YouSaid · the spoken record

Dylan Patel

lines on the record: 170
first: 2024-10-02
most recent: 2024-10-02
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“ILT and Sam, maybe. I don't know if that's AI. That's not AI. Anyways, like there's like tremendous opportunity to bring breakthrough innovation simply because there is so many layers where things are unoptimized, right? So you see like all these like, oh, single digit, low double digit like advantages just from like RL techniques from like alpha ghost type stuff like or like not RL from alpha go, but like like five, six, seven, eight year old RL techniques being brought in. But like generative AI being brought in could like really revolutionize the industry, you know, although there's a massive data problem.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“It's all design, right? Most of it's design. Manufacturing is not really seen much of this yet, although there is starting to come in.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“So, you have all the abstraction layers, but then it's like one is there's a lot of breakthrough innovation that can happen now stretching across abstraction layers, but two is because there's so much inherent knowledge in each individual one, what if I can just experiment and test at a thousand x velocity or 100,000x velocity? And so some examples of where this is already like shown true is some of NVIDIA's AI layout tools, right? And Google as well, like laying out the circuits within a small blob of the chip with AI. Some of these RL design things, some of the, there's a lot of like various simulation things.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Most of the equipment in semiconductor tool fabs runs on Windows XP, right? Like each tool has a Windows XP server on it or all the chip design tools have CentOS like version 6, right? And like that's old as hell, right? So like there's like so many like areas where like why is this so far behind at the same time it's like so like hyper optimized that's like the the tech stack is so broken in that sense”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Because it's all siloed within companies. B, it is there's a lot of human element to it because a lot of the knowledge, like as John was saying, is like apprentice master, apprentice master type of knowledge, or I've been doing this for 30 years and there's an amazing amount of intuition on what to do just when you see something to where like AI can't just learn semiconductors like that. But at the same time, there's a massive amount of talent shortage and ability to move forward on things right so like the technology used on like like”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“This is obviously true. This is the Word of God. And every engineer in the entire industry, tens of millions of people, like literally, this is what they were driven to do. Not every single engineer didn't believe it, but like people were like, yes, to hit the next shrink, we must do this, this, this, right? And this is the optimizations we make. And then you have this abstratification every single layer and abstraction layers every single layer through the entire stack to where people, it's an unholy concoction. I mean, keep saying this word, but like no one knows what's going on because there's an abstraction layer between every single layer. And on this layer, the people below you and the people above you know what's going on. And then like beyond that, it's like, okay, I can try to understand, but like, not really like.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“I think it's like God came and proclaimed it. We will shrink density 2x every two years. Gordon Moore said he made an observation and then it didn't go nowhere as it went way further than he ever expected because it was like, oh, there's line of sight to get to here and hear. And like, and he predicted like seven, eight years out, like multiple orders of magnitude of increases in transistors. And it came true. But then by then, the entire industry was like.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“The stratification of just like manufacturing is absurd. Like the tool people don't even know exactly what Intel and TSMC do in production and vice versa. They don't know exactly how the tool is optimized like this. And it's like, how many different types of tools there are? Dozens. And each of those has like an entire tree of all the things that we've built, all the things we've invented, all the things that we continue to iterate upon. And then here's the breakthrough innovation that happens every few years in it too. So if that's the case.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“It is like a free thing. The paper publishing industry is like abhorrent everywhere else and you just cannot download IEEE papers or like SPIE papers or like other organizations. And then two, at least up until late 2022, really early 2023 in the case of Google, right? I think what the Palm Inference paper up until the POM inference paper, before that, all the good best stuff was just posted on the internet. After that, it's kind of a little bit clamping down by the labs, but there's also still all these other companies making innovations in the public. And like, what is state of the art is public? That is not the case in semiconductors.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, and then he was noticed. He was like, who the hell is this guy? And why is he posting this? I thought everyone who knew about this was at Google already, right? It's like, come to Google, right? That can't happen in semiconductors, right? Like, it's just not like conducively like, it's not possible, right? One, archive.