Huawei and the US-China Chip War — #44
Welcome to Manifold. Today, my guest is TP Huang. TP is a Twitter, or should I say X poster covers areas of Chinese technology and China, U S tech competition. I have followed his writings for some time. He is very insightful and often when I'm thinking through consequences. Of some new information that comes out, for example, today, we're going to talk about Huawei's made 60 launch and what this says about the chip competition and the AI competition between the US and China.
I notice that of all the different sources that I read on the Internet. TP is often one of the few people that puts together the pieces and extrapolates correctly into the future. In looking at his past analysis, often what he has suggested will come true actually does come true. So, I'm very excited to have TP, uh, on the podcast.
He works, uh, in a quantitative field and has a background in computer science. TP, welcome to the podcast. Oh, hi, Steve. Thanks for welcoming me on the show.
Great. It's great to have you, and I'm really looking forward to this conversation. maybe you can just tell us a little bit about how you got interested in this subject, how you follow this subject, because I, I think you're bilingual, and maybe what are some of the best resources? Yes. So I got into this subject because one, I'm Chinese American. I was born in China and then I, you know, uh, grew up outside of China.
And then, I came to America for work and during this time I have, uh, you know, a lot of, uh, connections to both, both countries. And I found the entire tech war to be extremely interesting. So I made it a personal priority to read up as much on the topic as I, as I can. And I would say that, involved a lot of, going through all those, sources, the Chinese language sources I can find online of different companies, investigating every company that that's listed on the stock market that are semiconductor related in China. And, uh, reading their research report that analysts put together.
I also have some, insider sources, I would say, in the Chinese semiconductor industry. Not many, but a couple that helps me out a little bit. And I've spoken to a lot of people with, on the subject just to make sure that I'm thinking in the right direction. And I think over time, just by reading through all these different sources and knowing, both China, Chinese society and the American society and how we think a little bit and the competition that we have, I think over time, I have been able to have a, you know, a picture of. Where the competition is right now, and what are some of the areas that are things I should be focusing on, and others should be focusing on, you know, going forward. Can I ask, uh, to what extent is it kind of a purely intellectual interest of yours? So for me, I, I like to follow this stuff just because, wow, this is, these are really important events that are going to shape the future of our world.
I don't actually trade, like, financially on my insights in this area. I just kind of do it as an intellectual exercise, the same way I would follow some, you know, sub discipline of physics or some other science. I'm curious, uh, whether that's true as well for you. Are you, or are you, did you short Apple right before it tanked? No, I actually, you know that on my, on my Twitter profile, I actually post the most about EVs and BYD.
So I am long BYD, but more because I have a, I think they're going to do well. But in general, I do this as pure of a academic just interest level kind of topic. And also because I, you know, as a Chinese American, I really do hope that, uh, you know, we can work through things between the two countries. Yes, I think it's, you know, we could have a positive sum game between the U.
S. and China, not a zero negative sum game that we're involved in right now. you know, I, I, I should mention that I think I actually became aware of you originally because of EVs, because, and, and, and to my audience, I would love to have TP back if, if he has a good time talking to me for the next hour, I would love to have him back to do a whole nother episode on EVs because EVs and batteries, uh, are another area that I've been following. And if you want to follow that, you have to understand what's happening in China. And, and that's all actually what led me to. TP in the first place and then only later did I realize, Oh, this guy's also following what's happening in the chip war and, and his insights on, you know, what's just happened in the last six months have, I think, have been very good, but I would love to have him back.
Uh, and we may have another episode on EVs and batteries. That would be great. Oh, good. So good.
I've booked you. I've booked you now for another episode. Thanks. okay. So TP, let's talk about what just happened. I mean, it was literally like, I think a week ago, right? And, we can set the stage a little bit.
My audience probably knows some, you know, has some general understanding of this, but it's always good to, like, lay down some specific facts. So, U. S. has imposed very aggressive sanctions on Chinese companies, uh, related to the manufacture or even in some cases buying of advanced semiconductors chips. And Huawei, which a few years ago was actually either number one or number two in global smartphone sales, had pretty much fallen off the map. And people were questioning whether they would even survive as a company in the handset business.
