SPEAKER_00: If that's happening every two to three years, that means theoretically that you might buy more of them. But that does not mean that you're throwing the old ones away. That's ludicrous. That would be like saying I had a Corvette C7 and the C8, eighth generation came out. So I took the C7 and I parked it in the garage. And yeah, I just took the air out of the tires and the battery and I put it on a cinderbox and that's it. It's over. That car can't be used ever. It's got 30,000 miles on it. That car is going to be driven for another 10 years. What are we talking

SPEAKER_03: about here? It's not that egregious to think that Oracle believed in 2020 that they would get five years of life out of one of this depreciating assets. Then the next three years, they said, yeah, it's still five. And then in 2024, they said, you know what? We think it's six. And the Microsoft one is more dramatic. It goes from three years of useful life to six. So they doubled, but everybody's winding up at five and a half, six, six, six, five. And the year before it was four and a half, six, six, six, six. So Amazon brought it down a year. These are all kind of

SPEAKER_06: within 30%, right? This Week in Startups is brought to you by Northwest Registered Agent.

SPEAKER_08: Starting your business should be simple. With Northwest Registered Agent, you can form your entire business identity in just 10 clicks and 10 minutes. From LLCs to trademarks, domains to custom websites, they have got you covered. Get more privacy, more options, and more done. Visit northwestregisteragent.com slash twist today. Crusoe Cloud. Crusoe is the AI factory company. Reliable infrastructure and expert support. Visit crusoe.ai slash build to reserve your capacity for the latest GPUs today. And Pipedrive. Bring clarity and control to your sales process with

SPEAKER_06: Pipedrive. The number one CRM for small and medium-sized businesses. Supercharge your sales today. Start with a 30-day free trial. Pipedrive.com slash twist.

SPEAKER_10: All right, everybody. Welcome back to This Week in Startups. I'm your host,

SPEAKER_11: Jason Calaghanis, here with me, Alex Wilhelm, again. And I am in Vegas. If you're watching the video on YouTube or Spotify, I'm in my suite at the Venetian, thanks to my friends at the Venetian for their hospitality. I'm here for F1 with the All In Podcast doing a hot lap. And then they did all

SPEAKER_13: these, like, really nice things for me. Look, they built this, like, little chocolate thing for me for... Oh, there goes all the chocolates. You can eat this whole thing. I can eat the track,

SPEAKER_15: I guess. And all the little chocolates on top of it. Is that a roulette wheel made of candy?

SPEAKER_20: Yeah, basically. It's a roulette wheel, but it's also, like, a track. They redid the Venetian. They

SPEAKER_22: redid the poker room. And they've got, like, a live poker room where you can do live shows. So we did a live show where we played a little poker with Jason Kuhn, Alan Keating, if you know that, Maniac, and Phil Hellmuth. So we're having a blast. Sounds good, man. Venetians are gorgeous. They redid the poker room there. You would like it. The poker room has, like, a nice spread of one, two, three, five. You can get in all kinds of different games. Good crowd. And very classy. You know, they got a nice espresso machine. They classed it up. You know, most of these poker

SPEAKER_08: rooms, they leave a little bit to be desired. My local poker room, it looks like the back office of, like, a 1970s accounting firm. There's no curb appeal whatsoever. I really want to go back to Vegas and play cards. But I haven't gone to Vegas since I quit drinking, to be honest.

SPEAKER_22: Just haven't made it back since. So. Good for the games. Yeah, I mean, you got kids, too. I mean,

SPEAKER_11: it's nice to see a show. You know what the most amazing thing is? The restaurant scene in Vegas is absolutely absurd. It's the best in the world now because what they do is they're in a competition. People book their hotels based on the restaurants. So last night we ate at a restaurant, C-O-T-E. It's a Korean Michelin-starred restaurant from New York. They built an outpost here. Amazing. Like, six-course tasting menu. Unbelievable. There's a famous Peking duck place. I love to go when I'm in Hong Kong, but I don't travel to Hong Kong like I used to because I've been a little bit critical of, you know, the government over there. And I don't need to, and I can't do business there anymore. It's kind of a bummer, actually. I kind of liked it when we were partners. Maybe that'll happen again in our lifetime. Mott 32. Turns out Mott 32 just opened in Dubai, and they opened one here at the Venetian. So now they're collecting the best restaurants. I had dinner at Carbone when I got here, on Wednesday. That's over at the Aria. So they're in this competition to see who can get the top restaurants around the world to open their outpost here. And then they have now, like, they'll put more effort into the restaurant here. Like the quality of the meats, the quality of the

SPEAKER_03: produce. It really just goes to show you the value of IP. These restaurants are, you know, kind of IP

SPEAKER_22: machines. So they opened the second Carbone here. And then when I was in Riyadh, the hotel I was staying in, which was a very nice hotel, had a Carbone in it. And I was like, what? And then I found out the PIF, the Sovereign Wealth Fund from Saudi Arabia had somebody told me, oh, yeah, they invested in the restaurant group that does Carbone, the very elite New York institution. So then literally, I ate at Carbone in Riyadh, Carbone in Vegas in the same 30 days. You start to think about, like,

SPEAKER_07: which one was better? Come on, do a side-by-side. Honestly, if you took the experience from each,

SPEAKER_11: they're sourcing everything from the same place. So the Parmesan cheese, the salami, the pasta, the rigatoni, the spicy rigatoni is like a very famous dish. It was exactly the same. If you were to, you know, swap it out, it would be the same. So it's kind of like just a really interesting lesson for entrepreneurs. There are many different markets where you can bring your product. So if Carbone can exist in all these different markets, then they opened up at the Bellagio, a Carbone on the water there, and they changed the menu a little bit. So now there's two Carbones in Vegas in two different properties. Both, I think, are owned by the MGM group. And so I was talking with the CEO of the Venetian, a really nice guy, a fan of the pods. And he was explaining how now people sometimes come for a show. Sometimes they like the games. But most frequently, people are talking about coming for the food. They're perfecting specific dishes. So when I was talking about Japan on the show last week, they will do like one dish really well at most

SPEAKER_22: restaurants. And that's what they do. It's a soba place, or it's a, you know, sushi place, but it's not like a sushi place where you can get soba, you know, it's a tempura place, you know, but you can't get sushi, vice versa, as opposed to in America or experience.

