The AI chip squeeze, Dirty Soda mania, OCC approves Palmer Luckey’s Erebor | Diet TBPN

0:00It's white suit day, the market's up. For the rally. Don't check what the market did yesterday. Check what it's doing today. You never doubted, did you? You never doubted. You never doubted when the fear and greed index was on full fear. You're post-econom. Scared for its life. It happened. Big week. Bottlenecks. Bottlenecks. I mean, we're going to talk about the clog code moment, the clog code psychosis, sort of the software singularity. There's clearly signs of a takeoff. It feels like a slow takeoff, but there's a whole bunch of sort of recursive compounding elements that are starting to form.

0:37A recursive, you say? Yeah, a recursive. Literally recursive. Like the models feed back into themselves, give them more tasks. We saw this with Gastown. There's a whole bunch of stuff going on in orchestration that's interesting. And so I wanted to sort of reflect on like if there's going to be a break, if there's going to be a damper on the party, if someone's going to pull away the punch bowl, who's it going to be? The semiconductor industry or the energy industry? At the start of this year, I said it was going to be the year of energy. I still think it's important to think about energy because that it will be a bottleneck.

1:04But this year, it's about chips. And not even just because of AI, just for overall human flourishing. Totally, totally. Yeah. And yeah, so we had some great conversations this week. Sam Altman. We talked to Dylan Patel. We talked to Sholto. We talked to, you know, we reviewed what Ben Thompson was saying. Dwarf Keshe interviewed Elon. So there's a lot of new data points about how people are thinking about the trade-offs between semiconductor fabrication capacity and energy production capacity. So I wanted to sort of like crystallize like where I think the debate and consensus is right now. There's been this like tick tock going back and forth in the AI supply chain.

1:39What's the key bottleneck to growth? It does feel like only if you have to work at an AI lab to really feel the bottleneck. For most people, they're just like, I open the chat app, probably chat GPT. I ask it a question. It gets back to me in a reasonable amount of time. I'm not really hitting rate limits. The rate limits come when there's big moments, the Studio Ghibli moment. I think people have always felt the rate limiting is very real with using Anthropic products. Yes. Which is why I asked Scholz yesterday, what are free limits going to look like? That's why you asked him that. They've consistently been compute constrained.

2:14At least they've talked about it more than most, and users have talked about it quite a lot. So I'm just quite curious, what is the free experience going to be like Sunday? If somebody downloads the app, sees the ads, they're like, I want LLMs without ads, even though no other popular LLM actually has ads yet. They're pretty upfront with you about the fact that if you're on the Opus model, you'll hit your rate limits faster. And even just a couple prompts deep, it will take a second and kind of compact the conversation so it can keep talking to you. I know they're increasing the token context window, but it feels like it works fine, but it is a smaller user base.

2:52Although I'm very interested to see where it goes in the App Store. ChatGPG is still at number one. Grok is oddly doing incredible in the App Store. Way higher than Axe, the social networking app. Yeah, for no obvious reason. I mean, there's just not a lot of hype about Grok. They're doing okay on benchmarks. Obviously, Elon has a very solid playbook for scaling and hiring. Yeah, but to me, I don't care as much about hype. Yeah. When it comes to the App Store, There's plenty of apps on the App Store charts that have no hype. One of the top apps is called free cash. Free cash. It's number two. But that one makes it.

3:27Above Gemini. I understand why that one is. Is free cash. Get paid real money. Yeah. Hot on Chatshpete's heels. You know, we have the OpenAI Elon lawsuit. Yes. As part of that, what if part of the settlement agreement is OpenAI has to give 4-0 to Grok? I mean, they might just wind up. Yesterday was seriously one of the craziest experiences. is if you weren't watching our SAM interview live, and honestly, the entire show, there was thousands of messages from people, from 4-0 soldiers saying, keep 4-0, keep 4-0, throwing hashtags. You don't see hashtags that much anymore, but they were throwing hashtags around, keep 4-0.