Now, of course, they're quite strong in 5g networking and infrastructure, but, but in the smartphone business, handset business, people were questioning whether they would survive. So I think both you and I were surprised when they announced the launching of a flagship phone. Which is built using some very advanced seven nanometer Chip technology. So maybe you can just react to what I just said. Yes, Steve. So, I think if we go back a little bit, obviously the, uh, the sanctions on Huawei itself started before, uh, the October 2022 sanctions, those were, you know, while we got put on the entity list at a very early point, maybe 2018, 2019 range, and then at a certain point, there was more rounds of sanctions that, prevented companies like TSMC from actually, Making the chips that Huawei itself design.
And, uh, the last round of those sanctions around, I would say 2019 or 2020 basically, uh, stopped Huawei from making any kind of 5g phones. their phone business only survived because, the U S government, uh, basically gave permission for. Qualcomm to, to sell some of their flagship, uh, Snapdragon SoCs to Huawei for both their.
Phone business and, uh, pad business. And, uh, obviously there's, you know, I think they're, they're laptop that are still using Intel chips. So they, they still, they still did buy some American chips during this time. But, because they were cut off from the 5g, uh, space, they were becoming increasingly non competitive in the 5g market in China.
Uh, whereas before the, the sanctions, they were actually. Uh, competing with Apple for the, the top end phones in China, you can see after the Huawei sanctions, I think the, the number was that Apple sold 34 million phones in China in 2019 and it sold 58 million in 2022, something around there. So there was about a 20 million handset pickup for iPhones after Huawei got, you know, strangled in this case. So that was quite, the blow to Huawei's bottom line and, uh, heading into this year, they said they were actually turning things around and, uh, finally, uh, reverse the sliding revenue declines. But, uh, in terms of 5G, we still didn't really know where they are.
And, uh, we also know that even though they were still turning a small profit, but they were spending so much money on the, you know, R and D, and they were giving a lot of bonuses to their employees. That they were actually bleeding cash. So they, you know, financially, they weren't at the best position at the start of this year.
And, so that's kind of brings up to the start of the year. It was also at this point around, uh, January or December that I first heard about, uh, return of Huawei 5g phone rumors from a pretty good source. And at the point at the time, that person actually said the Huawei.
S. O. C.
Which is, you know, system on chip is what the small, uh, main chip that you use on your smartphones. Uh, he said at the time that the S. O. C. The Huawei will use is actually a stacked. 12 nanometer chip, so, you know, stacking or advanced packaging is, you know, something that, uh, chip design or chip makers use to, I guess, improve the performance of chips, if you can, if you can think about it that way, and the idea was.
People thought that China itself was not able to make any more 7 nanometer chip or they, you know, they never had a, a really solid process was high enough yield for commercial usage. So, 12 nanometer made a lot of sense at the time because, it seemed like that was the best. Technology that was available to Huawei domestically at the time. So I actually tweeted about this news and that surprised a lot of people. But, uh, it's, it turned out to be wrong because they ended up with a more advanced process. But, uh, you know, they, you can see that they have been working on this for, quite a while.
And I also had known from, uh, from my sort of, You know, source that SMIC, I think I'll just call them SMIC from now on, just, it's easier to say so SMIC had, was actually producing seven nanometer, uh, wafers. They were, uh, they were doing seven nanometer, chip making, and they were doing quite a bit of it, but nobody knew at the time who the customers were. It wasn't, uh, obvious from just looking at their usual customers that we know.
That had been, using the 12 nanometer or 14 nanometer process, who was actually, doing buying the 7 nanometer production from SMIC. So we just, I had, my intuition at the time was Huawei, but you know, obviously that was, uh, not confirmed for a long time. And throughout the, throughout the year, as, uh, we went on, you start seeing Huawei's, uh, chip design business. starting to pick up more steam, you start seeing like Huawei really, uh, promoting its AI technology.
It's, you know, I send NPUs and you start seeing more companies out there working with Huawei and you start seeing, uh, some of their former well known Kampoong, server CPUs, starting to make it way, make its way into market again. So the system. At that time, I suspected they were getting new productions from SMIC, but I was, you know, I wasn't, I was, I didn't know for sure. So I didn't really want to report on that. Quick clarification on that.