SPEAKER_11: All of this is to say, if you can perfect something in one market, the markets are now becoming very similar. We have a global culture now that has emerged. So the people in Saudi, Dubai, Tokyo, Vegas, New York, you know, Europe, they're all kind of becoming connoisseurs of great products. And if you can make a great product, it can make it to those different cities, whether it's Uber, Airbnb, or Carbone or Mott 32. Speaking of global expansion, part of why I'm bringing this up is we had something work inside of our organization, Foundry University. It really clicked with people. And so we're going to be launching Foundry University in Tokyo. Started in America, we've done 11 cohorts. Then we brought it to Riyadh, and I was just talking about that. And last week, I announced that we are going to have it start in Tokyo in January. So we have the details for you. Applications

SPEAKER_03: are now open for the 2026 cohort. Startups are going to get to be part of basically a three-part experience. Let me just say in plain English, and this will help them expand their reach. We're going to have a bootcamp in Tokyo, January 12th to 16th. Then the program will move to online from January 19th to March 6th. Finally, you're going to come to the US for a week, March 9th to the 20th. So we've designed this so you get to have the Tokyo experience. You get to hunker down, get work done, and then

SPEAKER_11: you get to come to Silicon Valley and meet investors and meet founders there. We're also looking for mentors. Some of my friends, if you're an angel investor and you want to work with the 35 companies,

SPEAKER_03: you can join us there. To apply, it's very simple. launch.co slash Tokyo. There are a couple of unique things about the program. The program is for companies that are in Japan, Japanese founders or focused on the Japanese market as well as the US market. So go ahead and go to launch.co slash

SPEAKER_28: Tokyo. I'll see you in Tokyo. I'll be there for the launch. Just for folks out there who are listening

SPEAKER_08: and heard us talk about the Riyadh program last week, Jason, you said that in Riyadh, it was mostly for Saudi founders, but you could also join if you weren't Saudi. They were focused on

SPEAKER_11: Saudi nationals because they're trying to build an ecosystem there. But if you're willing to spend

SPEAKER_22: time in Saudi, great, come. Similar for our friends at Jetro in Tokyo. If you're going to spend some time there or address the market or you're a Japanese national, you'll have a preference there coming in. And you can see the picture there. I had the incredible hospitality. If you look on the website, I just gave like an opening keynote where I explained what the program would be. And they had 200 people show up for this thing, 150, 200 people. And they had like 100 or 200 people on the wait list trying to get in. But there is a big founder community there. And there is a very large group of Americans who are based there now. So I met with, I won't say which ones, but there's a half dozen Americans who I had lunch with who are active investors from America, spending either half time, full time, or majority time in Japan and still investing in the US. So we're really seeing this globalization where you can be in any market, you can address any market. And there's investors in every market. And I think the future of startups is really going to be, you know, folks starting in whatever market they want to go after first, but then going global, just like Carbone can exist in different places. They can build an incredible product. That product might need to be adapted. For example, in Saudi, they had mocktails, you know, and like some very flavored, really nice mocktails without alcohol in them. And because alcohol is not approved in all regions in the

SPEAKER_43: Middle East yet, but it's starting to be. And so you can't have an effective sales team without an organized sales process. Trust me on this. I've seen thousands of startups up close and personal and man, a messy, chaotic sales process. That is a major red flag. This show, This Week in Startups is about helping founders. I want to tell you about the tool I use, Pipedrive. Pipedrive brings together your entire sales process in one clear, easy to understand and centralized space, giving you one place to check on all your customers and the deal information and to keep you and your team moving faster and working together as a single team. Teams using Pipedrive close an average of three times more deals every month. You could average three times more deals every month by joining over 100,000 companies that use Pipedrive. And when you use our link, you'll get that 30-day free trial. No credit card or payment needed. Go to pipedrive.com slash twist to get started. That's pipedrive.com slash twist. I have been using this product for close to a decade. I don't have time to babysit the sales team. I need them to have the right tools. And that's why I use Pipedrive.

SPEAKER_22: You have to adapt for the local culture. I'm sure if Carbone was in Japan, they might have some nuance there. Maybe you have to take your shoes off when you go in, like you do in some homes or restaurants, depending on how you sit. So there you have it, folks. And I will be there. I'm going to be in Japan again. I'll be doing the show for two weeks from Tokyo in January. Which is going

SPEAKER_50: to be a lot of fun because I think that's also when I'm going to be on paternity duties. So we're

SPEAKER_52: going to have some fun scheduling times. Easy peasy. All right, what's in the news? Let's get some

SPEAKER_08: news done here. After we finished the show on Wednesday, Jason, NVIDIA dropped its earnings. I think we should spend just a minute explaining to folks what happened so they understand the lay of the land. Just in case people didn't read the news, NVIDIA beat on revenue. It beat on earnings per share and it beat on guidance. And then it shares rose and then it shares fell. Jensen actually made a comment about this in an internal meeting that Business Insider got a recording of. And he said, quote, if we delivered a bad quarter, it's evidence that there is an AI bubble. If we delivered a great quarter, we are fueling the AI bubble. So a little bit of damned if you do, damned if you don't from NVIDIA. But overall, a very impressive print. I'm curious what the vibes are amongst the group chats regarding the current AI bubble discourse, Jason, now that we have NVIDIA's

SPEAKER_03: numbers in hand. I mean, the number one thing people are talking about is obviously AI buildout is going spectacularly. Obviously, NVIDIA can't keep their chips on the shelf. Obviously, the market got a bit overheated. So people throwing money into late stage startups or public companies without really understanding the details is people's concern. There was an interesting tweet from Gavin Baker that I retweeted. Essentially, the gist of it is NVIDIA came out, I believe, in this quarterly earnings and they addressed the open AI investment. And they addressed also, I guess, their potential

SPEAKER_11: investment in Anthropic. So what the themes are now showing is, and what people are starting to discuss, you know, walking from casino to casino between playing games or, you know, in the group chats is which of these deals are actually guaranteed? And I asked you that question. Hey, I don't know if you remember on a previous show. I was like, is that guaranteed? Can you look it up? And it's like,

SPEAKER_58: we can't find anything. Okay. As you speculated correctly, we'll find out in coming quarters when they announced because they're going to need to disclose this. This is from Gavin Baker. It's a

SPEAKER_60: bit long. I'm going to read it verbatim, everybody. Just bear with me. We expect to continue investing