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“But he's cracked. He's like Omega cracked at like what he does. What he does, you could pick him up and drop him into another part of the AI stack. First of all, he understands it already. And then second of all, he could probably become cracked at that too, right? Whereas that is not the case in semiconductors, right? One, you like specialize crazy. Two, you can't just pick it up. You know, like Schultz, I think what did he say? He just started like”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“18 year olds who are just cracked at AI, right? Already, right? And like there's high school dropouts that get jobs at OpenAI. This existed in the past, right? Pat Gelsinger, current CEO of Intel. Went straight to work. He grew up in the Amish area of Pennsylvania and he went straight to work at Intel, right? Because he's just cracked, right? That is not possible in semiconductors today. You can't even get a job at a tool company without at least a freaking master's in chemistry, right? And probably a PhD, right? Of the 75,000 TSMC workers, it's like 50,000 have a PhD or something insane, right? It's like, okay, this is like there's like some, there's like a next level amount of like how specialized everything's gotten. Whereas today, like, you can take like, you know, Sholto, you know, when did he start working on AI not that long ago?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Like, cool, like, even if I could make my Mac chip, I can't make the chip that drives the display. Like, you know, you just go down the list, like everything, no fridges, no automobiles, no weed whackers, because that shit has, my toothbrush has fucking Bluetooth in it, right? Like, why? I don't know, but like, you know, there's like so many things that just like poof we're tech reset.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Until four. Yeah, yeah. I mean, yeah. So you have like TSMC always builds new fabs. That old fab, like tweak production up a little bit more and more and new designs move to the next, next, next node. And old stuff fills in the old nodes, right? So, you know, ever since TSMC has been the most important player, and not just TSMC, there's UMC there, there's PSMC there, there's a number of other companies there, Taiwan's share of total manufacturing has grown every single process node. So in like 130 nanometer, there's a lot, including many chips from like Texas instruments or analog devices or like NXP, like all these companies 100% of is manufactured in Taiwan, right, by either TSMC or UMC or whatever. But then you step forward and forward and forward, right? Like 28 nanometer, like 80% of the world's production of 28 nanometers in Taiwan. Oh, fuck, right? Like, you know, and everything in 28 nanometers, like, what's made on 28 nanometer today? Tons of microcontrollers and stuff, but also like every display driver I see.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“There's like 2,000 plus chips. Every Tesla door handle has like four chips in it. It's like, what the fuck? Like, why? But it's like shitty microcontrollers and stuff. But there's like 2,000 plus chips even in an ICE vehicle, like internal combustion engine vehicle, right? And every engine has dozens of dozens of chips, right? Anyways. This all shuts down because not all of the production, there is some in Europe, there are some in the US, there's some in Japan. They're going to bring”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Of chips for making new cars gone or sequestered to make military shit, right? You can no longer make cars. And we don't even know how to make non-semiconductor induced cars, right? Like this unholy concoction with all these chips, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“I think day one market crashes a lot, right? You got to think about, I think the big six biggest companies, Magnificent Seven, whatever the heck it's called, are like 60, 75% of the S&P 500 and their entire business relies on chips, right? Google, Microsoft, Apple, Nvidia, you know, you go down the list, right? They meta, right? They all entirely rely on AI. And you would have a tech reset extremely insane tech reset, by the way, right? So the market would crash week, a day in, a couple weeks in, right? Like people were preparing now. People are like, oh, shit, let's start building fabs with fuck all the environmental stuff. War is probably happening, but the supply chain is trying to figure out what the hell to do to refix it. Six months in?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“You know, the AI really, really, really wants new processed nodes because of A, power used is a lot less now, higher density, higher performance, of course. But the big deal is, well, if I have a gigawatt data center, I can now, how much more flops can I get? If I have two gigawatt data center, how much more flops can I get? If I have a 10 gigawatt data center, how much more flops can I get? Right? And like you look at the scaling, it's like.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“So there's the simplistic argument that, like, oh, moving a process node only saves me x percent in power, right? And that has been petering out, right? When you moved from like 90 nanometer to 80 something, right, or 70 something, right? It was like, you got two x, right? Denard scaling was still intact, right? But now when you move from five nanometer to three nanometer, first of all, you don't double density. SRAM doesn't scale at all. Logic does scale, but it's like 30%. So all in all, you only save like 20% in power per transistor. But because of like data locality and movement of data, you actually get a much larger improvement in power efficiency by moving to the next node than just the individual transistors power efficiency benefit. Because, for example, you're multiplying a matrix that's like, you know, 8,000 by 8,000 by 8,000. And then you can't fit that all on one chip. But if you could fit more and more, you have to move off chip less. You have to go to memory less, et cetera, right? So data locality helps a lot too.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“But like, anyways, like, you know, there's a strong argument that, like, Funding the next node would not be economically viable anymore if it weren't for AI taking off, right? And then generating all this humongous demand for the most leading edge chip.