Dylan Patel - Inside the Trillion-Dollar AI Buildout - [Invest Like the Best, EP.442]

0:01 Ramp is the only platform built to make your finance team leaner, faster, and better, saving businesses five percent annually on average, so you can stay focused on growth. Ram customers grew revenue 3.2 times faster than the average American business. Visa, Versale, Cursor, Stripe, Notion, 11 Lab, Shopify, and 70,000 other businesses all run on RAM. Mine does too, and so should yours. Learn more at ramp.com slash invest. FelixByrogo is a personal finance agent that turns a single prompt into finished client ready work using your firm's own templates, context, and standards. Send Felix an email like, take these comments and turn them for me, or update my tracker with the context of these emails.

0:39 Or run the ability to pay math on this buyer and Felix sends back finished PowerPoint decks, Excel models, and sourced research. Felix works the way your team already does, delivering work quickly and accurately around the clock. Learn more at rogo.ai slash feelings. The best AI and software companies, from open AI to cursor to perplexity, use WorkOS to become enterprise ready overnight, not in months. Visit workOS.com to skip the unglamorous infrastructure work and focus on your product.

1:12 Hello and welcome, everyone. I'm Patrick O'Shaughnessy and this is Invest Like the Best. This show is an open-ended exploration of markets, ideas, stories, and strategies that will help you better invest both your time and your money. If you enjoy these conversations and want to go deeper, check out Colossus Review, our quarterly publication with in-depth profiles of the people shaping business and investing. You can find Colossus Review along with all of our podcasts at joincolosis.com. Patrick O'Shaughnessy is the CEO of Positive Some. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of Positive Some. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions.

1:54 Clients of Positive Sum may maintain positions in the securities discussed in this podcast. To learn more, visit psum.vc. My guest today is Dylan Patel. Dylan is the founder and CEO of Semi-Analysis. At Semi Analysis, Dylan tracks the semiconductor supply chain and AI infrastructure build out with unmatched granularity.

2:16 Literally watching data centers get built through satellite imagery and mapping hundreds of billions in capital flows. Our conversation explores the massive industrial build out powering AI. From the strategic chess game between OpenAI, NVIDIA, and Oracle, among others, to why we're still in the first innings of post training and reinforcement learning. Dylan explains infrastructure realities like electricibling and companies using diesel truck engines for emergency power. While making a sobering case about US China competition and why America needs AI to succeed.

2:44 We discussed his framework for where value will accrue in the stack, why traditional SaaS economics are breaking down under AI's high cost of goods sold, and which hardware bottlenecks matter the most. This is one of the most comprehensive views of the physical reality underlying the AI revolution that you'll hear anywhere. Please enjoy my conversation with Dylan Patel. I was gonna lay out this idea of going through the past, present, and future of compute as like the big, big idea for our conversation.

3:10 But since it just happened, I don't think I've heard you talk about it anywhere. I'd love to start by asking about this whole open AI and video thing. Sounds exciting, seems vague, not really sure what's going on. Maybe you could explain it. to us as you see it and what the strategic implications are of the big announcement. I think it's very, very simple. You've got open eye.

3:28 paying Oracle lots of money. You've got Oracle paying NVIDIA lots of money. You've got NVIDIA paying OpenI lots of it. Spider Man meme. No no no I mean it's it's that's not actually what's happening. If they pay each other then their market caps all keep going up. What's really happening is Open air has an insatiable demand for compute. Не комп'ю прос.

3:49 The build up of Business. You have to have the cluster before you can run models on it for inference. You have to have the cluster to train the model. That's good enough that it unlocks new use cases, which then can be adopted, and there's an adoption curve there for any new use case.

4:03 So you have to have all these things like sequenced. Given this is a game of the richest people in the world or rather the biggest tech giants in the world, right? It's it's Zuck. It's all the biggest people in the world. It's Elon, right? Google, Larry, and Sergey is like constantly in the business now again. There's Very much a risk of opening eye being Too small to matter.

4:22 Which is crazy to say because they've got eight hundred million users, but Where's the revenue? Where's the compute? They could easily get swamped in terms of how much compute they have. If they don't move fast enough and if they don't have the most compute. Or among the most compute. They will get beaten.

4:37 The magic of OpenAI was that they just spent way more compute on a single model run on GP three and four. And they had the foresight and the vision and the execution. But they made that bet and they were about to secure it. And at the time it was like Meh. Right? It was a few hundred million dollars. Whatever, right? That's a ton of money, but like Now it's sort of like Mark Zuckerberg sees how much compute he's gonna have to get.

4:57 Even though he has This insane cash flow that He's like, Oh wait, I need to go sign a deal with Apollo for thirty billion dollars on this data center. in Louisiana, this mega data center I'm gonna build. It's like, wait, why didn't you just fund this with cash flows? You have so much cash flow. It's like

5:11 That's just the physical data center. Now what I'm gonna put in it is like so much money it's The amount of capital that people are gonna have and are dumping into this is insane. Google was slow to wake up. slow to pivot their data center operations or slow to do everything.

5:24 While they could have way more compute than anyone by a humongous degree. And then they have like how much they allocate to search and generative search is not really necessarily competing with Open AI. It's the mega models. So if you have this tremendous vision of what's gonna happen with AI, you know that it takes a ton of compute to build them, you know.

5:40 Pretty much the amount of compute you could dedicate to these models is limitless and they will get better. Now it's a log log scale, right? I you need 10x more compute to get to the next tier of performance. You might think of it as diminishing returns, but what if the next tier of performance is like A six year old versus A six year old. A six year old you can't get to do much, right?

5:58 And this is not exactly the way to think of AI, but This is the conundrum that open AI is in. They have to race with the giants. These giants are trillion dollar businesses. So how does opening I get there? Well, it's partnering with Microsoft. Well, that's soured some, right? It's partnering with Oracle. Well, Oracle can do a lot, but

6:13 Oracle doesn't even have a balance sheet like Google and Microsoft and Amazon and Mera This is very much like The Pascalion wager nature of all of this with the tech giants, Oracle can be part of it, but OpenAI needs allies. They need people to.

6:28 Effectively spend the capex. ahead of the curve and trust that they'll be able to pay the rental income. Cause that's what it is at the end of the day, opening eyes. committing to five year deals. These five year deals cost X amount of money. It's ten to fifteen billion dollars.

6:40 per gigawatt of data center capacity that you pay a year. And then that 10 to 15 billion dollars for a gigawatt of data center capacity, you're paying that for five years. Okay, that's fifty to seventy five billion dollars of cash that goes out the door to open AI for one gigawatt of capacity. And you talk about what Sam's saying is like, hey, I need 10 gigawatts. More than ten gigawatt. Then you end up with this. challenging aspect of like how do you pay for that? And hey, that's only the rental price.

7:03 If I were to actually do the CapEx, because it's front loaded, right, it becomes who is the balance sheet for this? That's the reason these deals are coming about. Oracle is making a massive bet. Larry, he's getting good margin off of it, but he's making a massive bet that this capex that he's gonna pay for Open AI. Will actually be paid.

7:18 'Cause they signed a three hundred billion deal with open AI. It's like Your revenue is like fifteen billion ARR this month, maybe, on a run rate basis. It'll get to twenty by the end of the year, pretty clearly. Maybe it's like sixteen now, but It's very tough to get to How do you pay three hundred billion dollars of revenue?

7:34 If the bet works out, they've just made a hundred billion of profit. Pure cash profit. It's crazy. But If it doesn't work out, they've got this huge And they've they're starting to raise debt.

7:42 There was a small deal they signed recently. But they're gonna start raising more and more debt. Now NVIDIA's kind of got the same conundrum, right? It's like Google and Amazon are doing these deals, whether it's to other vendors for TPUs or for training, whether it's Anthropic or others. They're trying to court open AI. They're trying to court other companies.

8:00 How do I get into this game, right? Okay, fine, I can rely on Microsoft somewhat, I can rely on Oracle somewhat, but at the end of the day, If I want GPUs to be king, part of it is just like my chip is the best. But part of it is also Who's gonna pay the CapEx up front? Google and Amazon will pay the CapEx up front.

8:16 If it's for TPUs or training. They won't pay the CapEx up front. Necessarily for that same capacity of GPU. So you've got this like challenging aspect, and so that's where this NVIDIA and opening ideal comes from.

8:28 I wanna dig into the underlying assumptions driving this on the training and inference side because obviously there's the willingness, like Zuckerberg just needs to go down the hall to CFO to get access to all this capital. He doesn't even need to go down the hall. He can just make it so he's got the voting share. But I wanna make sure I understand your thinking on the underlying two sides of this. One, which is like your view on the diminishing return curve on just I don't think it's a diminishing return, right? I think that's important to recognize, right? Given it's a log log chart.

8:58 Skilling laws are. Given there's no model architecture improvements, you just throw more compute data model size at it, it gets better at this pace. But you're confident that that will continue. Everything has shown that it will continue and it's continued over more than ten orders of magnitude. G V D five is not not necessarily that much bigger than the four O. Right. And four O is smaller than four. What's changing sort of the paradigm of how you spend the compute.

9:21 And also like if they made a bigger model, could they even serve it? No. They did four point five and it was terrible. No one could no one could serve it. It was actually like quite a bit smarter, but they couldn't actually serve it at any reasonable cost and speed. Anthropic has the same issue. I wouldn't even call it an issue, but like

9:35 All of their revenue comes from force on it. Doesn't come from four point one Opus, which is the better model. It's bigger, but it's slow. Because the hardware's not caught up in terms of inference speed for that. And so no one wants to use a slow model, right? The user experience sucks. Yeah. But as far as like

9:50 If the model gets better. At each scale of hardware spend. I would say all the tech giants believe it. I believe it. I think a lot of people in the financial community are like this is Fricking scary. Yeah. Because the moment it stopped, you know, wherever you were on the wrong, right? If we went from fifty billion dollar spend to five hundred billion dollar spend, well, that five hundred billion dollar spend is never gonna have R O I. It was one thing if fifty billion didn't have ROI, but now this five hundred doesn't have ROI. It's a big problem. So anyways, one could think of it as as diminishing returns because if you when you go from fifty billion dollars of spend to five hundred billion dollars of spend

10:17 You only move up one tier of model capabilities. In absence of major algorithmic improvements. I'm holding those sort of off to the side for now. But that iterative performance improvement in the model It's like a six year old versus a thirteen year old, maybe, right? The amount of work you can get a thirteen year old to do is actually quite valuable relative to a six year old.

10:35 And the same applies to like a college intern versus someone who graduated and has even one year of work experience. Yeah. Cause there's a learning curve for kids coming out of college all the time. While it may be an order magnitude more of compute, the amount of value If you made a company full of high schoolers And you had to refresh them every six months so they didn't learn too much, right, and become really good. It would be really hard to create a valuable company. The most you could do is like

10:57 Dig trenches and Do yard work, but then All the time these kids wouldn't even show up, right? Like as a function of like how valuable of a business could you do if you had unlimited high schoolers that refreshed so they didn't build knowledge versus college students versus 25 to 30 year olds. the value of that business that you can build, even though incrementally it's just five years between each of them. It's a drastic value change.

11:17 Where do you think we are today? Like which level are we at, do you think? Depends on the domain, right? Like for software developers, like I think we're we're really pretty good. And that's where we're seeing the most value creation happen. Anthropic have gone from like a billion or less of revenue to Seven to eight.

11:32 It's the fastest revenue ramp we've ever seen. For anything. That's basically all code related. Right. I mean, like you know, some of it's their own clod code product. Some of it's, you know, cursor, some of it's GitHub Copilot, which also offers anthropic models and has f since the beginning of the year. It's windsurf, it's all these different avenues to access.

11:50 The same thing. And these companies aren't all doing the same thing. There's tweaks and nuances to how they're doing things differently, but it's all Code and You know, in that sense it's like If I had Thirty year old senior engineer.

12:02 at Google, and if I had infinite of those, all it cost was CapEx for chips, and the operational cost is actually quite low. Then you could build businesses worth insane amounts. You could have A replacement for the two trillion dollars of wages that go to all the software developers in the world today. Or rather you could augment them and build, you know, twice as much or five times as much or ten times as much.

12:21 If you could augment them,'cause these things don't like just run on their own, right? It's more of like a force multiplier to the existing person. So the value creation potential is is there. It's obvious. If you've coded it all in your life. I mean it even works for VBA. It's not that great for VBA. So I know a lot of people in this audience probably know VBA. But like it's it's not even that terribly bad for making macros, but The value creation potential there is incredibly high. So let's capture it. How do you capture it?

12:46 This draws back to the OpenAI NVIDIA deal because I think Most people in the market don't quite get it, right? They're like, Oh, this is just like round tripping. It is to some extent. They agreed 10 gigawatts of capacity, NVIDIA'll do a hundred billion dollars of equity investment into open AI in the form of cash. And NVIDIA gets return capital. The first chunk of the deal. in the press release is one gigawatt. Ten million dollars.

13:07 So pretty straight line. One gigawatt to build, as we established earlier, is like fifty billion dollars. So NVIDIA's paying Ten billion. OpenAS to come up with other forty somehow. go to the markets get a loan or get someone else to put a loan, right? There's the infrastructure funds that are trying to get into this.

13:22 You know, these commercial real estate people are trying to get in this, there's some way where they'll be able to figure out other people to front the capital. And then come up with like a deal much like it is Oracle, but OpenA has to do more of the work in terms of setting up the cluster, the software, the networking, et cetera. Nice thing for NVIDIA is Of that fifty billion they capture, maybe thirty five billion of that is CapEx that goes directly to NVIDIA. So year zero. OpenAI slash its partner spends fifty billion dollars on the data center.

13:46 The timing is not exactly that. They spend fifty billion dollars on the data center. Thirty five goes into NVIDIA. NVIDIA's gross margin is 75%. So You know, again I'm gonna make it simple numbers. Let's say it's ten and forty. Ten billion cogs. Forty billion revenue, thirty billion dollars of gross profit.

14:00 If we fix the numbers, it's effectively like half their gross profit from that deal is going directly to open AI in the form of an equity investment. The twenty five percent that's Cogs. NVIDIA's paying for that, and then they keep the other half of the gross profit on their balance sheet or do buybacks, whatever they want to do with it. So NVIDIA's not necessarily like They are like round tripping some of this. What effectively is happening is OpenAI gets the opportunity to pay for a big chunk of it in equity.

