Ep. 022 - Market Drawdown, Historic Bubbles, Funding The Buildout, AI Politics (Doug is Back)

Welcome to the new Semianalysis uniform. This is what we wear every day in the New York City office. I come in, I clock in, I put on the Jensen Rap shirt. They're just apparently they're sold out or sold out everywhere in Taiwan, but it's just like, come on, you gotta admit this goes hard.

Amazing. Shout out Timothy. Thank you, buddy. Hello everyone, welcome back to Semi Analysis Weekly. I am joined this week by Doug O'Laughlin, who for some reason hasn't been on the podcast for a little while. Doug, what's going on, man?

I thought I wasn't allowed to come on anymore, but apparently that was, uh, that was never the case. I don't know, I just been— I've been doing a lot of things here at Semi Analysis, um, and it's just been, you know, it's been a crazy season. And, uh, yeah, so Doug has been busy, but he's back.

Yeah, today we're going to talk a little bit about the drawdown that's going on in stocks right now. Memory, optics, lots AI names are down. We're gonna, um, either fan the flames or provide people with some, uh, soothing positivity. TBD on that one. And we're gonna basically juxtapose that, uh, drawdown with some of what's going on in the labs, like Sam Altman warning staff that their new pre-trained Doug is going to be RSI in 6 months. Um, I'm just continuing to build a ton of models. I mean, Anthropic as well, rumors of, uh, new versions of their models coming out. We saw Opus 5 get released, GPT-3. Get Doug's reaction to that and some moves around the industry with people leaving. So, okay, let's, uh, start with a lay of the land, man. Um, stocks, drawdown, like, what's going on?

So let's just talk about— so I mean, okay, I've been doing this spiel just a wee bit. So pretty much up until very recently, like let's say by June 30th at the end of Q2, it was pretty much the best performance in semiconductor history. And we've had a little bit of an unwind since then. And what you can say is a lot of it you can argue is quote unquote technical, meaning there's like reasons because of factors and people maybe overleveraged and some stuff like that. But the reality is it's like what goes up really, really fast often has a little bit of gravity and we're just kind of paying for the massive almost crazy momentum rally we've had, but it's gone a little sloppy and it's been very interesting. The tape recently, um, South Korea is pretty fucking crazy. Every single day the stock markets essentially hit limit, like they literally limit. There's a tweet right now, um, where it's like, how do I even do my job? The head of HR lost everything and everyone's super depressed and like because their stocks are all down. Um, and it's, it's kind of just kind of crazy. Um, if you look back in the history of Asian financial markets, this ironically seems to happen a lot more often than you think. Like, there's a— like, probably one of my favorite books of all time is The Great Taiwanese Stock Bubble. I think Taiwan on a per capita basis had like a 100x bubble, like just, just incredible numbers. Like, just everything went to like 1,000 times. Like, you know, you're talking like a bank trading 500 times price earnings, like just crazy stuff everywhere.

When was this?

I was in the— it was in the late '80s.

Okay. So are you comparing the current situation in Korea to the late '80s in Taiwan, or vibes are similar?

Vibes are similar. The bubble in Taiwan is truly historic. Like, I just don't think anything— I don't know if— okay, just put it this way. If we were to hit that bubble in Korea today, I think Korea's stock market would be like worth 15 to 20 times more. Like, it was just— it's such a crazy bubble. I don't think we should ever be able to possibly recreate that. Like, it just, it just shouldn't happen. Like, if they can go off, I guess. But I do think some of the lessons learned and some of the behaviors that happened during that bubble are often indicative of this one. Yeah, I'm not going to call it a bubble or whatever, but like maybe this crazy run-off. What happens, you have these like blow-off tops that get absolutely crushed. And at the absolute top, everyone's buying as much as they can. They go super levered. They're buying 2x ETFs. People are like literally, you know, getting second mortgages on their house to buy more stocks. And this is actually what happened in Korea. And Koreans particularly, for some reason, have this like crazy hit rate of always buying the top of every— like they buy the top of every cycle. They were buying banks in 2007. I'm not joking. They're buying CNBS in 2007. They're buying SaaS companies in 2021. They have like, they have this like 20-year track record of always buying the top and Korea just went ultra mega YOLO into itself. I think it's a—

can you clarify a little bit about what's going on here? Like, you think the fundamentals are different in this case?

Like, so on the fundamental side, I think they're good. But I think the problem is it's, you know, things are never as bad as feared and they're never as good as you think they will be. And I think what happened— SK Hynix missed earnings today, missed consensus. And the reason why is because they shifted more to LTAs. It's ironic because they were in the United States and when they were doing the ADR, they were kind of dumping on Micron for doing LTAs and effectively getting a lower price. But one of the things that is happening in this memory cycle is like, you know, memory prices, say DRAM and even NAND, right, has probably 3x'd over last year. And so next year they're not going to 3x again. They're going to go up 30 to 50% or something like that. But, um, finance brains are just absolutely broken. It's all about rate of change. And historically, in the memory cycle, when you have that rate of change go— when the second derivative goes down, it's usually the end. Because it doesn't just go up to a 30% price increase, it usually goes negative 50% price increases. So what happens is they invest all this capital, all the bits come online, and then once upon a time they're like, dude, these factories, we can't even run them hot enough because we're running out of chips, there's so much demand. And then they build another factory and they're like, dude, why, why was there so much demand then? There's so little demand now. They were doubling, or there people people were double, triple ordering in order to get, uh, their orders in. And so this is like a typical tail whip, a bullwhip in, in a cyclical thing. But I do— we do think that this cycle is bigger and longer and stronger than past. So I think what's happening is people are looking out, and the price increases that the memory companies are showing in, in terms of like future price increases, because the LTAs are lower and more conservative than what the market hoped for, and they were like euphoric, they were, you know, way over their skis at the top, they're levered up. And so pretty much it's not as good as your dream in your mind could be, and you're levered up 3x, and the stocks are against you, now they're down 2 or 3 times, right? If you're 2x levered and your stocks go down 50%, you don't have any more stocks, right? And so many people were that levered.