SPEAKER_08: in strategic partnerships in the third quarter of- This is Jensen. This is Jensen speaking. Yeah. In the third quarter of fiscal year 2026, we entered into a letter of intent with an opportunity to invest in open AI. In November, 2025, we entered into an agreement subject to certain closing conditions to invest up to $10 billion in Anthropic. There is no assurance that we will enter into definitive agreements with respect to the open AI opportunity or other potential investments, or that any investment will be completed on expected terms, if at all. The timing and magnitude of these and other investments we may make will depend on various factors, including the ability of our partners to successfully develop and deploy AI infrastructure. There can be no certainty as to the timing or amount of capital. We may ultimately invest in these or other strategic partnerships, and we may be limited by our available liquidity and capital resources. In other words, Jason, this is all handshakes and good vibes, nothing locked down, and they can all kind of walk away

SPEAKER_03: if they want. Yeah. Essentially, you know, the opportunity is doing a lot of work in that sentence and may, I think, do a lot of work in those sentences. Lots. They obviously have an agreement. It's probably more than a handshake. It's probably written in fairness to them. It's probably a memorandum of understanding or, you know, a letter of intent. You know, we call it a letter of nothing in the startup community. So when people are like, oh my God, we got a letter of intent. We're like, you've got a letter of nothing. People will do these as favors for each other. Now, when big companies do them, okay, well, there's a little more at stake. So if, you know, you're a cloud provider and you write, we're going to get tens of billions of dollars in build out. Those are the ones that I really want to see. So this is the one where NVIDIA is sending money to open AI and Anthropic as an investment. Obviously, those two companies are buying compute, whether they're buying it themselves or buying it through a cloud provider, you know, doesn't matter. They're going to be buying compute. So that's where the round tripping concern comes in. Is this a round tripping? So there's one interpretation of this. Okay. They're making sure people understand that these two things might not be tied together. In other words, you're, we're agreeing to invest, you're agreeing to buy. Now they're kind of backtracking it, right? This sounds very different than the initial announcements. Now, part of that could be that the press ran with these, which is why here on the show, I'm always like, can I get the details on this? Where are the details? Because the press does run with things sometimes and the companies might run with things sometimes. Social media might run with things sometimes. And we're like, oh my God,

SPEAKER_11: look at all this money. It's all option. It's all optional. So this chart of all of this round tripping, et cetera, we now need to redo that chart. And we need to know

SPEAKER_22: if it's a dotted line, which means it can be broken, or if it's a solid line, which means it has to

SPEAKER_56: occur, right? It's an actual commitment. And just, you're referring to this famous chart right here,

SPEAKER_03: I presume, right? Yeah. And so at the center of this is NVIDIA at a $4.5 trillion valuation. And what you see is Oracle, Intel, CoreWeave, Microsoft, OpenAI, AMD, XAI, Figure, Mistral, we had on the program recently. So they're all sort of explaining where the money goes and investment services. So the red lines on here are hardware or software. I guess the blue is investment, et cetera.

David Friedberg: It's complicated, but now if we have dotted lines, we'd actually know. So I think in some ways, this is a good way to let a little air out of the bubble and remove the concerns around round tripping. That's my two interpretations here. I also think Jensen's a little pissed off at Sam Allman. Okay. Tell me why. I think, you know, if you're going to offer to invest all this money and then the next week,

SPEAKER_03: you know, he's buying chips from AMD and, you know, dropping these other announcements. I think he maybe thought maybe he was part of a tighter, more narrow relationship set. And I think Sam just did, how many deals did Sam announce over from the summer till now? 574 deals. No, no, but it was, it was literally, I think, closer to 10 than five, right? I mean, we should actually do a little research on it because it seemed like he was announcing one or two a week, right? Investment, the secondary, SAIL, AMD, NVIDIA. I mean, it was pretty Amasa's investment.

SPEAKER_08: Oh yeah. It was notable. It was absolutely notable. But the thing is, we've talked about how people at OpenAI, when Sam was ejected for a couple of days, had struggles with him, working with him. And there were discussions of elements of truthfulness. Then we have OpenAI and Microsoft. That relationship eventually kind of breaks. Now we're seeing OpenAI and NVIDIA cracks there. You know, Jason, if

SPEAKER_11: everyone you date is a jerk. Yeah. Might be you. I mean, I think he makes, he's a great, he's a consummate dealmaker. I mean, I've known him for, I guess, close to 20 years, I guess. He's always

SPEAKER_03: been like an operator, a networker, a dealmaker, really knows how to get into the room where it happens, understands how power works, understands how to take power. That's how he wound up on the top of the pile of a crazy scrum that is known as OpenAI. And it's working out. I do think when we look at this, the company that's going to have the most market share decline because they were first up the hill is going to be OpenAI. So their percentage of AI jobs from APIs and their lock on the consumer is going to go from essentially a hundred percent of consumers, right? They had the first chat product. It's going to go down to less than 50%. And I don't know what their percentage is now of consumer searches. And I don't know if we're tracking them in a robust or refined way because some of the companies are private, Mishra, Anthropic, XAI, and a long tail of different LLMs. And I don't know if Google's taking the time to break out Gemini searches or the AI results that

SPEAKER_11: you can click over to and how many people are defaulting to AI versus search. They said in their

SPEAKER_80: Gemini 3 announcement to Jason that there's over 2 billion users now of the AI overviews and search.

SPEAKER_08: And they said that their Gemini mobile app now has 650 million monthly active users.

SPEAKER_03: Hold on a second. Wait a second. So they have 2 billion people using AI search.

SPEAKER_81: Launching a new company is all about finding your first customers and then just learning how to solve

SPEAKER_43: their problems. And that is going to put you on a relentless pace. And that means you're going to be releasing new products and features, hopefully at a brisk pace or better. But my Lord, the complexities of working with AI, it's going to slow your developers down. You're going to have to install new GPU drivers. They're going to be provisioning clusters. All of this detail-oriented, labor-intensive work can distract you from working on the thing that you're building, your product. But now there's Crusoe. They are the AI factory that worries about your backend so you can focus on your code. And their hardware has been designed specifically for AI development, so they're going to make sure that your models run with unparalleled performance every time. Head to crusoe.ai.build to apply and receive $100,000 in credits for virtualized NVIDIA GB200 NVL72 on Crusoe Cloud. Pending availability, okay? That's C-R-U-S-O-E dot AI slash build to apply and get those $100,000 in credits.