For their AI Slash server side chips. Do we know that those are on a 7 nanometer process? so we do not know for sure that the current ones are 7nm, but we know that when the original SN910, you know, MPUs came out and when the Kenpoo 920, uh, CPUs came out, they were using TSMC, 7nm process. Got it. So let me just clarify a few background things for my listeners. So number one, uh, One of the things that caught my eye about Huawei as a company Uh, this is many years ago, maybe 5 years ago or more when the Mate 10 was their flagship. Now they're up to the Mate 60.
uh, Huawei is well known as an infrastructure provider for telecoms. And so they are the main provider for the infrastructure that provides 5G. connectivity and, but that's mainly sold to carriers, carriers like, uh, Deutsche Telekom, for example. when they got into the handset business, I realized that not only were they making world class handsets, but they had a chip design arm called HiSilicon. HiSilicon, and HiSilicon designed, I believe, the first flagship level. Uh, I don't know if you would call it the S an SOC at the time, but anyway, a CPU used in phones.
And I remember I bought, uh, one of their early models just to see if the, that early cure and processor was any good. And it was actually quite good. So Huawei has strength in chip design and, uh, as both T. P.
and I mentioned at one time, they were probably the strongest competitor to Samsung slash Apple in the, uh, handset business. Now, the other thing that they do is design, for example, AI chips. And so HiSilicon has an AI chip, uh, called, uh, it's an NPU, so Neural Processing Unit, or also sometimes you would call that a GPU. And that competes against NVIDIA's, for example, A100.
so it's, it's quite a powerful company with capabilities across. A huge spectrum of technologies and possibly now getting into chip into actual, uh, semiconductor fabrication. Not, not clear yet. We don't know. But anyway, that's a little bit of background on Huawei's, uh, technological capabilities. Yeah, they're definitely a very impressive company from a chip designing point of view.
if you talk to the people on the street, like involved in the semiconductor business in China, they would say that HiSilicon is by far the most capable chip designing, uh, chip designing firm in China. And they actually lost some people to different other companies, uh, because of the sanctions. And some people, I guess, felt they lost faith in, just... The entire business model, but they, they're still very, very strong. Yeah, so just to clarify, so HiSilicon is a world class chip design house for our arm for Huawei, but they were dependent on TSMC to actually physically fabricate.
The chips and so when the sanctions hit, uh, in a way, there was not nothing for high silicon to do because they couldn't get their designs manufactured. I do want to mention that 1 of my postdocs, uh, a former theoretical physicist who worked with me for several years, his 1st job when he left physics was as an AI researcher at Huawei's. I don't know how I always still has a Silicon Valley lab.
But when he started working for Huawei, he was working in the Silicon Valley lab, and, uh, he's since risen to, uh, the role of director of AI for one of the major cell phone companies in China that, uh, and, and as you may know, a lot of the photography. The photo, the capabilities for photography on handsets is due to A. I. And so there's a huge number of people that work on image processing and low light imaging, et cetera, et cetera.
And these are, these are people who are specialists in a certain type of A. I. So I, I do have some have some insight into what Huawei is like as a company through my, my former post on. so let's come back to the announcement.
TP of this new phone, and maybe you can summarize what are the most like important features or capabilities of this phone, uh, that, that are going to impact the future. Yeah. I saw, I think like when people look at this phone, I think too much focus is on the seven nanometer itself and not on what this technology tells us. So the first thing is, uh, in generally, when you see a country like as big as China make progress, it doesn't make progress only in one area.
It's going to be making progress, uh, most likely at similar paces in different part of semiconductor industry. So I think that the first thing it tells us, if you. You know, if you had a certain view of where China's semiconductor industry is now, you should rethink where their semiconductor industry was back October of 2022 when the sanctions were originally levied. So what this tells us is that in the chip making side of things itself, it's at 7 nanometers. But what does this tell us about the other aspect of it's, uh, Semiconductor industry. For example, what does it tell us about? It's chip making tools.
How far advanced are those areas? Where is it in the memory chips industry? Where is the rest of the chip design houses in China? Where is it advanced packaging industry? how far is it progressing in China? What we call third generation and fourth generation semiconductor materials like, uh, gallium nitride and, uh, silicon carbide and also gallium oxide. These are all very important topics, a very important technologies, uh, for people to think about. And, uh, the, the one thing that, there was two things that really surprised me.
