I'm not crying because of the AI researchers leaving, I'm crying because Onion died, I promise.
I swear to God, if Uncle Roger finds this video, I'm gonna cry.
This is about to be sweeter than Panda Express.
Hey guys, welcome to In Context Cooking, where we take one dish, taste it and try to recreate it with minimal help.
Today, we have a very special guest, the founder and CEO of Semi Analysis, Dylan Patel.
Welcome, Dylan.
Hello, thanks for having me.
Yeah, thank you for being here.
I guess the question to start off is how would you rate yourself on a scale of 1 to 10, of 1 being awful?
10, 10, 10.
So you're an amazing cook.
Lux, lux, lux, lux, lux.
As a cook, how would you rate yourself on a scale of one to ten?
Probably like a five or six.
Okay, five or six is not bad.
Maybe three.
Let me revise.
So low expectations and then over delivering.
The presentation is great, just not the plate.
Okay, honestly, we'll work with three or five, whatever.
But yeah, I guess in front of us, we have a bunch of ingredients.
Do you have an idea of what we're going to make?
I feel like it's kind of obvious just looking off of it.
Eggs, rice, chicken.
This is like... very typical refrigerator stuff, leftovers.
But then the right side, you know, the peas really throw me off.
The ginger, though, and soy sauce.
Okay, so this is like fried rice.
Okay, yeah, so today we have... chicken fried rice, which we'll be recreating.
This is restaurant chicken fried rice, so we'll try to recreate it as close as possible, based off tasting it.
So here's a spoon for you.
Cheers.
Cheers.
Mm.
That was very good.
I'm shocked it's still warm.
Yeah.
I assume this has been sitting here for an hour.
Yeah, we nuked it a lot and then kept it as is.
So, yeah, just to get started, do you want to introduce yourself?
I know that you went to Minnesota briefly after, and was it you were a beekeeper for two years?
You did a lot of these kind of, like, side quests.
You have, like, a lot of lore.
But now you're the leading voice on chips and everything, whether it be hedge funds or even people in AI.
So I did live in Minnesota briefly after college.
I'm from rural Georgia.
Yeah, I did beekeep for like a year and a half, basically.
I feel like I've just sort of went through a lot of life, just next step, next step, next step.
Doesn't seem like there's a clear, immediate path.
Yeah.
Looking back I can spin a narrative like oh, obviously I would be doing this because my interest when I was eight was this and my interest when I was 12 was this, but moderating forums related to chips.
But I thought it was just a serendipitous thing.
But then eventually, everything culminated and blogging and doing, consulting and doing research, being interested in AI and data science, being interested in chips.
And then it all sort of like culminated in like, oh my God, everything blew up all together.
And so I guess right time, right moment.
Maybe some foresight to have your passion be the thing that everyone cares about now.
Yeah, no, I think it turned out very great, you know, doing amazing.
And rumor has it, you started your Substack because you read Doug's and thought you could do a lot better.
And he was like, hey, you should start a Substack.
Is that true?
Or is there more behind that story?
Yeah, so what happened was I had an anonymous blog on the internet for many years.
I was moderating Reddit and all these things anonymously around like hardware NVIDIA Intel AMD, this kind of stuff.
I was posting all this stuff.
And I had an anonymous Twitter in Silicon Twitter right, which people on like Teapot and like tech Twitter don't understand.
And so I was doing this anonymously.
Doug started posting and I was like well, this is interesting, but like I think I could do a lot better.
And then he's like, dude, why are you like posting on WordPress, right?
Like do it non-anonymously, do it on Substack. and start charging for it.
But then I was like too, I was like, I'm not gonna charge for this, I'm so great.
And then one day I was like you know what?
Screw this, I'm gonna start charging for it, because Doug told me to many, many times
And then a few years go by, and then Doug joins the company.
So it was really a great moment that he told me to do it, because otherwise I might still just be like, a niche anonymous blog where I'm still just doing random consulting rather than actually, you know, a company with 60 people.
Gotcha.
So at that time, you're still doing consulting, right?
And you just had a separate blog?
It was consulting related to the blogging industry.
Gotcha.
Okay.
Yeah, that's amazing.
I guess now we could kind of look at the ingredients and maybe try a little.
So here we have ginger, garlic, carrots.
We're just going to eat the whole garlic.
Yeah, we could all take a piece.
And then this I think is, and we also have tasting spoons.
I think this is sugar.
Yeah, so sweet.
Okay, let's see, i really hope this is a salt.
Okay, it's sugar.
Yeah, so this is sugar and we'll know, because everything will be laid out for you exactly like this.
Um, so we'll have to eyeball and we'll be going fast this.
I don't think we need taste.
I'm pretty sure this is baking powder.
Oh, What kind of fried rice does have MSG?
Yeah, that's true, that's true.
What's the YouTuber?
Uncle Roger, yes.
But yeah, we have soy sauce here, baking powder, sugar, onion, of course.
If he discovers this, he's going to roast the fuck out of us.
Yeah, he will, but hopefully, you know, he'll give us some grace.
But okay, and then eggs, salt, yeah.
Wait, is that the tactic?
Intentionally don't put MSG in your rice.
Yeah.
So rage bait Uncle Roger.
Yeah, this is also short grain rice, which is also a red flag, but we just have it.
And this is day old.
So this is the important ingredient to have it day old, not fresh.
Why does that matter?
It's a little drier and it'll make the fried rice, just like the grains, be a little more separated compared to being a mushy mess.
But yeah, I think we're good.
Are you ready to show off your cooking skills?
Sure.
Great.
Let's get started.
Dylan, are you ready to make some fried rice?