SPEAKER_03: And OpenAI was at 850, I think, the last time, or 800 monthly active users. That means Google has

SPEAKER_22: two and a half times the number of people using AI on a consumer basis. How come nobody's talking about that? This is incredible information. Yeah, I'm just double-checking. So AI overviews now have

SPEAKER_08: 2 billion users every month. The Gemini app surpasses 650 million users per month, and more than 70% of our cloud customers use our AI. That's from the Gemini 3 announced. But I think people aren't counting them as MAUs or Ds in the same way, Jason, because you can use them by accident. I can do a search and Google might give me one of these, but I don't know if that's as strong an intent signal as actually loading chatgpt.com, for example.

SPEAKER_03: If we're going to use, did you start at an AI portal? Fair enough. 650 million users of the

SPEAKER_07: Gemini app. Is that correct? Unless Google's lying to us. That's what they said. I'm just confirming

SPEAKER_95: with you how they phrased it. The Gemini app surpasses 650 million users per month. And what

SPEAKER_22: is the latest from ChatGPT on their number of users per month? 800 million per week. Per week. Okay,

SPEAKER_03: so we give them 900 million a month, give them 10% more. So that means the Gemini app is already two thirds of ChatGPT. This blows my mind. How many of the 2 billion clicked on the AI result? 10%? 20%? 30%? As opposed to just serving it up? Let's say 30%. Let's be generous. Okay, so 600 million. Okay, 600 plus 650 is 1.2. And you get 10% overlap, 20% overlap, you're at a billion. Yep. Google's ahead. I mean, this is like major news that nobody's reporting on here. Nobody's just taken out a calculator and done this. Let me just state this clearly. If this information is correct, Gemini now has 20% more users every month than ChatGPT. Gemini is beating ChatGPT

SPEAKER_08: in consumer adoption of AI. Just flat out. I would say the Google AI family of products is beating the open AI family of products at reaching the most number of consumers. Just to be a little bit,

SPEAKER_101: a little bit simplified. So Google's winning consumer AI. That's an interesting question. I don't know

SPEAKER_08: if I would, I'd go quite that far. Winning implies to me a revenue and usage calculation. And I don't think we know enough about Gemini's monetization. But it just in terms of consumer adoption,

SPEAKER_104: they're winning. If we use the term consumer adoption, I think you can see it consumer adoption. I mean, Google doesn't need to make money from this. So this actually just dovetails neatly with

SPEAKER_50: another story, Jason. Before Gemini 3 came out, I think open AI had a little bit of inkling about

SPEAKER_08: what was coming. And so the information got a memo that Sam Altman sent out around open AI. And he said that basically Google's progress could quote, create some temporary economic headwinds for our company. And then he said that we know we have some work to do, but we're catching up fast. I expect the vibes out there to be rough for a bit. So I think that we are kind of in the open AI nadir. Like they're getting beat up on infrastructure costs. Their rivals are launching impressive models. They're losing some of their consumer edge. They're not winning the API game Viz Anthropic. And so I think they're struggling a bit. They're doing great as a business. They're going to reach 20 billion annualized revenue by the end of the year, Sam says. But certainly the open AI of January 1st, 2025 is not going to be the open AI of December 31st, 2025.

SPEAKER_108: And I think the, if I were to steel man this, the defense that the open AI team and investors would

SPEAKER_03: say is the AI results from Gemini are being forced upon these consumers, right? Because they do have it in Google Docs now. So I wonder if they're counting those as Gemini searches when you do something in Google Docs. Did they just say any Google Docs searches that were done through the AI button are part of the overall Gemini? It would seem fair to do that.

SPEAKER_111: It would seem fair.

SPEAKER_03: But this would go back to the Teams versus Slack. If you remember when Microsoft Teams came out, they turned it on for everybody who uses Office 365, whether you want it or not, it was bundled.

David Friedberg: Therefore, if you were to click on it in the interface, and I don't use Office 365, I'm on Google Docs, I'm using Slack. But my understanding was it's just all automatically in your interface, and you're going to wind up clicking on it at some point. Therefore, their active user count compared to Slack's, and there's like this dramatic chart of how, boom, they just got everybody. But this was always my thought was, whoever has the guts

SPEAKER_03: to put AI first in their product with an established product is going to win the day. Meta has some number of people on Instagram doing AI first searches. When you do a search on Instagram,

SPEAKER_102: you cannot do a normal search anymore. It is by default, an AI search, right?

SPEAKER_50: Instagram is the worst app in the world. I just want to say that I use the desktop

SPEAKER_08: interface on this and Meta, it's just garbage. Like, try again. Two things, Jason. First of all, here's one of the charts you're mentioning with Slack versus Microsoft Teams. This only goes to 2012, but it shows the hockey stick-like growth of Microsoft Teams' DAUs compared to Slack, which was a major narrative up until Slack's eventual exit. And then the other thing, just based on what we're talking about, consumer usage, this is a top apps for iPhone ranking for the US. And if you look at the top of this, you'll note that it's ChadGPT and Google Gemini. Yeah. It's kind of impressive how popular they are. But if you look, Grok's number six, and Sora's number nine. These AI products are incredibly impressive. And as you point out,

SPEAKER_118: repeatedly on the show, this is new downloads. This is not a usage chart. This is still downloading.

SPEAKER_03: The app store rankings are based, I think, in a blended algorithm that benefits the new,

David Friedberg: so that it doesn't become self-perpetuating top list. Because people will go to the top 10 list or top 100 list to listen to music. And then those artists stay in the top 10. So new entrants do get a little bit of, I think, a push. Same thing happens in podcasting. Somebody comes up with a new podcast, they always tweet and they're like, oh my God, we're the number one podcast in the world. And it's like, okay, you're the number one new podcast this month. Therefore, Apple gives you a little shine.

SPEAKER_49: They do the same thing on their podcast rankings. Before we move on from this entire hardware conversation, there's a detail that I wanted to just bring out of the earnings call, Jason.

SPEAKER_08: We've talked on the show about GPU depreciation. How long will this hardware last? Is it just going to be a couple of years? During its earnings rundown, the CFO said, most accelerators with CUDA, which is their software suite, and NVIDIA's time-tested and versatile app architecture become obsolete within a few years. Thanks to CUDA, the A100 GPUs we shipped six years ago are still running at full utilization today powered by a vastly improved software stack.