14:22 And NVIDIA's lowering their prices without lowering their prices effectively. But and they're getting ownership of a company. But NVIDIA comes out great because they're they're getting the Capex dollars up front. So all they're really doing is they're saying half of my money that's in this Sure it does make its way to me somehow, but

14:35 In reality, I still made half of that gross profit and the other half is equity in a company that may or may not be worth something, a company that may or may not be able to pay. hundreds of billions of dollars of compute deals that they've signed, in which case they'd be bankrupt. It's about the highest stakes like capitalism game of all time. And it's so interesting to think about when it might run out. You mentioned like if we hit that final point and we don't see the return. We're kind of toast in a big hole. But I'm also curious about the other side of ability to serve and just demand for like today's models by inference.

15:03 That's that LSI is. Token demands doubling every two months or something crazy. Obviously, there's all these reasoning tokens that are really exciting for some of the longer thinking models. How do you think about the growth of the pool of demand for entrance tokens themselves. Even in today's models, like even if we just like stop things and fix things and we'll leave that other side of the equation just for a second. What's your model for thinking about that today? What most interests you? So the thing I like to call it is tokenomics. I stumbled upon the word actually it's like a crypto build off. So I'm I'm trying to make tokenomics SEO.

15:35 direct to you know us talking about tokenomics and then hopefully you talking about tokenomics like everyone talking about the tokenomics twenty more times. Yeah. It's gonna be in the title now, right? You've got some powerful SEO, we kill the crypto bros for this. But tokenomics Economics of the tokens, right? How much compute is being spent? How much is the gross profit? What's the value being created by these tokens? That's that's the end of the day, what's relevant here, right? NVIDIA keeps saying AI factory, which produces intelligence, that intelligence has value. Let's say you have a gigawatt of capacity, what can I serve?

16:05 Well I could serve А тазен тайм. Of A model that's really shitty. I could serve one times of a model that's good. And I could serve like point one times of a model that's amazing. Now multiply that by whatever factor.

16:17 How many users, what's the number of tokens outputted. I could do X number of tokens, X times one hundred, X times a million tokens, right? Depending on the model quality. This is sort of where The whole

16:28 GPD five thing comes around. OpenAid had a challenging thing, right? They're like, Hey, we have A couple of gigawatts of capacity, effectively. Right. By the end of this year, roughly a couple of gigawatts of capacity too. How do they

16:41 Maximize their serving capacity of this. One avenue is We continue to serve big models and we make bigger models. And the tokens are more expensive, but this log log scale is really challenging because Yes, the value is

16:54 Way more. But the cost is way more. And then the real whammy is the user experience is way worse. If I serve a massive, massive model. It sloped.

17:03 And users are fickle. You need the response to be faster than they can. Yeah. So there's this like user experience challenge, but really in the end, it's like For a given model level I think there's a saturation point of how much demand of intelligence there is. You can only have such large child army, right? Of of like people digging trenches or like Coney twenty twelve, whatever it is, like this is very cancelable, but you know. But you can have a much larger army of

17:28 or business of like the larger level of intelligence. When you think about what could I have done with GPT three? GPD three, even if we paused there. Pause the model capabilities, right? You know, obviously the cost to serve a model quality of GPD three, that's like two thousand times cheaper now. And then GPD four, same thing, right? People were freaking out about Deep Seek because it's like five, six hundred times cheaper.

17:48 GPT OSS came out, and that's even cheaper than that. For roughly the same quality. Actually, I would argue the GPT open source model is actually a little bit better than GPT four OG because it can do tool calling. And anyways. The cost of these things tanks rapidly with Algorithmic improvement. Not necessarily model getting bigger.

18:05 But at X level of intelligence, you can only serve so much demand. The flip side is It takes time for people to realize how to use it. So when G V D three launched, no one cared.

18:15 When G V D three point five launched, it was like still most people didn't care. Chad GPT launched with G V D three point five, people cared a little bit. GP four launched on Chad GPD, then people cared a lot. But A model tier of GPT three point five or three still can be very useful in a lot of the world. Now it's not useful for like

18:30 A lot of use cases, right? Like for coding it was terrible. For copywriting, it's okay. There's some level of use case. And it happens to four, but it takes time for that adoption to happen. And so you've kind of got this challenge of like If I pause on a model capability, Then I end up taking way too long for adoption and also

18:46 How can I get people to adopt it if I don't let people use it? So OpenAI had this tremendous problem with GPD four O. Four turbo was smaller than four. And four O was smaller than four turbo. What open eye basically did was they made the model

18:57 As much smaller as possible while keeping roughly the same quality or slightly better. So four to four turbo was like the model was less than half the size. And four turbo to four oh, like four oh's cost is way lower than four. And they just kept shrinking the cost. Now five. What could they have done? They could have gone, Oh, we'll go big step.

19:14 They actually tried that with four point five. They screwed up some things because it was really hard to get, you know, a hundred thousand GPUs to work properly. There's challenges there. Also, they hadn't figured out the whole reinforcement learning paradigm at that time. The scaling laws are like It's a chart of quality versus compute. But that compute breaks down into how much bigger do I make the model.

19:33 How much more data do I put in the model? And if the internet only has so many tokens You're kind of screwed. There was potentially a cliff until reinforcement learning happened where you can generate data and train the model to be better without the internet having that data. So they kind of had this problem of

19:48 Yeah. X amount of compute. You can service your users, but hey today If people want to use my API, I rate limit them. Because I can't actually serve them all.

19:55 I have to rate limit the people who have chat GPT free pro. and max whatever the twenty two hundred dollar deals. There's like different rate limits. You can only do deep research so much. I have multiple chat GPT accounts because I use deep researchers like You kick off a bunch, you read it, and you're like, Wow, I learned a ton. Move on.

20:11 So you have this challenge of like you can't actually serve your user base. Enough. So how are they ever gonna move up this adoption curve? So then as open AI, what's your choice? Do you go from four oh to five? Do you make the model? Way bigger and not to be able to serve anyone.

20:23 And plus because you can't serve anyone and it's slow to serve, the adoption curve doesn't really get going. Or do you make the model the same size, which is what they did for GPT five. It's basically the same size as four oh in roughly the same cost. That's actually a little bit cheaper, potentially. And then you just serve way more users. And get everyone up the adoption curve more. And then you can instead of putting them on a bigger model, you put them on models that do thinking.

20:45 If you've used GPT five thinking or GPT five pro, there's more intelligence there. This is the whole conundrum they have. And this is where the whole tokenomics thing comes into play. The question you had, I wanted to level set it, right, which is How do you serve these users? The demand is growing so much. I'm not doubling my hardware every two months. Yes, this CapEx is crazy, but I'm not doubling my hardware every two months.

21:06 But I'm doubling my tokens every two months. So there has to be enough of a cost decrease. And there is. at a given level of intelligence. If you could snap your fingers and change a dial somehow that would most unlock and unleash more development, is it just inference latency because then we could do bigger models and serve them much faster in a way that consumers would enjoy?

21:25 Is that the main like bottleneck to be attacked? All of these things are curves. And it's a trade off, right? Everything in engineering is a trade off. So you have Inference latency versus cost on any given hardware. GPUs can do lower latency

21:37 To a certain extent, but then the cost is way higher. Or you can do really, really high throughput and the cost is way lower. And the company. They set the dial where they think it makes the most sense. Yeah. And there's other types of hardware which aim for their curve to be at a different spot. Maybe the GPU curve is here, but latency, you know, over here, you know, you're in very diminishing returns. And so actually someone made a little curve right here. That's like okay, maybe that's a useful point, but actually the market cares about this point.

21:59 So anyways, there's a curve of like who cares about latency. I think if I could just press a magic button Is it capacity? Is it latency? That's a tremendous like question. I'd probably still say capacity slash cost is more important than latency. Really? I think existing levels of latency. Are fast enough for a lot.

22:16 If the latency was ten x lower for GD five. Then they could have made a model that was ten X bigger. I sort of at the speed. But then you would have the same capacity issue. If you can have your cake and eat it, which is all the capacity in the world and the lowest latency in the world. Yeah.

22:29 You'd make the models way better. It's the physical realities of if I'm at OpenAI, what do I choose to do? Do I invest more in The model that people can use, and that's fast. Do I invest a lot in the model that most people

22:40 won't use because it's expensive, first of all. And even those that can't afford it will often go back to the regular one. I have access to cloud four point one Opus. I still use Sonnet way more. Just'cause it's a better experience. Right. Slow. My time's worth something.

22:52 If you had the magic button. I think opening I wouldn't have been afraid to like make a model way, way, way bigger. And a terrible user experience. And as a result, we're just gonna probably have to wait a little bit longer to see what the bigger models are in practice. to see what consumers actually do with them because it's just gonna be too hard.

23:08 It's not necessarily even bigger, right? Like there's this whole concept of Over parameterization. I e If you just throw more parameters. In a neural network.

23:18 Equate it to humans, right? When you had a vocab test or you had some test, you memorized before you understood. And it wasn't until you did multiple repetitions and in different forms that you actually understood the content rather than just memorized. It takes cycles. When you do an LLM, it's the same thing, right? If you throw some data at it.

23:36 It will memorize it. Before it generalizes. It's this concept called grocking. You grok the subject, i.e., it's like the aha moment. Yeah. The models do the same thing.

23:46 They memorize it up until then they understand it at some point. And if you make the model bigger and bigger and bigger. Without the data changing. You just memorize everything and actually it starts to get worse again. Because

23:56 It never had the opportunity to generalize because the model was so big and there's so many weights and there's so much capacity for information. The challenge today is not necessarily make the model bigger. The challenge is how do I generate and create data In useful domains so that the model gets better at them. Nowhere on the internet.

24:13 To show you how to fly through a spreadsheet. Using only your keyboard and all these like functions and all these things, right? Like That's that's a repetition, that's bars, but there's no data on the internet about this. So how do you teach a model that?

24:26 It's not gonna learn it from reading the internet over and over and over again, which you and I could never do. And so it hasn't a level of intelligence that we can't do. We can't read the whole internet. But it can't do basic stuff, which is like Play with a spreadsheet. How do you get it to learn these things? That's where this whole thing.

24:39 Reinforcement learning, paradigm. Giving it environments, specific environments to learn in and then fold back in. Right, exactly. That's where There's sort of a challenge in terms of Building those environments.

24:49 So there's like forty startups now in the Bay toing these environments and you know. Questionable whether or not any of them will make it or what will happen. And then these companies are also making their own environments, but these environments can be anything and everything. So it's like as simple as like Here is a fake Amazon. Amazon terms of service ban chat.

25:06 Models and all these things. But here's a fake Amazon full of items Figure out how to click around and purchase items. figure how to compare the two items and pick I've generated a list of deodorants. Three of them are fake, one of them's real, one of them's not the one I want. Here's the prompt.

25:19 And you know, it tries many things and you know vary the prompt and all these things, but eventually, you know, it it's bought the right deodorant and you've succeeded and you fold it back in. That's a simple thing. Or it could be. Clean this data. Right, here's this table. Ton of dirty data in there. Oh, there's like colons and stuff. There's an address in one column.

25:34 How do I separate out the columns so the address is like Street address, city, zip code. And it'll try a bunch of stuff, but like, hey, maybe it can't do that yet. You give it an iterative like here's addresses, here's different formats, and you slowly iteratively teach it. So another example is like You're in

25:48 Again. Whether it's a tic tac toe or a call of duty or a math puzzle, whatever the game is. And that's what a lot of these environments initially have been is like math puzzles. Do this math poll. Oh well, I can't do this one because it's too hard. Here's an easier one. Oh okay, I can I can spin on this one. Okay, I'm better enough. Okay, now I can learn this one. iteratively step through those to where

26:05 From Q four of last year. to Q two of this year. These things hill climbed up math puzzles like crazy. And a lot of that was not Hey, I just know the math. A lot of that was here's how I use Python

26:16 Two. Write something that does the math for me. And now these things are actually quite good at math. The environments can be Super varied.

26:23 Um and it doesn't need to be something that's like clear, cut and dry. It can be here's here's A medical case. What's wrong with it? And then there's like and then you have another model say, Well, here's the instructions on how you would grade the result of a case. Well it looks like they didn't even try this or didn't even look up this. Okay, you did that wrong.

26:39 So these environments can be very, very complicated. So building those out. is a challenge, right? It was one thing to say I'm taking all the internet data. I'm gonna filter it some. Math of the model. There's tons of engineering challenges there, for sure. There's a different set of engineering challenges that take time to build out.

26:55 In those two, like in pure raw internet pre training world and in this new like environments world, what inning are we in in each of those, would you say? Like how far into the potential benefits? This is where like the whole like Oh well then Dylan, what you're saying is you never need to make models bigger again. Right, because you've already run out of data. And until you figure out how to generate

27:14 Tons and tons of data. That's great. But actually we haven't. You know, we've seen another angle where it's mostly just been pre-training scaling, right? Is VO3 and Banana Nano, right? These Google Image and video models. And Genie. And like all these Google image and video models. And that's that's purely like scaling on multi modality, right? The models still aren't that great at video and audio and images. They're fine. They could be a lot better.

27:36 Um so there's like angles of scaling there, right? Because when I said we've run out of training, we run out of the text. Yeah. There's tons of video and image and audio. It's just so expensive. So, you know, like we we didn't get to that. Maybe late innings on text. mid innings on pre training methods. Yeah, we're quite on. And then the other angle is

27:56 Just because You've used the text doesn't mean you can't learn faster. You know, it's like take a class for them again, like you know, this is like Machiavelli I don't even know if Machivella's the right word, but like you take a class, you give them all a book, you tell them to read it once, and you test them all. It's like one kid's gonna get a hundred and one kid's gonna get a forty, right? It's just the reality of life. And If you read the book out loud to them, the kid who got a hundred might get a thirty.

28:17 And the kid who Got a forty, might have got a sixty, right? So there's like these different parameters. And when we talk about model architecture, the same thing happens there. So it's not like you stop training new models. It's not like you don't have algorithmic improvements or smarter kids, right? You know. So it's not like pre training is done. It's the base of everything. So you want to keep having gains because any gains on pre-training, i.e., the model learns a little faster or the model's a little bit smaller for the same quality. feeds into the next stage.

28:41 Which is this whole post training side. Which will Subsume the majority of the compute. At some point. And inning wise, is are we in the second inning of that? Like I think we've like thrown the first ball.

28:51 Wow. I think my favorite thing my brother just had a baby. He had his third month birthday like a couple days. I'm gonna go see him in another week, but like this baby will literally stick his hand in his mouth. He's like Calibrating the senses on his fingers by sticking his hand in his mouth because his tongue is the most sensitive thing. He doesn't know he's doing it, but like that's how he's calibrating. He's like, Oh, that's me.

29:10 Oh, I can touch and feel. Right? It's like How does the model learn these sorts of things, right? It's like you just have to try stuff and fail. And we're so, so early. Think about how much we see throughout our life and how much of that information we throw away. We throw All of this information away.