Well, three things actually. One thing was that the chip itself. Came out to be way more advanced than I expected. Why do I say that? Because of, uh, a couple of things. my original expectation was that while SMIC was doing seven nanometer, it was a very early seven nanometer process.
So, uh, Huawei, if it designs, uh, you know, a flagship or any kind of SoC. It will not be at the same level as Snapdragon 8 Generation 2, and it's also likely to be stacked, which means that, it will allow it to have a better performance than your usual 7 nanometer, process. but it turned out that actually it was not stacked.
the chip itself was about the same size as the previous, uh, Kirin. Uh, 9, 000 S. O.
C. And it had its own five G. Integrated five G. Modem. It had its own, uh, M. P.
U. G. P. U. And, uh, you know, the S. P.
S. All the things that you would want to having like a flagship S. O. C.
That's really powerful. And the performance itself actually ended up being better than what I expected to in terms of things like power consumption. It was, obviously not at the same level as you would see from a four nanometer or a five nanometer process, but it was better than, uh, snapdragon 88, which was, you know, came out two years ago and, it had really good peak performance. And, uh, they were able to use what's called hyper threading in their CPUs to get it to, uh, have, uh, performance that are as good as Snapdragon 8 Gen 1.
And while that process itself utilized a lot of power, you know, overall, it's, people are not going to be using their phone at max. Uh, power or max, uh, requirement for a long period of time. So, even if it's a little inefficient at the higher end, it's not something that's going to degrade the performance so much that it can't be a flagship phone. So, so obviously the, the SOC itself was actually very impressive based on what kind of process they had.
It turned out later that SMIC's process was not as advanced as I thought. When I first saw the, some of the test results. So that actually spoke more for the, how well designed the chip itself was. So the SOC itself was, it was impressive. On top of that, they also had their own 5G, uh, RF front end.
So, this will be for things like when you, when signals come in, there are switches that, will then go through power amplifiers, which then goes through, like, what they call BAW filters. And then, uh, you know, gets converted to signals that you, you process in your 5g modem. And, this is an industry or this, this industry itself has been dominated by American and Japanese companies for a long time. For something like BAW filter, which basically is essential to get the 5g signals. And I think Skyworks or Skyworks or, Broadcom have something like 87 percent market share. I can't, I can't remember right now which one it is.
And, there's basically very few players outside of, you know, American firms in this. So the fact that, uh, Huawei was able to find like a local source for this. Product was very surprising. So, I want to come back to the 5G modem and the RF chips that you just discussed, but before we leave the fab fabrication question, so the 7 nanometer process that SMIC, and first of all, I think it's SMIC, but I think you also think it's SMIC, but I'm not sure how fully established that is. they had to use something called multi patterning because the lithography machines they have available to them are only deep ultraviolet.
The U. S. is not allowing ASML to sell Chinese companies an extreme ultraviolet EUV lithography machine, which is preferred for the smaller feature sizes. So it was a challenging process to be able to produce seven nanometer level chip features using DUV through something called multi patterning. And there are questions about. What the probable yield is for this process.
So I think it's fair to say China will probably be stuck with DUV lithography for at least a few years. We'll talk a little later maybe about their own indigenous DUV lithography machines and maybe future EUV machines. But let's just, for the moment, just focus on the following. Let's assume they're stuck at DUV. Uh, what do we know about the yield and the impact of that on the cost of seven nanometer chips made in China right now? Yeah, so there has been a lot of speculations on this topic. Certain people online have quoted as low as 10%, which I found to be very hard to believe.
In general, I would say that, uh, there was an article today by Dylan Patel on this topic and he explored it and, uh, the people that, uh, looked at the, Kieran Neistat and S from Tech Insight also got interviewed. I think, you know, I would say the yield is over 50 percent for sure at this point. And, it's one of these things that over time, as they get better at it, the yield will keep improving. And, in terms of like how high it is, I don't think anyone has a, you know, a, a real knowledge, unless you are with Huawei or SMIC. So, but my personal perception is it's probably in the 70s by this point, because.
Of the, the rumor order size from Huawei. So they are, going to be cut off from the American chips, uh, supply most likely. So next year they're planning something like 60 million phones.