You're wiped out.

And yeah, and like, so the entire cost fee right now is down 40%. So if you're over 2x levered, you're completely like, you're completely wiped out. And so we're seeing just crazy margin calls and then that kind of like, that kind of becomes a self-fulfilling thing where everyone who has a lot of money on the table looks at it and says, damn, this pile of money is shrinking, maybe I should just sell. And then, then that like exacerbates it. And part of this too is like, dude, we have like the run-up to this was like truly historic. And I don't know, being cautious in any way during that period of time, everyone would like pretty much laughed at you. And now, I don't know, now people are going to probably freak out and they'll overshoot on the downside. This is, this is what markets are all about.

Yeah. Do you think there's some smart money on the sidelines right now that is still looking at this and going, demand is strong, these LTAs are going to continue, these are really healthy businesses that are completely sold out for the next number of years, and it's going to take a little while to bring more production online anyway?

I think so. But I think the reality is now there's going to be this overhang of, okay, well, price increases are not going to go up as much as they did. And then that multiple that they capitalize it is going to be lower. So I'm not saying it's over. I don't think it's over. But I do think when you have that big boom bust, it usually takes a little bit of time to retrace out. That or we're just going straight back. But like, like, okay, sorry, actually, let's bring this all the way back to the Taiwanese stock bubble. I was up late last night and I was literally looking at stock charts of the Great Taiwanese stock bubble. And during the Great Taiwanese stock bubble, you had 20— yeah, as you do as a normal person, you had two 40% retracements. So it's not unheard of having this giant retracement and then it continues to go higher. But, um, it's— I mean, it's pretty traumatic and there's a lot of pain right now.

So, well, okay, let's talk about the, the people who are, um, uh, calling top and selling and trying to short right now. Are they have different approaches to, you know, calling BS on all of the demand? I think we can go through a few of them. One that was hot in the news recently was the Intru introduction of Chinese memory to the ecosystem. CXMT, YMTC, obviously big IPO. What's your take on Chinese memory coming into specifically the memory makers industry right now?

So I think on the Chinese side, historically, everything they touch gets dumped. And that's why people are scared, because they have an ability to ramp a lot of capacity. And even if it's at a worse yield, It's not like China is trying to make a 10% gross— or it's not like China is trying to make a 50% gross margin. They usually historically make 10% gross margins. And so they're more than happy to dump to win market share. You know, the correct framing for how Chinese companies like compete against each other is really it's provinces competing against each other for GDP output. And they're not actually competing against each other for, you know, profit margins or EPS or shareholders. Right. The shareholder is the government. The government incentivizes production and each province kind of competes against each other. So CXMT is incentivized to produce. But I do think the CXMT LTAs are a little bit like, I don't know, I think they're clearly number 4 in the market and it's a shortage. So they're going to make a lot of money in the meantime. And Apple, for example, is down to use CXMT memory because Micron, they're accusing Micron of price gouging. Right. And so No crying in the casino, Tim Apple. Uh, well, I guess he's retired. No crying in the casino, Apple. The reality is you have to buy it at the market price. Um, and so I think CXMT might ruin the party, but the reality is there's still just more demand than supply. And the real question here is, how do we know the demand is going to be strong? Because like, okay, now I'm just going to go on a rant. Okay, so demand and supply, they are always yearning toward each other for the crossover of price. But in a new market like this, they're blindly searching. We don't know— we do know the supply curve. The supply curve is relatively easy to understand. The demand curve, we actually don't know, right? We know that coding agents and stuff like that means that there's a lot more demand. We know that chatbots and the value of like knowledge work means that there's more demand. But we don't know if it's 10x more demand, like, you know, maybe DoorCash is like, uh, it's like 100x more demand, or if it's, uh, 50% more demand. And also supply is getting better the entire time. So supply is just going to ramp. It's going to ramp blindly into this curve no matter what until one day it meets demand. And the question is, when you're something super in demand like it is right now, you're sitting there and you're like, okay, I want to make a data center today. I want a gigawatt cluster. If I make a gigawatt cluster, I'm going to land Anthropic tomorrow. I need my shit to come on time. What am I going to do? I'm going to double order my equipment. I'm going to triple order my memory because, hey, if I get memory and it's a little bit late, I could just sell it to someone else. It's a shortage everywhere. So during semiconductor cycles on the upside, everyone's double ordering. And so the factory looks at it and there's like, oh my God, there's so much demand. And they almost always are overbuilding for a level of demand that doesn't come. And then that's— and then historically what actually happens is there's usually some kind of economic wobble, uh, like a financial crisis or the Fed freaks out and then everyone just kind of flips out and then demand weakens for like a quarter. Supply ramps because, you know, you can't stop a factory from coming online. These things take years. So your supply is going to ramp this entire time, and then demand sneezes and you have a factory that isn't— that's like, you know, your factory would go from 100% utilization to 50% utilization. And the only way you can make money back on your factory is like you want utilization up, so you just, you just cut price. And that is essentially the semiconductor market in a nutshell. And so the real question that's harder than anything else is where is demand? 'Cause, you know, if supply is like, you know, supply can ramp 2x or whatever over 2 years and we— Semi Analysis exclusively focuses on that. The real question is where is demand? Like that's the, you know, the trillion dollar question, right? And we think it's strong.