29:26 Do I remember what I had for lunch yesterday? No, but if it was amazing or bad, I would have remembered that, oh, I don't like this or I like this, right? There's all this information we throw away. And These models, these environments. Yeah, we're generating tons of data.

29:39 And throwing most of it away and training the model, but it's like infinitesimal compared to what humans have done. And so I think There's so many environments you can put the model in. There's people who even think you don't get to the magical AGI until you embody it. I e you put the model in something that can interact in the real world. Elon and XAI, like they're a bit more along that angle of like they think embodiment is required to get to artificial general intelligence because you need the model to be able to say like

30:03 Pick this up. Or like, Oh wow, this is like a rotate thingy, which you could never get from like just watching a video about it. You wouldn't get the concepts of it even. I think we're so early in the Reinforcement learning because that's what humans are. We're reinforcement learners.

30:16 Let's say we fast forward we're in the seventh inning of that or something like this. What do you think the way that the average person will most feel that difference in terms of the utility of the model? It'll be very different like modus of using it. It's one thing to like Ask for information or ask it to organize information.

30:34 Versus It just doing things. Those twelve year olds, you need to really direct them how to dig a hole. 'Cause a lot of them haven't dug a hole. But you're talking about order me this vitamin and just like it's just done.

30:44 Right. And we're actually like not too far away from that, I think. If you try and research electric toothbrushes, like this is something'cause you know, your electric toothbrush, I leave it at a hotel all the time. And I've been obsessive about this. Like in twenty twenty one, I made a spreadsheet of all the electric toothbrushes'cause based on how many I Cs were in each one of them, right? Like this one has a Bluetooth IC. Why? I don't This one has a display I see. Like it it has a color display I see. Like what's going on, right? Like I said, I made a spreadsheet of all this. And so like I don't know, it's like this weird like little thing that I do, but I've been finding like How I research which toothbrush I wanna buy now, I bought a oral B Io like series nine or whatever. These models now can figure out exactly what you want.

31:20 More than ten percent of Etsy's traffic is straight from GPT. Wow. Amazon blocks GPT, but otherwise it would be really high. People make purchasing decisions through GPTs, they just don't make the purchase. OpenAI is head of applications or C of applications. was at Shopify and created the shopping agent.

31:37 It's very clear this is how they monetize and this is like something we we wrote about like on the newsletter once. The models are gonna purchase for you. They're going to do actions for you. And then the company that does those actions for you, the model that will be able to take some sort of take rate. Even if it's like point one percent, even if it's one percent, it's two percent. It'll be like a credit card transaction. Visa is the most amazing business in the world because of this, right? And chat could be that too, if I'm making my decisions on purchasing. All sorts of things. I mean, I already almost outsourced like what am I gonna eat to like the front page recommendation of like Uber Eats sometimes or Google Maps reviews, it's like

32:09 Oh, I'm feeling like pizza. It's like, Okay, well the first one is one I'm gonna go to, right? I already outsource a lot of decisions. It's not too much further till I've like completely outsourced a decision and a purchasing intent. That's Made Meta and Google such amazing companies is they figured out

32:24 how to get the thing you want to purchase in front of you. All their work on recommendation systems is figuring out what you like. How to keep you on the platform longer, whether it's YouTube or Instagram or Bite Dance Right with TikTok. Or it's Hey, here's the ad of the thing you'll probably click on and buy because that's how I get paid.

32:39 And everyone likes to claim they don't like pay attention to ads, but you do. Before asking even more holistically your view on where we're going. There's a third category which is the reasoning part of the equation. So we've got pre training, we've got R L and environments, post training. What about

32:52 raw time spent reasoning and where that going as its own independent part of the overall scaling law. So I think the scaling laws again, like if you zoom out, that's not actually what the like original paper is, but in spirit, sure. Scaling laws are more compute, better intelligence. And that could be bigger and bigger model, each iterative token is better. But again, when I talk to you and I Whatever like word garbage I spew out. If I went back and I wrote about everything I talked about in this, I could make it way more condensed. It could be way more clear, potentially, right? Now the benefit of podcasts is a lot of times people driving. It's fun. Yeah, exactly. Right.

33:26 They're walking their dog and they're listening, whatever it is. They're working out. But like an important Thing here. Is that

33:32 By putting in these environments, you're teaching it like humans. If I asked you To go figure something out, you might not necessarily know the answer right away. But I know you could probably figure it out in a given amount of time. That's reasoning.

33:46 You're spending more brain cycles. the magic again of like intelligence of humans, of people is not that they are the best at information retrieval. GPTs are amazing at information retrieval. We're really good at because we've been trained in these environments, which is our world.

34:00 at figuring out how to do things iteratively. And so reasoning and these oral environments are linked together. If I'm telling a model, hey, do this math puzzle, it's not just spewing out like oh the answer's one, oh the answer's two, oh the answer's three. Okay, the answer was actually seven. And when I got there, I trained it again. It's like okay, now it knows next time. Oh, the answer's six, seven, or eight. No, it's like seven. Okay, great. It's not like now it instantly knows the answer. It's actually like Oh here's this puzzle. Oh, like these numbers. Oh, this line Sudoku, these numbers add up to this.

34:26 Oh, it has one through nine, but it's missing eight. Okay, it's eight. Right, like it's thinking through it, right? Like you and I would solve the Sudoku. Now, eventually, when you get good enough at Sudoku, you could probably just like spit out an answer. You could do it in your sleep. This reasoning time is a way of spending more compute, more brain cycles. On the task. Without actually

34:45 Scaling the model. And then the model becomes more versatile, right? Because humans have A rate. If I just held a match against you and you didn't notice it, you could immediately jerk. The rate at which you operate is Hundreds of hundreds of hertz.

34:57 Your body can actually take actions at like Hundreds of actions per second. If you look at like a fighter pilot's reaction time. What reaction can they do is like completely like primal instinctual, right? Very little Thought is put into it.

35:10 This alien intelligence that we're trying to make. Is it immediately going to one shot the answer always No, but like, you know, at times it needs to.

35:20 At times it needs to be able to tell me exactly the answer in like two seconds or half a second, or whatever action it needs to take immediately. But a lot of times also needs to think through the problem, go and do stuff. That's why you hire interns,'cause you're like Yeah, I know this data exists. Here's the format I kinda wanted on.

35:37 Go figure it out. And then they they spend a whole summer doing something you could have done in like three days. But like great, they learned a shit ton. These models need to go through that progression. And so when I think about reasoning RL, it's a lot about how the human psyche and intelligence works. There's a caution of like trying to make it too much like humans. 'Cause it's not the fundamental substrate is not like humans.

35:56 the processing is not like humans. Our brain is very different from a how these ALUs on a chip work, like The scaling of these things is very different. The raw speed, the amount of words they can everything is so different, but the same time it's important to like Reckon back to what actually makes people, you know, smart. On the topic of like embodiment. Continuing with the human analogy, how do you think about things like short and long term memory in a human versus raw model capacity or something?

36:22 What role does that analogy of memory I don't mean literally like semiconductor memory, but like memory in a model. How do you think about the importance that that will play and where are we in that? The magic of transformers. was attention, I e I calculate Everything in my context lamp I calculate the attention to each other.

36:41 Basically in a vector space, like King, queen, there's the vectors, there's like dozens of vectors for each number. And king and queen are actually exactly the same on a ton of stuff, but then it's the opposite on one number, because one's a male, one's a female. And then that will have a lot of other like ramifications throughout other literary stuff, like what adjectives do you put with a male of you know this vectors, oh it's like regal and you know, like powerful and could be ruthless, whereas a queen could be like Dignitary or whatever. When you think about

37:06 How that applies to Humans were terrible. But Exact recall. Absolutely horrible.

37:13 I could tell you a sentence. And tell you to repeat it. Yeah, it's like six numbers the average person can remember or something like that. Right. But like you get the gist of the sentence. If I told you a whole paragraph, you'd get the gist of it. You could repeat the meaning of it to someone. You could translate that meaning. So models Very different, right? Fundamentally transform our attention has been calculating the attention to everything to each other and getting the models to actually be able to recall.

37:35 That's been a training data problem, but like you can get the model to repeat exactly what you want anything in its context length. It's like a needle in the haystack is the like it's a benchmark that people did for a while because models had to get good at that. But now models are just like amazing. But what they really suck at is having infinite context. The real word is sparse. You've taken this entire world.

37:54 And you've encoded it in such a small amount of data that lives in your brain. And it's so sparse. But you understood how to grab the fundamental reason and put it down there. Whereas models They haven't been able to create something sparse yet.

38:07 How do you reason over the context of Infinity. Humans, maybe we have like a short term memory and a long term memory. I think it's a lot more blurry than that. There's no like clear line, Oh, this is in my short term memory. Oh, this is in my long term memory. It's like It is much more blurry. As we go back and back and back, it's more and more sparse.

38:24 Right. If we think about, hey, what do you remember as a kid? The most crazy thing in psychology, I remember when I learned it, I was like, Wait. My memory of what I Did as a kid with my Dad at this like thing, right? Is fake.

38:34 It's me remembering it. And inventing the picture and me remembering that picture like successively. But like the actual memory of what happened is like morphed a little bit. Over time. The way humans collapse information is super, super dense, but we are able to extract all the relevant information out. Now, models, there's a ton of research going on in this domain of long context.

38:55 How do I get longer and longer context? Without blowing up My model cost. This is a big challenge with reasoning. This is why we had this HBM bullish pitch for a while, right? As like you need a lot of memory when you extend the context. Simple thesis. The fundamental algorithm needs to change and improve over time iteratively to get to

39:13 something like this short and long context of memory. That doesn't necessarily mean the model has to work like we do. Why can't the model just reason and have A database that it writes stuff in. Or like

39:25 a word document that it writes stuff in and then it takes it out of its context, works somewhere and like roll calls back, it's like, Oh yeah. We don't do that, right? Like you and I refer to our notes, we refer to our calendar, we refer to our text, refer to anything. All the shopping list, right? Like Great I know I need food for dinner. I go to the store I'm like, I need a shopping list'cause otherwise I'm gonna buy like stupid shit.

39:44 So The model doesn't necessarily have to fundamentally work the same way as humans, but there is that challenge of like how do I train the model to operate over the context length of a human. How do I train it to interact with these databases and these Word documents that it writes to because it's never gonna learn that from pre-training. Has to learn that from an environment, but these environments have to be like architected in a way where the model knows it can write stuff down and refer back.

40:08 And so one of the first things OpenAI did was deep research. Everything is not in deep research's context. Deep research is working for like forty five minutes. It's outputting millions and millions of tokens. And it's creating this amazing And it's pretty good research.

40:22 I would say a lot of like memos that you read from people are like on par with like deep research. At least like a junior How they did that was they enabled it to be able to write something down elsewhere and have this Recall and And effectively use language to compress information that it looked at, put that off to the side, use language to compress other information off to the side, use language to compress other information off to the side, and then looking at all this compressed information and writing something.

40:46 That's sort of what deep research is. So How do models get there? I'm not sure, right? I think it's a fundamental research challenge. It's why these companies need millions of GPUs to train on. Not for Oh, I'm gonna make a million GPU model. But because I need to try a bajillion different things because I don't know what will work.

41:02 What's gonna work for humans is so different from what works with models. When there's any number of parameters or things you could tweak that could end up like Changing how it develops. And how good is it at if I do it this way versus that way, right? That's the whole point of ML research. Constantly trying stuff out and trying to get better and better. If I add all of this up.

41:19 And hold the mirror up. I would put you in the category of unbelievably bullish on what these things are gonna be able to do in ten years time or something like pick your time frame. Am I calibrated the right way? Like amongst everyone you talk to who you respect and think is I'm much more bearish than a lot of people actually, which is the crazy thing. So help me understand that distinction. Like where are you one through ten amongst the people that you respect, ten being the most bullish? And then if you're not a ten, what's the difference between you and the person who's a ten? I respect you but I know I'm way more bullish than you and I respect

41:49 Mark Zuckerberg, but I know he might be I don't know if he's more bullish than me, but I know Sam Altman's definitely way more bullish than me. He says we have Artificial general intelligence in less than a thousand days. Or Dario, like I respect him immensely, but he's way more bullish than me. My roommates.

42:03 One of them's an anthropic ML researcher, and one of them is another podcaster, Dwarkash. They're both way more bullish than I am. Even they are not as bullish as like some researchers in this field. But then if I go talk to someone I respect like a famous investor. Right. Like any of these famous investors. I don't want to name one because I'm scared of them. But there's all these famous investors, right? They're not more bullish than me. And the stuff I'm saying sounds like crazy shit.

42:26 Some of it though is timeline. I'm actually even more curious about the upper limit. The upper limit I think I'm among the most bullish you can get because That's what I mean. Yeah. Like upper limit of this is that this will just be smarter than humans. I don't think that'll happen like anytime soon. Even if that doesn't happen anytime soon, there's so much valuable stuff that can be done with these models that economically we will skyrocket. There's so much value that can be created in the world just by hey, if the models know how to do Cobalt to like C and Python migration of like mainframes. Migrate everything from mainframes to cloud.

42:58 The world is how much more efficient? Making all these random applications and like automated reports and like stop using Excel as a database, but instead like you can make a real database and and manipulate stuff in Excel, but like you know, there's all sorts of humongous business efficiencies that could happen or automation that could happen without the model ever being We could literally just pause it at like six months from now time frame of like how good it is at software development. And it would be like God send in terms of like how much efficiency and value can be created. for the economy and it doesn't ever have to get to like digital God level. Now I do believe

43:29 Eventually, eventually. Now is that is that is that Is that Ten years? Is that five years? Is that a hundred years? Is that a thousand years? I don't know.'Cause there's there's so many Unknown unknown.

43:42 I know how many unknowns there are, but like there's so many unknown unknowns and so Like I mentioned, right, like these babies are putting their freaking hand in their mouth to calibrate and then later they put their foot in their mouth. And they're like, Oh, that's my foot. Oh, here's the senses on it. And then they can pick up stuff in their hand and they no longer have to put it on their most sensitive part of their body.

43:57 Because they know what it is. Or they're like, Oh, this is a speck on the ground, what is it? It's not food. But now I know what it feels like inside my hands and I've calibrated, right? It's like The models have not gotten there yet. It has no idea how to do this.

44:09 Digital God is like Well, one, I like kind of believe in embodiment. Like you need non-digital God, you need a physical, and you need the capability of like having touch and feel and all that to truly Have an experience like humans and be smarter than us in every way. But you know, that's so far away. What do you think about what physical intelligence is doing?

