You're about to listen to yet another Value Podcast with your host, me, Andrew Walker. I guess before we get into it, it means a ton if you can rate, subscribe, review the podcast wherever you're watching or listening to it. But for today, I think you're really going to enjoy this one. We have Doug O'Laughlin. He is from SemiAnalysis, he's from Fabricated Knowledge. He is just an absolute, absolute expert on all things semiconductor, AI, power, all of that. Just an absolute expert. He is at the center of the field. We have a really fun conversation. And here's a bonus: he's really effing good at corporate governance, spring-loaded ads, all that type of shenanigans where if investors are really reading through the 10-Ks and stuff, they might find a management team that all of a sudden flips really bullish or signals that they're going to sell the company. Obviously, nothing's guaranteed. You can see the full disclaimer. Nothing is investing advice. See the full disclaimer at the end. But we have a really fun conversation on the back end about some of the crazier corporate governance signals and everything we're seeing. So, that's the podcast today. I'm going to get there in one second, but I'll just roll right into the advertisement. This podcast is sponsored by AlphaSense. AlphaSense is obviously a fantastic sponsor of mine, but one of the reasons Doug is coming on the podcast is because A, he's a friend, but B, he is doing a webinar with AlphaSense that is going to go live. I'm posting this podcast on Tuesday, October 28th. It will go live Tuesday, October 28th. So, if you kind of like what you're hearing from Doug here, I'll include a link in the show notes. I'll include a link on the blog, all that sort of stuff. You should go sign up for the blog because again, here's a little secret, SemiAnalysis Doug, if you were a random pod shop guy who wanted to talk to them about what's happening in semiconductors and AI, this podcast is free, the AlphaSense webinar is free. I promise you, Doug's time ain't free. You would be paying quite a good bit of money. So, look, anytime Doug speaks, I listen. I think he's super thoughtful. I think you're going to enjoy this, and if you enjoy this, I think you'll enjoy the AlphaSense webinar. So you should go check that out, see a link in the show notes. And with that said, we're going to hop on into the podcast. All right, hello and welcome to Yet Another Value Podcast. I'm your host, Andrew Walker. With me today, I'm happy to have on from Fabricated Knowledge, SemiAnalysis, everything, Doug O'Laughlin. Doug, how's it going?
Good, man. I'm worried that I have to follow up one of the best episodes of Yet Another Value Podcast ever, but we'll try. We'll try. Maybe I have another 30-bagger accidentally, you know.
I'm laughing because the last episode I published was me just talking to a camera for 30 minutes. So, I assumed you meant, being the narcissist I am, I assumed you meant the episode of me talking, not Doug is referring to literally the best pitch on Yet Another Value Podcast, episode 166, AppLovin. Doug pitched it at what, like 15? It's at 500 or something.
Yeah. Something stupid. I don't even know. I don't own anymore, so it's like whatever. It's just I see it, I get sad. Yeah. Honestly, disbelief. Yeah.
I remember that one, too. I think that I was like super delayed on that one and then we were like, 'Fuck it. We'll do this podcast on that one.' So, here we are. I'm a little sick today after the SemiAnalysis retreat, but we're here to chat about AI.
I remember being quote-unquote sick in college a lot.
So, to be clear, to be clear, I was actually hungover. Like dude, I probably drank like eight days, eight of the last 10 days if I had to guess. So, yeah. Yeah, I deserve to be sick and I have been hungover, but I know distinctly what is hungover and what is sick. I am sick and actually someone on my team is sick and so everyone got sick and this is a bioreasoning bottom pitch and that's where I'm at.
All I hear is when I hear 'drank eight of the last 10 days' and Doug again at SemiAnalysis, I just hear, oh man, the AI guys are partying so hard and the value guys are just so sad all the time.