SPEAKER_102: Six years ago, NVIDIA is using their same hardware. Yes, the A100 GPUs.

David Friedberg: So Michael Burry, you may have seen, and you can pull up some of Michael Burry's tweets. Michael

SPEAKER_03: Burry likes to delete his tweets. I don't know if he's deleted them, but if you follow the Michael Burry account, famous from the Big Short, he is accusing the big tech companies from committing fraud, literally thinks that this is fraud and that they're cooking the books because they're taking the depreciation schedule for servers from three, four, five years to four, five, six, seven. Now, we had a very granular discussion about this on All In and Freeburg went deep into the generally accepted accounting practices. You have a new piece of hardware here is my take on it. How long do you get to spread out the lifetime value of them? So when you buy a building for your office space or you buy an airplane as an airline, instead of just taking a charge for a hundred million dollar airplane in year one, you say, this has a 20-year life. We're going to take the hundred million and theoretically, five million per year. It might be more in the early years and it may taper off a little bit, but essentially there's some lifetime. What people are finding is with these NVIDIA chips, some people thought, well, we're just going to rip them out because NVIDIA is doubling the power of them every 18 months or 24 months or maybe even better. So therefore they're going to be ripped out and new ones are going to be put in and they're just going to be turned off in year four, year five, and nobody's going to want them. Well, it turns out like there might be jobs that these are

SPEAKER_00: perfectly suited for and they might last in year five, six, and seven. So then the question becomes,

SPEAKER_03: when are these consuming so much power? Because there's also power and there's like the facilities that they're housed in. So you have power, water, staffing, cooling, all that stuff is involved in the maintenance of them. So let's say you get to year seven and the maintenance cost is higher than the appreciation of a new one in the first year and the cost of swapping them out isn't that expensive. So therefore you get to some reasonable moment in time where you would just turn these off. Now what they're finding is, well, maybe those Sora jobs where people are making goofy five-second videos and eventually becomes five-minute videos, maybe these old ones will be just sitting there ripping

SPEAKER_11: those for people or making new advertisements for people or indexing their iPhoto or their Google photos in the background so that when they open up their photo app, you know, the AI is already done

SPEAKER_00: for it. Being a founder is a lot of work. You have to focus on your fundraising. Then you got to hire

SPEAKER_43: the perfect team. Oh gosh, product market fit, finding your first customers. But all of this means nothing if you're not actually a real business that's structured properly, giving your investors and clients the confidence to partner with you. At launch, we're constantly recommending Northwest Registered Agent to our founders because for just $39 plus state fees, they will act as your registered agent. That means they take care of all the paperwork, making sure your business remains compliant, protecting your privacy, and more. And as a founder, hey man, there are so many organizational odds and ends that you have to worry about, so why not have a partner who can focus on all those little details? Then you can obsess about your customers. And Northwest Registered Agent is even going to help you set up a phone line, a professional email account, and find you a great domain name. So here's your call to action. Go to NorthwestRegisteredAgent.com slash twist to get your

SPEAKER_00: company set up the right way. Michael Burry is correctly saying, hey, if you double the lifespan of these and these are your major expense and you're doing this massive build-out, well then you, it might be

SPEAKER_03: the equivalent of 10% of your earnings, 20% of your earnings. And he's kind of insinuating, and I don't

SPEAKER_00: know if he's explicitly used the word fraud or cooking the books, but let's see what he says here because that's kind of what people are interpreting this as, is that there is shenanigans.

SPEAKER_67: I'll use the technical term shenanigans, but let's see what he says. So Michael Burry says,

SPEAKER_08: understating depreciation by extending useful life of assets artificially boosts earnings, one of the more common frauds of the modern era. Again, this is quoting here, I'm not saying this, massively ramping capex through purchase of Nvidia chips and servers on a two to three year product cycle should not result in the extension of useful lives of compute equipment. Yet this is exactly what all the hyperscalers have done. By my estimates, they will understate depreciation by 176 billion between 2026 and 2028. By 2028, Oracle will overstate earnings 26.9% meta by 20.8%, etc. It gets worse more coming later on. Here is the table that he showed, Jason, and it shows the depreciation schedules for compute, essentially chips, if you will, inside major hyperscalers, Meta, Google, Oracle, Microsoft, Amazon, and you'll note that they've risen over time. The reason why the six year point from the Nvidia earnings call matters so much now is because most of these companies, Google, Alphabet, Oracle, and Microsoft are all running six year depreciation schedules on their GPUs. So it seems that they might not be nuts, but I do think that what Michael Burry is saying is that if these do burn out in one to three years, then these numbers they're putting out are in fact

SPEAKER_11: I think he's making a, he's confusing something here. If you look at the original tweet in that first

SPEAKER_03: sentence, he's saying the, I'm sorry, in the second paragraph, Nvidia chip servers are on a two to three year product cycle. So that's the product cycle. Okay. They come out with a new one

David Friedberg: every two years or like new, it might not be a new version because obviously they're versioning these things up like a new platform, like a new, the new version. Hopper to Blackwell. Exactly. And so

SPEAKER_03: if that's happening every two to three years, that means theoretically that you might buy more of them,

SPEAKER_00: but that does not mean that you're throwing the old ones away. That's ludicrous. That would be like saying I had a Corvette C seven and the C eight eighth generation came out. So I took the C seven and I parked it in the garage and yeah, they just took the air out of the tires and the battery and I put it on a cinder blocks and that's it. It's it's over the car can't be used ever. It's got 30,000 miles on it. That car is going to be driven for another 10 years. What are we talking about here? I

SPEAKER_03: think he might be making a story here that isn't in touch with reality. And then using the term fraud is like, that's a big, big accusation. It might just be when you pull up that chart one more time. Thank you. And you look at that table. It's not that egregious to think that Oracle believed in 2020 that they would get five years of life out of one of this depreciating assets. Then, you know,

David Friedberg: the next three years they said, yeah, it's still five. And then in 2024, they said, you know what? We think it's six. Okay. It's 20, it lasts 20% longer. That'd be like saying, oh yeah, that Corvette I'm talking about. You know what? We change the oil and we change the filters on it and we maintain the tires and we rotate the tires. So we think it can get another 20% of life out of it. And Meta going from, I don't know why Meta was so low at three or Microsoft was so low at three.