Absolutely.
Okay.
So the first thing, let's grab the chicken, and then we just need to marinate it or velvet it, rather.
So take the cornstarch, it's one of the white powders that we saw before, and the baking powder.
And then add like a spoonful, maybe half a spoon of cornstarch to the chicken.
And then just a hint of the baking powder.
And so, yeah, once we add that.
A pinch or what are we talking?
I just did a pinch or a very small amount with my spoon.
And then we can add a little bit of salt and soy sauce.
I think this is salt.
Make sure it's salt and not sugar.
Yeah.
Okay, there's a little bit of salt.
And then soy sauce.
And so what we want to do is we want to mix it kind of vigorously so it just looks like You said salt and soy sauce.
Yeah.
Just like a little bit of each to give it some flavor.
So we're gonna marinate it.
And yeah, we just wanna mix it vigorously so that it looks like a slop of chicken. but i guess one question on my mind that you've all discussed about is end game scenarios with arrakis or in this case taiwan to ngsmc i guess like have you thought about what that would look like bro wants to make fried rice and talk about china politics let's go indeed that's vaguely racist Just vaguely, just vaguely.
Vaguely racist, yeah.
Okay, so endgame scenarios are kind of insane, right?
There's a variety of things that could happen.
Yeah, such as?
So in some cases, right, like it's like, you know, status quo is the best, right?
You know, no war happens.
I'm done with the chicken, by the way.
No war happens.
And just start cutting the onion to like small dices, but continue.
We have no war, no invasion, there's no blockade, sort of status quo.
China continues to industrialize itself with the industry that Taiwan has, but there's no major event that occurs.
That's sort of option one.
Option two, which is what a lot of people seem to think is the best, which is Taiwan actually, more stringently, claims it's independent.
At least a lot of Westerners.
And then that seems like actually a poor option, just because that potentially causes China to move much more aggressively on Taiwan.
There's another set of options, which is like, okay, well, what if Taiwan moves closer to China?
In a political way, right and so like.
That's an example of that is like hey, like they elect the KMT And the KMT ends up winning.
So there's two parties in Taiwan the DPP, KMT.
Yeah, and If the DPP wins, they have won for the last decade and they're sort of more pro-US, more anti-China, more Taiwan independence.
And then there's sort of like the last scenario, which is just a full on invasion.
Yeah.
Or at least a political coup or takeover of some sort.
And so there's like a variety of options possible.
What I think is most likely is that there is some sort of political coup or action that destabilizes Taiwan in some way, but doesn't necessarily lead to an invasion, a full scale invasion.
And so sort of this is the best of the both worlds for at least China, right?
Like they don't have to actually enter an invasion, but they get to continually creep more and more on Taiwan without actually having to deal with the repercussions of a war and subsequent blockades, and so on and so forth.
The like kind of galaxy brain thing for an American to want is for actually the Taiwanese government, the pro-US party, to lose right.
So you want KMT to win.
And if KMT wins, then that means China will be placated more.
You don't have Taiwan fully move into China's orbit, but you do have the government sort of placated in China.
At the same time, even if the KMT wins, it's not like TSMC starts disobeying American export restrictions, because the way American export restrictions are upheld is that Taiwan utilizes American banking systems, American equipment industries, and so they'll still have to uphold any us export restrictions, so sort of like.
You know, china's placated by the fact that they have a friendly government in power, and yet china doesn't actually have any of the chips and the us continues to get to access the chips I guess for expert controls as well.
So I know that right now there's like stances.
For example, Dario right is very anti-China having access to things.
And I think supposedly one of the top researchers, Xunyu, also left because of that.
There could be maybe a lot of- He left Entropic?
We left one of the top labs.
Got it, got it.
I guess do you have opinions on like, if the US takes, How much onion should I dice, by the way,
You can just cut a dice, like, half.
Okay, okay.
And then, after you do that, just start cutting the carrots to, like, a similar size.
Do you have any thoughts on the danger of maybe a lot of Chinese talent fleeing due to the very staunchly anti-Chinese stance that a lot of these AI labs or top figures may take?
Or do you think this is a non-issue?
You know, I'm not crying because of the AI researchers leaving.
I'm crying because Onion died, I promise.
No, but it would be a travesty, right, if a lot of Chinese researchers left American labs.
You know what, I think it's probably like or at least a third of researchers at labs are Chinese.
So obviously there's like a level of like, you know, why are you, don't antagonize too much.
At the same time, there is a level of like you know, this is this is the greatest technology to ever fall into humanity's hands.
You know, obviously we think we're the good guys.
And by we, I mean Americans.
And so Americans think that they should control it.
And maybe maybe Anthropic thinks they're the good guys, not Americans as a whole.
But whatever the moral justification is whether it's hey, we're the good guys and AI's gonna be super powerful and we're the only ones who can steward it, or hey, AI would be a great weapon, so let's make sure we're the only one with that weapon.
There's certainly some level of control that needs to be had.
Now the question is, where does that control start and stop?
Right.
Because one could say, OK, well, let's just control the chips or let's just control the AI.
But then it's like OK, well then you let them buy the chips and they're able to.
They're able to do everything they want to do anyways.
Yeah.
Right.
And this is this has existed across many technologies.
Right.
You know China, China's great engineering.
You only remove one piece of the puzzle.
They're able to re-engineer that last bit of the puzzle.
You know, that's one context, right?
And so that is like some people's argument, right?
So, for example, if you look at Nvidia or you look at like David Sachs, I think their argument is like you know hey, we should not let them have the models, but we should let them have the chips and everything else, because then they're still relying on American ecosystem, American talent, American technology, American platforms.