Look, I want to talk about so much stuff. Before I get there, quick disclaimer, remind everyone. Nothing on this podcast is investing advice. Always true. Particularly true today because Doug and I are going to talk about a thousand things. And I just want to remind you, even though the last time Doug stepped on this podcast, he pitched a 30-bagger. That's not investing advice. See the full disclaimer at the end of the podcast. Doug, tons of stuff we want to talk about today. I want to talk corporate governance. I want to talk AI. You want to start with AI just because that's probably the buzziest stuff and then we'll flip to corporate governance.
Yeah, let's do it, man. I mean, as you know, AI is everywhere. AI stands for artificial intelligence, and it's everywhere all the time. I'm just memeing, but yeah, it's pretty hot, dude. You want to hear my bubble pitch, honestly, because it feels...
Okay, this is something that I feel like every day I feel like I have 20/20 vision and I feel like I'm lucid and I'm like, this can't possibly be happening, yet it continues to happen. This capital cycle is just incredible and I really do believe that if we look at, because I've been, and after I read the railroad bubble books maybe I'll have a different take on this, but the one thing I read a lot of telecom bubble stuff when I wrote a really good telecom bubble thing I want to say like two years ago and I think the one difference between the telecom bubble and this is how much debt is involved, which until very recently is like effectively zero. So Oracle kind of showed us what's possible. There's a lot of debt available and whoever gets the next job at the Fed is the guy with the lowest number. US Treasuries are only a 5% discount to the Microsoft 10-year. And so you kind of have all these things. Oh, then there's like even more stuff. I saw today that the government wants to not have as much capital requirement growth for whatever the risk capital and you're just like, wow, everything is kind of coming together. OBBB is effectively an incentive to invest more today. There's just so many things. So because you have the double depreciation, every single possible incentive for you to invest in a capital-heavy GPU data center today is happening today and everyone wants to do it. So here we are.
Doug, I'm so glad you started with that because the first note I had is you wrote after the, as you said, after the Oracle deal was announced, you wrote the bubble is starting and exactly what you said. You said look, everything to date, even though everybody says bubble, bubble, bubble, everything has been out of cash flow, right? Facebook, Microsoft, Google, all these guys. Yes, they're investing tons, but they're doing it out of operating cash flow. Oracle was the first person who said, 'We're going to go, we're going to take out debt. We're kind of going to bet the firm, right, by doing this.' And once one person does it, guess what? It ain't stopping there. Everybody's going pedal to the metal. So I guess, and then as you said, so we've got the people are starting to use debt instead of equity. That can blow the bubble up huge. I mean, you and I were laughing before, President Trump literally, if you're an executive and you go in, the first thing he's going to say is how much are you investing into data centers in America? Right? He asked a CEO and nobody was like, 'Uh, we don't do data centers.' So, I just feel like it's all speed ahead. Everyone's just pushing this to a big bubble and everyone can party. Well, yes, there might be a big hangover as you have right now. There might be a big hangover in three years, but it just feels like we've got a long way to go.
Yeah. It kind of terrifies me that that's the conversation starter, but like there's just no way that you can't have a, I mean, it's like you can't go bankrupt on a debit card, right? And what we've been doing up until now has been all debit card financing, meaning it's like cash in the bank. And...
Maybe securitize Nvidia GPUs if you're CoreWeave or something.
Yeah, maybe securitize, but even then, CoreWeave never built on spec, meaning they only build if they have a contract. So they would go out, they would win a contract, then find the financing. And with the DDTL loan, which is like the delayed draw down term...
Yeah. Specifically focused on the GPU financing side, that was only done, it's like a flexible revolver focused on GPUs after you win. So it's still, you know, it's different than what would really be kind of crazy, meaning everyone goes and builds on spec. And so we're still kind of in this like early, relatively early stages that everyone has been doing this on the beautifulness of how much cash they've raised from the most valuable businesses in the history of capitalism. Dude, all the hyperscalers are still cash flow positive and I think the thing that may be slightly underestimated or underappreciated, and this is something where you see this and you're kind of like going crazy, for example, you're like, oh, private price-to-earnings is actually not that high compared to history because it's fine, they're all like mid-20s but they're growing 20-30% EBIT and you're like, well, this doesn't seem that crazy. But maybe the one difference that I think is maybe underappreciated is price-to-earnings is a really, really good metric when everything is super capital-light, but right when something goes from capital-light to capital-heavy, price-to-free-cash-flow blows out. And so you have these price-to-earnings, like it's pretty hard for price-to-earnings to go up from, or sorry, to really expand, or rather the earnings don't flow in mechanically because you have these ginormous cash outflows that, like, you know, and that will be, sorry...