4:08So I think maybe if Elon can negotiate for that. So thank you to the TBPN army, the day ones. We'll never switch up on you. We'll never let the money change us. We'll never forget where we came from. Yeah, Bobby was going to... You guys in the chat were doing overtime work, keeping things somewhat sane. There's not much you can do in that situation. And yeah, it'll be interesting to see 4.0's fully going offline. It's already very hard to reach. You have to go into turn on legacy models in the settings. It's going offline right around Valentine's Day. I think February 13th. It's just a fascinating, weird time that I don't think we've ever seen...

4:43Because people were upset when Facebook launched the feed. And they were like, we want to go back. and they made these groups and the groups got a lot of application ironically some people were upset when Microsoft stopped investing as heavily into Clippy that's true that's true but I was never logging on to a twitch stream and saying I got hang bring back Clippy the notable thing is they didn't seem to be able to process that it wasn't just a show about Sam yeah even after Sam had left they just kept going yeah which is very very strange anyway back to the bottleneck right now it feels like chips are the more important piece of the bottleneck to talk about Sam Altman put it this way I asked him like chips versus energy what's the bigger bottleneck right now he says it goes back and forth but right now it's chips it's different at different times it may get solved on its own normal capitalism may solve it but I think somehow deciding as a society that we are going to increase the wafer capacity of the world and we're gonna fund that and we're gonna get the whole supply chain and the talented people who make that happen would be a very good thing to do And so why do we have a chip bottleneck to begin with?

5:45Semiconductors have been doubling and you know, we've been on this Morse law curve What's interesting is that the the semiconductor industry should be better equipped to avoid a bottleneck because it's already been on an exponential Whereas energy production has just been like flat sort of like a malaise for a long time Getting on getting that unstuck is hard, but I think that's a problem for 2027 potentially So the chip bottleneck comes down to consolidation power plants data centers cooling technology There's a bunch of suppliers in each of these industries, and you can parallelize them, and you can steer resources from adjacent areas to focus on AI projects specifically.

6:19Even a company like Boom Supersonic can turn into a turbine manufacturer, and there's a lot of other industries that are able to move over. The semiconductor industry is used to doubling the amount of transistors made every year or two. Part of that is more advanced nodes. Part of that's more capacity, whereas the energy industry in America wasn't built for that kind of growth. So initially, people weren't creative. They were like, let's build these combined cycle gas plants. But now we've realized, yes, there are three main manufacturers of turbines. And for dual combined cycle, you've got IGTs, but you've also got medium speed reciprocating engines.

6:55Turns out Cummins can make about a million diesel engines a year, and those can generate electricity. If I don't care about aesthetics and I put them in West Texas, easy. Back to the leading-edge fabs, it's a completely different beast. These fabs cost tens of billions, 30, 50, 75, 80 billion I've seen to build and it takes three, four, sometimes five years to go from breaking ground to actually getting up to producing volume. And we have like the perfect example of this because TSMC announced a plant in Arizona in 2020 and in 2025 it's still not producing at volume. It's doing really well, it's great, but that's five years and it's not just like oh yeah it's as effective as what's in Taiwan.

7:35There's this bottleneck within the chip bottleneck which is the TSMC supplier ASML and they ship around 50 EUV machines per year maybe 50 60 each one cost 350 million dollars and leading-edge fabs need dozens so if you want to build a bunch more fabs you need a bunch more toolmakers and ASML has its own supply chain for different lenses and glass and all sorts of stuff they work with Zeiss Trump a whole bunch of different companies and that supply chain is not very diversified so you have another bottleneck even deeper in the supply chain. And then even after you get the fab built, there's still at least a year of processing, engineering, maybe 12 to 18 months where you actually work to get to high yield production.

8:11We're seeing that in TSMC Arizona. And TSMC just has decades of intellectual capital locked in the heads of engineers that they can't easily transfer or parallelize. And so this has made TSMC the real bottleneck. Hyperscalers are pushing CapEx numbers into hundreds of billions of dollars. So TSMC controls 90 % of the advanced node market with Samsung and Intel far behind. And this is why Ben Thompson, aside from the geopolitical concerns about TSMC, is really urging tech companies to wake up. And so he says, the reality that hyperscalers and fabless chip companies need to wake up to, however, is that avoiding the risk of working with someone other than TSMC incurs new risks that are both harder to see and also more substantial.