The other argument is like hey look, this stuff, if they don't have access to our models, but they have access to our chips.
That alone gives them shoot.
That alone gives them all the control and capabilities they need to basically be on par or just slightly behind right.
And so we see that in the current regime today, right?
China has effectively great access to chips, not as much, not completely unfiltered access, right?
They either have to rent it, or they have to smuggle it, or there's some chips that are allowed.
And China's not that far behind, right?
You look at Kimi K25 agent swarms and it's like well, this is not worse than Codex by a marginal amount.
Maybe 5.3, right?
But 5.2, it's not worse than Codex by a large amount.
Mm-hmm.
And so there is, there is like an argument to be made that like hey, current regime does not have the US leading in AI by enough.
Right.
And that's sort of the argument that, like Adario would make, I think, is that look like current regime of export controls.
China is still way caught up.
China is still not that far behind, and it would be disastrous if they are in their eyes, right?
And so the question is, how does one write that circle, right?
So one could either A antagonize China more, ban more things.
Does that risk a Taiwan evasion, or does that lead to something say catastrophic?
Does that alienate researchers in America?
That's one argument.
Another one is like, well look, we've got like two years, right?
You know, AI-27 bros.
I'm not fully there on terms of AI27, but I'm pretty bullish on AI generally, and so then the other agreement is like look, we only have a few years until the capabilities of AI accelerate GDP growth, and it's by the way.
Am I cutting something else?
Oh, you did mash sticks, huh?
You can start cutting.
We need to mince the garlic and then the ginger.
But the garlic and the ginger has to be a lot finer.
But yeah, continue.
I felt a little awkward just talking and not doing anything.
So we're just cutting all the vegetables.
Yeah, sounds good, sounds good.
The one stance that a lot of folks have is that, yeah, let's be more anti-China.
Cause there's only a few years left.
And so, like you know, until like super powerful AI systems, and those super powerful AI systems will make next generation AI systems.
Right.
And that's like.
That's like a lot of the um argument that like, let's say, the Dario's of the world would make.
Um, and I'm sympathetic to that argument to some degree as well.
Right.
Um, if we start looking at like Hey, um This year, Amazon's spending 200 billion, Google's spending 180 billion on AI infrastructure, primarily right.
This is 4X what they were doing just not too long ago.
And if they get returns on that of any degree, then we're talking about trillions of dollars of economic value being added in just the next handful of years.
And so the risk here is that, like okay, whatever AI is capable of doing, it's obviously adding hugely to the economy.
You wanna do anything and everything you can to slow China down.
We've never had an explosive growth like this.
We're on the cusp of it right now.
The AI industry maybe does 50 bill of revenue.
But we're seeing it explode, right?
Anthropix adding two, three billion of revenue a month now, versus they were just adding a few hundred million of revenue a month earlier.
So clearly we're in the takeoff period.
And so the argument there is that like, let's just limit them completely.
So they don't end up with all these super powerful AI systems.
I'm a little bit like.
You know it's hard to rationalize.
You know, in every specific like argument, just because, like I could argue or I could steel man any of the arguments, right.
Actually, okay.
No, maybe China should be sold our chips because, at the end of the day, they're still relying on our chips now for their AI systems.
And they have less incentive to invade Taiwan.
And if they invade Taiwan, then all of a sudden the whole party stops and we can't do anything.
And when you think about what's the capability of China versus the US to to have a vertical supply chain in chips or AI or in really anything.
China has by far the most advanced supply chain in semiconductors, if you just look at China itself, right.
In which case, if the lifeblood of AI is compute to some extent, to a large extent, then China would win if we didn't have Taiwan, right.
Yeah.
And maybe that time scale would be way longer, because Taiwan is so far ahead in the production capacity and China doesn't have the equipment ecosystem like the rest of the world does.
But at the end of the day, that's exactly sort of the argument that one would make is hey, if you push China too far now, they might invade.
They might invade Taiwan, and that ends up with this catastrophic scenario.
Double-clicking on something you mentioned already, with these big AI labs and even hyperscalers, somewhat overextending on future spend with these data centers, Would there ever be something that would prompt you to get a little concerned?
Because it seems right now we're in a very acceleration type of moment for AI, where still adoption for what even like cloud code and a lot of these great tools, isn't as mainstream.
But are there things that if you see happening in the next what six months or year, that would kind of concern you in terms of where the future lies of these companies, maybe overextending a little too much, and whether concerns of being in a bubble may actually have some merit to it?
I think sort of not answering your question but talking about something else I want to talk about, which I love to do.
You mentioned these doomsdayers and these believers.
I think the biggest risk is actually just the general public hates AI.
Right.
You know, I think, I think, if you go literally anywhere, the general public absolutely hates AI, so much they they have literally no, They don't.
Yeah, I mean, like you know, you go to your random artist and they like hate AI.
You go to your random, like, person in rural America, they're like, screw AI.
It's like, you know, taking all the water.
You know, completely, like, nonsense arguments, but, like, it doesn't matter.
Yeah.
So when you look across the ecosystem, you've got that problem.
And so like, and then you've got the doomsdayers, right?
There's also the general public, which like doesn't quite understand AI, or maybe they do, you know, and they're just like so worried about AI, like taking her gerbs and like all sorts of other things, right.
And so I think, I think that's that's another aspect of this.
That's like quite interesting is, you know, general public hates it.
Yeah.
Doomsdayers.
You know look, I'm a bit of a to live in the moment person to like think you know.
Hey, what exactly is alignment mean?
And, and does AI kill us all?
Um, you know, obviously there's huge risks to that.
Uh, that's not what I'm an expert in, so I don't, I don't really care to opine too much.