SPEAKER_03: And the Microsoft one is more dramatic. It goes from three years of useful life to six. So they doubled, but everybody's winding up at five and a half, six, six, six, five. And the year before it was four and a half, six, six, six, six. So Amazon brought it down a year. These are all kind of within 30%,

SPEAKER_08: right? Yes, absolutely. We do have the kind of rejoinder from Michael Burry here, Jason. I'm just going to pull that up for us. He argues in a November 19th kind of update to his argument that the idea of a useful life for depreciation being longer because chips from more than three to four years ago are fully booked confuses physical utilization with value creation. Just because something is used does not mean it is profitable. Gap refers to economic benefits. Airlines keep planes around for overflow. Sure. A100s take two to three X more power per flop. So they cost two to three X more than H100s. Yeah. NVIDIA claims H200 is less energy efficient than blackboard for inference. Okay. So his argument is that, yeah, everyone pointed this out to him, but maybe these chips won't be profitable. Here's the thing. The chips will be turned off when they are no longer gross margin positive. And so far we have not seen that. And as you pointed out, Microsoft for four years now has had this six year depreciation calendar. If it was wrong, they would already have some early warning signals and we'll be telling investors about that. So I like that Burry here is poking at things. I just think he's a little bit overly certain before enough of the facts have come in. But then again, that's how you make money.

SPEAKER_39: Has he talked to the people? I mean, Shemaf made this point on all in like, has he actually gone

SPEAKER_03: and visited the data centers and talked to the people who run the data centers and looked at the jobs they're running? Because if that airplane in his analogy is flying people on Thanksgiving and Thanksgiving and holiday tickets and July 4th tickets, moral day, if those extra planes are incredibly profitable during those peak seasons, then you just do the math on it. Like, is it worth keeping this one around for the overflow? In these cases, the overflow jobs, like I was saying, are

SPEAKER_00: things that might be running in the background that need to get done. And then you're going to do whatever the newest jobs are in three years on the latest generation. Like maybe they're going to be doing synthetic data. Maybe they're going to be doing self-driving data, like just more complex

David Friedberg: things. Doesn't mean people don't want to make bulldog, you know, superhero videos for their daughters still. And those would get done to those jobs. And there is the backstop of the power consumption because power consumption and the staffing of data centers and the cooling of data centers and the HVACs and all that, the water consumption, because that all exists, that's a backstop against

SPEAKER_00: keeping this stuff online unnecessarily because those bills keep coming in. That's not part of the depreciation schedule. That's just cash off the book. So there is a real incentive to turn off things that don't work. Like for the airplane, I guess, would be the

SPEAKER_03: maintenance contracts on the engines, et cetera. You still have to maintain the engines, et cetera. Every certain number of hours, the airframe, et cetera. Yeah.

David Friedberg: You have to have a hanger, right? You have to have somewhere to put it. So like at some point, they might be like, you know what? We don't use this plan that much. And it's costing us more

SPEAKER_03: than we make on Thanksgiving. Anyway, I think he's off on this one. I do think it's great that

SPEAKER_08: he's stress testing it. All right, moving on. Really quickly, I want to talk about Nano Banana Pro. This is actually Gemini 3 Pro image, but we did a little test, Jason. We want to get your take on this. So we gave it a prompt. This is Oliver and I, shout out to our friend. The prompt is, create an infographic of the top five barbecue spots in Austin. And we have one version from-

SPEAKER_149: This is great. I haven't seen this. I can give you my-

SPEAKER_08: Yes. So actually here is the Nano Banana Pro image. I want your take on the information,

SPEAKER_80: because I know you know Austin barbecue and your rating on the overall design.

SPEAKER_110: So this is very interesting. So they were able to make this without giving it the data and the

David Friedberg: text that's in the image. It went and did the research as a language model and then also did this beautiful image. Okay. So let's start. First of all, it's a good looking image. If you were to make

SPEAKER_03: this in a magazine, this could in the 90s and early 2000s, if you had a magazine that was doing a feature, this could be an infographic in the magazine as is. It might not be one in a Condé Nast magazine, but every other magazine, it would be totally fine. Time Out New York, whatever would be fine with this

David Friedberg: as an infographic. It would have been improved. So that's mind blowing. And it would have taken a week

SPEAKER_157: for a designer to make this. Yes. So, and that designer would have been a 50, 60,

SPEAKER_03: $70,000 a year designer for a mid-tier magazine. Maybe they would have been a 70, 80, 90 if they worked at a Condé Nast. Actually, you probably don't get paid that much at Condé Nast, but putting it aside, this would have cost $2,000 to do internally, internal costs. So now we go to the

David Friedberg: data, Franklin Barbecue, Leroy and Lewis, Interstellar, La Barbecue, Terry Black's Barbecue. I'm just

SPEAKER_00: looking for spelling errors. And do you see any spelling errors in this? I didn't see anything

SPEAKER_56: that caught my eye when I went through it before the show, but I still don't see anything. I'm just

SPEAKER_94: more terrified about the idea of fatty brisket, which is not my jam. Yeah. I like a little, they

SPEAKER_64: call that wet brisket in the business, but looking at these, Terry Black's, I always say, get the beef

David Friedberg: rib and the brisket there. That's my order. It is cafeteria style service. It is consistently excellent. It is spacious and they do have a quick moving line. It does have family legacy. That's all true. For La Barbecue, Michelin star, I don't know. Female owned, I don't know. Locally sourced meats, I actually do know that. And they are known for their brisket. I don't know if it's known for the fatty brisket, but they are known for the brisket. This peach tea glazed pork belly from

SPEAKER_67: Interstellar, I have heard about that and I'm certain they have the Michelin star. I do remember hearing that. Leroy, smoked beef cheeks, sounds amazing. And for Franklin, they are considered the best

SPEAKER_03: brisket in Austin. This is unbelievable just from the design to the accuracy. I'll tell you the other

David Friedberg: thing here that's interesting is that is what Terry Black's looks like. The illustration is close to the vibe and Interstellar and the barbecue. I think that is also the vibe. I think Interstellar is like a strip mall. Well done. I mean, apologies to infographic creators and graphic designers

SPEAKER_03: everywhere. If you're a graphic designer, you need to know how to use these tools and know how to edit

SPEAKER_07: from third base, essentially. Yeah. Like you're starting on third base. It's literally like-

SPEAKER_08: It is not as good when we do the same prompt through OpenAI. Just so people know the process here, I asked ChatGPT 5.1, my version of it, to give me the same thing. It gave me a written list and then said, would you like me to design an infographic? I said, yes. Jason, this is what it looks like in comparison. And I made this this morning. So this is absolutely, you know, state-of-the-art from what we have in ChatGPT. It's a mess. So a couple of things that are wrong here. One, beef briskexit comma ib is their specialty. Oh my God. Scratziad. Yes. And you can see there's overlaps between the images. It's inconsistent about what's shown. Scratziad. Go ia yul if possible.