44:27 Attacking the whatever you wanna call it, large movement model or large robot model or something. Yeah, yeah, yeah. I mean like what what are they actually doing today is like holy shit, it's so simple in terms of like to a human. To models, it's like picking this up is freaking hard. Like how much do I squeeze my pinky versus this finger versus this finger versus this finger? I don't know. Like but you pick up a glass of water and you tilt it and it's like This is impossible for a model today. And it's likely like at the level of dexterity, fills a wine glass and I'm swishing it. Think about how simple that is. You don't even think about it, but like you instinctually pick up a wine glass and you swish it and it lets the aroma out and you smell it, but it's like Oh, that little swish is so much tactile feedback and movement. And it's like these models can't do that shit yet. Like nowhere close.

45:04 But it doesn't need to be that good. It doesn't need to be able to swish a wine glass and not break the wine glass and put it back down and tilt it perfectly and not spill it. It doesn't need to be able to do any of that. to be tremendously valuable. What it needs to be tremendously valuable is pick this up and put it down here. After knowing what it is. So there's so much value that can be created just by being Really good at like getting data. Yeah. Yeah, yeah.

45:23 The robotics world is huge. I think we're Warming up. We haven't even left the dugo dugout, right? Like we're we're like nowhere close to the scaling on on robotics. There's a ton of like The data flywheel needs to get going there. One of the most interesting subplots of this whole world is the talent wars. And

45:39 A cool idea is that as these things get better, maybe we begin to automate some of the research function that people formerly would have played. Do you see a world where like we're squeezing down the fewer and fewer number of people that really matter that will have all the impact on where we go in terms of like net new research? And that means that all this crazy spending that's happening at Meta or elsewhere makes a lot of sense that maybe even those numbers should be higher or something like this? I think it's like tremendously hilarious that people are like oh my god this person's getting paid a billion dollars. Or like, oh my God, this person's getting paid a hundred million dollars. Hilarious to me that it is infeasible. It's like how could this person possibly be worth that much? Well, they're running the experiments on chips.

46:17 That cost A hundred billion dollars. If every wasted experiment they do, if they just used like a third of the compute and their ideas and their impact on it wasted the compute, it was an idea that was already done, or like There's so much wasted compute. Call it wasted, it's trying stuff and failing. But like

46:33 None of us know what to try and what not to try. And these things are so complicated. There's like a group of people just trying different stuff on the existing data. How do you mix it? What order do you feed into the model? How do you filter it? There's a different group of people that are doing what's the architecture, there's different people working on long context, there's different people working on every single aspect of the model. that like if you just make them a little bit more efficient that they come up with the idea that's five percent more efficient. Well, fantastic. I just save not only five percent of my training time.

46:59 I also say five percent across my entire inference fleet. 'Cause we're so far away from like These models being anywhere near as efficient as a human brain. And we know it can at least get as efficient as us. Maybe the compute substrate isn't the same, but like whatever. Adding more people to the problem doesn't make it faster.

47:13 Because there's so many Things you're trying. Лорнинг став. Is something that You run these experiments, you learn something, and then you implement it.

47:23 And then you try a bunch of experiments, right? You tweak these knobs these ways. in a hundred different ways, and then you see the trend line and you're like, Oh, so actually I should tweak it this way. Let's implement that. There's so much like gut feel, there's so much Reading data, understanding it, re implementing it into these things. And if you add people, you're gonna slow it down. And in a sense, a lot of Meta's problems before

47:43 they did the superintelligence thing was that they just had too many people that weren't led by leadership that was amazing and they had a lot of failed experiments and wasted time doing things that didn't matter. There's a tweet from one of my friends. Uh open AI. He's pretty famous on Twitter. His name's Rune. He made a tweet about like

48:00 I get viscerally angry every time I think about how many H one hundreds Meta's wasting. It's like it's such a funny tweet because it's like, Well yeah, they're wasting a ton of compute. They were, you know, maybe they still are, but you know, like Everyone's wasting compute, right? Open AI's wasting tons of compute. Cause what's the Pareto optimal model architecture of the house? Another thing I saw Rune say recently, which is so interesting was Why don't we just go make even more ridiculous offers?

48:23 around the people that have process knowledge for things that we want here in the US. in other countries. Like if we're getting pretty good at the Arizona fab that we've built. And we think that we can sort of extract the process knowledge from the people, why don't we like go aqua hire like all the best people in Shenzhen or all the best people in other places in the world. Do you think it starts to escalate to that level, like so much is dependent on the process knowledge of a relatively small group of people? the talent war should actually be it shouldn't be meta and open AI. It should be like

48:52 the US, maybe through Meta and Open AI, and people from all over the world, like do you think it starts to get that extreme and should it? That's almost a function of why Intel has fallen off a lot, right? Is like You have all these geniuses and nanochemistry and PhDs and all these like random things, whether it be chemistry, physics. All these like incredibly smart people.

49:12 But there's a whole class of incredibly smart people that never went that way'cause they're like, Oh, those guys are making like two hundred K Why would I do that? I'm gonna go to Google and make 800K. And now I'm gonna go to Meta and go make 100 million, right? Like any like smart 18 year old is gonna be like, fuck that. I'm doing this. Why do the smartest doctors and I don't mean to say the smartest doctors in a general sense, but their skews uh really smart population of doctors that want to be dermatologists and anesthesologists. Like, is that the most valuable thing for them to do? No, but those are the two professions that give you good working hours and great pay. Yeah.

49:42 Not to say that the general doctor is not smart as them, but if you took the population of general like family doctors the random doctor and you took the population of dermatologists, newest coming out of school, the ones that are being dermatologists and anesthesiologists are way smarter, or at least scored better, were able to get into the field that was coveted. So Yeah, talent wars. We've sort of been through this process of like

50:02 It's always been human capital and capital goods, sort of those two vying off of each other. And for a long time with mechanization, industrialization, We had The human capital decreasing as the industrial capital increased. That got to a point where especially in the seventies it really started to tank as the ability to globalize and and all these things started to really hit the US and and that's why you have all the lot of the population level dynamics and income inequality that we have today that like is very bad for the psyche of the US and stabi stability of it. But you now have like we're in such a age of

50:34 Actually, like manufacturing things is pretty commodity. Like most of the value doesn't come from the manufacturing of it, it comes from the creation of the idea. One thing Jensen told me, which I thought was like amazing, right? He's like, you know, Dylan The reason America is rich, like people have it all wrong. The reason we're rich is because we've exported all the labor, well, we've kept all the value. And that's what NVIDIA does, right? They've exported the labor of making their chip.

50:55 An Apple, right? Everyone. It's it's done in Asia. Those companies make money. Not as much money as Nvidia and Apple. Right. All the gross profits are going to them.

51:03 And then they're either reinvesting it or buying back stock or whatever, however they allocate the capital If like as you said, the process knowledge is so valuable, weren't we doing this? That's a great idea. Um, not mine. Yeah, I know. I mean I think I think the challenge is How to choose people. is really difficult.

51:20 Judging someone as smart or not. For some roles, someone who can talk the talk. They're great. People just automatically assume they're great. Because they can talk the talk.

51:29 But you know how many people suck at talking and are really frickin' good at doing, yeah. Yeah. But then you don't know. But then there's people who talk about being able to do Better than the person who's doing and like these tests are never as good. How do you select? And this was a big challenge for

51:44 Not uh So some of the criticisms are like they didn't get all of the best people, they actually got a lot of like bad people. It's like the Cope. From like OpenAI and Enthropic, and you know, these kinds of companies are like, No, no, no, they didn't get our best people. That's what Sam said, right? He's like, they didn't get our best people, it's okay. Meanwhile, he did have to do counter offers internally. As far as the process knowledge, I think the ML researchers are an extreme.

52:06 Of how much value one can do, but my favorite analogy. that it came up with recently. Is that ML research is the exact same as semiconductor manufacturing. In the sens of

52:16 There's a ton of jobs in semiconductor manufacturing that don't exist in ML research, but it is a ton of Tune A thousand different knobs. Oh, you put the wafer in this tool. You're gonna change the pressure of the chamber when you're doing the deposition. Oh, you're gonna change the mix of the chemicals flowing in, which chemicals you're putting in. What speed you do it at? Do you do it for thirty minutes, do it for thirty one minutes, do you do it for you know obviously it could splice this way down. There's so many

52:39 knobs on every single tool. And you have a thousand functions. And process knobs. Right. Process knobs on each tool, plus it's like the sequence of them all. You frankly cannot test everything. It's impossible. It's too large of a search space. Just like designing a chip is too large of search space. You have a hundred trillion transistors. How you gonna possibly try every single thing? Impossible. You just have to have enough intuition

52:58 Like pick that point, pick that point, pick that point, see the data. Oh okay, I think the answer is here. Obviously once you you think the answer's here, you test here. A different person might have seen these three and then said, Okay, the answer's actually here, not here. The data is like fuzzy. It's like somewhere in the center, but like, you know, it's like this this whole like idea of like ML research.

53:17 You spend a lot of time on compute training. Doing What effectively were useless things besides teaching yourself what's the right thing to do and what's the wrong thing to do. And semi-connected manufacturing is the same way. And actually all process manufacturing is the same way. If you're iterating super fast and you're trying to get better and better and better, or you're optimizing a process on a chemistry or whatever it is, you try, you fail, you learn, you do.

53:41 In semiconnector manufacturing, maybe it's just running tens of thousands of wafers. And so your RD cost of your main fab that is running the RD. is very very high And it's producing zero economic value besides that it's teaching you how to do the next node, which then you can deploy at volume, and that is what actually makes the money. I wanna go back all the way to where we started and ask about what I'll call like the wellspring or the fountain of power in this whole ecosystem. So I wanna understand how you think about

54:07 Who has the power and how to keep or generate power. As a business, I mean Maybe talent is like the very beginning of the chain and he who has the talent, like on a long enough timeline, has the power or something like that. But also there's structural stuff like just the industrial scale of some of these things that just takes forever to build or whatever. How do you think about Even smaller zoomed in examples like, okay, cursor is unbelievably popular, the revenue is insane.

54:30 So much of it goes back to anthropic, like who is the power in that relationships? How does that dynamic change over time? It just seems like the power dynamics are so fascinating in this world. And I'm curious where you think it comes from in the first place, like where it exists today and where it will go in the future. Well, you know, way back when uh monkeys were you know, before we became humans we were territorial, so you know And me having two bananas is makes me better than you. When we think about the power structures, right, like you mentioned a really interesting one. Does anthropic hold all the cards in this cursor relationship?

54:59 Cursor has like Nearly a billion dollars of revenue now. If you do current month times twelve. That's a ton. But their margins are what they are and they're sending

55:07 Most of it back to anthropic. Some people say their margins may be negative. I think they're slightly positive, but regardless, they're sending most of it back to Enthropic. The gross profit dollars are at Enthropic right now. Yeah. And but then Enthropic is taking Anthropic is taking all the gross profit dollars and putting them into compute for training. So then all those gross profit dollars are going to like Jensen laughing hilariously. Well maybe Jensen or maybe like in fr Amazon or Google who's sending it to Broadcom, right? Like the gross profit dollars are going to the hardware layer from all of this, for sure. Does anthropic have all the power? Like the common view is yes from a lot of people, but then it's like well

55:41 Anthropic only makes the model that's generating the code, there's a lot more in the system. Cursor gets all of the data. They get all of the users. They get how do they interact with this. Enthropic doesn't get that. They get prompt to send a response. Yeah.

55:53 No, they have plot code, which is like taking share and it's very different than cursor, but like They get prompt response and then like cursor is like, Oh well, I'm training embedding models on your code database. There's actually multiple models that I've made. I've made the embedding model, I've made the auto complete model. Oh, I can switch the entropic model to open the eye model whenever I want to. I'm only using the anthropic model'cause it's the best one.

56:12 Oh, and because I have all this data, maybe I can train a model Not for everything better than you, but for the segment better than you. And it's so it's like The power dynamics are it. It's weird, it's friend of me, right? Everyone's a friend of me, right? Same as opening Microsoft, the most crazy power dynamic that's going on in the world where they signed a MOU that said they had an understanding of like what the deal would actually be for them converting to for profit. Like

56:35 What is going on here? Like this sounds like the most non announcement announcement ever. The power dynamics of this all It's the most fascinating soap. Opera ever, right? Like one of my friends was telling me about K pop demon hunters. I don't know if you've heard of this. I have a nine year old daughter, so it's all I hear about. You've seen it a lot. I I had just heard about it and they're like, Oh, let's watch it. I'm like

56:54 What? Whatever. Um like it's like there's drama, but like this this real world power drama is way cooler than this like Which parts of the drama interest you personally the most? Like where do you think the stakes are the highest in the various like subplots? There's a few different ones. The Microsoft OpenAI one is

57:11 Absurdly interesting because At one point, right like twenty twenty three, it was like Microsoft's gonna own the world. Twenty twenty four a lot of it too. And then like H two twenty twenty four and Microsoft backed down a lot. They pulled back'cause Amy Hood and whoever else at Microsoft, Mip Sundar, whoever were like Maybe we don't need to be on the hook for a three hundred billion dollars we're not gonna build out three hundred billion dollars worth of compute for open AI. Like that's they can't pay for it. Yeah.

57:34 at least had to go through their head when they cut back. And so they paused a bunch of data centers. They said We don't need to be the exclusive compute provider. You can go to Oracle, it's fine. And they relinquish this power. Now Oracle has that deal.

57:46 OpenAS sends like twenty percent of their revenue to Microsoft or API revenue or something like this, and then Microsoft has this like 49% cap profit structure on OpenAI. And then there's like this whole like IP sharing, like this deal, it's like really hard to understand. the mechanics of the open A and Microsoft deal even. So you have this whole power dynamic and they're trying to renegotiate this. The whole deal is like, Oh, when we have AGI, you no longer have IP rights and it's like

58:09 If you ask someone twenty years ago and you put them in front of Chad GPT It's like this fucking AGI. Like it knows everything. And it can have a conversation. I can't tell it's not a human. Actually, I can tell it's way smarter than a human. But now it's like ah whatever, I can't do exposed. So the the thing the the bar always moves. No matter what the level of intelligence is. And for me it's gonna be like when the thing puts its hand in its mouth and it's like, Yeah, this is me, this I'm a human, right? Like, you know, that's sort of like the sentience, the consciousness of it all, right?

58:32 That's one power dynamic that's like crazy. Another power dynamic is the one around NVIDIA and the hyperscalers. Invidia's the king. All of the gross profit is going to them today.

58:43 Pretty much all of it. Sure, TSM G makes some, sure, SK Heinex makes some. They have to invest a ton in CapEx. Sure, Broadcom makes a bunch and yeah, Broadcom makes a ton of gross profit off of these companies, but like NVIDIA makes by far the most gross profit in the industry. It's not even close. They're king. And they wanna continue to be king and they wanna make sure GPUs continue to be most used.

