that a user can just tell it where it wants it to go, what payload to execute. Um, it's not, you know, it doesn't have to be teleoperated. Uh, you know, a single user can control multiple vehicles is one of the things which is really cool about this tech.
I imagine there's like very high demand for both teleop and full autonomous mode though, right? People want to be able to port in for a particular mission and then also be able to just try and draw waypoints on a map.
So, one of the things that I think we see is when people first start to use a vehicle like this, they're like, 'Oh, like I want to tell it and I'll be able to precisely control where it goes.' But they quickly realize that you can actually just tell it where to go on the map, tell it, you know, what orientation you want it to be in, and it will just go there and do it. And so, um, it's actually safer, more efficient, and allows you to get like much more force multiplication if the operator just relies on the autonomy to get it there. And so what we've seen is warfighters really embracing that and starting to experiment much more with that. So, uh, we think autonomy really is the future.
Any plans to make a dual-use version for Malibu? If you got a bunch of surfboards, you want to get them down to the beach, you load them up and say, 'Hey, I'll meet you there in 5 minutes.' Some natural light. Some natural light.
I mean, sounds awesome. We are actually dual-use. We are using this for things like wildland firefighting and other things. So, who knows? I mean, maybe pretty soon we'll be able to get you guys one.
The day party market, the Aman taking you around the hotel property. Uh, well, thank you so much for coming on the show. Uh, fantastic news. Congratulations, and we'll talk to you soon.
Incredible. We appreciate what you... Thank you so much. Really appreciate it.
Have a great week. We'll talk to you soon. Goodbye.
Let me tell you about Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. Uh, there's a scoop. There's a scoop on the timeline. Perplexity is quietly building an AI coding tool that takes on Claude Code and Cursor and Codex potentially. Uh, it's meant to build software end-to-end. The tool is being internally used for now under the code name Teammate. Perplexity has been sort of bouncing around from different... So Steve here says, what happened to the Bloomberg terminal killer? They were working on that for a while. Then they launched Bloomberg Computer which was sort of a computer-use agent and saw incredible growth, like you saw the ARR numbers went way up. It feels like one of those examples of a company that people were sort of like, I think one of the challenges, and then all of a sudden there's some new good news.
Perplexity has a ton of revenue. They're taking a ton of shots on goal, just like every other company to be clear. But I think the difference is that Perplexity doesn't have a product that's being used heavily by tech insiders. And so people just see them as like running around with like... uh, they... it feels like they don't have something like that's a hit in enterprise that they can fall back on while they keep taking shots at some of these more enterprise-y products. Yeah.
And so, uh, yeah, it's interesting. I... it feels like... Fun Prince, remember they fired up the Perplexity Investment Fund investing in other startups.
I don't know what they invested in.
I don't know what they invested in, but... but he has good access.
It's possible everything in that fund is just on fire because the market is doing so well. So many companies are getting marked up. That could be a pretty interesting portfolio to dig into at some point.
I think that this is more... this teammate makes me think that it's maybe something else. But I could see them seeing Lovable's growth and having a lot of overlap audiences and Replit and saying like, hey, we want to make, you know, we want to make a more consumer-leaning product. I think it'll be hard to really get a foothold.
They are more adjacent than you think. You go to Perplexity to look up some information. You get a nice, you know, UI design on the responses. The Joe Rogan partnership which seems like it's crushing, and Joe can be like, 'Hey, Jamie, build me a new CRM during the show.'
I don't think Joe Rogan has any interest in a CRM. I don't think he's in the market. I think he does everything over text message and you have to be invited. So, you know, if you're going to be on the show... before we have our next guest, Grok 4.5, that's right, is out.
Is out. Some reviews, benchmarks.
Looking good on the benchmark goes. Uh, I wonder how does Michael Truel have a SpaceX AI badge yet? Is the Cursor team involved in this heavily because I know they were doing some consulting work while the deal closed. Of course, the deal was announced, but they still have to go through post-merger integration. I wonder how deep that team was integrated, but uh, excited to...
I think they're still going to be running under the Cursor umbrella for a long time. I think it's much easier to sell Cursor through the enterprise than...
And they have Composer separate model, but yeah, Cursor did say that we've partnered with SpaceX to train Grok, so I think that there was a lot of...
I really wonder if there'll be any gap in capabilities, like in the spiky intelligence sense of Grok will be good at a particular thing, writing or something that's a little bit more X-native, Twitter-native, uh, news, social media, like that's been the advantage there potentially, and then Cursor will be more data-driven, code-driven, and there'll actually be sort of a bifurcation in the models, or will we just see a consolidation where both models wind up offering like basically the same spiky frontier as every other model.
Uh, anyway, excited to dig into that. Always love a new model launch. But we have our next guest in the waiting room. We have Will Mayer from Cult Holdings. Welcome to the show. How you doing, Will?
I'm doing well. How are you guys?
Thank you so much for taking the time. Great to have you on the show. Great to have you on the show. What do we got in the background here? Oh yeah. Is that real?
That is real. Yeah. That's like various artifacts from the past. So, yeah.
Very cool. Uh, Webby Awards. I think the spring. Isn't that a Webby Award?
Congratulations. Uh, give us a little background. Give us a little bit of the journey. Introduce yourself for those who are watching who might not be familiar with your work.
Yeah, of course. I am Will Mayer. I'm the co-founder and high priest at Cult Holdings. Uh, background came from making skateboard films at a young age when I was like 14 for companies like Vans. Became deeply obsessed with brands and what they represented. Moved from there into advertising, working with brands like Nike and Netflix, and then started working more and more with startups, incubating companies, and fell in love with cults and religions and believe that many of the best brands are some form of a religion.