Um, but, um, as far as like, you know, is it a bubble or people are people, are we, are we doing too much?
Like what's going on?
Is there a bubble?
Are we doing too much?
The question is, if AI model progress slows down, then of course we're in a bubble.
That's obvious.
But the way that progress is accelerating so fast, right?
You see it. month on month on month, right?
I mean, you know, the new models are coming out every month, every week, almost.
It feels like nowadays.
Or new capabilities, right?
Just think about like this year so far, right?
You know, obviously Cloud 4.5 came out and Cloud Code came out last year.
But adoption really upticked in the beginning of this year.
And so we saw a huge uptick.
Just in January it went from 2 of commits on GitHub to 4 of GitHub commits were done by Cloud Code.
That does not mean, you know, a lot of people use Cloud Code without having Cloud commit for them.
So you're like, you're talking about like or, and then they use Codex, they use Cognition, they use or sorry Devon, they use all these other platforms.
They use GitHub Copilot, they use Codex.
You know we're probably at 10 or so of total code is being committed or written by AI, if not more.
But at least for Cloud Code itself, it went from two to 4% in one month.
This acceleration is like, I think it's the thing that we've all been waiting for.
Cause like fine people are using ChatGPT, that's great.
Fine people are using like ImageGen, that's great.
These were not like things that add trillions of dollars to the economy.
These are social networks.
These are like, help me on my homework.
These are like chill things.
But now we're in the stage where it's like no no no, these are trillions of dollars of economic value that could be added.
And if you think about worldwide software developer wages 2 trillion in wages you end up with a pretty incredible amount of Spend that could happen.
I feel like it's my worst interview ever because I'm so focused on chopping garlic and not cutting off my fingers.
I wonder if people are gonna trash me for my lack of claw grip.
Because I've tried to do it, but then I failed and I've done more dangerous cutting grips.
I do wonder, how vicious is the audience?
We'll see.
But yeah, you're talking about cloud code adoption and GitHub commits being significantly more, especially this past couple months.
And like the value that cloud code will have towards the economy?
Do you want to elaborate more on that or was that a finish?
Yeah, yeah.
I mean, I think like there's quite a bit of, I'm definitely claw gripping now, you know.
I've got to make up for it now that I'm cognizant.
So I think the adoption of AI has been so accelerated over the last month.
You know, just think about everything that's happened in the last month.
We've had Cloud Code happen.
We had Cloud Bot.
We had a Malt Book.
Now we have, you know, we had Kimmy K 2.5 Swarms.
We had Codex 53, which is, you know, a significant step up as well, and seems better in some specific areas.
And these are, there's so many areas, right?
And it's like, at least internally, like at my company, we've completely flipped over, right?
Like about a third of the company's engineers, about third of the company's hedge fund people and about third of the company's like passionate individuals.
Um, you know, so the ex hedge fund people, they are, they're all in on cloud code now too, right?
They scrape data, they do pro, they do financial modeling, they do pro forma financially modeling with cloud code as the assistant.
Um, and so there's a variety of like, sort of like um, You know, I think like we've hit like sort of escape velocity and all these things.
And so you know, over the last month or over the last two weeks, we've had, you know, the hyperscalers report earnings.
And everyone's stocks have gone down, right?
Google announced 180 billion of CapEx.
The stock went down.
And then Amazon announced 200 billion of CapEx and their stock went down like I want to say like 10.
So the market hates it, but they don't realize.
Like you know, these CapEx decisions are because they see the light at the end of the tunnel, if you will right.
The amount of adoption is just insane.
So, now that we've had these companies report earnings and they've disclosed what their plans are for the year and they're much higher than almost anyone predicted you've got.
You've got the market just hating it.
And so that brings, like this interesting conundrum, which is that like okay, the market is mad, they're spending all this money on compute capex, but these companies know much better than than than you, right like and by you i mean the investor right um, in reality, they're spending this much because they see insane amounts of demand right um, you know, anthropic doesn't just add two billion dollars of revenue in one month.
Uh, you know, without having huge demand.
And they're doing it at positive margins, right?
And they're doing it, and when they go to everyone, they're like look guys, we need more compute, we need more compute, we need more compute.
And so, After three years straight of the AI lab saying we need more compute, the hyperscalers are now saying it's not just we need more compute because we want to train bigger models and we want to do more research.
We need more compute because we need to serve our users right.
We need to add hundreds of millions of dollars or billions, actually billions of dollars of compute right.
If Anthropic add two and a half billion of revenue and their gross margin is 40, they added like one and a half billion dollars of compute in one month right.
Just to serve that, you extrapolate that line out a little bit and you're like holy crap, they actually need hundreds of billions of dollars of compute.
Okay, fine, well, we need to build this all, right?
Because it's a front run of you build it and then they can rent it.
So the bet here is that growth will continue to accelerate and the amount of money that Anthropic and OpenEye makes will just continue to go up.
Yeah, exactly.
And I don't, you know, like, look, the party can stop at some point.
Like, you know, that's for sure.
At any point.
You know, I think my favorite thing was I tweeted about like people were like oh, who expected this CapEx?
And then I was like, well, we did, right?
And then someone replies, the whale watcher told you that you're going to see whales.
Wow, surprising.
I was like, wow, that's a pretty good reply to me.
But anyways, it's like, obviously I'm the whale watcher here.
We see the CapEx coming.
So the market doesn't like it, but it's clearly obvious that it's needed.
And now the discussion is sort of like.
I think a couple of years ago I said the hyperscalers would have no free cash flow.
Ie, they would not have any.
They would not be generating profits and buying back their stock in a short amount of time.