Let me push back on that because I think a bull would say, hey, Doug's right, right, this huge AI supercluster that Google's building is a lot more capex-heavy than Google Search, the best business in the history of the world. But I think a bull would push back and say, 'Hey, right now they're taking all the losses upfront for these giant AI spends.' So yes, in the future, like cash flow isn't as good, but they're bearing the losses now. So when you say 20 or 25 times for Google, that's probably 10x for Google Search plus, you know, minus 10 for all the Google moonshot bets plus minus five for the AI, I think would be my big pushback there.
Okay, so that's probably fair. Hey, the cash flow, you know, you cash burn today for reward tomorrow. And that works when we know supply demand. There's more demand than supply today. Like feel very empathetically, yes, on that. And so, yeah, I agree with that. Like, hey, you're like, oh, they're under-earning and they'll over-earn or they'll earn their right share tomorrow. That's what accounting is for. But I just think it's really crazy that it happens all at the same time and the capital intensity shift happens right as double depreciation kicks in. So you effectively get to expense this stuff. So it looks, you know, so it's just this weird dynamic that everyone's price-to-earnings are really low, right? When everyone's business model changes into a more capital-intensive business and you could be like, well, what's the big deal, dude? Capital-intensive businesses tend to trade at low multiples of earnings. Like that's just how it works.
As someone who generally buys capital-intensive businesses, I know, I know you're really where it hurts. [laughter]
Yeah. So I mean, the whole thing is really interesting and if we're talking about like, well, and actually while we're here because I'm going to just, you know, I'm going to do some takes on the Google side. The thing that I think maybe, because you definitely don't follow this space probably as deeply as we do. The one that is definitely hitting the newswires a lot more right now is this recent Anthropic TPU deal. And we have been, SemiAnalysis called this out in like, I don't even know, I want to say July or something like that. We were really early on that and we were very excited on the TPU opportunity because it's the second best chip of all time. And so if you want to do the real bull case on Google, it's, you know, a trillion for the TPU, you know, whatever GCP accelerates and that's like AWS, whammo, and search is free, bro. That's the real bull case. I'm just kidding, but like I really think, yeah, I think...
TPU, I mean again, I'm a mouth-breather, but you're saying the TPU has the possibility to be competitive with Nvidia is basically what you're saying.
It is definitely competitive with Nvidia. It is number two. It is number two. There's no other...
Number two does not necessarily mean competitive right there. Bing is number two to Google. It is not competitive to Google. Is this Bing or is this, you know, Android and Apple?
Okay, so let's, it's time for some AI rumors. Okay. The best pre-training team in the world is a TPU. Everyone says that and one of the reasons why is because TPU is a much more stable, a stable version and scalable version for pre-training specifically. So there's a lot of advantages and the tech tree that the TPU has gone down has a lot of resilience and advantage for like software that makes it very, very attractive at scale. And I'm going to try to do this justice, but pretty much the NVL72, which is 72 GPUs all tied together, has a really big blast radius problem that's like kind of complicated and it doesn't really work. Meaning, if one of the GPUs fails, they're like, 'What the hell do I do with the other 71?' 71 of them will get stopped or paused. The 3D Torus, which is like, you know, you're going to have to freaking Google this, but it's very complicated. Like it's a 3D cube where all the...
You just gave the bull case for Google, right? You said Google this while you were pitching the Google TPU.