59:04 But also they can't buy anything. They weren't even allowed to buy ARM when they were like a nobody. They were like pretty much a nobody on the grand scheme of things, and they weren't allowed to buy ARM. They totally could not buy any major companies. about my startups, like about a start up that I was like a seed investor in and like an advisor in But like they can't buy

59:21 A real company. So what do they do with all this cash flow? Sorry, but you're a loser if you just do buybacks. Like that just that's admitting that's admitting you can't get higher returns on your capital. Which is fine, like Meta, Apple, Google, they were mature companies for a while. Guess what? Those companies aren't gonna do buybacks ever fucking again, right? Or not like ever again, but for a while.

59:39 They think there's better ROI for their capital now. In NVIDIA If you look at Jensen, he's like he's always like flirted with buybacks, but like mostly he's been like reinvesting in the business. But you can't reinvest that much into the business. He's doing demand guarantees. He's doing like all this crazy stuff now. Yeah, right, right. He's using his balance sheet to win.

59:56 Yeah, try and win more. Which is an interesting dynamic. I don't know if there's ever been anything like this in terms of the anti competitive nature of this, where you Backstop clusters. Corre recently got a deal with NVIDIA where it was like they backstoped a cluster. Now, Corvey would have never built this cluster because it's for like short term demand and renting GPUs on short term is like a terrible business model.

1:00:16 You want to do long term contracts. And you want to do long term contracts to people with balance sheets, that's the golden goose. But that doesn't exist so much, so you do long term contracts of people who don't have a balance sheet, like OpenAI. But if you can't do that, then you'll do short contracts with people who do have a balance, right? Like there's this whole matrix of like who you run GPUs to, but from NVIDIA's interests, it's like You know what I really love is when venture capitalists fund a company

1:00:37 And then seventy percent of the round is spent on compute. I fucking love that. Right? And that's what's happening with all these companies. Whether it's physical intelligence They're spending a lot on like robot arms and shit too, but they're also spending a lot on compute. Or it's like any other startup that's raising cursor or whoever, right? And even if it's not directly.

1:00:54 It's indirectly going to GPUs. They love when people spend their entire round on GPUs. Will be really good is if it wasn't like a two year deal. Or three year deal. For that compute. If it was

1:01:05 You can spend seventy percent of your round on one training run. Leave a company with these ideas. Gather the data, do the training run. And then you have a product. And you show how good the model is, and then you try and raise again. That's what would be really great for NVIDIA, but no one wants to build a cluster who's predicated on that as a business model. That's crazy. So they have to backstop a cluster do that. Or hey, OpenAI might go to

1:01:24 Their own. Some ASIC from another company. They might even buy TPUs. They might even like go to Amazon. They don't really care. They're not beholden to Microsoft anymore. Trying to serve a product to a customer. Yeah. And they want to build the digital God and they want to serve a product, right? Make revenue, right? So

1:01:39 They don't have to go to NVIDIA. NVIDIA is the best option. But you would be really, really helpful is if I could But going back to earlier part in this discussion is I get the compute up front and I don't have to pay for the compute for the first year. It'd be really good if I could do that because then

1:01:53 For a full year I can do training, I can subsidize inference. I can do all these things that build up the user base. And then I can actually pay for it. I have a year of a gigawatt. To figure out a business model. Whether that is serving free tokens and then implementing

1:02:07 This purchasing purchasing stuff for the free user. A lot of that is like almost no fee initially purchasing and then slowly raising the fee over time. Or it's I have to serve this model at worse gross margins or negative gross margins initially, but then eventually I can serve it at positive gross margins because the models keep getting cheaper.

1:02:24 Or it's I train the next generation model that's so much better than everyone else, and then I'll win all the business for that level of intelligence because I'm the only one with a eighteen year old. You guys all have fourteen year olds. Who are working for you. They can do whatever they want with this allocation. It's not an allocation of capital per se, allocation of compute. They get to decide what they allocate that compute to. And NVIDIA's helping them by effectively front loading it if they can find a capital. And that company's like, Oh yeah, NVIDIA's backing this too. There's all these other things. It's much more reasonable for someone to say, Oh yeah, I'll pay the capex because I know the first year's already going to be paid, because you've got that investment from NVIDIA. What about the next four years?

1:02:57 If you ask a bunch of investors who are like students of economic cycles through history, like Carlo De Perez type stuff. They'll say that the concern is that every shortage is followed by a glut. And we always overbuild on long lead time big CapEx projects. And you've got multi gigawatt power being installed, you've got all this crazy stuff and semiconductors. At some point.

1:03:17 It just gets overbuilt. All the stuff we talked about earlier feels like We're not really close to that. Like there's so much fitting demand. If the models don't improve, yes, we will overbuild, right? Like it's pretty simple. It's like yes, there'll be like supply chain things where switches from one supplier to another and like that's a lot of the stuff we focus on at my company. But at the end of the day, if the models don't improve, we're absolutely screwed. And if this lasts another year and then it happens, like the US economy will go into recession. Straight up because of this. And probably Taiwan as well, and probably Korea as well.

1:03:46 Because there's so much build up and revenue flowing through To us for this. But When you look at these other things like The bubbles of the past, some of them were just silly nonsense, right? Like tulips.

1:03:56 Silly nonsense, right? Crypto. Complete Ponzi scheme, right? But then there's other stuff that's like This was real, right? Like the UK like spent like some absurd percentage of their GDP on railroads for like a decade. Six percent of them and crazy. Yeah. We're nowhere close to six percent of our GP, like holy shit. That was like okay, there's tangible but it's like oh well we over did overbuild because like how many goods are there to transport?

1:04:18 You must build these roadroads. To reduce the cost of transport so much because you have no clue when the demand stops. And you've overbuilt. And because there's ten people trying to do it at once, you're obviously gonna overbuild at some point. Same thing with fiber. And like a lot of the argument against this is like, well, no, but this time it's the strongest balance sheets in the world. It's the world's most profitable companies. They can st all pull the plug at any point. Microsoft the plug at one point before they're like, Oh shit, no, no, plug it back in. They recently plugged it back in. They're like, Oh, wait, we're starting, we're restarting this, we're going out into the market, we're signing deals with Nebus for GPUs. Like I don't remember how big the deal was. It's like nineteen billion dollars for Nebbius. It's like

1:04:49 If they had just not pulled the plug on their data centers, they wouldn't have had to do that and they wouldn't have had to pay those profit dollars to Nebus. But the Nebbius made the bet that the demand is there and they were right. When you think about this, it's like What is the level of demand where this stops, right? If scaling laws continue. Of course, there's a adoption curve, there's a pace, there's realities with capital, there's realities with supply chains, things take time.

1:05:10 adoption for businesses takes time. But If you like boil it down to it, it's like your demand for Thirty year old senior engineers at Google who know how to make and program anything

1:05:20 is effectively like I don't know what's fun to say infinite, but it's two trillion dollars of value. If I could have an intelligence as smart as a Google senior engineer. That's two trillion dollars of software value. Because that's how much I pay the world pays to software engineers today.

1:05:33 And you just go down the list of every other use case. If you have just a simple physical intelligence robot that can recognize headphone versus water and versus phone, right? And pick up the right thing and manipulate it properly and put it in the right spot and sort it. That's worth How much to the distributions supply chain, right? I don't know, but a lot. We don't need to get digital God for there to be immense value.

1:05:51 But the interesting thing here is that you know it's like human capital, capital goods. All of these other revolutions have been capital goods that reduce the amount of human capital you need. Whereas this is just creating human capital. In a sense. So I like get everyone bowled up.

1:06:05 We're on this podcast. I don't know if you've heard the curse, right? It's like uh if you talk about the stock on this podcast, it goes down, right? It's like we're pop of the bubble right now because the limit of AI is infinite. For the record, we went and did the math one time because I was sick of hearing about this curse. And it's just market performance. It's not your other podcast. I talked about applied materials and the stock was up like 70% the six months after. There you go. Yeah, I broke the curse. What do you think about all the companies in the middle? We've talked a lot about NVIDIA and then like people at the end serving applications. What about these companies like Together and Base Ten and Fireworks and you mentioned Nebus, like all these interesting middle layer players? Are there amazing businesses to be built there, do you think? Are they temporary patchwork to make the system work and serve demand? Like what do you think of that? The cloud business model, right? Like uh let's let's say neo cloud business model. So there's sort of you mentioned inference providers and neo clouds. The neo cloud business model. is absolutely amazing or terrible, depending on how you do it.

1:07:02 It's terrible if you like sign Short term contracts, and you just hope and pray you have short term contracts forever. And actually, initially your short term profits have amazing cash flows. You bought a GPU and you put in a data center and the power and all that, the cost per hour over a six year period. For Blackwell is two dollars. Let's just call it for simplicity's sake, it's two dollars. It's not exactly that, but And if I sold it for six months I could get like three fifty or four dollars.

1:07:25 Holy shit, that margin's insane. But what happens two years from now, three years from now, when I'm still selling six month contracts or one month contracts and the next generation of NVIDIA chip is out and it's ten X faster for three X the cost. Okay, so now Naturally the price of this should tank. The only way to do it is Hey, I I have a long term contract of

1:07:42 The other end of the spectrum is what Nebbius just signed. I'm signing nineteen billion dollars to Microsoft. They will pay me no matter what. They're The market literally believes Microsoft will pay its obligations before the US government'cause it's like literally a cheaper bond rate, which is like insane to me, but whatever. This ninety milyen dollars Has a huge gross profit.

1:08:02 And it's not exactly three dollars and it's not two dollars, but like the margins here are really good. Nebbius is gonna make at least six billion dollars of gross profit off of this. I would do that all day. And Core We did. Until Microsoft stopped going to Court Week.

1:08:17 But Corbez turned around and they found other customers and all these things selling to Google. And selling to open AI. But now oh open AI is you can't rely on their balance sheet. I still have amazing margins when I sell to Open AI. but they don't have a balance sheet. So how can I be sure that they're actually gonna pay the thing that they've signed up to?

1:08:33 In theory, this contract is worth a ton of money. And Corey's books today are all the contracts they've signed are mostly Microsoft. Mostly money in the bank. Open eye contracts like what if they can't afford to pay for this? Okay, now there's like there's a bigger risk. And then there's a longer and longer tail of like these businesses. So like Yeah. You absolutely can make a ton of money. There have been more recent deals with crypto miners, Google and Fluid Stack,'cause Google's

1:08:53 really short on data center capacity. People want to use more TPUs, they can't reserve them all themselves, so they're gonna sell TPU systems to Providers. They're back stopping the deals with Terra Wolf is one of the companies. I can't remember the other one, but there's two companies they've signed deals with where they're

1:09:08 Backstoping the data center. plus like selling the TPUs physically to another company and then they're being deployed and then they're getting rented and Google still makes all the money, but like Those companies, you know, yeah, that's great as well. But then there's a long tail of like, is the enterprise demand there? Who's taking the risk?

1:09:24 And it's like OpenAI is taking the risk'cause they're betting their entire company could go bankrupt if it doesn't come. Oracle's taking a risk because they're signing up for three hundred billion dollars of contract, okay, two hundred billion dollars of hardware spend. across data centers and chips. And of that.

1:09:36 They're gonna have to go get debt. So they're on the hook and they'll probably be able to pay for it if it happens, but they'll just be like Their P V will tank. If Opening I can't pay for all the hardware that they bought. Luckily for them it phases in over time and

1:09:47 And then you go to the inference providers and it's like There's a business to be made here too, right? Like I'm serving models, I'm serving them efficiently. Maybe Roblox comes to me and they want to put an LLM in their game. Another company like Shopify wants to put an LM for customer service and Yes, they could do it themselves, but actually inference is a hard thing, especially as you get to larger and larger models and more complicated models and all the things or all these different use cases where people want to serve models and maybe it's just open source models and maybe it's fine tuning of those open source models, which those companies can help you do, or you can do and they can serve for you and they have scalable, reliable capacity.

1:10:19 There's businesses to be made here, but there's also like I'm selling tokens to random people who are trying to build SaaS apps and N SF and maybe they run out of runway. Okay, that funding doesn't directly go to NVIDIA, but you go through some steps and it's going to NVIDIA for some value chain. Yeah. And NVIDIA's holding no risk, everywhere in the middle's got a lot of risk. Back to the other s side of the equation, the app side.

1:10:41 stuff we're gonna use these models to do. At the significance of this switch from like deterministic Code. to a much different thing. And it seems like what we're doing is the thing we always do. Apple used to call this like the schemorphic era, where you just Basically use the new technology to do the old thing you used to do. So we're making engineers better.

1:10:59 That would be like the obvious current example. Seems like we haven't yet gotten into the world where we're gonna start using this this technology to do things that we couldn't do before with deterministic code. I'm curious how you think about that side of like pushing the envelope. Wha why is that? Like I feel like

1:11:14 That's exactly what we do with it, right? Is the cost to develop things Is so high that you can't do it. Or the cost to like have someone go buy stuff for you. It's like okay, great, you might have an executive assistant. And you can tell them to like Do this.

1:11:29 But like the vast majority of people don't. And now GPT's on the cusp of doing that. Go buy this for me. And they'll find the best thing and they'll buy it. Right. And you just trust them enough. It takes time to trust them. But like these things Tech is the most deflationary thing in the world ever, right? In terms of quality of life. It's

1:11:44 It gets cheaper way faster than the revenues go up. But the revenues still go up. That's sort of like the fundamental basis of semiconductors, of tech, everything. Are we doing things that we can't do before with tech with AI? Sure. I mean like The Covid vaccine was

1:11:58 Created with AI. Yeah, drug discovery. There's like entire briefs about How it was done with AI. And guess what? If like another pandemic happened, I bet it'd be even faster to discover the vaccine. If there's a vaccine for it or whatever.

1:12:09 There's all these protein folding things. There's all these Optimization things. There's AI for material science and AI for all these other aspects of society. There's optimization, maybe it's not in your face. Right. It's like Oh my God, AI just made this drug.

1:12:23 No, I mean AI worked with the researchers who made the COVID vaccine and so we didn't have to all be stuck inside forever or whatever, right? Point being we're it's already happening. Right. And the whole like use the new thing to make the old thing faster, it's like Sure, but if I go back three years

1:12:38 How many people would it have taken To deploy A image recognition model. That looks at every data center in the world and looks at what's the pace of construction's in and what equipment they have. This is something you do. This is something we do, right? It's like it's like how many people would that have taken? I don't think it would have been possible. My business model, like this is the second highest revenue product for us.