You can also transfer your onions and some veg to the bowl white bowl on your left, I believe, or right.
Yes.
So on my cutting board, I just have garlic, ginger, and the green part of the scallions.
So leave the green part for a garnish for later, so just keep it on.
I guess at what point do you think the market would be satisfied or okay with this?
I think the market is going to get really mad at the hyperscalers.
They haven't really yet, but we saw the signs of it at the beginning, about mid last year, right?
For example, Oracle peaked when they announced, within a week after announcing, that they were gonna do 300 plus billion dollars of deals with OpenAI.
And the market really peaked around then.
And then, like, since then, they've gone down.
And like other darlings that were like doing AI infrastructure, like CoreWeave, have also peaked right.
So now we've got this like interesting conundrum where now the hyperscalers are starting to say how much CapEx they're gonna do.
Yeah.
And so when we think about like hey, what's gonna end up happening is you know, these hyperscalers are gonna keep spending, right.
What is their biggest advantage, right?
It's that they can build the most infrastructure in the world.
They've built the organization to build infrastructure faster than anyone else.
Wow, this like got cooked super fast.
Show the camera.
It's just eggs, bro.
I just don't generally use an induction, right?
Yeah, they heat up very fast, so just as a heads up.
Yeah, so the hyperscalers have been the most profitable companies to ever exist in humanity.
Whether it's meta through ads, whether it's Google through search, whether it's Amazon through AWS and Amazoncom, So on and so forth.
Right,
Microsoft through Windows plus Office 365 plus Azure, right?
They've all been the most profitable companies.
Am I gonna continue with the onions and such?
Yeah, or so, are the eggs cooked?
Yeah.
Okay, you could put the eggs on the plate.
Yeah, already done.
Okay, and then now add some oil, and then yeah, add all the veggies.
Yeah, yeah.
But make sure not to add too many onions, because there'll probably be more onions and carrots and other things.
Just try to have like an even balance of vegetables.
Yeah. yeah and then yeah so just get some color on the veg um do you do you uh tend to um you tend to do the carrots and onions at the exact same time yeah like doing them all together is probably just easiest so carrots onions and the white part of the scallions okay um So the hyperscalers have been the most profitable companies ever.
And now they're about to face sort of this interesting conundrum, right?
There's a huge innovator's dilemma here, whether it's Meta doesn't own the platform.
Wherever people's eyeballs are is where people are going to spend their cash.
Or in the case of Google, AI can disrupt search.
Or, in the case of Microsoft productivity suite, is where they make all the money Office 365 Windows, et cetera.
But things like Cloud Code, Cloud Bot, and future iterations of it just take a little imagination.
We'll displace those immediately, right?
Same with Amazon, right?
AWS is a general purpose AI infrastructure or sort of infrastructure play, but there's a lot of risk with everything else, with everyone getting disrupted quite heavily.
And so they've got this dilemma where they could get disrupted heavily.
At the same time, they've also got this challenge with regards to potentially being beaten, right?
So they have to invest hugely in AI.
They have to try and win AI.
And if they don't, then they're really, really screwed.
But right now the demand for AI is insatiable and they can get pretty good returns just by building infrastructure and renting it out to the labs.
But they'll obviously get way better returns if they have AI models in house.
So they need to spend like crazy to do this.
And, at the same time, if everyone else is like sort of it's like a, it's pascal's wager, right?
Um, if i don't spend like crazy and others do, i lose, right?
If i don't believe in god right, digital god coming then and others do and it happens, then i'm a loser, right?
Yeah um, And so they've all got this dilemma.
And the only solution is I have to spend more and more to keep up in the race.
And so this year's announcements of 180 billion of CapEx from Google and 200 from Amazon, which is 4x what they were just doing a few years ago, is quite intense, but in addition to that, we're looking at this skyrocketing in the next few years.
There's no reason why Google will have any profit in 27 at all, in terms of cash flow.
They will just spend every dollar they make on AI infrastructure.
And I think that's at least my belief and AI models and so on and so forth, because that's basically my belief.
And the market hasn't fully woken up to this realization.
We've been saying it for a couple of years.
In fact, we even did a piece last year which was like how much debt can the hyperscalers borrow right?
Because at some point, they have to lever up on building capacity.
So an example of this is Meta.
Meta is not as large as Google and Amazon, but they wanna be in the race.
And so they've already started taking some debt on to build their data centers.
Now, obviously they have a tremendously profitable business that could pay it off.
They just have to stop spending the money on capex.
That's not necessary.
But Zuckerberg has woken up and fully realizes this.
How much do I want to cook the onions and carrots?
Just get some color on it and then just put it in the plate with the eggs.
OK.
And so you've seen some hyperscalers, such as Meta.
They're already taking debt on for their largest AI cluster in Louisiana.
They're taking like $40 billion of debt on for that.
But they're in the market to take on much, much more.
Google and Amazon haven't taken on debt yet for AI infrastructure, but they will. right yeah and and so i think i think people really realize and panic and and and probably this year when they see all of these companies doing exactly this right are they gonna you know what ends up happening uh when the most profitable companies that have ever existed which have compounded at double digits for uh over like a decade and a half Now all of a sudden say, we're not going to, we don't care about profit anymore.
We're just building pixie dust, right?
We're building digital God.
And if you believe in it, great.
If you don't, then, you know, tough luck.
And all this CapEx predates infant revenue, right?
Because you need to have spent the CapEx, brought on the clusters and all that before you can have the, before you can ever have the.
The revenue come online right.
And then the revenue starts off at lower margin, right?
Yeah.
Also it's time to cook the chicken.
So just add some oil and then we'll cook chicken.