SPEAKER_162: Thwa craft meats. What is that? Anticonditional sides. Thwaite craft meats. Anticonditional sides.

SPEAKER_08: It's not reliable, central. It's relable. But the point is, this just goes to show how good the new Google model is. And I think this goes to underscore the earlier points about their growth in consumer, their rising app store rankings, and the fact that OpenAI is kind of just not winning

David Friedberg: right now, which feels weird given where we were most of the year. If you're going to look at where

SPEAKER_03: gains will be compounded, the person who is first up the hill, who has to spend all their time fighting to raise money, fighting to build out infrastructure is going to spend, let's just take these three buckets of founder time. One third spent fundraising, one third spent building infrastructure, one third building the product. Now we move over to Google. Infrastructure is taken

SPEAKER_00: care of and fundraising is taken care of. It's called profits. So that team over there now has but one thing to focus on. The output of their image and the output of their language models,

David Friedberg: which is spectacular. And so this is super important for founders to understand when they're making a company. If you were spending your time to my points earlier about outsourcing things, there might be

SPEAKER_03: an argument for having your own servers and saving money in a robust business, or maybe you have some competitive advantage because you don't have to pay overages for your bandwidth and you're a bandwidth driven business like YouTube and they would do well to put up their own servers and not have to pay somebody for the overages. Putting all that aside, the more you can focus on the thing, the better.

SPEAKER_00: And that's why not having side quests, not having things you're doing that are not the core thing is so critical because eventually a competitor does show up. And this is why Zuckerberg was such a formidable competitor. MySpace was trying to keep their servers up and running. They were trying to deal with the politics inside of News Corp and the integration. And Zuckerberg was just like, I'm just going to focus on growth and making the service faster and growing the product. And that's it. And then he ran them over. So you got to focus on what matters.

SPEAKER_49: Quick recap before we move on to Kaoshi. On Monday, Grok 4.1, excellent.

SPEAKER_08: Tuesday, Gemini 3, excellent. Wednesday, Grok 4.1 fast, well-reviewed. GPT 5.1 Pro, GPT 5.1 Codex Max. Thursday, Nano Banana Pro. What a week for AI models.

David Friedberg: There's no training wall. What a time to be alive. I mean, literally everybody firing on all cylinders.

SPEAKER_03: This is why people are looking at the space and there is an America first, America only movement that is gaining steam in America. The reason is people are starting to believe on the right and on the left. We're getting into this horseshoe theory that my warning that job displacement could happen faster than we think it will and that I am concerned about it. Now, I'm not a doomerist. I'm just

SPEAKER_00: saying, hey, I just think it's a little bit faster. And people inside my circles, a number of people, pull me aside and say, can you stop saying this? This is like, you just, people want to be in the room where it happens. I've had multiple people in the industry. Hey, can we keep the music playing here? Like, please do not scare the public that jobs are going away. But I just showed that image. And I just said like, if you're running a magazine, why would you have a graphic designer on staff when you could just do that? That's more than enough to get the point across. If you had three

SPEAKER_03: designers on your team and one of them did illustrations, you would just get rid of the illustrator. That's it. Period. Full stop. And so here we are, folks. Get concerned. Get curious. Somewhere in between those is the truth. I'm curious about why this is happening and I am concerned. I'm not a doomerist. I'm not saying ban AI. To your point that like five things came out in one week and like we're going into the holiday week. Like, I don't think this is slowing down. And AI is going to make these AI tools better. Disgraciad version that Chat Chippy T made. If you were to use AI and just say, hey, we made this and Gemini made this. Why is our software suck compared to Gemini?

SPEAKER_138: AI is going to start fixing those problems. That's where the compounding gets really weird.

SPEAKER_08: In the open AI memo or conversation that the information reported, Sam does stress the importance of working towards super intelligence and eating some losses along the way to get there. Because I do think that there is a moment in which one company reaches this idea of a self-improving AI, Jason. And it might be that the first person there has a long-term advantage over their rivals. But right now, dear God, it feels like everyone's at the cutting edge and I love it. Moving on. Kalshi is going to raise again. TechCrunch reports that Kalshi is raising new capital, this time a billion dollars at an $11 billion valuation. Keep in mind that Kalshi last raised 300 at a $5 billion valuation in October and raised at a $2 billion valuation in June. So what the hell is going on? Well, I went and pulled the data. So Jason, from The Block, a crypto-focused news and data service, this is their reporting of Polymarker and Kalshi volume on a monthly basis. Now, Kalshi is in blue if you're looking at this on the visual version. Polymarker is in red. What matters is that after some very modest growth through August of this year, Kalshi and Polymarker have grown very rapidly in the last couple of months, Jason. And this to me explains why investors are salivating to get more capital in because they're seeing incredibly impressive growth

SPEAKER_180: from maybe about $2 billion in volume between the two companies in August up to $7.5 billion in October.