1:12:57 Would not have been possible. If it weren't for AI vibe coding, like being able to dig through permits and regulatory filings, being able to run image recognition on satellite photos, like this would not be possible. This business is not possible without AI. Am I using it directly? Like sure I'm scraping through the regulatory filings and permits through it with LLMs and then manually reviewing it with people on

1:13:20 Doing that the same with like the images, how images, yes, there's a lot of stuff that, you know, the image recognition model does, but we just also look at them a lot. And then it's like compiling them and selling a spreadsheet that you get bi weekly reports on like all the data centers or what's changed or like, hey, actually this Amazon data center, the fans are starting to spin. So actually there's revenue going on from this Amazon data center. So we can forecast Amazon's revenue. It's like, oh okay, like this is relevant. I don't think this would have been possible.

1:13:43 Just a few years ago. There's demand for it because everyone wants to track this and it's so important, but it's like it begets each other. And I think like, at least in my daily life, I just like I don't think I could have taken that step from where I was in a business, which was still a research provider. But like that is a monumental jump. And like being able to do it with Three people out of the gate?

1:14:00 Versus like fifty. Or a hundred. Like I don't know how many people it would have taken, but I don't think it's possible. And it's like mainframe migration is something people have always wanted to do. Amazon leading Oracle took fucking twenty years. And they wanted to do it 20 years ago, and they have their highest revenue products after EC two. The next four were like database products at AWS, and yet they still freaking used Oracle's database. Because it's hard.

1:14:24 Mainframe migration can be way faster, or like or migration from one tech stack to another can be way faster. You can make your business more efficient, add more automation. I think As far as like yes, the tech exists. Go to all the businesses around the world and it's like They aren't using the leading edge of what they could. They aren't using what a twenty twenty company could have done without AI.

1:14:44 No one is doing that. And if they did, they'd be so much more efficient. All of these things just take too long to build, they're too expensive to build. You have your existing processes. How do you hand them over? How do you switch them over? How do you teach people to do this? AIs can help you with all of this. You can take the pessimistic view of like Oh, we're just doing the same things, but it's like the value here is humongous.

1:15:02 If it's tokens on one end, we haven't talked much about like watts at the very beginning and power. What are your thoughts on like what is going on here and how humanity is responding to this crazy new demand for just raw power? The first approximation is that it's not that much power yet. data centers are like three, four percent of the US power. Data centers period. Of that two is

1:15:22 regular data centers and two is like AI data centers. So like That's nothing, dude. Like that's literally nothing. It's just we haven't built power in like forty years. Or like we've transitioned from coal to natural gas more and more over forty years.

1:15:32 It's mostly we just don't know how to and there's these regulations and like There's not enough labor and like the supply chains for like Gee Vernova and their dual combined cycle gas reactors are not there yet. And same for like Mitsubishi and oh like this random UV curing process for transformer coils is like There's only this much capacity and it takes two years to build'em. It's a supply chain thing, it's a like lack of labor thing. It's not that it's actually that much yet.

1:15:57 At the end of the day, it's like, Okay, wait, wait, you're telling me OpenAI is making a data center with two gigawatts and that's like the entirety of the power consumption of like Philadelphia. That's insane, right? But like In our Slack. We used to get like excited about finding like a couple hundred megawatts new data center.

1:16:10 Now it's like if it's not a gigawatt The guy who leads that team He was like, Oh, it's just five hundred megawatts. I also agreed immediately. Then afterwards I was like, Wait a second, dude, that's like a lot of power. That's like how much wait, five hundred megawatts is twenty five billion dollars of CapEx. Like, come on, like it will once you put in the GPUs and everything, right? It's like that's a ton of money, but like snore because there's so many of it happening. And so when you think about what happens to

1:16:33 the country, what's happening here is like We're learning how to build power again. We're getting the supply chains to do it again. We're reshaping the grid. There's all these challenges with these AI data centers. with regards to demand response and making grids unstable.

1:16:48 AI workloads because they change so much so fast, especially training, you can just cause like brownouts. Or blackouts. Especially if the grid doesn't have enough inertia or if you're not putting enough things to dampen it in between the workload and the grid. And even if it's not destroying it, the grid runs at like fifty nine hertz or whatever, right? If you skew it up and down too much. these transient power responses, your refrigerator will break down sooner.

1:17:11 The motor's in it. And you might not even know it because the data center's nearby. So like All these things. Because maybe it's not like turning off the power, but the hertz is not perfectly fifty nine. It's like oscillating too much and then the motor in it and the coil windings and all this is burning out faster because of this variation. Like a car, right? Like if you were to floor a car on floor, instead of just like accelerating slowly, like think of it the same way. You're flooring it, on letting go, flooring it, letting go.

1:17:33 There's so many like third order effects here. The funnest one is just that We're building Power, right? And it's like Whether it's

1:17:41 Gas, which is a lot of it. Whether it's through efficient dual combined cycle reactors or it's like random generators that are not nearly as efficient, single cycle, or even worse. Diesel generators, there's company that's putting a bunch of truck engines in parallel. Diesel truck engines, because the industrial capacity for diesel truck engines is huge.

1:17:57 And no one's tapped it yet. So why don't we just put a ton of them in parallel? And create this power generation thing right here, right? And then you're generating power with a bunch of diesel truck engines in parallel and then you're able to power a data center, right? Like, okay, great. Because I can't get turbines. There's all these like crazy things people are doing.

1:18:12 Elon buying some power equipment from Poland and shipping it to America because he needed that power equipment, but Whatever, couldn't get it here because the supply chains were weird. I'll just get it over there. Any lax capacity in the supply chain is Being eaten up immediately.

1:18:26 And then everyone's like, Okay, let's invest. So GE's like I'm gonna double my turbine production. Like, holy crap, okay, that's awesome. Admit to Bishi's doing the same thing and you know, you go down the list, it's like my transformer supply chain is expanding like crazy and like fully sold out so I'm gonna go to Korean guys.

1:18:40 And that's fully sold out. So I'm gonna figure out how to get the training stuff in, even though it's not exactly like what people want to do, right? It's like there's all these. Weird things. But the funnest one is that like electrician wages have like doubled for mobile electricians that can work on data center stuff. If you're Down to move to West Texas. It's like two thousand fifteen again and like being a fracking guy.

1:18:59 You don't need to be super duper skilled. You can go to West Texas and make a shitload of money off of fracking. Go to the Permian. But there's not enough of those people. That's why, right? Like if there were enough electricians in West Texas, if there were enough electricians in America, we could build these data centers faster. All these little supply chain quirks and everyone's supply chain is different because the way Google makes their data centers different from the way Vantage makes their data center, which is different from the way that EdgeConnect makes their data center, which is different from the way QTS makes their data centers, which is different from the way Amazon makes their data centers. The other two companies I mentioned, like rent data centers too.

1:19:29 The hyperscalers mostly. All these companies make data centers differently and so their supply chains are not exactly the same. And so you get all these weirdnesses in all these different supply chains. And it's really fun, but it's also No one really knows it because everyone who tracked the supply chain or like knew it. You go talk to like power people, it's like on one end of the spectrum is like Dario.

1:19:49 And then you take a few steps and it's like the average ML researcher, then it's like Me, and then it's like you in terms of how bullish we are on AI. And the guy at the power utility's like over here. There's like a few more people. There's like the standard New York semis investor, then there's the New York Not semi's investor, and then there's like

1:20:04 the Sequoia guy who thinks that A has been a bubble since twenty twenty three. And then there's this utility guy. You know, this utility guy is like, I'm not building power. Power doesn't go up. Whatever. And then you have like the regulations around it. How can I build a data center in this density because

1:20:19 Now all of a sudden the grid's like so this has happened in Texas or caught and it's happening in P J M. Northeast kind of area ish. These two grids are putting these rules where Hey, we're gonna actually say, Hey, big loads. We can tell you 24 hours or 72 hours beforehand, we're gonna cut off half your power. Which is fine, right? Because like We need to'cause we need it for something else. Yeah, like people people need to have their homes powered.

1:20:39 So anyways, like it's like in Texas and in PJM, you can cut half the power if you give them a notice. This is SB six in Texas, and PGM's considering it. If you do that, then you need to turn on the generators that are there on the site. It's often diesel generators, maybe it's gas. maybe it's like hydrogen stuff, there's all sorts of weird stuff people try to do just to ramp up power.

1:20:58 For that period of time, but then all of a sudden, oh crap, the density of my generators means that I fail the air permit. If I run the generators for more than eight hours a month. So now what do I do, right? It's like there's all these like weird regulations, even if it's like Texas. There's like so much interesting stuff happening because We've decided to build again.

1:21:17 It's really fun that uh Watch it and then watch the supply chain and try and like, from my perspective, provide the data so people can trade on or provide the data so people can adjust their supply chains. industry wise, right? Um and you know, people who Go to your audience, they can trade on it. If I were to line up all the stages of this between the US and China, so you know, power, semis, models, applications, et cetera, where do you think the most interesting differences are? Like what are the storylines between us and China at those various layers of like the AI stack?

1:21:47 That are the most interesting to you. When you look at China, they're very formidable Competitor. I think if we didn't have the AI boom. The US probably would be

1:21:58 behind China and no longer the world hegemon. By the end of the decade, if not sooner. I think most people would have agreed with that. And a world where the US is not the hegemon is A bad one for Americans, at least.

1:22:10 You know, I'm sort of like a bald eagle like carrying like America and like it's bad for the world. But you know, like that we're not the world hegemon, because then we can't spread freedom. Without AI, like we're definitely just gonna lose. Our supply chains are slower, they cost too much. Our debt is like unsustainable. Our economy's not growing fast enough to maintain the level of debt.

1:22:30 We're over consuming relative to what we produce. Um, the financialization, there's All this like dirt. in like the US in terms of like social instability, partially because of income inequality, but Also largely because of the visual nature of income equality and the tendency of people to flaunt their wealth more because of social media and how that hacks people's brains. And then also like because the algorithm serves people different content, we're drifting further and further apart in culture, right? monoculture of everyone watching the same movies in the fifties and forties and thirties versus like now

1:23:02 You and I are pretty similar and our feeds are completely different. So think about someone who's not in this world. The feed is like insanely different. Our memes are different. I think the wheel us would literally fall apart if we don't do something. And by do something I mean like

1:23:15 AI has to dramatically accelerate GDP growth. Because once you start talking about dividing the pie, you're screwed. It has to be growing the pie and The US really, really needs AI. China's view is like I think it's like a little bit different, right?

1:23:30 They Don't necessarily need AI to win. If AI doesn't take off like a rocket. Whatever that is, even if it's not AGI, it's like you can do the software two trillion dollars of value of software helpers, right? Whatever it is. China was always played the long game. They always like, yeah, we're gonna really screw over our consumers by

1:23:48 Taking all this wealth and dumping it into EVs and losing money on gaining market share. And the EV industry has still not created positive market value for the Chinese economy yet. Absolutely will. At some point. I mean, I think it hasn't created it yet. Maybe it started to, but like Rather we wanna like kick all of the European and American and Japanese and Korean car companies out of the global

1:24:10 Automobile market. And they're doing that. They've always played this long game. They did it with steel. They've done it with like Rare earth minerals, they've done it with solar panels, they've done it for producing phones, they've done it for P C Bs they've done it for so many freaking industries.

1:24:23 Incrementally, they're just gonna continue to do that and then they're gonna win because they work harder and they're on average smarter, right? I would say on average in the sense of their government is more intelligent in the way they allocate capital and the way they think about things. We're smart in the sense that We get to brain drain the rest of the world and we get to maybe allocate capital more efficiently because of our more free market capitalism in some ways, maybe

1:24:44 The whole point here is that I don't think most people have as pessimistic of a view as this as I do. I also view like if we don't have super powerful AI systems. We'll run out of Easily accessible, like

1:24:54 nickel and cobalt and oil and natural gas, and we won't be able to make solar panels efficient and fast enough. And everything will just start to get more expensive and the pies will reduce and we'll also tear each other apart in that way. So I have like a very pessimistic view that if we don't accelerate, we die. If that's your worldview, then like we really need to win AI. And China's worldview is sort of like, you know, they want to be the world hegemon, like who doesn't want to be the world hegemon? But there's only two countries in the world that can legitimately do it and legitimately are trying. Right, the US and China.

1:25:21 The way like the Chinese AI ecosystem thinks about this is We don't necessarily need to have the biggest compute cluster. They're still not Pilled enough on that. When open AI is trying to make a two gigawatt data center full of G B two hundreds and G B three hundreds.

1:25:35 And they'll have another five gigawatts of Vareruben or whatever, right? Like all these different chips. And those chips are way faster than the chips that will sell China. So that's the chips China can make themselves and China's deploying less of them. The dearth of compute is huge. We're kind of doing what China's done historically, just dumping tons of capital into something and the market becomes interesting. And the beneficiary is like, Oh, if OpenAI you know, they have eight hundred million users today.

1:25:58 When they have three, four, five billion users across the world. which is possible of chat GPT and whatever applications they come up with. Then they can start to make money, right? It's sort of like YouTube lost money forever, but now it's the platform for watching videos across the world, right? And ChatGPT will be the same thing. So sort of that like aggregation theory. argument, but then there's also the other argument of like, you know, what if we're wasting the capital. China doesn't necessarily think of it the same way.

1:26:22 But they are still incredibly pilled on like, well, we wanna be able to make everything ourselves. Right, so make all of the chips ourselves. We don't actually care that much about making all the chips ourselves. Sure, Trump's doing the tariffs, sure, we have the Chip Sact. Those were drops in the bucket compared to how much money China's releasing into the semiconductor ecosystem. Has been for the last ten years. They're made of China plan in twenty fifteen, five year plan.

1:26:44 And then the five year plan in twenty twenty, and then it just continues like they've dumped At least like four or five hundred billion dollars into this ecosystem. through SOEs, through certain tax policies, through certain like land grants, through provincial governments through The big funds.

1:26:59 Which is like government. Venture funds. They're just called the big fun. There's all these different ways that they've dumped capital in the poly, whatever you want to call them, whether they're the government, PLA or not. Like I think there's an argument there, like So they've dumped so much more capital into semiconductors than we have.

1:27:14 in an unprofitable way because they want to build that ecosystem. And over time, you know, it's like if you take any country in isolation China is the one that has Everything at the highest level on average, right? Sure, they're like thirty years behind on jet engines or twenty years or ten years, whatever it is, but they don't need to go outside of China for any of the materials besides like raw materials. Whereas like the US needs like titanium from here and like, you know, blah blah blah from there, right? And the same applies to their semiconductor ecosystem.