Okay.
Sounds great.
Yeah.
Yeah.
So saying investment before the revenue comes online.
Yeah, so there's a timeline lag when revenue comes online.
There's a timeline lag in terms of when you rent the infrastructure versus hey.
You have to train the model before you can ever start to actually get the AI service revenue right.
And so you've seen this with all the vendors, right?
There's huge spend for OpenAI, Anthropic and Google on training models, and the others are doing it too, like Amazon and such, before they ever end up with enough revenue, generating revenue from the models, from the services that they sell on top.
And so the market is just going to really hate this.
Yeah.
And I feel like that's going to lead to.
Despite the fact that everyone in San Francisco is going to see revenue skyrocketing, they're going to see all the amazing capabilities.
But we live in a bubble, right?
If you told...
We put out some research that was like hey, 4 of commits on GitHub are cloud code and everyone SF is like that's too low right.
It's like 100% of mine.
Maybe it's 50% for all the boomers, right?
And it's like, no, no, no.
We've got a lot of adoption to go.
And so people are gonna like, sort of like, see all this amazing model progress and revenue growth and adoption and SF, but then, like in New York and in London and like Hong Kong and other financial capitals of the world Singapore, etc.
People are gonna see the exact opposite and They're gonna see the most profitable companies ever are destroying their business model to build capacity in something that maybe necessarily doesn't have returns.
And likewise, they're going to see the general public fucking hates AI.
And you likely see a real backlash to AI from both the financial class and the normal people of the world.
Does it matter though, if the public hates AI, if it provides a lot of value towards enterprises and companies?
Isn't the main value and profit coming from enterprises more so than the general public?
Exactly, and I think that is the big fear, right?
You know, we've already had decades of people being like hey, income inequality is bad and the value of labor has been falling right.
The value of labor used to be way, way higher as a percentage of the economy.
But as we've mechanized, as capital has become more and more important, as machinery has grown, we've sort of had this major change, which is that, You know, capital is taking more and more of the earnings.
Capital's taking more and more of their earnings.
And so people are really mad about that.
And now we're gonna start seeing huge job loss too, right?
Like.
Hey, like Turns out, there's shitloads of software developers just out of school who can't get jobs.
Okay, fine.
But what about the two million people who drive cars for a living?
Okay, well Waymo works well.
Tesla Robotaxi's starting to be deployed.
Zooks is starting to be deployed.
We're starting to see really the beginnings of all that.
We're gonna see.
You know, the stock market maybe does well, or the GDP is gonna look good, but then normal people aren't gonna be accruing much value from it.
And so more and more, you know, and eventually, like the financial markets, will not do too well either, because software's imploding, because hyperscalers are gonna invest all their capital.
And you're going to end up with this major, major fear and worry for everyone in the industry.
Or sorry, everyone in the world.
And there's an AI backlash, right?
And I think that's going to be the hottest button issue of the next election.
If not the midterms.
And it seems obvious to me that any party that wants to win should just become the anti-AI party.
Because life as we know it is changing.
Taking a little detour, where do you think the alpha is or the bet is?
Because you said that Nvidia is kind of covering their basis with the Rubin CPX, the graph chips, standard GPUs.
And there's a lot more startups out there that are very specialized and even, like a lot of YC companies, are like popping up and kind of tackling this industry.
Do you think it'll kind of turn out to be a play where like nvidia, at the end of the day, still reigns and um like crushes every other company that tries to take away the market share?
Or do you think there will actually be a lot of value accrued to these more specialized, smaller companies um, in the in the chip space specifically?
Yeah Yes, I think that's a really strong debate that people are having.
How much value accrues to NVIDIA?
How much value accrues to the model companies?
How much do they start to really You know, do smaller chip companies take charge and win?
And it is really an innovator's dilemma in the sense that, like you know hey, why did Intel and AMD not win in AIG fuse?
Well, it's because they were making money off of CPUs and Nvidia was focused on parallel computing.
And now you've got sort of the same question, which is will NVIDIA be able to innovate on all the things that needs to be innovated?
Or will... There's a lot of scraped stuff on the bottom.
Yeah, yeah.
Well, it's like there's stuff stuck in the pan, you know?
So I'm trying to scrape it off.
We can also get a new pan.
I have two new pans.
Actually, that would be amazing.
The final thing, we just got to now add everything together.
So first, add some oil to the pan once it's dry, and then add the garlic ginger.
But have it kind of lower.
Have it what?
Have like the temperature be lower, like not too high.
Oh, really?
Okay.
Yeah.
NVIDIA is, they've kind of got this like innovator's dilemma.
The nice thing is they embody Silicon Valley spirit more than maybe any other company, which is Andy Grove right.
Andy Grove from Intel.
Only the paranoid survive.
Only the paranoid survive, right?
And I think Jensen Huang is like one of the most paranoid people in the industry, right?
He's constantly like freaking out, changing internal things, like you know, in a good way though, right.
Like truly founder mode.
All right, the aromatics are very aromatic.
Okay, then just add all the veg.
Don't add everything because proportions, but add the onions, carrots, eggs, peas, and then chicken.
And then once you have that, then add the rice and like should mix everything together.
And then at the very end, you're going to add soy sauce and then some sugar and salt to adjust. yeah but jensen's very jensen's very paranoid um and that makes him like an amazing founder um and ceo um and so you you have all these people freaking out but it's like the moment he sniffed wind of the open ai cerebrus deal He immediately went out and was like, okay, I wasn't actually building this technology because I didn't believe in it, but now I do because OpenAN is trying to use Cerebrus.
So I'm just gonna go acquire Grok right, like you know, it's like that's like why he did it right.