8:50So there's a huge risk if you say, you know what, I'm gonna go and place a huge order with Samsung or Intel, or I'm gonna I'm gonna take a huge risk and be the anchor customer of their new cutting edge, leading edge fab. But Ben Thompson is saying there's a risk to not doing that. And he says, except again, we can see the harms already. Forgone revenue today as demand outstrips supply. Today's shortages, however, may prove to be peanuts. If AI has the potential, these companies claim it does. Future foregone revenue at the end of the decade is going to cost exponentially more, surely a lot more than whatever is necessary expense-wise to make Samsung or Intel into viable competitors for TSMC.

9:31You really got to wonder what conversations are like between Jensen and TSMC right now, given that TSMC is not, they're certainly not going out and making, kind of going risk on, right? They're staying relatively, probably more conservative than than some of their downstream customers would like. Anyway, let's move on to the hyperscalers. Compound248 says, poor Jassy, Andy Jassy. He's gonna learn something about Amazon today. When Google announces a crazy number, it's because it's playing offense. But when Amazon announces a crazy number, it's because it's playing defense, ipso facto. So Amazon said it will spend$200 billion this year on AI build out, but this is worrying investors that the company's colossal bet on artificial intelligence will pinch profits while it waits for the investments to pay off.

10:22The company reported spending$130 billion on property and equipment in 2025. Analysts expect those expenses would reach$150 billion this year, but Amazon's saying, we're going to$200. We're going all in. And a lot of frustration from Amazon shareholders. Obviously, They had the fastest growth in 13 quarters, and the stock is still down 10%. Looking back over the last five years to early 2021, only up 23 % over five years. Yeah, just not as sexy as a narrative as the rest of the hyperscalers. Zuck, obviously, is a huge beneficiary of advancements in AI. We see that with the accelerating ads market.

11:06Microsoft, even though, yeah, it's getting beat up a little bit now, They have like that massive position in open AI that feels like fully in solid place. They got the IP. They got, you know, a real hold on the AI question. Google's obviously in a fantastic place. Yeah. NVIDIA, clearly. Yesterday was just such a strange day. Reform Trader says software stocks are cooked for not investing in AI. Mag7 stocks are cooked for investing heavily in AI. Hope that helps. It is crazy. Oh, this is the image. I love this image. Wait, what's this from? Matt Damon in, which one? I don't know, in Rounders. I watched it last night.

11:42We were just talking about this. That's Rounders. I thought Goodwill Hunting for some reason, but this is a summary of the 2026 CapEx numbers from the hyperscalers. Google is gonna do 175, 185 billion versus 120 estimated. Meta's doing 115 to 135 relative against 110 estimated. Tesla's going to 20 billion from 11 billion estimated. and Amazon, they were estimating 145, but they're gonna do 200. And so they're going all in. One of the top comments here is very funny. It says, what exactly is meta buying? They don't do anything with AI. That's not true. Open up reels, it's AI. Even in the Cheeky Pine Elon episode, you know, Dwarkesh and John are pushing Elon on the data centers and space thing.

12:32And I thought that Elon could have done more to really get people basically the same thing. The TAM is an order of magnitude bigger than anyone is contemplating. And we are investing for that future, right? It was a lot of like trying to just justify putting it up there. Yeah, in the short term, yeah. Yeah, when they were saying like, hey, there's a lot of land in the US, we can blanket the US with solar panels. We do a lot here. And so again, I think people are getting kind of caught up in the details. If you're gonna spend, you know, you would hope AWS making this kind of investment that they have more conviction than Tyler does on the opportunity, right?

13:09I don't know if that's possible. Tyler's in the white suit. He's looking good. Yeah, looking good, Tyler. Yeah. Round of applause for Tyler. This is a classic Elon thing where, you know, he will lay out a vision that takes 10, 20, 30 years, but he says it's going to take 10, 20, 30 months. And so you have to sort of grapple with the short-term but still you can't lose sight of the long term because we did get rockets that land and we did get space, internet, and whatnot. Yeah, I think he said 36 months is his estimate for when we start getting compute in space, like probably 30 months, I think.