So yeah, I mean, I have my opinions on that. Like in my view, I think demand is pretty obviously very strong for a very long period of time here. I just look at my own internal usage. I look at how others are using it. And I think to me, if you're calling a future where demand stays flat, does not increase that much, or starts to decrease, you really have to believe that the models are not going to be getting better in the future. And I see zero signs of this. Like, I only see signs of the opposite. And so it's, I don't know, it seems hard to imagine a future where next year or the year after demand is suddenly caught up to by the supply curve as you're describing. Maybe.

So I'm going to— can I do— can I do some devil's advocate here? Because I do agree with you as a guy who now spends a crazy amount of tokens a week or whatever, and I'm like probably hooked on tokens. The two things that people will say is that the technology gets better at a faster rate than people can use it. So for example, let's say the real killer application for, for AI happens to be data entry. And Kimi K3 is good enough. And effectively, we make faster and faster and faster cars and better and better and better products, but the real demand curve that matters gets saturated by a product that we already mastered. You can argue this was the internet, right? Um, the reason why the internet's like, you know, the internet bubble, they, you know, they're like, oh my God, where the demand is doubling every 90 days was like one of the common refrains. It wasn't. And then also, uh, the technology behind making the lanes faster got literally got 2x or 3x better every single year. And so then all of a sudden, after everything happened, one strand of fiber became 500,000 times more performant. And then they're like, wait, I don't think we actually need, we can't actually fill this fiber demand. So that would be like the pushback, right? Whereas like, okay, the models get 100 times smarter and 10 times cheaper. And that level of intelligence saturates the big market. I think that that's probably the biggest bear case I could think of. And I think I'm really wondering if that's probably true for some parts of the economy, right? Like we talk about this, right? Like you don't really need, you know, you're like, send her my div, you know, like does Kimi need to, you know, does Fable need to do that or can Sonnet do that, right? Yeah.

Order my burrito and stuff. No, I think absolutely not. But to me, there's still 100 to 1,000 times more people who are currently not using any of the, uh, currently good enough models that will learn to use them over time. That's one. Number two is I just see a whole variety of other use cases. It's like internet was connecting people to the internet, but AI is coding. It's also chat and research. It's also video generation or image generation or drug discovery. Or material science. I mean, somebody just starts making superconducting elements with AI or doing all sorts of other material science work.

What's that?

Well, I think it's worth a lot of spend on GPUs, for example.

Yeah.

And I think that in many cases, the fact that we don't have 10 startups all working on weather prediction right now for farming or for leisure or for whatever is almost purely because Everybody's distracted by coding. And meanwhile, coding in and of itself, the, the claim that, uh, centering a div is gonna run out because everybody's gonna have a website for themselves. Fair enough. But coding also, uh, represents a whole bunch of other tasks that are so much more, uh, economically valuable intrinsically. Than centering a div. And you can pursue those in this. I mean, there's rumors of Sam Altman talking about RSI, right? And, you know, maybe—

I guess, okay.

Yeah.

I'm gonna push back, like, and I don't believe this. I'm more on your side of the camp. Okay. I'm just, okay. So the question is, will that still fulfill and satiate and all this, like, Will it push and do all this economic good work? And we build all these data centers. And plus also, like, I think the problem is like the chips get faster every year. We also get better at using the chips and then the models get faster and the models also get better. So you have like 4 rounds of deflation to just make really good products. And so all of a sudden you can look up after 3 years and you're like, wow, every person in the world has a, you know, an 8x B200 to do whatever, every single token, every single day. And then we're like, ah, we've run out.

Like, you walk through those 4 that you say are deflationary. Let me tell you the inflationary ones, right? The models get bigger every year. The models think more every year. More people use the models and more people use the models per day. Like, like each individual user uses the model. You can look at this as it's happened. Kimi just tripled in size and now it can only fit on a B300 or an MI355X single node. Um, can't run Kimi K3 on a hopper, right?

Um, yeah. Yeah. I was gonna, I was actually gonna say that. I was actually gonna talk about that. Yeah. I was actually, I, do you wanna talk about that? Cause that's a, that's like a spicy take. Is that like, we talked about the, um, oh, oh, sorry. Two things. One last thing, because I know we're like, whatever, we're bantering on the, on the demand thing. The thing that I think that changed my mind more than anything else truly though, was last year in coding agents in Claude 4.5. Because I don't know about you, if there's never been a clearer moment where you hit some level of intelligence on the curve and an entire new market showed up, you know what I'm talking about? Like, I could not do this shit before, and then the day after it came out, you could do it.