Yeah, there you go. Sorry. ChatGPT this. You know, it's an all-to-all mesh that you can make slices of TPUs and so you can scale this very, very big. And so they're like kind of two different philosophies on how to scale into really, really, really large clusters, but they are both very competing and valuable philosophies. And I think TPU could be very price competitive. It is just, it's straight up a margin question. I would argue that at some price I would definitely take TPU over NVLink if all things being equal with software. And it's actually a product that runs inference production at scale, training inference production at scale today. Like it's a real competitor. You can argue it's, you know, it's number two and I think at the right price it would be number one for me. Midjourney, for example, allegedly uses all TPUs. So there are companies at scale today who are stoked to use TPUs and I think that's going to be a big and new interesting competitive vector that I think people don't appreciate or haven't been appreciating because, yeah, Google's been, I mean, I've tweeted about this for years. Google's go-to-market, it's like, you know, you don't want to bet on those PMs because they get killed by google.com. It's like one of my favorite websites. But I think something has changed in the last year because in the beginning they really were like, 'Oh, we're going to have the best AI. It's going to be done on the best infrastructure that's ours and you're going to only be able to use it in our Google properties online.' And I think they've been willing to open the aperture by being willing to do Gemini on Siri, willing to sell TPU externally. They're willing to open it up so that they can have more customers. And then that means they can have a real chance of being a true number two.
Killed by Google. It's one thing to kill Google Friends or Google Meet or whatever when you know it would be a nice business but it's not. It's another thing, like AI is existential for them. Every big tech company has clearly realized that and I'm sure they had a vision last year but at some point you start to realize, hey, we can't do this all internally, like it's too much of a scale game. It's too much of a moat. I will be honest, that was super interesting but we are like at the far limits of my technical knowledge. So even though if I was a better interviewer and better at this I would dive into so many more questions, but I'm going to back up to my dumb-dumb brain and ask you something else. For the past year, if I would talk to, you know, I'd be more value investor than SemiAnalysis growth investor. When I would talk to a value investor friend and they would say for the past year, AI bubble, AI bubble, AI bubble, they would all point to the models are stalling out, right? Inference is selling out, the models are stalling out and everything. I don't know where we are in that, but I just want to ask you, if I got an AI bear on here and the first thing they would say is the models are selling out, they're not improving, we're hitting the limits of scaling. What would you say to that question?
Okay, so that's a great question. It's pretty hard to answer. So in December specifically, 'scaling walls are done on the pre-training side' was probably pretty valid. I think there's a lot of hype right now in the Gemini 3 is supposed to coming. I was told Gemini 3, which is the newest and best model, it's going to be the newest and biggest model in existence and that I think comes out sometime this month. So that's the big proof is in the pudding. A lot of people are very excited about it. We'll kind of see, but there's been multiple scaling vectors of how these models are getting better and I think we become a lot less focused on AGI, which is going to, you know, superhuman intelligence that's going to zap your little brain because you can't even understand, to almost like essentially all white-collar information knowledge is free or the marginal cost goes to your GPU. And that's really the focus of where things are going right now. Specifically, what the entire industry is very excited on is RL. RL means reinforcement learning. And so in the same way you and I learn how to do stocks good or bad on a very low-end size data set of like, you know, our history, career, whatever, we're like, dude, this setup really works for me and I do a really good job. Pretty much the goal is that you do training on these models too to be like, 'Hey, this setup does,' or like your job is to check out, you know, check out this book from Amazon in the least amount of clicks. And so, you know, in the beginning it literally clicks everything. It does all the kind of crap and it doesn't, but you can kind of just say, 'Hey, no, bad job, bad job, bad job. Good job, bad.' And then all of a sudden it gets to the point where it can click on all the things and check out your book really, really easy. And anything that is verifiable, meaning that there is an outcome to be desired and you can effectively guide it along a path to figure out the answer, is probably a solvable problem. And you're like, well, that works for an Amazon book, you know, or something like that. And Amazon already is blocking all the agentic stuff, by the way. That's like going to be a war, I'm sure. But you can say that. The reason why people really liked OpenAI to begin with was the original RL thing they did using reinforcement learning was making these superhuman Dota players. So like the...