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Viable is 2 nanometer just for one player TSMC, you know, ignore Intel, ignore Samsung, just because Samsung is paying for it with memory, not what their actual profit, and then Intel is paying it from it from their former CPU monopoly.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Worth it? Every two years you get a shrink, right? Yeah. Like Clockwork, Moore's Law. And then five nanometer happened. It took three years. Holy shit. And then three nanometer happened. It took three, or no, sorry, is it three nanometer, five? It took three years. Holy shit. Like, is Moore's Law dead, right? Like, because TSMC didn't. And then what did Apple do? Even on the third year, or sorry, when's three nanometer finally launched? They still only Apple only moved half of the iPhone volume to three nanometer. So this is like, now they did a fourth year of 5 nanometer for a big chunk of iPhones, right? And it's like, oh, is the mobile industry petering out? Then you look at two nanometer and it's like going to be a similar like very difficult thing for the industry to pay for this, right? Apple, of course they have, you know, because they get to make the phone, they have so much profit, they can funnel into more and more expensive chips. But finally, like that was running out, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Unless more technology spawn, like AI now makes one nanometer or whatever. There was a long period of 16 viable, right? So like right before AI is viable.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Sorry, there's like a video that's like Ham is disgusting. It's an unholy concoction of meat with no bones or collagen. And like, I don't know, like, he was like, the way he was describing two dandy meter is kind of like that, right? It's like the guy who pumps his right.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“I'm sorry to completely derail you, but there's this video where it's like this is an unholy concoction of meat slurry. Yes.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Way to look at it today is it's super super stratified in every industry has anywhere from one to three competitors and pretty much the most competitive it gets is like 70 share 25 share 5% share In any layer of like manufacturing chips, anything, chemicals, different types of chips. But it used to be vertically integrated.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“It's like it's the Andy Grovian mindset, right? Like, shout out to the base Intel, but like only the paranoids survive, right? Like paranoid Western companies do well. Why did Google really screw the pooch on a lot of stuff? And then why are they resurging kind of now? It's because they got paranoid as hell, right? But they weren't paranoid for a while. If Huawei is just constantly paranoid about the external world and like, oh, fuck, we're going to die.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“They have the espionage without a doubt, right? Like ASML has been known to have been hacked a dozen times. Right, right. Or at least a few times, right? And they've been known to have people sued who made it to China with a bunch of documents, right? Not just ASML, but every fucking company in supply chain. Cisco code was literally in like early Huawei, like routers and stuff, right? Like you go down the list, it's like everything is, but then it's like, no, architecturally, the Ascend 910B looks nothing like a GPU. It looks nothing like a TPU. It is like its own independent thing. Sure, they probably learn some things from some places, but like it is just like they're good at engineering.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“There's all sorts of stuff. They're filling water in their rockets. Clearly they're like incompetence. Like, look, if I'm the Chinese military, I want the Western world to believe I'm completely incompetent because one day I can just destroy the fuck out of everything, right? With all these hypersonic missiles and all this shit, right? Like drones and like, no, no, no. We're filling water in our missiles. These are all fake. We don't actually have 100,000 missiles that we manufacture in a facility that's like super hyper advanced and Raytheon is stupid as shit because they can't make missiles nearly as fast, right? Like I think that's also like a flip side is like how much false propaganda is there, right? Because there's a lot of like no Smick could never. Smick could never. They don't have the best tools. Blah, blah, blah. And then it's like motherfucker. They just shipped 60 million phones last year with this chip that performs only one year worse than like what Qualcomm has. It's like proof is in the pudding, right? Like, you know, there's a lot of cope, if you will.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“There is like that crazy conspiracy, not conspiracy. You don't know what the hell to believe in China, especially as a not Chinese person.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“I think Hualwe is so fucking cracked that it's hard to say that, right? Like, Kuala out competes Western firms regularly with two hands tied behind their back. Like, you know, like, what the hell is Nokia and Sony Erickson, like trash, right? Like compared to Huawei and Huawei is not allowed to ship, sell to European companies or American companies, and they don't have TSMC, and yet they still destroy them, right? And same applies to the new phone, right? It's like, oh, it's as good as a year-old Qualcomm phone on a process node that's equivalent to like four years old, right? Or three years old. So it's like, wait, so they actually out engineered us with a worst process node, you know, so it's like, oh, wow, okay, like, you know, Hallway is Hallway is like crazy cracked.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Huawei was the largest TSMC customer for like a few quarters actually before they got sanctioned. Huawei makes most of the telecom equipment in the world, right? Phones, of course, modems, but of course accelerators, networking equipment. You go down the whole video surveillance chips, right? Like you kind of like go through the whole gambit. A lot of that could use seven and five nanometer.