And you were the prototype, right? Like, like there was a time last— this time last year when the technical staff at SemiAnalysis, call it under 10 people, were using coding agents. And then sometime in November or December, You and Dylan said every single person at the company needs to learn how to use this thing. And now we have like 90 users of this. So that's 10x right there.

Yeah.

From 9 to 90.

Yeah. And like, I think the thing that's interesting is like some of the work gets better, blah, blah, blah. But anyways, smarter model, new capabilities. That's the, um, oh, sorry. That was the first part. The second part, do you use more tokens?

Sorry. From the time when you started to use it in November versus the time today, do you use more tokens than you used then?

Per user on average? Yes, on average, yes.

I think like 10 times more tokens than you were initially, right?

I'm unsure because that first week I literally did 14-hour shifts for a hot second. Like, my initial coding psychosis was pretty hard.

Like, that was subagents, bigger models. Maybe do it by spend, not actually—

yeah, you know, no, you know what, you know, by spend, 100%. If we do by spend, not even close, right? Because something like, okay, Well, do you remember the conversation in Claude Code? Remember when it first came out? Everyone remember that one random ass paper called Gastown was like, oh, I'm going to create this self-healing whatever agent harness that would do this crap. And everyone's like, this is crack cocaine, but I kind of vibe with it. I would argue we were kind of there, but you're like, you're seeing subagents, tool usage, you know, like agents, like, you know, agent, like, you know, multi-agent systems, like the fact that you fan out tools like this, like you're already, you're seeing the realized version of this. But I think back then it was impossible to market.

Thinking mode, right? They're, they're doing more. So in our experience, that was 10 times more users almost overnight. And then over a span of like 3, 4 months, people probably 10x their individual usage, leading to our company going 100x on AI spend, if not more.

Yes, yes. Um, now the question is, will every company do this? Probably not at our level, but I do think a lot of companies have a lot of of wood to chop.

Well, your argument is this will happen over time, but it will happen on the cheaper models because they will have budget constraints or something like that.

Yeah, yeah, something like that. And okay, well, actually, so sorry, okay, two things. Let's talk about the H100 thing because I think that's really interesting. We talk about these new models, it's very clear to me that the old chips will become worthless. You know, everyone's like, oh, the H100 is an appreciating asset.. But at some point in time, we're going to be, you know, it's going to take like 100 H100s to inference one of these models. And you're just like, dude, just let the old girl go. Just get a B300 or, you know, a GB200 instead. Right. That's my speculation. And we'll see. We'll obviously be writing in notes about it. But it's like the true confirmation of this trend is if there is a pricing divergence between B200 and B300.

Okay. I completely disagree with this, but, um, I'm interested in hearing you flesh it out. So, uh, maybe the most fundamental reason I disagree with this is because I don't know anybody who's ripping out H100s to replace them with B300, GB300, or Rubin, specifically because the data centers are so completely differently designed. So you would need to justify retiring a Hopper data center because there's no demand for the chips. Above the price to run them. The, the, the very cost, opex on energy and people to maintain the data center to justify, because you can't mostly for the bulk of the market, you can't replace a Hopper data center in the same footprint with Blackwell or Rubin without like completely ripping things out or just like knocking it down and building new, right? Yeah. So it's a very, by the way, That is a potential scenario if there's so much demand for the new stuff that people literally just like don't have power permits and land. And so they just like knock down an old data center.

So, so in a frictionless world, that is true. But we live in a world with friction and the friction is getting worse when it comes to new compute. Is that fair? So, so I agree with you.

That's the first thing. Now, the second thing is the, the, the, Okay. So the camp for GPU prices go down is model progress stalls, roughly. And GPU prices go up if model progress continues, meaning demand continues. That's the rough thing. Now there's one X factor that I don't think is being considered that I want your take on, which is if there is government intervention on the Frontier Labs. To me, if that happens, GPU prices go down. Especially the old stuff.

Explain why you think that.

Because they're going to restrict who can have access to the latest and greatest, and that's going to restrict how much demand there can be. And therefore, there's going to be more demand for alternative other stuff, um, that's going to cause people to use—

I feel like there's a lot of— too— what if— there's too many what-ifs on that one. I agree, sort of, but that's like such a top— like, okay, so I, in my heart of hearts, I'm a bottom-up I'm a bottoms-up guy, meaning that the world changes, you know, one mind at a time until it becomes mass consensus. You know, people choosing them at bottoms up, like, you know, product-led growth, right? Like this— I don't believe in CTOs. Like, I do believe— but see, you don't believe in the great man theory of history, Doug? We've had this— I'm not gonna have this conversation again. Come on. I've had— no, I'm not— I've had this one too many with too many people on Semi Analysis. I'm not gonna—

you're sitting in a Jensen world.

Sure, we could have a—

not believing in the great man theory of history.

He was just— he's just Adams pushed along by the greater Moloch. Yeah, right.

Anyways, right place, right time.

Yeah, a little bit. I mean, you know, I'm going to be pragmatic. I feel like there's many variables. You know, it's hard to say that the only variable was him, right? Like, I'm sure there are some great guys who are like, you know, I just think it's some— and in my experience, in my experience, Really great horses make the rider look better. Now, I will admit, Jensen's a hell of a rider and he had a hell of a horse. So, like, you know, it's very hard.