I was going to use that as an example. Yeah. Yeah.
Yeah. And so these superhuman Dota players are better than humans. And you're like, 'Okay, well, you know, they won't be able to be like in these complex things.' Like, that's a pretty, I mean, I've played enough Dota to be like, 'This is really complex, and I'm going to be bad at it, and I know I'll never be better at it than these bots.' And so that's probably enough. That's probably enough. Meaning, you use RL to kick the [__] out of a lot of like verifiable domains and tasks and you just scale it to the moon. And that's the plan going forward. And you'll see these like multiple, there's like multiple roads of progress. Pre-training is slowing down but it will get a little bit better. Test-time compute is probably not slowing down but no one really wants to spend the compute budget and so everyone's like, 'Give us more compute budget.' And then all this post-training and RL stuff is definitely not slowing down and that's where people are excited. And so what happens is this new vector kind of hits an asymptote and then all the stacked vectors get better in aggregate. It definitely is happening faster, meaning the returns to scale are slowing down a lot quicker than like AGI cases. But I think that you look at the history of improvements in technology, it's pretty damn fast, dude. I mean...
I agree with almost everything you said. Let's just say that. Almost everything. Everything. Whatever. I do have one question. You said the returns to investment are slowing. How are they measuring returns to investment? Right? Because I would even hear an OpenAI guy talk about the returns on investment stuff and I'd be like, how the fudge are you measuring this because you don't have any, it's not like they have notable revenue or anything, right? So I'm sure that they are getting better and you'll hear all these guys talking about their returns on investment in AI. The only one I really believe has any is Facebook and maybe Google because I know Facebook's using it for a lot of internal training and targeting stuff, but I hear returns on spending coming down but still very positive. I'm like, I don't know how they're measuring it. So yeah, what else?
This is, okay, so this is actually a really good question and something I got to pump my stuff. Inferencemax.ai, we have this thing. It tells you how many tokens you can get and how much cost.
If I can pause you, I mean, people don't understand. We've got Doug on. If you were a pod shop trying to have this conversation with Doug, this podcast is free. It would not be free. It would be a lot more than not free. So this is like highly valuable info we're getting here.
Okay. So yeah, so inferencemax.ai kind of talks about how much tokens at how much cost you can make. And that assumes it's paid. So, inferencemax.ai, we made this benchmark. It's really cool. Shows AMD versus Nvidia. Kind of shows how AMD is catching up. Blah, blah, blah, blah, blah. We're not going to go into that, but the takeaway is if you could sell the tokens that's happening on one GB200 in a year, and Nvidia took this and they turned around this in their marketing material, one GB200 which costs like, let's say, $3.5 million makes you like, you know, $5 million a year. So it's like well over one-year payback period. The raw unit economics of this stuff is extremely high, like insanely, incredibly high. The problem is how much are you able to give it to actually get paid the entire stack. That's the real issue.
Let me push back in one way there. So, I hear you because you're renting out. You take the Nvidia GPU, you buy $3.5 million, you rent it out to, well, CoreWeave would be the one renting out, but they're renting it out to some AI startup, right? That's how they're going to generate their return. I think where people would say, you know, all the classic value investor things where they've got the surge protector plugging into itself. What they would say is, hey, they're renting it out to an AI startup that has no revenue. So how is that AI startup generating a return on investment? Because that is where you kind of get into the dot-com bubble, like it's all just a house of cards, circular reference, blow up on itself.