So it's like you know there's there's a bit of like um.
You know, the moment he sniffs anything, he changes course in tune, updates his priors um, and i think that's like really impressive.
And so as you step forward to like hey, what about um?
You step forward to like okay well, what does that mean for his hardware roadmap?
Well before he was like making one, you know, just a few kinds of architectures and chips, but primarily it was all like very similar, right.
It was a large GPGPU and it was like having.
It was like the best memory, the best networking, everything sort of the best as possible and sort of like one size fits all, with the main line of like A100 H100 B200, right.
But as we look to Reuben and beyond, Jensen is really like starting to fully embrace heterogeneity.
Right, much like this fried rice right.
There's no one individual ingredient that shines above all, right?
You've kinda gotta have a little bit of everything.
And so this is sort of like what Jensen's believing here.
So he's got this CPX chip, which is made for context processing, pre-fill.
It's pretty good at video and image gen as well, but it's not really good at latency-sensitive applications.
You know, they've of course got their main line of GPUs, and now they've got these grok chips right.
So you know, Nvidia's sort of got every single aspect or type of chip possible now within his company.
And he's continuing to try and like innovate and move as fast as possible in all these things.
Yeah.
And so when we think about like hey, what ends up happening With NVIDIA in this case?
NVIDIA knows they will lose because they have a business model deficit, right?
Google, Amazon.
They get to vertically integrate and vertical integration always saves tons of money.
So he has to be better than everyone by not just like a little bit, by a ton to justify his margins.
Otherwise, the vertical integration of his competitors will win out.
And so this is sort of like I think the big challenge.
And I think the story is not finished, right?
NVIDIA will remain on top this year and next year, based on what we see, but others will gain some ground.
And the question is what happens in the long term?
Um and and honestly, you know the the, the cards are up in the air, right?
Um, no one has the, the right to win, no one has a destiny to win.
Uh, things are moving so fast.
Whoever, whoever does the, you know, innovates the hardest, will win.
Not, not necessarily like oh, you know, and and i think moats are as shallow as they've ever been right, because how fast things are moving, the size of the numbers that are being thrown around now right,
It's hundreds of billions of dollars for each major hyperscalers.
The size of the numbers are so large that you can just go and justify hiring anyone, any talent.
The moats become much smaller.
And this is sort of like, pretty big deal with regards to, you know, does NVIDIA win or not, right?
Yeah.
I guess what do you think's the biggest bottleneck for speed, to keep us from going as fast as possible?
Is it memory?
What do you think is the main bottleneck?
You know, I think I have zero wok hay, by the way.
Yeah.
I don't think I'll have any wok hay with me.
Dude, I swear to God, if Uncle Roger finds this video, I'm going to cry.
I'm like, he's like, he's going to be like no MSG hiya, no wake hiya.
Induction furnace.
What are you doing?
No, but anyway, sorry.
You know, what's the biggest bottleneck to speed?
You know, I think you cook fried rice much faster if you have a wok.
All right, I'm done, I'm done.
I'm not good enough to justify that.
There's like a hundred other mistakes I've made, you know?
But like I think the biggest bottleneck to like, hey, why only $200 billion this year for Amazon?
Why not 500, right?
I think there's like a number of limiting factors and it's sort of like year by year it's been different right.
In 2023, it was definitely all related to chips, right?
Semiconductors co-ops, which is chip on wafer on substrate.
Driving up production of this was very, very difficult.
And then, as we step forward to 2024, as you step forward to 2020 oh, I didn't throw any sugar or soy sauce, no wonder.
Oh, did you serve it already?
No, I didn't.
I almost did.
Okay.
Yeah.
Final step, just sugar, soy sauce to taste.
So taste it as you go.
I don't know how much sugar to do, but... Dude, this shit's about to be sweeter than Panda Express.
Oh, shit.
I completely forgot about the soy sauce.
Anyways, in 2023, it was co-ops, it was semiconductor supply chains.
As we step forward to 24, 25, it started to become data centers.
Energy is a bigger deal in 25, 26.
But as we step forward, supply chains are fast and they react quickly.
So this current whole thing of like, oh, data centers are the shortage.
Yes, data centers are a shortage.
Yes, power is a shortage.
At the end of the day, actually, there's a lot of other shortages around too, right?
You know, and when you think about power okay well, like if you were not creative right, and you just relied on grid power well, there's only three companies that make dual combine cycle reactors.
But if you step forward to, like, oh, okay, well, what if I make, what if I want, um...
What if I take something else, right?
What if I take aeroderative engines?
Okay, there's a few more vendors.
What if I take industrial gas turbines?
There's a few more vendors.
What if I take... what if I take medium speed reciprocating engines, right?
Which are these sort of?
Like any company that makes diesel engines, there's dozens of them.
They can make medium speed reciprocating engines and I can connect those up to make power for the data center.
Right.
So when I look at when I look at like, hey, what did who sort of broke these bounds?
Right.
Elon was the first one to sort of say, well, no, I don't care about the actual rules.
Let me just like let me just put power generation on site with low quality mobile turbines.
Right.
Not turbines even right.
Industrial gas engines, you know, reciprocating engines, et cetera, et cetera.
So Elon broke all these rules and now the whole industry's reacted fast enough because there's so many suppliers, right?
And the lead time to ramp up production of these things is ultimately not nearly as long as it is in the semiconductor supply chain.
So going back to your question of like, hey, what's the big bottleneck?
Well, the big bottleneck is now back again to semiconductors, right?
Semiconductors are extremely cyclical.
The buildings that chips are made in are the most complicated buildings people make.
You know, they are...