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“The reason why I was bringing up Shanghai, they're building seven nanometer capacity invasion. They're building five nanometer capacity in Beijing, but the U.S. government doesn't care, and they're importing dozens of tools into Beijing. And they're saying to the US government in ASML, this is for 28 nanometer, obviously, right? This is not bad. And then obviously you know, like in the background, yeah, we're making 5 nanometer here, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Going back to the yield question, right? Like, oh, theoretically, 60,000 wafers a month times 50 to 100 dies per wafer with yielded dies. Holy shit, that's millions of GPUs, right? Now, what are they doing with most of their wafers? They still have not become skill-pilled, so they're still throwing them out like, let's make 200 million Huawei phones, right? Like, oh, okay, cool. I don't care, right? Like as the West, you don't care as much, even though Western companies will get screwed, like Qualcomm and media tech Taiwanese companies. So obviously there's that. And the same applies to the US, but when you flip to Sorry, I don't fucking know what I was going to say”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, yeah, as a multiplicative effect, right? So yields are bad because they have hands tied behind their back, right? Like A, they are not getting to use EUV, whereas on 7 Nanometer Intel never used EUV, but TSMC eventually started using EUV. Initially, they used DUV, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“I think it's like 99. If I think it's like if it's six sigma of perfection and you have your 10,000 plus steps, you end up with yield is still dog shit by the end, right? Like, yeah, this is multiplication.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Wafers a month, right? Yeah. Doing the math, right? The chip die size and all these things because Puabai also uses chiplets and stuff so they can get away with using less leading edge wafers, but then their yields are bad. You can roughly say any, you know, something like 50 to 80 good chips per wafer with their bad yield, right? With their bad waiver. Because it's hard, right? You know, you're.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“So there's like two points parts there, too, right? Like, so the way the US has sanctioned SMIC is really like stupid kind of is that in that they've like sanctioned a specific spot rather than the entire company. And so therefore, right, SMIC is still buying a ton of tools that can be used for their seven nanometer and their, you know, call it 5.5 nanometer process or 6 nanometer process for the 910C, which releases later this year, right? They can build as much of that as long as it's not in Shanghai, right? And Shanghai has anywhere from 45 to 50 high-end immersion lithography tools is what's like believed by intelligence as well as many other folks. That roughly gives them as much as $60,000 wafers a month of 7 nanometer, but they also make their 14 nanometer in that fab, right? And so the belief is that they actually only have about like 25 to 35,000 of seven nanometer capacity.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Even ignoring sort of even once the centralization question is solved, right? Because that's clearly not happening today for either party. And I would bet if AI is as important as you and I believe that they will centralize sooner than the West does. So there is a possibility, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“And then domestic chips are the other question, right? 600,000 of the Ascend 910B, which is roughly like 400 teraflops or so So if they put them all in one cluster, they could have a bigger model than any of the labs next year, right? I have no clue where, oh, the Sen910Bs are going, right? But I mean, well, there's like rumors about some, they are being divvied up between the major Alibaba, bite dance, baidu, et cetera. And next year, more than a million. And it's possible that they actually do have 1E30 before the US because data center is not as big of an issue. 10 gigawatt data center is going to be, I don't think anyone is even trying to build that today in the US, like even out to 2728. They're focusing on like linking many data centers together. So there's a possibility that, hey, come 2028, 2029, China can have more flops delivered to a single model.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, and then a 28 model, you know, 28 model in the works, right? Like they totally could just with foreign ship supply, right? Just a question of centralization. Then the question is like, do you have as much innovation in compute efficiency wins or what have you get developed when you centralize? Or does anthropic and open AI and XAI and Google all develop things and then like Secrets kind of shift a little bit in between each other and all that. Like, you know, you end up with that being a better outcome in the long term versus the nationalization of the US, right? If that's possible, and what happens there. But China could absolutely have it in 26-27 if they have the desire to. And that's just from foreign ships, right?”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“They're not centralized though, right? So that's the big question Are they centralized? And then also, you know, there's the belief, I'm not sure if I really believe it, but like, you know, prior podcasts, there have been people who talked about nationalization, right? Okay, now you're talking about Well, I think there's a couple No, but I think there have been a couple where people have talked about nationalization, right? But if you have, you know, nationalization, then all of a sudden you aggregate all the flops. It's like, no, there's no fucking way, right? Yeah. China can be centralized enough to compete with each individual U.S. lab. They could have just as many flops in 25 and 26 if they decided they were scale-pilled, right? Just from foreign chips for individual model.