1:27:41 Sure, the US and Taiwan and Korea are way ahead. Like three, four years. But then they also have the accumulated capital base of all of the existing equipment and all of the existing fabs, but they need to import from all these different places because it's a global supply chain and so China's like much more concerned today about being insular than being the best in doing this like aggregation theory.

1:28:03 Which is sort of maybe what chat GPT is kinda doing. I wouldn't say OpenAI has fully piled on that view. And there are also like a lot of people who view the like, Oh, we're gonna make AI that makes AI better, that makes AI better, that makes AI better, and is A GI. The other view is like, you know, hey, we're gonna make AI so much better that it can do software engineering and we have trillions of dollars of value. China doesn't fully believe in those things yet.

1:28:22 I think. On a total basis. But because they're so talented. And they have an insular supply chain. Yes, they purchase some stuff from the foreign world, they rent stuff. They have bite dances.

1:28:33 I think the third largest user of GPUs in the world. After open AI. And probably meta. although Bite Dance may be bigger than Meta, the third largest user. Of GPUs in the world, or second, maybe even Byte Dances. They have all the other major Chinese tech companies, they have all of these amazing graduates. They don't have a talent war.

1:28:49 Companies don't poach from each other. Deep Seek engineers make a lot more than other engineers, but they're not making ten million dollars, even though they may be worth it. There's this like real big Perception. Difference.

1:29:01 China could build. Way faster than us. If they wanted to build a A two gigawatt or five gigawatt data center, they could probably smuggle a lot of chips. If they wanted to build a ten gigawatt data center, I bet they could build it in like a few years.

1:29:13 Whereas the US is not gonna build a single 10 gigawatt data center for a while, right? Like the total capacity of an open AI will be like 10 gigawatts in a few years, optimistically. China's not they don't have the best chips, quali's speed rating. Trying to get better and better and faster and faster. They don't have the best memory. They're trying to get better and faster there.

1:29:29 They do have the most power. They can build stuff way faster. We're impressed at how fast Elon does stuff. Elon's slow compared to China. And I think he knows that. Which is why he's maybe like the one who's like actually using the Chinese ecosystem more in terms of like the battery facilities making in China and

1:29:45 He probably recognizes it too. There's like these major differences in like viewpoint and approach. And I think we could dive into like any part of the supply chain, but the important thing is like philosophically we're so different. on what we're trying to do here. 'Cause China wants an insular supply chain, they want to have supply chain security. We talk about wanting that, but we don't actually put the money behind it.

1:30:05 the slot machine of where the American capital's being allocated. It's building the biggest data centers. It's training the best models. Yeah. Whereas in China the capital's being allocated into growing the EV supply chain, growing the semiconductor supply chain, catching up in all these areas. The US sure we want to catch up, but Maybe Jensen was right. That like what you wanna own

1:30:23 Is the end customer Thing. They're doing the same thing they've done forever, which is like prepare at the base level and be behind at the customer side. But then like you get to the point of like, okay, well what happens in like three, four years, even if the US AI is amazing.

1:30:41 The doomsday scenario of like China decides to blockade Taiwan or even invade it or create some political instability. Or try and so like people talk about like Cambridge analytics and like Russian trolls, whatever, like China could do a billion times that into Taiwan, especially with how good AI is now, and somehow subvert it. or coup or blockade or whatever, and we no longer have Taiwan. US economy kind of free falls'cause we can't make refrigerators. Without Taiwanese chips. We can't I mean they're American company's chips, but made in Taiwan. We can't make

1:31:13 You know, cars. We can't make AI data centers. We can't grow any of the cloud. That means we can't deploy any more SaaS applications. What the hell can we do? Back to going to acquire all the talent. Get'em over here. Do not have I I think like that that's sort of like the catch twenty two of this all is like if you push China too hard, they totally will, like

1:31:31 You back someone into a corner, they're gonna start swinging. Or if you just bump into them at their shoulder and then keep walking, like it's fine, you know, like you pickpocket them, you run away, like it's fine. But if you push him into a corner, it will like blow up. But then there's all these like dynamics. Like if you don't push them a little bit away from having the chips Because they have all this talent.

1:31:50 Probably half the AI engineers in the world are Chinese. Whether they're immigrated to America or not. You look at the list of meta's superintelligence, like it's like they push like eighty percent Chinese people, right? And like how many of them are from China versus being ABCs, right? American born Chinese. Like you could look it up, but it's like push China too hard. They have the talent. They could go crazy. They could If we no longer have Taiwan, actually China could build a way bigger cluster than us, and if compute is all that matters They could do all of these things.

1:32:15 And they own the means of production for everything. There's this like Challenging aspect of giopolitical risk. Well that's why people don't want to invest in T SM C, but it's like almost like You can't invest in Amazon or Apple or Google or like

1:32:28 Microsoft, if you If you believe Taiwan has risk and so it's like YOLO, invest in TSM C. I know a lot of people's PMs are like, Oh, you can't invest in TSM C because geopolitical risk. And it's like no dude, you can't invest in fucking Apple. Who is your favorite AI bear? Like Someone that

1:32:46 is far distant from you on Just their perspective on the direction of this whole thing that you nonetheless like and respect. The most. I'm having a hard time because everyone I like think of is just like more bearish than me, but they're not like bears. There's some of the like AI researcher like gods.

1:33:03 Yan Lacan and like these kind of people who are AI bears. I I respect them. I like Their ideas. I think they're completely wrong. And their argument is what? Like if you had to summon you know, the ways we're doing this won't work. LLMs on scale or it's like okay, yeah, yeah, auto regressive like it's like okay, auto regressive pre-training on the internet doesn't work to get you to AGI. Not gonna lie, a lot of people believe that. Yeah. Maybe I believed it like twenty twenty three, right? Like let's be clear, like or twenty twenty two, like

1:33:28 Or not AGI, but like super powerful AI systems. That's obviously not the case. Just pre training on web scale internet will not get you there. So he was completely right on that. But then like he'll turn around and be like R L systems and all these things are not the right way either, right? Like you know it's sort of like The no buts.

1:33:42 There's some investors that I know who like think this is bullshit, but they're just making tons of money on it anyways. Think it's bullshit in what sense? Like it's like like it's an overspend and this is a dumb way to do it and like Yes, I bought Oracle before earnings because We see these deals happening whether it's through our data or some other means, they saw these deals happening with open AI. But we don't think OpenAI can pay for it, but we know the market's perception will be this and therefore the stock will go up. And so we'll own it.

1:34:07 There's investors who like kinda have that viewpoint and they're like kind of roofless. I guess I would respect them to some sense, but it's more and more the level of evidence that's there that this stuff is gonna get super powerful Hard to not, right? Again like the AI bubble is going to pop because it's podcast, man. I assure you, it's just a market return. It's a great talk.

1:34:28 Startups interest you the most. Like real startups relatively young not to market with their product yet, like that sort of thing. I'm especially interested as potential acceler to all of this. Like people that are Attacking some interesting bottleneck. So one of the startups

1:34:43 It's the most recent investment I've made. It's called periodic labs. Mostly open AI people. It's a Google guy and you know the couple of material scientists. The area of AI that we've all been talking about is like large scale web training, RL, all text, all digital God. But you know what would drive a shitload of value for the economy besides automating programming of everything is like If we just like came up with a battery chemistry that was like twenty five percent more efficient.

1:35:07 Like holy shit. You know, like the main cap against like us all having like face glasses and things like that. It's like batteries are not good enough. And the power dissipation, but the batteries like sick terrible. So you have to do make all these compromises. But if I could have the processing power of like a laptop. On my face.

1:35:24 We'd be way further ahead and like and then if we all had like these like super powerful machines attached to our face, we could do inference on things and recognize and interact with the AI at much higher speed and velocity, and and that would like dramatically improve our productivity, right? Like things like this are like so gated by Hard tech. moving faster. And so what periodic is trying to do is they're taking this RL paradigm, but they're trying to do it with like Real world, right? Like Test chemistry for something.

1:35:49 Here's an optimization, here's something that the model spit out. You can do it in like all the like CAD type programs, not not exactly CAD, but like computer aid design programs and test it, simulators and test it. But then you also want to test it in the real world and then feed that feedback back into the model. And so you do this like chain of Instead of Purely being Digital.

1:36:08 Which is why, right? RL is like really hard because you need to generate a bunch of responses, test, and then train the model. So the flywheel was so freaking fast. The flywheel in the physical world is so slow. You mean I need to make a chemistry. I need to try this. I need to test the thing. I need to input it back in and keep calibrating and keep doing this. It's so much more expensive. It's so much harder to do. But actually there's a ton of low hanging fruit there, I bet, in terms of like large scale experiments, whether it's Drug discovery is one angle, right? But design of like

1:36:36 Well most complicated chemistry ecosystem in the world is semiconductors. There are tons of things that we know. Or the next thing to build. And now the next five years of work is actually building it. Right. And it's like next next five years is actually getting the chemistry to work, getting it to be a low enough yield, a low enough defectivity. Or it's we can't we don't know what material could do this.

1:36:54 Do superconductors exist? Do they work, right? Like these are the sort of things that are like the flybridge. We should be investigating, yeah. Yeah, yeah. I think that that's that's one. I think like in the hardware world, like just in the pure hardware space attacking some other interesting bottleneck. When we talk about like Where we are in tech. Semiconductor manufacturing. is super space age, right? It's like the most complicated tools we make in the world.

1:37:15 including like tools that cost like half a billion dollars. Right? Like and and they're super, super like amazing feats of engineering. Then the software behind them all is like really shit, right? So it's like, you know, you could you could celebrate all that. Um In the hardware world, the biggest challenge is that like I'm not really a big bull on the accelerator companies. I've never been. Defined how accelerator companies. Right, like companies competing with NVIDIA. I'm competing with NVIDIA with TPUs, with training um, with AMD. Not a big bull on those kinds of companies. There's just too many things to do. It's too capital intensive. There's not enough of a revolutionary leap. There's too many predicated things.

1:37:49 I wish it could happen, right? It'd be fun. Maybe it does happen, but it'll take hell of a badass thing. But I think there's a lot of individual parts of the supply chain which are not space aged. NVIDIA Space Age, yes, it's the biggest value owner today. But their supply chain has so much

1:38:04 Hold shit. And whether it's their supply chain or the hyperscaler supply chain, transformers have not changed in like fifty, hundred years, right? There's a guy building one company in that space. Solid state transformers, right? Like things like this. Yeah. So there's all sorts of interesting things there. on the power delivery side all the way from super high voltage A C all the way converting all the way down to you know 0.8 volt or 0.4 volt DC that goes into the chip. There's so many interesting companies in that space because there's so much innovation to be done. And there wasn't that much of a need to do innovation before.

1:38:33 You don't get a high return on it. Now you do. Yeah. Another area is like networking between chips because As we extend context length, the memory requirements become bigger and bigger. And yes. New memory technologies would be awesome. But D Ram is an industry has so much invested capital, goods, so much

1:38:49 So existing factories, it's really hard to attack. But Networking is less so. And there's more breakthroughs that can be done in networking. That

1:38:57 Okay, maybe you don't have better memory technologies, but you've tied the chips closer together so you can use each other's memory on the problem. So the simplest one is like N VL seventy two from NVIDIA Blackwell is them condensing a lot of networking into one, you know, sort of the chips are networked together. But there's so much more that you can do in terms of the optic space, bridging the gap between electrical connectivity and optical connectivity. Cause like NVIDIA created Blackwell, they had a ton of manufacturing problems and challenges with it. for their supply chain.

1:39:24 Balance sheets went up for, you know, various companies in the supply chain were building servers and stuff because they're trying to figure it out. AI data center deployments were slowed because of these challenges, there's reliability challenges, because these things are connecting to each other at absurd bandwidths. Every chip in the rack could connect to every other chip in the rack. Uh one point eight terabytes a second, right? Uh that's that's If you think about how much data that is, right? Like

1:39:45 The amount of bandwidth like It's so Hi. For connecting these chips together. Like you can't fathom what a terabyte a second is. You can't fathom what a gigabyte a second is.

1:39:56 Oh, okay, a gigabyte a second is like a video, right? Like or like a megabyte a second. But actually that's a million bits of information. A byte second. Okay, you can understand what a byte second is, because that's eight bits. Okay, I'm transmitting eight bits to you back and forth every second. That's pretty fast. That's what it used to exist. And it's like where we are now, there's still tons of innovation left to be done there. I think in the hardware space, I think I'm super bold up on

1:40:20 Advancing the techniques Like I think part of the reason Intel is behind is also that data sharing internally was terrible. Just within the fat. The lithography team doesn't want to share their data with the etch team and that data can't leave the fab and go to an AWS data center to run correlations and all these other things. So you don't learn from the experiments you do fast enough. Now TSMC is not perfect here either. They won't send their data to a cloud either, but like, you know, this experimentation experiment.

1:40:47 Analyze the data. Figure out the new experiments. Cycle. It's slow. And how you break that is actually like

1:40:54 Changing these companies' culture. Trying to do? But also it's a lot of it is like building better simulators. Simulating the world more accurately.

1:41:03 And so there's all these world model companies. And like world models that that dives into the software realm again, right? But actually some of these world model companies are actually just focused on So so world models generally are like Hey, I'm going to simulate the world. I'm going to walk around. In The common one I think is GE3 that Google made, right? Where you can walk around and at Cape State and you walk around the world and you can like see cars driving and like they're talking about like, but it's like actually what a world model could also be is just like simulating molecules, but not through classical methods, right? It's not computational fluid dynamics.

1:41:32 It's the model experiencing this enough and then running training a model on physics and then feeding that back through and doing it through a AI method instead of and so world models can be doing any sorts of things. You can make a world model to train robots how to pick up cups. Or you can make a world model. That is Simulating some chemistry in a chemical reaction.

1:41:51 Or a fire, right? Like you can do all sorts of different things. So there's a lot of world model companies out there. Some of them are really interesting, especially when they're targeting the physics and reality of the world. I love robots, robotic stuff as well. I think there's so many like venture scale investments. But also I think like Most of the cool innovation is just happening at big companies or already existing companies, right? That's just the nature of it all. Actually, TSMC is doing the most cool innovation and NVIDIA is doing the most cool innovation and like Anfinal is doing cool innovation. It's like all these companies are doing cool innovation. Could we do like a uh quick speed round where like I say some company just give me like

1:42:25 You know, a sentence or two on like your impression of them. How you feel about them in this moment. Yeah. Start with open AI. Oh yeah, super awesome. That's it.