13:47But then so there's kind of the, that's the shorter term thing, but then he did talk about terawatts in space, right? So the US is currently half a terawatt, but we're gonna be putting terawatts in space of compute. He's sort of like falling back to first principles and just talking about - Yeah, he keeps repeating like, oh, if you just look at like physics. Yeah. But it's kind of this catch-all term he uses when he doesn't really know what to say, it seems like. Yeah, well, yeah, it just seems like it all checks out from the physics calculations, but when you map that to the economic realities, the human capital and financial capital realities, you're just looking at a very long timeline and he doesn't like talking about thinking in decades.

14:24It seems like the short-term like bull case for space data centers is basically just like, you know, regulatory, right? It's just gonna be way too hard to find the actual land. You see India already saying, hey, we'll just take a slice in 20 years. That's how AGI-pilled they are. They're like, come build a data center here. You can be tax-free for 20 years, but then we're gonna be making money in, what, 2046? They're ready to rock. None of you nerds know physics. They were feeding him drinks. Chill out. It is actually crazy. I mean, we've never seen a Cheeky Pine episode where they actually drank.

14:56They did drink on this one. That was the main feedback from last year. Yeah, it is funny. Yeah, it feels like Elon took down three pints or four pints or something. You did the math, right? So we'll talk about this later. There's some posts, but, you know, they seemed a bit smaller than normal pints, maybe. So is it really? But I counted three, like, full glasses. I mean, I think Elon, Dworkesh asked a question about this, which is like, oh, it's so crazy that, like, SpaceX, they have this super ambitious mission, but they keep seeming to find these, like, good businesses on the way, right? So first it's like Starlink with Falcon, and then with Starship, it's like, do we really need that many Starlinks?

15:29But if you can get space data centers, if you're putting up terawatts of compute, then you need massive Starship. OTP in the chat says, dude just turned 21 and he can already eye a plant. He's been studying the last week. $660 billion of CapEx this year on AI data centers. To put a number like that in perspective, this is more than what we spent on the US Interstate Highway System,$630 billion. More than what we spent on the Apollo Moon Program,$257 billion. More than what we spent on the International Space Station,$150 billion. It's more money than Walmart's revenue last year,$650 billion. It's the equivalent of spending$1.8 billion a day,$750 million an hour, or$1.2 million a minute.

16:14This year alone is without a doubt the biggest project in the history of capitalism, and we are spending all of it in one year. France announced an initiative earlier this week. They want to lead in AI research. So through their France 2030 program, they have invested more than 30 million euros in this initiative. Macron says science has found its home. and sean frank says france is investing 30 million in this new ai initiative that's how much google will spend in 90 minutes no joke every 90 minutes google will spend 30 million on capex this year one company wow one company they can hire ilia for like an hour an hour one hour consulting call take him is breaking down the nvidia news the stock is up 6.22 percent nvidia shares barely moved in after i was trading last night yeah this is what i said i was like Amazon just announced that, hey, we're going to spend$200 billion.

17:10Yeah, what are they going to buy with those? I mean, they will buy Tranium and Inferentia, but they're going to buy a lot of NVIDIA. Anyways, let's move on to Dirty Soda. Dirty Soda. We got to talk about the Utah mom. Have you heard of Dirty Soda before? We got the world hooked on Dirty Soda. I had never heard of Dirty Soda before. Have you heard of this? Yeah, yeah. This is big in Utah. This is big in Utah? Yeah, yeah. Have you been to Utah? No. How do you know that it's big in Utah, then? Exposed. My friends send it to me on Instagram. They're like, look at this. You have friends who live in Utah?