Okay. So, yeah, let's talk about it because I think in aggregate a token that is sold is profitable in the aggregate by a meaningful amount because TSMC makes a 50, makes a 60% margin. Nvidia makes a 75% margin. A neocloud makes a 40% margin just buying those and renting it out. And then all of a sudden OpenAI makes negative like what, 50% margin. So I think if you aggregate it, it actually is, there is like, you know, whatever the token sold is actually pretty profitable. And I think specifically to OpenAI, like you're probably right, they can't afford it, but on a unit, so how you justify your spending all this stuff, you got to improve the technology, the capabilities is like no end, but on a unit economic basis, specifically going back to that rack example, if you were just to buy a rack and imprint the models and sell it to people, even on a subscription service basis, it's pretty, it's pretty [__] profitable. Like the, like I think, you know, assuming that it's a DeepSeek model, like you can kind of do the entire global infrastructure of OpenAI, all the free guys for like, I don't know, man, probably like a few billion, a few billion, 10 billion. So, let's just say you're supporting a billion active monthly active users for 10 billion bucks and you're telling me you can't make 10 billion bucks off a billion users. I feel like that's a skill issue. They'll be able to monetize that. That's not like a crazy, crazy number to me.
Is OpenAI, and I ask this from personal experience because I'm actually going to do a post at some point, but a lot of apps that I would pay the app store $10, $50, $100 per year for, you know, tracking apps, whether it's food tracking, workout tracking, other stuff, I just pop it all into OpenAI for free. Is AI insanely deflationary?
Yes. Yes, I think it is. So, let's actually, this is something I've had this whole debate for a long time. I think AI is insanely, ridiculously, stupidly deflationary with the exception of if you like build...
If you're trying to power your lights.
Yeah. Yeah. Well, yeah. But you have to keep paying for energy. I think the next leg that makes everything a lot more healthy is if you sell things as services and that's why we've been very bullish on like this agentic purchasing thing. So my favorite, you know, ChatGPT-5 setup for monetization is like a post we did on SemiAnalysis, pure dude, pure schizo posting, straight up. Fiji works at Instacart, gets ChatGPT Instacart to work, she leaves Instacart to go work at ChatGPT and she's like, 'Uh, we are definitely going to be monetizing this stuff soon.' And so that is an example where I think you can massively have a very profitable business that is done on a service basis that isn't deflationary and has high margin, meaning that like, hey, we're pretty programmed to pay take rates if it's convenient. So, I want to go buy, you know, my entire, you know, make me a keto weekly shopping list and purchase it on Instacart, right? 150 bucks or something like that. Well, we say we'll just do 100 bucks. So, it's like a raw number and it's like, okay, well, you know, Instacart will take some platform fees. They'll share it with, you know, let's say 10%, they'll share some with ChatGPT because it does increase all this stuff. Hey, I want to have a, you know, a vision checkup and it does like all that top-of-funnel stuff where you just take a vig on it effectively. It's like, dude, that I think is a very valuable thing and you could be massively deflationary, but it's a service and it doesn't like, you know, you don't deflate the entire economy away because that's the real concern, man. Like we have these things that are, it's kind of terrifying, but this same thing has happened over and over again in the history of time. Hey, railroads were an order of magnitude better than canals.
The classic, right? The queen didn't want to bring sewing machines in because what happens to the poor sewers? Well, guess what? You get with the times or you get run over.
Yeah. Yeah. So it's going to get run over. It's going to be really deflationary and then on the other side of it at some point in time it blows up and there's like all kinds of new stuff. You know, I hope it's like not, we're not like just all talking to our AI girlfriends and there's something more, less to play, you know, more positive than just...
My eight-month pregnant wife. She knows I use AI a lot and she had a friend who actually had a friend who got like really into it and she was like, 'Please tell me you're not using AI as a girlfriend.' I'm like, 'Girl, have you met me?' But I can understand the concerns. I have two more questions on AI and then let's hop into some of the corporate governance that I think you and I both have a special place in our heart for.