Anyway, this also doesn't make sense to me because you're implying that government intervention is a single person as opposed to a groundswell of minds, whereas I pretty fundamentally believe right now that a lot of people in the US really don't like AI. And I think that is not being priced into the—

that is not priced in. I agree. We— and how this will be priced in, I think, unfortunately, is the midterms. I'm, you know, so this is the vibe. So we get, we get a little vibe, like we can do a little bit of wind testing in about 2 months from now. And I think, so my guess is it is not a top 5 priority, or no, it's not a top 3 priority, but it is a top 5. Is that fair?

Health care for most of the candidates.

For most of the candidates.

Yeah.

Everyone is going to talk about it, but no one is going to be like— no one's going to platform on it. Does it make sense? And I think that that's where— okay, so usually what happens is if someone— if no one platforms on it, that's where corporate interests win.

Yeah. And corporate interests are currently stopping, for example, the RAISA bill in the Senate. Which, uh, you know, passed the House like 300 to 20, and yet is stuck in the Senate while people lobby against it. So yeah, for those who don't know, RASA, the Remote Access Security Act or something like that, that would restrict China from remotely accessing the Frontier GPUs, um, like an Nvidia GB300. Uh, they passed that bill in the House, it's stuck in the Senate right now. So Like, if you're saying if it's not a top 3 priority, then the lobbyists win over the will of the people, slash the interests of the candidates.

Yeah, pretty much.

Because dollars from 4th or 5th priority to number 3, where it gets some attention.

Oh, I don't fucking know, man. I mean, you tell me, man. I'm pretty interested because speaking of which, I think we literally just started this commissioning today. We're going to do survey work specifically focused on this problem because I think it's interesting. Like, I think unfortunately, I hope, I hope you get a a nice little survey in the mail and you fill it out.

I think a lot of— so here's my view.

I know surveys are broken.

Yeah. I mean, I haven't really thought this through, but to me, the— let's say that the concept of AI is not forefront in people's mind, but it's used as a scapegoat to excuse other things that people care about. Like you mentioned healthcare, like you mentioned people's financing or housing. A lot of energy, climate change, literally all the stuff that people actually do care about. I think there's a big chance that AI slash tech in general, or tech people who a lot of people have watched back Trump financially, will be scapegoated for other issues that people care about, like inflation and the economy, or like Yeah, 100%.

It is a whipping boy for what you actually care about. So I know it's important, so you have to attach it.

Yeah, I think I disagree with you that I think probably cost of living is the number one issue in a lot of people's minds, and that's what they're going to vote for.

Yeah, number one.

And that AI will be like a sub-agent component of—

sub-agent of the cost of living debate.

Yeah, yeah. It's not going to be like, okay, we should decide whether we sponsor AI and push it forward. It's going to be like, I care about the economy and you should blame tech bros and AI is probably going to be a big part of the midterms. I think like there, I think we'll see.

I mean, I think tech bros and AI. Okay, 100% agree. But I guess I'm just curious to see what is going to really resonate and stick because there's also like, you know, the horseshoe theory. You go super far on the right, you come to conservation, right? Like, it's like, oh, my ranch is being, you know, the EMF, you know, and the noise from the data center is killing my calves in the farm next door. So I'm pretty curious on this one. I'm gonna be honest with you, I don't have a strong enough view to be like, yeah, this is definitively what will happen. Um, usually I'm like a wait-and-see kind of guy, but we're doing a lot more work in this because— okay, so, so now I'm gonna bring up the last part. The last part of what, uh, what would actually kill it is a slowdown, midterms, whatever, stuff becomes illegal, we regress a little bit. But also, the thing I keep thinking about here is the problem is the future for all these technology booms always comes true, but it's the timing of cash flows that is the issue, right? You spend $1 trillion to get $100 billion, and it actually does become $1 trillion one day, but it comes 5 years later. And at that point in time, you're like, dude, I don't have enough money to keep this thing going. Does it make sense? Like, I think that that's going to be the real The real issue is that like, okay, we spent $2 trillion and all of a sudden, you know, AI rips, AI's at $300, $400. We'll say we send $5 trillion and AI's $500 billion in revenue.

You believe that OpenAI and Anthropic believes in the final pre-train coming soon because they're going to IPO and have the final funding round?

Yes. And then when that happens, the final pre-train, effectively, whatever the final, final pre-train happens, it is really good. It's the best it's ever been. It makes a ton of revenue. It grows very quickly, but it doesn't grow at a rate that's enough to pay the bills. So you built a house that you cannot pay for. And so you're paying for $5 trillion of investment on a $500 billion thing. And you're like, wait, wait, wait, that's 10 years of spending. You're saying that's revenue, not profit.

You're saying this in spite of your, I think, pretty deep understanding of these companies' financials and how profitable serving the existing models is right now.