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“To some extent, the sanctions have made China believe in the importance of the semiconductor industry more than anything else.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“This thing, and they make that. And then now, when you take out another two pieces, it's like, well, I have my domestic industry for this. I can also now make a domestic industry for these. Like, you go deeper into the tech tree or what have you.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“And then they constantly say, like, if you restrict us, then China will have their own industry, right? And the way I like to look at it is like chip manufacturing is like 3D chess or like, you know, a massive jigsaw puzzle in that if you take away one piece, China can be like, oh, that's the piece. Let's put it in, right? And currently this export restrictions year by year by year, they keep updating them. Ever since like 2018 or so 19, right? When Trump started and now Biden's accelerated them, they've been like, they haven't just take a bat to the table and like break it, right? Like it's like, let's take one jigsaw puzzle out, walk away. Oh, shit. Let's take two more out. Oh, shit, right? Like, you know, it's like instead if they, you either have to go kind of like full bat to the freaking like table slash wall or chill out, right? Like, and like, you know, let them let them do whatever they want. Because the alternative is everything is focused on.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“You know, just taking lithography, right? If your goal is to restrict China from building chips and you just boil it down to, hey, lithography is 30% of making a chip or 25%. Cool, let's sanction lithography. Okay, where do we draw the line? Okay, let me ask. Let me figure out what where the line is and if I'm a bureaucrat, if I'm a lawyer at the commerce department or what have you, well, obviously I'm going to go talk to ASML and ASML is going to tell me this is the line because they know like, hey, well, you know, this, this, this is, you know, there's like some blending over.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Restrictions are flipped the other way, right? They can build better chips in China than we restrict them in terms of chips that NVIDIA or AMD or an Intel can sell to China. And so there's sort of a problem there in terms of the equipment that is shipped can be used to build chips that are better than what the Western companies can actually ship them.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Amount you can smuggle is quite small, and then the sanctions have limited the chip performance to a point where it's like, you know, this is actually kind of fair, but there is a problem with how everything is restricted, right? You want to be able to restrict China from building their own domestic chip manufacturing industry that is better than what we ship them. You want to prevent them from having chips that are better than what we have. And then you want to prevent them from having AIs better. The ultimate goal being, you know, and if you read the restrictions, like very clear, it's about AI. Even in 2022, which is amazing, like at least the Commerce Department was kind of AI pilled. It was like, is you want to restrict them from having AI is worse than us, right? So starting on the right end, it's like, okay, well, if you want to restrict them from having better AIs than us, you have to restrict chips. Okay. If you want to restrict them from having chips, you have to let them have at least some level of chip that the West also that is better than what they can build internally. But currently, the...”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Was Hong Kong headquartered? They just moved to Singapore, but that's effectively in a realm where the U.S. can't sanction it, right? A majority of these other companies are in similar places, right? So advanced packaging capacity is very hard, right? Advanced packaging is useful for stacking memory, stacking chips on COAS, right? Things like that. Then the step down is wafer fabrication. There's immense capability to restrict China there. And despite the U.S. making some sanctions, China in the most recent quarters was like 48% of ASML's revenue, right? So, you know, and like 45% of applied materials, and you just go down the list. So it's like, obviously, it's not being controlled that effectively, but it could be on the equipment side of things. The chip side of things is actually being controlled quite effectively, I think, right? Like, yes, there is like shipping GPUs through Singapore and Malaysia and other countries in Asia to China. But, you know.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, so in the world, I mean, the world we live in is they are not restricted at all in like the physical infrastructure side of things in terms of power, data centers, et cetera, because their supply chain is built for that, right? And it's pretty easy to pivot that. Whereas the US adds so little power each year and Europe loses power every year, the Western sort of industry for power is non-existent in comparison, right? But on the flip side is quote-unquote Western, including Taiwan, chip manufacturing is way, way, way, way, way larger than China's, especially on leading edge where China theoretically has, you know, depending on the way you look at it, either zero or a very small percentage share, right? And so there, you have equipment, wafer manufacturing, and then you have advanced packaging capacity, right? And where the US can control China, right? So advanced packaging capacity is kind of a shot because the vast majority, the largest advanced packaging company in the world.”
2024-10-02 · Dwarkesh Podcast · @Asianometry & Dylan Patel — How the semiconductor industry actually works · IDENTIFIED FROM THE TRANSCRIPT · source