They have multi-year timelines.
They require not just all the complexity of electricians and plumbers that data centers do, but they actually require a lot more complex because there's all sorts of chemicals and precursors and so on and so forth that are going through the data center or through the fab.
And so, you know, people have just not built enough fabs.
And then, you know, that's ignoring all the complicated tools, right?
These tools cost hundreds of millions of dollars in some cases.
They're the most complicated thing people make.
And so you end up with wow.
So the challenge here is not just it's ramping up production of semiconductors.
And so now we've entered an age, especially in 26, but as we go into 27 28, you know, when we look in 2026, Google would buy a lot more TPUs, but they can't ramp production fast enough, right.
And so they have to buy tons of GPUs.
When we go to 27, it applies again, right?
Google simply cannot buy enough TPUs, and they have to buy tons of GPUs.
And when you look across the entire supply chain, no one is getting enough capacity of semiconductors.
Yes, they can put them in data centers.
Yes, they can get the energy through maybe ghetto methods like putting reciprocating engines, right?
Diesel reciprocating engines or gas engines, but maybe not the most clean or efficient thing, but they can do it.
And so you end up with oh okay, semiconductors are the shortage, but what's the bottleneck to building more fabs or to building more chips is more fabs, and people just have not built these fabs yet, right.
And that's, I think, the big bottleneck now.
And that's gonna persist through the end of the decade, or until AI sort of slows down.
How are you feeling, Dylan?
I turned around and I saw yours briefly, so now I'm trying to wipe the edges of my bowl so it looks beautiful, you know.
Yeah, no need to worry too much.
No, no, no, no.
People are going to judge me hardcore.
How was that, Dylan?
Was that fun?
It was quite fun.
I think there was the right amount of stress involved.
I think I maybe didn't share my thoughts as well as normal, but maybe that's more natural and fun.
Yeah, like a mix of things.
Great.
And this is yours.
Okay, let's try yours first.
Well, let's try the restaurant actually first.
Control, control.
Yeah, control.
Okay, cheers.
I always feel like you've got to cheers this food.
Mm, that was good.
Okay, and we'll try yours now.
Is that okay?
A lot of meat, I love it.
A lot of chives as well.
Cheers.
Mm, yeah, it's pretty good.
I like mine more.
Yeah.
No, it's very... It's like a lot stronger.
Like, it's very deep.
Mine's probably going to be bland compared to that.
So, I think...
You know, there's this, like, debate in the world, right?
French people, they don't season their food so much.
Mm-hmm.
They're all about the ingredients shining.
Then you got like you know equator, equatorial people write slop like Indian food and like Caribbean food and like
Southeast Asian food, just like throw on the spice, throw on the sugar, throw on the like everything.
And it's like okay, well then is this?
And sort of like the elitist French would say, that's because your ingredients suck.
You have to throw all the slop on there, right?
Overdo it.
Yeah, but I don't know.
I'm a slop son, you know?
Yeah, I mean, it is very tasty, so let's try this.
The palates are...
Yeah, I think compared to yours, it's like a lot blander.
It's very light.
Not gonna lie, You like yogurt better, right?
I can tell.
Honestly, yours is like... There's a mess-up of technique, right?
The bottom of my pan was getting burnt.
And, you know, we can't get a walk-A because this is induction cooktop.
But because the bottom was getting burnt, the smoke flavor, I think, was importing into the rice.
And I got a walk-A.
It's intentional.
Yeah.
Yeah, you can kind of taste it here, but it's, like, hard.
Very good.
Okay, are there any call-outs, things that you want people to know?
Are you hiring clients?
Yeah, I am hiring.
We're 60 people now.
We work with all the top companies in the world major AI labs, major hyperscalers, semiconductor companies, data center companies, etc.
We cover the entire swath from AI infrastructure, AI models tokenomics, usage of AI models, who's using them, what are they using them for?
What's the cost of running it?
All these sorts of things.
So that's the area we're really expanding into this year and last year.
And so I think that's the audience that also matches this audience that you have here.
So if anyone wants to track those things usage of AI, the person who was working on this before unfortunately for them, but also a call out they got hired by Anthropic right.
So sort of like.
The person who was working on this got hired by Anthropic.
As a sort of like please don't poach anymore of my people.
But I think it's a good thing.
It's a good pathway, right?
It's showing your work is public.
If you're going to kill it, you're going to kill it.
You have all the resources behind you of knowing and understanding infrastructure.
I think that's the big call out is we're hiring for that role.
We pay well.
We have health care.
We hire globally.
We're in eight or 10 different countries like US, Japan, Taiwan, Singapore, France, Germany, Israel.
Yeah, Canada, UK, right?
So we're everywhere.
Yeah, I think that's the allure.
We get to work with all the coolest people.
Great, that's amazing.
Well, thank you so much, Dylan.
It was a great time.
Well, thank you for having me.
Yes.
I really care about the quality of the chicken.
Okay, this is like chicken egg fried rice, right?
Yeah.
How do you assess?
I think mine's a little intense.
Yours is intense.
There's probably too much soy sauce.
That's fair.
I forgot to throw it in and then I just threw it all in.
YOLO'd.
How much did you have?
See, he left like... He left a little bit, yeah.
Yeah.
I would say, you know, like, I would take this.
I would take Allen's.
I would also take his.
I don't know, but.
I like Dylan's.
Okay.
I like Salty's.
There's one each.
Brandon, you're the tiebreaker.
I'm on the side of the slot.
Yes!
Wait, what do you vote?
What do you vote?
You voted for you.
Oh, I voted for you, too.
Wait, I fucking won!
You worked in a restaurant, bro!