Yes, but this only comes— so I don't think we're there yet. Like, this is like a chicken thing, right? Like, where it's like, okay, so the thing that matters is like how much. We don't have enough. It isn't enough. Like, we're still on the narrow path, I think, like where you can kind of blink and see where the revenue comes and it probably is good enough. To fit it. And you also have these hyperscalers who are super good for the money because they have these other businesses that gush cash and are like the most profitable in the history of time, whatever. But it just like kind of keeps hitting, you know, we talk about like blindly scaling the supply-demand wall, right? You know, like I think like, okay, legitimately we think it's like what we'll say 150, I'm gonna do round number 150 ARR for the entire ecosystem. Right? And we'll say $1 trillion of— $1 trillion of capex, we'll say, so far. Maybe it's not all on the ground, but $1 trillion of dollars have left the door. And there's like a lot of it working. So that's a 15% return on rev, but not on profit. So if we say it's 50% profit, that's an, you know, 7.5. That's not the end of the time. It's better than the rate of— maybe that's better than it cost for people to do, but that's not super profitable. So you have to believe that the $150 becomes $500, which is doable. And so that can probably get you to the $500, can get you to like whatever, $2, $3 trillion out the door. But then what happens is there's going to be a doubling where it just is really hard for it to double that quickly. And that's where I think the gap happens where you're like, okay, we just spent $5 trillion. Yeah.

Yeah. What do you think stops it from doubling? I think I've made the case that I think there would be the amount of demand from all the users and all of the different model types and whatever around the world.

There's also technology diffusion. You have to get your grandma to be vibe coding.

I'll give it a go. Um, I think she's up for it.

Grandma, I need you to make 12 agents. And so, like, so, so I guess incredible newsletter every month, man.

Yeah, she writes it to all of her, you know. Yeah, I, I can totally imagine the research being some AI stuff in there. Help with the design.

Yeah. So, so I think that that's probably the biggest mismatch between— and like, I don't have a strong view right now, I think. And I think we're nowhere near on the path where it's actually like, okay, because the thing is, the path narrows as you have more revenue and more capex. It's— it becomes a tighter path to walk. It is not a narrow path. There's a lot of cushion, I think, right now. There's a lot of companies make a shit ton of money, like Meta makes a crap ton of money. Yeah, their free cash flow goes negative. But if they, they wanted to, they could stop capex and then the profit would rip, right? That's okay, right? And so that path is not super narrow, but as you do more and more and more, you commit more and more and more, the stakes become higher and higher, and then the path becomes more narrow. And that's what I think would like— and then in that narrow path, you have to essentially demand people to use it. Like, people have to be using it like yesterday, right? And I think the problem is the decision makers and the people who actually adopt it are two very different worlds, right? You know, Zuck's sitting here and he's like, of course you're gonna have your Meta sunglasses and you have your Meta wearables. You're going to be in the— what you call it— in the metaverse. You're going to be using quadrillion tokens every single day. And then there's like a grandma in Nebraska who's like, honey, I don't really know how to get my new iPhone to work. And it's like you have to win every single consumer tomorrow. And I think the adoption curve is just— it takes time. And so I think the benefits, though, of the adoption curve so far is the internet is a really scalable distribution platform that really fits AI super well. And so most young people are pretty big adopters. Most, you know, most working age people are big adopters. Like I'm talking to people who are at funds who are in their 40s and 50s who use it every single day because they have to. It's a really good technology.

Yeah.

But the question is, will every single person use it? And are they going to all be token maxing? Because I think you have to get— you have to believe that.

Yeah. And I do. And I think we've both made the case on opposing sides here for a little bit. Maybe the more interesting question is on on the supply side, like the concept of being able to bring on enough GPUs or hire enough people in order to go and sell. Because it, like, to me, revenue probably doesn't necessarily— I mean, consumer is such a small part of the revenue of these coding agent companies in terms of ARR right now that I actually don't think it depends on the grandmas. I think it depends a lot on enterprise businesses and like defense agencies and intelligence agencies and like all sorts of other federal government agencies around the world actually adopting this stuff at scale. And, you know, I see a pathway to getting everybody in every bank and every telco and every retail company actually using this in their day job at work, as opposed to like having every consumer have a subscription to this sort of stuff. But maybe the more interesting thing is like, you're going to run out of gas on the ability to bring GPUs online, or to hire enough people in your like, B2C go-to-market sales motion or have them install enough GPUs in their private data centers to actually run this stuff so that you can get to the point where you can actually grow revenue to that point where you're saying, what, $100 billion ARR goes to $500 billion ARR in a year or 2 years?