SPEAKER_108: Amazing. It's becoming an integral part of understanding the world. And I think it's

SPEAKER_03: fantastic. We now have people doing research in the world and there's consulting firms and survey firms, people who do polling data, right? Then you have journalists over here, you have experts going direct and talking directly to consumers on podcasts. And so as a consumer, I've always felt it's your job to take all these disparate news sources, data sources, and try to

SPEAKER_00: figure things out. And now area where people are placing bets, they're putting skin in the game

David Friedberg: and trying to make a profit. Well, it really does help people understand where the market's going. So pull up Polymarker and the chances of a rate cut in December. You know, the stock market, as you were pointing out in the first story, we didn't get too deeply into it, has had a little mini correction, like 10% or so. Some stocks, 30%. But if we were to look at the Fed's decision here in December, and let's put it at like one week, so we can get a little bit of a better picture here. You had the stock market crash on, not crash, but correct a bit on November 20th, right? Today's the 21st, I believe. Yep. Yesterday was a bloodbath. The crypto market specifically has gotten really, has really corrected. So we believed that there was going to be no change. That was, looks like 70, almost 80% chance of no change in December yesterday, right? Yep. 67, 76, it looks like. 76%,

SPEAKER_06: yeah. And I think also this is after the jobs report, that was better than expected. But not

David Friedberg: ridiculously so. Then here, you look at today, now the chance of no change is down to 34%. So if you hover over today, it literally flipped. Now, why did it flip? Well, there was one of the people who work for the Fed said, you know, some things that might indicate. Right here. Here we go. Yeah.

SPEAKER_08: So New York Fed, John Williams said that we might actually need a rate cut in December. And people are kind of thinking out loud that the Fed doesn't communicate on accident, Jason. So if John Williams is saying this, they're probably trying to calm the market a little bit and make it clear that we're, there's a good chance we'll get a 25 bps rate cut in December 10th. Not financial advice, but I think

David Friedberg: the market's going to rip when we have these rate cuts. It's going to be super stimulating to the economy. And we already have a ton of money being invested in America through a lot of these international deals that Bessett and Lutnik negotiated for Trump in America. That investment's going to be starting to land at the same time that we're continuing to spend money. Japan's spending money. Everybody's doing stimulus. So this is going to be like, man, a lot of pent up energy going into spending. I don't know what that does for debt load. I guess we'll kick the can down the road, but I think it could be an absolutely crazy market. And then if you look at M&A, the fact that the meta deal resulted in a nothing burger, the meta FTC scrap over a monopoly in the

SPEAKER_194: personal social networking market, nothing burger, right? Just absolute nothing. And then the Google

David Friedberg: one was like, it was this. Yeah. It was like a tap on the wrist. They're like, here's a speeding ticket.

SPEAKER_00: So M&A is going to be on fire. I'm going to predict on the show right now, we're going to see a major M&A moment in the next six months. In the next six months, we're going to see a company on the scale

David Friedberg: of an Airbnb, an Uber, a Coinbase, call it the mid caps, not the MAG7. Between 50 billion and 500 billion. Let's say 50 to 250. I'll narrow it a little bit. It can't be as high as 500 billion. Those things are going to, all kinds of alarms are going to go up. But I'll say 25 billion to 250

SPEAKER_03: billion. We're going to see two or three of these get discussed and we're going to see at least one of them get pop. It's going to pop in the next year. So I'll say in the next year, well, in the next year, there'll be a half dozen discussions and we're going to see one of these pop. And it's going to be Amazon buying DoorDash or DoorDash and Uber merging or Uber buying and Waymo spinning out into the same company or Tesla buying Uber. I don't have any inside information here. Autonomy in cars, delivery of products, Amazon. Maybe, you know, somebody like Apple says, where's our growth going

SPEAKER_00: to come from in the future? What would be interesting for us to own? Yeah. You know, maybe it would be really interesting to own Adobe or Figma. Oh, if Apple bought Figma. Oh my gosh.

SPEAKER_04: Well, I mean, let's just, I don't have, again, I don't have any inside information, but the M&A teams

SPEAKER_03: are on. That I know because we have a portfolio and once in a while we'll hear, that's not somebody at my suite. Oh, I got to turn over your room. That's the M&A people are knocking.

SPEAKER_20: The knocking is starting to happen. It's going to be really juicy. 2026 is my prediction. Question

SPEAKER_08: from Arnody Rishav Rajain. Hey man, what do you think software engineers should work on considering AI? Answer this question, Jason, in terms of self-learning and what they should be adding to their roster if

SPEAKER_108: they want to be employable in the next couple of years? You know, the thing that is top of mind for

SPEAKER_03: all organizations is trying to make agents, which is to say something that runs like a cron job, we used to call them, on some set schedule, could be every minute, could be every second, that does something for the organization that would fall into chores. So if you can study chores that people do, whether it's in customer support or sales, you know, distribution, whatever those chores are that are going on and being able to study people in an organization and build an agent that allows this to occur on some regular basis. I was just saying the other day, man, I really wish somebody could make me an agent that just went to these 10 home pages and told me the top news stories on each and then made me a ranked list, a weighted list of the top stories and then made them on a graph so I could see, okay, the Epstein story is, you know, waning and the, this story is coming back. Oh, the Comey story is coming back. Oh, the unemployment story is back. It's spiked today. Oh, it went away. And then just tell me over time in some, you know, reasonable way, what is happening? And there's no Bloomberg terminal of that, but I could architect it really easy. There used to be

David Friedberg: meta lists that would do this. You know, pop URLs was a product that did it where you just see all the top things. Tech meme does this really well for tech. So anyway, long story short, I would look at agentic workflows, whether it's in the accounting department, the HR department, whatever it is, because people want to be more fit in their companies. They don't want to have fat. They want

SPEAKER_03: to have muscle. They want to be lean, mean fighting machines. All right. We're super excited about Startup

SPEAKER_43: Supper Club. Remember we were doing Founder Fridays. Well, our friends over at River, Rachel Lambert, she's the CEO of River. She realized through just studying all the people who were coming to Founder Fridays, that what people really wanted was to just share dinner. They wanted a very simple concept. So if you go to startupsupperclub.dnr.io, December 5th, they're going to start doing amazing dinners in New York City, Austin, and San Francisco. I might just randomly show up in one of those cities in the future. I don't know if I'll do it on December 5th, but I might. It's pretty simple. You take a little personality quiz. They do the logistics. You show up. You have a great dinner. You make new friends. Maybe there's a little after party. It's working. They tested this. People loved it. So get matched with other founders, investors, and just tech people who share your interest. The dinner reservation set, all you got to do is show up. Yeah. A lot of the groups, they'll meet at a bar afterwards. So New York City, Austin, and San Francisco, December 5th at 7 p.m. The link is in the show notes. Startupsupperclub.dnr.io. Check it out.

SPEAKER_209: Founders, investors, go make some friends. That's what life's all about. All right, everybody. That's

SPEAKER_43: another week in the books for this week in startups. We'll see everybody on Monday. Bye-bye.