First question. Actually, I'm going to skip the power stuff. I will just tell you guys, I'll ask one power question just so I can give this anecdote. A year ago, you and I probably talked once a quarter, but you can tell how highly I think of you because I remember the Oracle thing and I remember this conversation. I was doing a lot of work on the Bitcoin miner data center power play and you were doing it separately from me and I called you up and I was like, 'Hey man, like I feel like it's got a long way to run, but man, these things have really run and I think people don't understand like how much worse the Bitcoin miner assets are than like normal data centers. How much room is there to run?' And you were just like, 'Unlimited demand, buy them all to the moon.' And APLD is about an eight-bagger since then. I think that's the one we were really talking about. You can pick your share of anyone, but I'll just, I want to give that anecdote. I'll ask one question. How much more room can this power trade possibly have to run?
This is something, you know, we had a DM before we started this talking about like where I think things are maybe off-sides in the markets. My belief is power is much more off-sides than semiconductors because semiconductors are really scalable. Actually, TSMC is really scalable. They can add a ton, like they can add a ton more capital equipment and make probably double the chips. Taiwan can do that. Taiwan can make double the chips in like two years. I 100% believe that. We cannot double the power in two years. Just like straight up. I don't think people appreciate or understand that at all. And I think that's where the biggest disconnect comes from when I talk to investors, when I talk to people, is that like all the chips guys are like, 'Yeah, the orders are good. You know, Nvidia is going to crush, blah, blah, blah, blah, blah. Taiwan's going to figure it out. There's a lot of technological trees.' And they're like, 'Yeah, I think we can get to a megawatt per rack.' And you talk to a data center guy and they're like, 'Are you [__] kidding me? A megawatt per rack? Like, this isn't real.' And you're like, 'I don't know what else to tell you, man. Everyone else has bought in except for a data center guy.' And so there's a lot of really fast-moving data center operators who are doing very, very, very well in the space. I don't know if you have met Switch, like the Rob Roy guy, he's a billionaire. He was like, I actually remember this IPO very vividly from when I sat at my buy-side shop at like B. Riley Capital. I remember reading this and like, this guy's a crackhead and he was like, 'We're going to make high-density racks. That's the future.' And I was like, 'I guess.' And he has like always been about high-density racks. He's been crushing it because the second he heard about this he's like, 'Higher density. In fact, I'm going to three megawatts.' Like...
We're taking the dials to 11.
Exactly. He's like literally, 'I'm at an eight and you tell me I can go to 11.' He's going straight there. And so I think like the DLRs of the world are like still shell-shocked that we're at 120 kilowatts and then the really fast-moving players like let's say Vantage or Crusoe, they're all being like, 'How do we get to 500 kilowatts per rack?' And that to me, like 500 kilowatts per rack is kind of a crazy number. Like, I mean, you just start to, like my mental math here, six gigawatts is New York City. And so everyone is saying, and all the chips that are being purchased will support all this power. Like straight up, we know the power that it should support and the data center, the power side is just not moving fast enough. There's a lot of ways you can wring efficiency from the grid because the grid is focused on, you know, when we talk about peak versus trough and utilization is going up and that's like batteries, backup power, pretty much shaving the hottest days out of the year or the coldest days out of the year using peers. That is all working, but I still think even with all that, we are still talking about a lot of demand.
Just add P times Q equals more power from here. And I think, oh, the last thing I want to say specifically on this on the stock-wise. You know, the power, the miners probably still, like every weird little miner that hasn't run probably still works from here to whatever. We'll say IREN being the craziest or Applied and IREN being the...
crazier ones. But the other one too is like the incrementals on industrials is what really interests me the most. Probably one of my favorite companies, Comfort Systems. You can look at their print yesterday. Their revenue goes up 40%, and their EBIT goes up 80%, and their EPS goes up like 150%. Maybe not 150, 50%, but like all of these businesses, they're growing the fastest in their highest margin segment and they often are massively capacity constrained. And what that does for EPS is like relatively modest, and what I think it does from here, like you know, on the power and accelerates. And so you're like, dude, there's accelerating on these businesses with these crazy incremental EPS stories, and so how you underwrite it is very, very hard. Yeah.