17:42It's like soda, and they put cream in it and stuff, right? Okay. We'll dig in. I think we'll let the Wall Street Journal get the facts straight. A mom with five children, ages 6 to 16, in St. George, Utah. Nicole Tanner didn't realize she was on to something with dirty soda. That's what the creator of the chain, Swig, calls spiking. Coke, Mountain Dew, or Dr. Pepper with fruit, purees, and flavored cream served in plastic cups stuffed with pebble ice microplastic nightmare really I didn't know pebble ice was at risk no no no I'm talking about I'm just saying plastic oh the plastic cup her business took off a few years ago when pop star Olivia Rodrigo was in Utah filming the TV series high school musical and posted an Instagram photo holding a cup from swig now hashtag dirty soda yeah hashtag big chains want in McDonald's recently tested Sprite with lemon vanilla syrup dragon fruit taco bell swirls it's teal colored baja dream freezes with vanilla cream and calls it a mountain dew baja blast dirty freeze we forgot to ask sam altman if he's gonna issue a baja blast because he did the code red it's time for a baja time to baja blast he's back to the top maybe that's what the super bowl ad codex oh that would be blast codex collab campaign for the builders and drive-in chain sonic has encouraged customers to make it dirty by ordering creamer and mix-ins with their sodas okay while the heart of dirty sodas may still be utah and the mountain west states swig has expanded to around 140 locations this is insane to me to me this feels like a really tough business to be in long term because if if the product becomes popular everyone has it everyone has it immediately you have no real IP, right?

19:26Because you're just using sodas off the shelf. Tanner's main investor, Family Investment Office, Larry H. Miller Co., brought in a professional chief executive who has taken other companies public, and he is talking of an eventual initial public offering for the chain, which had around$100 million in sales last year. Tanner, I mean, these must be incredibly simple to run. It's literally a box. You have soda and creamer. We're doing what Starbucks did for coffee, but for soda, said Swig CEO, Alex Dunn. So Alex Dunn says, pull up here, 6 a.m. on the way to the gym, grab a big soda. I feel like you'd need more caffeine if you really wanted to displace Starbucks.

20:01Well, I was just going to say on the caffeine thing, isn't the whole point that it doesn't have, it's like not caffeinated, right? Because you're Mormon. Oh, okay. So it's a Utah thing. I did something recently that felt, it felt really wrong, but the result was good. I mixed a Yerba with a Mexican Coke. Ooh, okay. It was fantastic. Mataina? Yeah. And Mexican Coke. Yeah. That's fun. They said a few years ago, Olivia Rodrigo promoted it and it went viral. I thought this business was only a few years old, I guess. No, it's an overnight success. It started at 20 cents. Overnight success. That's pretty remarkable.

20:36After customers started referring to Swig's Dr. Pepper and Cream combination as dirty soda, Tanner latched onto the phrase encouraging customers. They got to pay future royalty. I won't stand by this. They should get future as an influencer. That'd be good. Yeah, he's like, when I said dirty soda, I meant. By 2017, Swig had grown to more than a dozen stores. The staff was spread thin, and Tana realized she lacked the expertise to grow further. She and her husband separated in 2020, RIP, and is no longer involved in Swig. I hate when I'm just reading this incredible story of entrepreneurship, and then they kind of interrupt the flow there.

21:15The White House has posted, biggest period bowl period run period ever period starting period now period did they post this scroll down yeah no so so they're sharing a fake screenshot okay from true social where donald trump says let the gains begin i mean he really did call the the bottom at least a local bottom of veracity and truth no i'm just saying like if you bought if you bought a video when he bought an video when he hit when this fake post was yes shared by the real account uh joe weisenthal shared yesterday silver down 19 percent anthropic must have launched a silver extension yes it's just chaos in the markets it's just absolute chaos everywhere one of the one of the most chaotic weeks of my my adult life yeah but it's been lots of green shoots lots of interesting projects lots of interesting applications and progress all over the place Derek Thompson said for me the odds that AI is a bubble declined significantly in the last three weeks and the odds that we are actually quite under built for the necessary levels of inference usage went up significantly in that period basically I think AI is gonna be become the home screen of a ludicrously high percentage of white collar workers in the next two years and parallel agents will be deployed in the battlefield of knowledge work at downright Soviet levels.

22:49And Kevin Roos over at the New York Times, host of the Hard Fork podcast, says, this is why everyone was freaking out about clogged code over winter break. Once you see an agent autonomously doing stuff for you, it's so instantly clear that all computer-based work will be done this way. This is why my serious AI policy proposal is to sit every member of Congress down in a room with laptops for 30 minutes and have them all build websites. Yeah, so we were joking about this yesterday with Sholto, like we maybe need more long weekends for AI adoption, but I wouldn't be surprised for a big company to like actually do something like this, which is like, hey, we're going to have the next Thursday, Friday off.