1:42:34 I mean we've talked about'day, you know, it was gonna be like random. I'm actually more optimistic on anthropic than I'm open AI. Why? their revenue is accelerating way faster because what they're focused on. Is

1:42:46 more relevant to that two trillion dollar software market. versus open AI is split between, yeah, they're gonna do that. They're also gonna do these other things, but they're also gonna do like target AI for science. And they're gonna also target AI for the consumer app and doing the like take rate thing, which all of these businesses could be amazing. And OpenAI maybe ex executes on all of them. Yeah. But Anthropic is definitely executing on the software side better. M D.

1:43:08 I love them. But they're pretty mid. Can they not be men? Why do you love them if a man? You know what you when you grow up like Building computers and like liking computers and like AMD's innovating and they're always like fostered this underdog mentality against Intel and against NVIDIA. Evil Intel and evil NVIDIA, you know, and like AMD's like, you know, the nice company that's like the underdog. And like they're always they've always got the oh, they're gonna take sure from them. It's like it's hard not to love them.

1:43:36 You know, like and I know so many people there and I like the I like all these major hardware companies, right? There's not one that I don't like as in terms of the people, but like A and B's got a soft spot because like I mean that was my first multi bagger as well. Buy it like two dollars. Like my first multi bagger, I was I can't own stocks anymore because of compliance, but like you know, like it was my first multi bagger. Yeah. Right. In terms of like I bought it at like two, and then I sold it at like ten and then I bought it at like nine and I sold it at fifteen and then I bought it at like nine and sold it at like fourteen and then I bought it at like ten and sold it at like thirty and bought at like fifteen and sold it like eighty. Like I just kept doing memories. It was like it was like great. I mean I was like I had like no money. Yeah. It was like it was like it was like high school and college and like the initial stages of working, but it's like you know

1:44:16 Sorry for the rant, but I fucking love A and D, you know? I also love NVIDIA. But X A I. They're in a real danger of not being able to raise capital. Of course everyone's gonna give Elon capital, but like The scale of capital required.

1:44:29 For him to keep up, he can get the next bet. He can get to Colossus too. This mega data center that he's building. Largest data center of the world when he built it. Three hundred thousand five hundred thousand black wells, right? Like it's gonna be really great.

1:44:41 But if he doesn't figure out a business model besides like porn bot. Which is what Annie is. Which also I think he's monetizing the wrong way. Like I think you could monetize it so much better. How? You've captured the zeitgeist with a cute anime girl.

1:44:53 that talks to you in a cute voice and like We'll rise you up. And and you've got like these users who actually fall for it. And it's like not realistic enough yet, but it will slowly get more realistic. You're selling like outfits for the same price. You should make it a random like, Hey, you have the chance to buy the outfit of her like looking like this one anime girl from this one anime.

1:45:12 Hey, you have the chance to buy this outfit that's her in a nun suit, and also we'll change our personality. It's like instead it's just like you can buy an outfit. But I think like he has to figure out like some business model beyond just this. Although I think this could be a big business. Like he he should partner with OnlyFans and make manifestations of the OnlyFans creator that are Annie. And then he subsumes the only phone's platform into X. The Everything app.

1:45:35 And it'd be like XXX. Like you know it's like you could you could just like Trojan horse like OnlyFans away'cause like the discovery mechanism for OnlyFans is like Instagram and Twitter, as far as I understand. And like you ha own one of them. And you could partner with the biggest OnlyFans creators to like get them over, right? They don't have to respond like they can also just like train a model that like acts and looks like them and talks to them.

1:45:56 Anyways, like there's all these different monetization methods that I don't think that's what you should only focus on, right, to be clear. My point being, and sorry, this is way more than two or three sentences. Ex AI. Can get to the next stage of compute. They won't have more compute than OpenAI, they don't have more compute than any individual company, Google, Meta, et cetera, but they will have the biggest individual data center.

1:46:14 They'll have a very focused team. And what they do with that. They have to do something. Like really big, otherwise they will fall behind in the race and Elon will not let that happen. Like he doesn't want that to happen, but he can subsidize and fund this round.

1:46:25 Well like As rich as he is Among the richest people in the world, maybe the richest. He can't go to A three gigawatt data center.

1:46:33 Unless he gets capital, which he can't do unless he gets revenue and fundraising. Oracle? Oracle's gonna make so much fucking money if you believe if you believe open air is successful. Then the question is like People who loan money to the neo clouds are idiots.

1:46:48 Like why would you ever loan money on a cluster? Either you get paid on your loan or you don't. You should own equity. Like people who loan money to Cor Weav. only in the early days are so dumb they should have just bought equity. If they believe the thesis, they should have bought equity. And then those who bought equity made tons of money, right?

1:47:03 And the same applies to like owning Oracle. Like maybe, maybe, maybe. They're gonna make a ton of money. If OpenAI is successful. But if you think open AI is gonna be successful enough to pay Three hundred billion dollars to them.

1:47:13 How many users do they have? And what's that IP worth? Like maybe and also like, you know, there's reasons you shouldn't own OpenAI like the Microsoft. stuff and like the risks around enthropic and all these things. But like In most worlds where Oracle gets paid. Three hundred billion dollars by opening eye. OpenAI is like a ten trillion dollar or five trillion dollar company or something crazy.

1:47:33 Well end of the old last generation best two business models, first being Meta. I think Meta's got the card to potentially like own it all. I don't know if you've seen these new glasses that they came out with the screen. As we go through the history of computing You have

1:47:47 It was like punch cards. programming. Then it was like DOS terminals, right? And then it was like oh you have GUIs and mouses and keyboards. Then you had touch. And the next Paradigm in human computer interface.

1:47:58 Is We don't actually have to touch it at all. We tell the AI what we want. And the AI will translate that into reality. Whether it's Hey, send an email to this person, send a text to this person. That's basic stuff. Then you can already do that of Siri or whatever, right?

1:48:12 But like oh go buy this. You're we're so close to all of these things, the input method into a computer changing entirely. And The only company in the world Who has the full stack.

1:48:22 From good hardware that is What Meta just showed with their glasses with the screen. That Plus the good models, plus the capacity to serve them, plus the knowledge and know-how around recommendation systems to know what content to put in front of the user. It's not just generating the content.

1:48:39 It's not just interpreting the user's word and taking actions, it's also putting the right content in front of the user. It's all four of these that you need. put in front of the user. Plus the capital. Plus the capital. I think Meta is so close to being the only company that can do that. But there's a lot of risks there too, right? So I I like that a lot. No. Google, to finish it off. It was pretty bearish, Google, like

1:48:57 Two years ago, but I'm I'm like Super bullish Google. Why I would change. They're waking up on every tri They're taking the TPUs, they're selling them externally. They're taking the their models and they're actually like competitive on them and they're training much better and better and better. They're being aggressive on infrastructure investments.

1:49:12 Um, there's still a lot of dysfunction throughout the company, you know, but they do have the hardware business that they can pivot into this. They won't be as head as Meta is. They won't be as good as Apple is, but like they do have Android. They do have YouTube. They do have like all these IPs they've search that can come together when we turn to that next interface of consumer, but also they can also dominate the professional sense too, potentially. Where's Meta, I don't think can dominate that professional sense. Only the consumer sense. And I think Google's well positioned to go Capture both markets.

1:49:42 Or a meaningful share of both. I feel like we've covered like an incredible amount of ground. Is there anything that We haven't talked about that you feel is like really critical to what happens in the future. That we didn't cover?

1:49:54 I think the question of everyone that I constantly get asked is like Okay, Dylan. You know, you're you're you're lucky. Your obsession is that you loved hardware and you like followed it and you followed the supply chain. And and you built this business on it, but like you don't follow the software side nearly as much and all the value's gonna get created there. When is that coin gonna flip over?

1:50:13 But I think the thing that most people don't realize I'm not saying hardware's gonna dominate forever at all, but like Software is not. The same as it was five, 10 years ago. You've had dramatic changes in software, and the business model is going to change as well. Go back like five years, three years, whatever when Sass was the darling. Let's call it October twenty twenty two, I think was when SAS peaked. Multiples peaked in April twenty one.

1:50:37 April twenty one, okay, okay. Multiples. Multiples, okay, okay. I just remember November twenty two is when shit like hit the fan. The the NASDAQ high was twelve thirty one twenty one. It was November. Okay, okay, it was November twenty one. Sorry. November twenty one, I remember SaaS started tanking. And at the time it was like mostly like they were overning and all these other things. Doesn't matter. The interesting thing about the business model is that it is such a good business model. When your R D is sort of this, it stays flat, right? And you grow a little bit, but really R and D doesn't flex that much. Your cogs are super low.

1:51:05 The flip side in a SaaS business your customer acquisition cost is quite high. And so when you look at what like certain companies have done when they've acquired a business is they've just crushed the customer acquisition cost or crushed SGNA. They made the business amazing. Whether it's like Broadcom with VMware and stuff, it's not really customer acquisition. They just had a bunch of wasted SGA, but like This SGA, this customer acquisition, that was most of your cost. RD was small, but not like crazy. And then once you hit critical mass, you just cash cash cash. Money, money, money, money, money. But software changes a lot.

1:51:36 When The cost to build that software that you have. Tanks like crazy. You look at non-US markets and the prevalence of SaaS, it's very different. I will bring up China as an example and a counterpoint. China.

1:51:50 Doesn't have that much of a SaaS business. Actually, their cloud business is pretty small too, right? Relatively to the US. Despite them importing tons of CPUs and storage historically, right? Most people just did stuff on prem and designed their own software because the cost of developing software in China was so much less than America. Vat the SAS business model didn't work. as well. People could just build rather than rent it out and buy it. And that creates inefficiency in the market. I'm sure those aren't the best of breed solutions always. Anyways,

1:52:17 That's what the software development costs may be, like software developers and Twenty fifteen in China were getting paid maybe one fifth of the US and they were maybe twice as good or something like that. So 10x lower cost of software. I'm I'm making up numbers. You know, they had ten x lower cost of software, and so SAS. Never happened. Cloud never happened. And at least as big of a way as it did in the US. And around the world for all the companies that have that same economic reality. And that's despite the outsourcing, right? To India and and Eastern Europe and to South America, et cetera.

1:52:42 You changed all of this with AI software development, right? And AI SaaS products generally. Not just AI software development. So there's two sort of coins here. So AI software development tanks the cost of building a competing software stack. Do you now move to a world where ex check can just build instead of buying slash renting? And the cost is very low.

1:53:00 For most functions, I can just vibe coded out. That level of what you can vibe code out at very low cost source. And I have a way more software developers in the software engineering market. Like there's a ton of people out there who don't have jobs. And so like that's one angle, building the SaaS product is easier. Two is

1:53:16 If you are a SaaS business and your customer acquisition cost remains the same, and most businesses in AI and in SaaS are gonna remain having a high customer acquisition cost sales is hard. Breaking into a cumbancy is hard. But now you add this AI part of it. You've now added a humongous cogs. Your cost of goods sold in any AI software is really hard.

1:53:38 And really big. And this is partially why I think Google also has an advantage. They have the lowest cost of goods sold for any token. Of any company because they have their own vertical stack on TPUs. But anyways, coming back to this, because you have this high customer acquisition costs and you have this high COGS. And then the cost of anyone developing it themselves or competitors in the market means you're gonna have a very fragmented SaaS market.

1:54:01 Or they're just gonna build it themselves, and therefore you never hit the escape velocity where your customer acquisition costs and your RD get amortized, and because you have such a high COGS. The amortization point means your net profitability. is actually much worse. And so I think The era of like software only businesses is really, really tough in the age of AI. Now Already scaled businesses can do great.

1:54:22 I think YouTube is going to have its glory days. And I'm sure it'll it'll always be amazing, but with the cost of generation of content falling and falling of and creating content, he who controls the platform is gonna win and win and win and win. The functionality you build within Salesforce is actually gonna be like way less, or like what you can build on your own. Like or whatever it is, I'm not saying it's a take on Salesforce itself specifically, but I think many software businesses will have a reckoning. with the fact that their COGS is gonna soar, their customer acquisition cost isn't gonna fall, and they have a lot more competitors. And so then they don't hit that escape velocity. And I think that's the thing that Maybe software. It's something I've I've like thought about. There's a couple of people in my company. Douglas O'Laughlin, he's the one whose idea this actually is. But he told me on this, along with some of the ideas I've had on like the cogs of tokens and tokenomics and way.

1:55:09 There are different people on my team whose idea it is, but I gotta steal it and say it. Something most people don't recognize. Yeah. Yeah, the fundamental economics of software has to change. Yes. And it's strange. This has been incredibly fun. I I love, love learning from you and listening to you and reading what you put out. I think you're just one of the most

1:55:28 energetic and awesome thinkers in this whole space right now. So thank you for all the work you've done. When I do these I ask the same traditional closing question, what's the kindest thing that anyone's ever done for you? It have to be my brother, everything he's done in my life. I've been a shithead my whole life and I still am a shithead. Every time he like pulls me back on path, he corrects me, he loves me unconditionally. I think my brother is probably the most he's done the kindest things for me, right? And I've been an asshole like so much of my life, right? Like unconsiderate and like everything, right? He's just always been there for me.

1:56:00 And always been like. Why? Yeah. If you're aware of it, it makes it interesting. No, but what it is is like I don't know, like the and maybe this is like the MO of like who I am and maybe that's why I'm like a good thinker, but like I I like Vibe really hard. And I'm in the moment really hard and I digest tons of information, but I'm very like bad at like Task orientation, remembering to do specific things. Like I'm very bad at those things. And thankfully I've like been able to surround myself in my life, whether it's through birth or not.

1:56:32 with people who help me with the things I'm bad at? Because I'm very bad at a lot of things. When I don't like call people or like be considerate of what they're thinking because I'm just vibing and I'm doing whatever, you know, I'm like focused in on like this path. That path ends up hurting someone else, right? Whether it's like, hey, I didn't call someone or I didn't like think about their feelings when I did an action or when I said something, but that makes me an asshole.

1:56:54 And yes, I should be more conscious of this. And I try to be, but it's like it's just one of the things I'm gonna wrestle with in my life. Forever. And a lot of times I don't even realise I'm being a frickin' idiot. Until my brother's like, You're a freaking idiot.

1:57:05 That's the kindest thing anyone's ever done for me as like my brother through my whole life. I love it. I love it. Wonderful place to close. Thanks so much for your time. Thank you so much. Yeah. If you enjoyed this episode, visit joincolossis.com where you'll find every episode of this podcast complete with hand-edited transcripts. You can also subscribe to Colossus Review, our quarterly print, digital, and private audio publication featuring in-depth profiles of the founders, investors, and companies that we admire most. Learn more at joincolossis dot com slash subscribe.

What you see above is a preview of the first minutes. One unlock costs 10 credits and covers this episode forever: full segment and word-level timestamps on this page, plus .txt, .srt, .vtt and word-level JSON downloads, as many times as you like.