Like, I mean, we're running it. We're, yeah, we're hitting, we're hitting like some physical, like, so some of the constraints that are not just like, you know, making the chips out of the factory, they're like relatively creative. We're running out of electricians in the United States. We're gonna have like a 100,000 person gap. Every electrician who is a Midjourney, like let's say a, I forget the name of it. It's like essentially like a, like a work, like a worksman or whatever. Like essentially a mid-level electrician is able to make $250,000 a year pretty easily. And if they really wanted to and they worked 18-hour days, I'm sure they can make $400,000, $500,000 a year. And these are people, you know, it's a trade. These guys are in demand, dog. Like, they're so— there's actually a really weird website that I go to. I'm not going to leak all of our alpha on this podcast, but there's a really good website that shows where it's essentially— it shows open jobs for electricians. And it's crazy because you can do a Wayback Machine and you can see the hourly rate go from like $15, $20 an hour to $50, $100, $200 an hour. So these guys are, you know, yeah. And the reality is like, it takes time. You don't just like wake up one day and you're an electrician. It's like a year. It's like, let's say 18 months of training. And, you know, another doubling requires an entire, like, you know, we've never trained that many electricians, right? And so another example is capital, which I think is probably the most top of mind one for the finance community is like debt, right? Let's say so far year to date, they've, the hyperscalers have raised $450 billion-ish of debt. That's like the biggest ever effectively. It's like the third after the United States government, China, and it's hyperscalers in aggregate raising capital. And the problem is it's like, there is a limited supply of money, like someone is buying the debt, right? When you issue a bond, someone buys it on the other side. And so in order for people to buy more bonds, they have to give them a higher rate. So you're hitting a supply-demand curve there. And the demand side, I really don't know what drives it. Well, I have a, I have a vibe as to what drives it. It's like life insurance, okay? Like retiree assets. You have to— and there's like a weird poetic justice that like the peak buying of annuities is right before retirement. All the boomers are in retirement. So the actual, the asset class SaaS is like larger, but can it double? Can it triple? I don't think so. And so that's going to be kind of one of the problems that we start to see is that in order for, for, um, the hyperscalers to raise more debt, they have to do more interest rates, and that's going to hurt. Um, you know, when you're a lender and you're saying, hey, I can give someone money, I can either buy these mortgage-backed securities for an American, you know, backed by the United States government at a, you know, let's say 5% rate for your mortgage— I think it's actually too low, whatever— for your mortgage Or I could lend to this hyperscaler who's a better lender than the United States government because they make a ton of money and they pay me 7 or 8%. So if they keep doing that, all the money goes this way. Essentially mortgage prices will start to increase. And so you have all these ways that like the system just can't handle another doubling. Like the scaling laws are scaling and they're like, great, let's make 2 times bigger model. But like, I don't think everything can scale at a 2 or 3 times bigger rate. So that's electricians and capital are 2, 2 of the ones that I think are creative and weird. But I think, I think if you give it time, it will, though. Like, they're going to issue a ton of money. I'm sure they can issue $1 trillion next year.

But truly, I'm actually fascinated by that. I really had not thought that through, but it totally is like pension plans that are funding this buildout right now.

Yeah, but pensions, pensions are a secular decliner. Pensions as a concept have actually gotten down over time. It's not like people are working, you know, people are working more in local governments than they used to be. Pensions are. And also most pensions are underwater. So historically, the concept of a pension has shifted over to like an equity participation plan, like a 401, right? Pensions have lost share at the expense of 401s. And 401s aren't really buying— they are buying stocks in this stuff, but like, you know, they're not really issuing equity. So like the really big dollars that get issued every single year, it's through debt. And so, yeah, like life insurance is a great, great source. And the life insurance is from annuities, and annuities get purchased when someone is about to go into retirement. So that's a source. Just general insurance, actually, health insurance, stuff like that. Essentially, you get all these premiums every year, you reinvest it in capital, and that's how you get— But you have to literally believe everyone just needs twice as much insurance. You're like, dude, no one's going to be— You know what I mean? That just doesn't make sense to me.

Well, I mean, it's an interesting balance because there is a lot of conversation in, you know, advanced economies, that there's an inversion in the, um, uh, age pyramid as many people live longer and retire and are building up this wealth for retirement.

Um, but they can spend it on data centers. They can spend it on data centers. You can spend it on data centers too. And then, and then importantly, the, the boomers who, you know, left the countryside. Okay.

Boomers, right? You can, you know, sell the house, right? Take that cash, you want to buy some life insurance policies, and those are going to fund the data center buildout. So we're trading houses for data centers, right?

And even better, it takes away your job at the end too. When it— so, um, yeah, you can't even own a house. You're now poor. You're poor and penniless, and you're in the permanent underclass because your boomer parents invested in life insurance and took away—

sorry, no, but the life insurance is also funding the data centers, which are also building the AIs and robots that are making sure you don't have a job either. So now you don't have a house, you don't have a job. Is it going to come for like retail food service soon as well? Like now you can't eat.

Yeah, dude, that's when the robots come, bro. No, I mean, that's being a little trite, but yeah, I don't know. It's pretty interesting. You just can't see like— I actually don't believe in that, guys. Yeah. Okay. A good example. Hey, another good example of this, actually my favorite one that kind of makes a lot of sense is Taiwan. Like, you know, we will just like, you know, we can double, like we've been able to just double the output of Taiwan over and over and over. And there are economies of scale, but like,, you know, I think like 20, I mean, Taiwan GDP is up like 25% this year just because like TSMC is just like cooking chips. Right. But if TSMC was to double again, and let's say there's some ratio of workers that needs to happen, like you're gonna run out of people in Taiwan to make the, the, the fricking chips. Like, I'm not joking. It employs almost directly and indirectly, I think it's 20% of the economy. So like the rest of it is like, healthcare, retail, and the government. There's really only one game in town. And so it's like, okay, what happens if— let's just say—

that's also for supporting TSMC though, indirectly, right?

Yeah, it's all— yeah, it's all for supporting TSMC indirectly. Yeah. So it's like, let's just say TSMC doubles, triples, triples, okay, triples the needs of workers. It's like, what are we gonna— like, Taiwan needs to literally have more babies to be able to like pay for that future, you know. Um, so yeah, it's just— it's kind of crazy if you think about it.

Yeah, I did just look it up. To be clear though, I mean, you're saying indirectly, I think that's doing a lot of work. TSMC has less than 100,000 employees and Taiwan has over 20 million people. So they can double up more times and find some people to employ directly.