23:31Hackathon. Yeah, basically. Congressional a whole other thing you got to go to congress sit all the conger all the congress people down and get them get them vibe coding what would they build i feel like they don't use a lot of software so there's not that much to build like they're so abstracted away from it it's all lunches and phone calls and dinners and meetings like there's not that much that actually this is the thing with like open claw there's some people asking like what are you actually using it for and people are realizing a lot of my a lot of my day is not you know the thing that i wish i could automate What?

24:05My mail at home. Okay, you can do that. Have you seen Earth class mail? There are a few of these virtual Okay, it sends it it sends it. So basically you forget your home address. You don't share it anywhere on the internet You never put it anywhere. You only use this other address. Mine was like 830 Market Street in San Francisco All of the all of the mail would go there if it's addressed to me It gets opened by a robot scanned and then you have a web dashboard but also it just goes to your email. And then you can actually say - So they've been AGI-pilled for decades. Decades. No, it was really important when you set up a business, you use a fake address or one of these virtual mailboxes, and then you can actually click a button, send this to me physically, and then you have a virtual representation of it forever, and you could run an agent over it.

24:53So maybe that's the next thing you pick up is a virtual mailbox, that would be good. I'm just saying I never want to open up a physical piece of mail again. Optimus opens it for you, takes a picture. Does a money spread. Does a money spread, walks to the post office, mails the next letter, mails the check. This is the future. Jensen is pushing back on the AI will kill all software narrative. Let's play the video. Software is a tool. There's this notion that the tool in the software industry is in decline and will be replaced by AI. You could tell because there's a whole bunch of software companies whose stock prices are under a lot of pressure because somehow AI is going to replace them.

25:32It is the most illogical thing in the world, and time will prove itself. Let's just give ourselves the ultimate thought experiment. Suppose we are the ultimate AI, artificial general robotics, the ultimate AI, the physical version of us. You could, of course, solve any problem because, you know, you're humanoid. You could do things. If you were a human or robot, Would you use a screwdriver or invent a new screwdriver? I would just use one. Would you use a hammer or invent a new hammer? Would you use a chainsaw or invent a new chainsaw? Ideally, they don't use it at all. A lot of disagreement in the timeline.

26:11But do you understand what I'm saying? If you were a human or robot, artificial general robotics, would you use tools or reinvent tools? The answer, obviously, is to use tools. Tyler? Okay, I completely disagree. So if you're a robot, you'll use the calculator, right? You'll just build the calculator if it costs you a lot of money to use the calculator. Yes, yes, yes. So I think digital agents will use Salesforce, right? But if Salesforce becomes very expensive or if it's already fairly expensive, at some point it becomes cheaper to just build it themselves and then you run it. Yes, yes. Yeah, especially when they're like, I can work the equivalent of 2 ,000 lifetimes today.

26:56Let's go over to the Guinness tally. Tyler has it. He counted it up. He sat down. He studied the Dworkesh Patel cheeky pint crossover episode with Elon Musk. And he counted three pints for Elon, four pints for John, and three quarters of a pint for Dworkesh. But there's a suspicious refill. So if you're tracking Dworkesh's volume throughout the episode, it kind of goes down, and he's at maybe like 80 % full. And then it goes back up. There's like an ad read, and then it goes back up. So the question is, like, whatever happened during the ad read, right? Did he just like slam it? True. And then he filled back up all the way?

27:33Or is it just kind of a partial fill? I think you've got to give him 1.75. Okay. Can we pull up the screenshot? They look like full pint glasses. So, yeah. Those look like full-sized pint glasses. Another photo of Elon holding the pint? Yes. Okay. Looks pretty small in his hands. A little small. And then let's click over to Timothee Chalamet. Timothee Chalamet, I mean, he's a smaller person, but it looks the same size to me. I think these are fair pint glasses. So Timothee Chalamet is 5 '10". Elon is 6 '2". Okay. So a slight height difference, but I mean, I feel like Elon is really mauling that.