They can, they can do it. I'm, I'm, I'm also knocking on the, like the factories that support the factories, support the factories, like the material services, all the stuff like that.

Yeah, I mean, but look at the percentage of GDP build in Arizona, right? And expand other things globally. Like, they're—

yeah, but there's no electricians in Arizona, bro. We're running out of labor. I think it's running out of lithium machines. Anyway, there's, there's a, there's actually a crazy thing. Well, to make a new fab, you, you need electrician. Um, the crazy thing, there's like, um, people are flying electricians via Cessna. To like, like, you know, to like little backwater places in order to like make it to job sites and stuff. There's a private— there's private flights just filled with like 16 dudes of electricians like, ah, you're here for your second shift.

So, okay, so this, this really reminds me of the oil fields in Canada. I have some friends who have worked on different oil projects, Fort Mac and stuff like that. Yeah, the, the boom-bust cycle of oil in like Calgary and northern Alberta and stuff is really fascinating as some of these communities kind of get destroyed as like, you know, uh, oil mining, whatever else is, uh, like, you know, kind of shut down. Um, it's gonna be fascinating to see how that happens where it's not even a natural resource that is drawing people to these areas. It's literally just like a permitting—

an information factory. Yeah.

No, but the reason why they have these data centers here is because of the ability to produce energy, which like obviously being close to the energy sources is good, but generally it's just like like they put up data centers where they can get permits most easily.

And yeah, that's which happens to be West Texas.

Yeah, exactly. Um, so, so it, but it's, it's not even like there's something fundamental where, um, you can just run out of lithium to mine and then, or gold, and then all of a sudden everybody has to turn around and go home. Uh, it's literally just like if a permitting regime changes, like if the political will of the people changes a little bit, then all of a sudden everybody in there is out of a job and things change a lot. I mean, I think we saw exactly this when we were putting out stuff on New Mexico. We had Ellie on a couple of weeks ago to talk about that with Jeremy. Yeah.

And the whole— but the thing is, okay, so there is— there, you know, like the invisible hand of capitalism does work a little bit. There's this also this interesting aspect where you think about it is if they own— so if they do this over and over and over, the logical conclusion is all these new jobs just get disappeared. And then some guy is going to be like, whoa, they took our data centers, they took our jobs. And so like, there is a self-correcting force in this as well. And because like, um, and this is an interesting poll that I saw the other day, is like, people who hate data centers often don't live near one, while people after having a data center in your community, especially for young people, it's a net favorable thing because jobs. Yeah, I mean, so like in—

I went to visit one in the Buffalo area, and just like everybody we met— I mean, of course they like having a job, but everybody was just super pro the whole site.

Yeah, because they're just like, because they're like, eff yeah, dude, I got a job, dog. A construction job is good. And like, I, you know, it doesn't take like super high-skilled labor. You don't have to have a PhD, you know what I mean? Like, it's just like, no, dude, it's like, it's an honest living, okay? And you're making— I mean, your people are making more money than they have been because it's clearly in demand. I mean, it's just, uh, what can you say? Supply and demand. When there's a lot of demand, supply's got to reach it. And so it's like, I think it's a net good thing if it's broad. And I think ironically, because these data centers are being built in the middle of nowhere, it's actually very good for broad participation, right? Like, you can argue the entire finance industrial complex and the technology industrial complex is extremely narrow. It's like, yeah, dude, San Francisco took, you know, is, is making me making an app so I can deliver something in a rural state, that is not broad participation in the economy. But building a building that is $10 billion in the middle of nowhere that has extremely high needs of electricity, high levels of service, extremely important, just like people are willing to pay for overages because it's extremely important for the strategy of these companies. That's broad participation because you you know, 1,000 people. I don't know what the math is. I feel like Jeremy knows. I want to say it's like 10,000 people for gigawatt or something like that. And so it's like, hey, if you bring 70 gigawatts, it's like, you know, that's a lot of jobs, 700,000 jobs. But that starts to move the needle, dude. If you do a— you know, if you do a—

in terms of revenue, right, there's many other jobs programs where you can spend less money and create a lot more jobs, but a lot of them are like not sustainable in the same way where this is obviously going to run for a very long time. People really like these data centers. And I think that generally speaking, a lot of people have problems with jobs programs when compared to investment from companies that happens to be directed to these communities, that happens to result in jobs.

Yeah, investment, it's more sustainable, it's more sustainable. So yeah, it's good stuff.

Okay, we gotta move to a wrap here.

Yeah, okay. Yeah, I was gonna say, I have a call I'm late for, so it's the reason why I'm not on this, on this podcast, bro.

Good to have Doug on. Good to debate a little bit. Yeah, we should get some hot takes.

Well, some, some modest takes, I think. Oh, we should get Joey on. I really want to do a Finance Week earnings recap. Okay, well, me and Joey can shoot the breeze.

We'll have to get Joey a competitively dynamic shirt as your Jensen Huang lightning leather jacket shirt.

But, uh, we have a Morris Chang— we have a Morris Chang So, okay, Joey could get a more— we can get a Lisa Su in here too. Be sick. Everyone, this is— we actually have— we have, uh, we have actually new company uniforms, so you're looking at it.

Okay, sweet. Thanks for— thanks for coming on, man. Good job.

Yeah.