That's a, that's a rumor that's out there right now. And on top of the 60 million phones they have, they're probably going to sell 10 million tablets, probably 5 million desktops. And they're also going to need to make, uh, server chips. And also AI chips, and obviously as the chips themselves get larger and more complicated, the yield gets lower. So when I did the calculation for these things, it would be very hard for them to produce that many chips or anything close to it, unless they feel confident that the yield for something as small as a 110 millimeter squared.
SoC is in the seventies, I would say, and I wouldn't be surprised if he gets to, you know, 80, 80s at some point based on the Dylan Patel, article. So, so yield, in my opinion, at least for the, the, the phone, the phone SoC itself should be okay. Uh, now in terms of costs, Can I just comment on that a little bit? the, the guy who's leading these efforts at SMIC, who brought a huge team with him, which included both Korean and Taiwanese engineers, was actually the guy responsible for getting both Samsung and TSMC right down to, uh, very small feature sizes.
So, so they have a lot of experience, right. Uh, in doing this, Oh yes, this guy's done it before. and I, just to re, just to. Uh, summarize what you're saying, at least on certain things, like, I think the number of cell phone orders, uh, that they're willing to take, uh, Huawei, and also the fact that they even launched another phone using, I think, the foldable phone, which is also using, I think, the same chip, it does express some level of confidence, like, they do seem to think they're going to be able to produce huge volumes of chips like the Kirin 9000s. And so then you would infer from that that they do, they're either, I mean, they could be making a mistake, but they themselves are clearly confident that they have some level of control over the yield.
Is that fair? I would say so. Got it. Okay. So, now one of the calculations you did, which I thought was the most impressive, uh, it caught me by surprise, is you, you said, you actually calculated, like, how much were they paying to Qualcomm? Like, if they launched a handset and, but they had to buy the, the, the, the SOC from Qualcomm, they were paying Qualcomm, you know, whatever, maybe 160 bucks or whatever it was.
Uh, but instead they, they keep that if it's their chip, right? The money goes back to Huawei if it's a Kirin chip. and so they could tolerate somewhat more expensive fabrication costs, uh, because they're now no longer paying effectively, like, the profit margin of Qualcomm. So maybe you could talk about those numbers. Yeah. So one of the things that surprised me was just how expensive the, uh, the latest Snapdragon SoCs came out to be.
the ones that Huawei itself bought are actually not the generation two or the generation three Snapdragon 8 SoCs. It bought the first generation ones, which probably weren't weren't as expensive. But if it actually wanted to stay in the smartphone business going forward without having its own flagship chips, it would probably have to pay the 160. per, I think it was 160 or 140, but the, the gen three actually I saw as high as $180 per ss o c.
So if you just do like a, you know, calculation head, if they were to sell, uh, 50 million phones next year, uh, 180, dollars per, per phone multiplied by 50 million, that will come out to be $9 billion a year. And, based on that, I calculated that even if, the cost of fabricating for SMIC was as high as a TSMC 5 nanometer process, just, you know, based on all the tools that they have to buy and, uh, you know, the amortization depreciation costs associated with it, they could chart like SMIC could charge the same price as what, uh, TSMC charges for 5 nanometer wafers. And, uh, why we could buy that and they would still be able to, uh, pay a lot of monies to their employees, uh, do all the testing and, uh, packaging. And still profit like a 50 percent margin on those things.
It was the mass was really mind blowing to me when I calculated it. Yeah, you were the first person I saw who actually looked careful at these numbers and I was actually a little disappointed in the Wall Street guys because the Wall Street guys should have been on this right away, but they, I didn't see any similar analysis for a while from these guys. Because that was a tradable event, right? If you, if you figured out like, oh, wait a minute, they could really do this. That's going to have a bunch of it. Like, I think you were ahead of these guys by like, at least 24 hours or something.
now, as far as I understand it, the, the, the cost per chip is scaling more or less linearly with yield, right? So if I get slightly less yield, I just have to run the process more one over the Delta and the yield. And and pay for that. So it increases the per unit cost. But like, if you're varying within 50 to 70%, sure, you maybe pay a little bit more for the chip, but you didn't have an alternative, right? Either alternative was either to pay a huge profit margin to Qualcomm for that chip or not have the chip at all because you're sanctioned by the U. S. Government.
Right? So it seemed like even if their yield wasn't that great, they still might have a business opportunity there. Sure. I think the business case is there. I think the bigger question is whether they have enough equipment to actually run the process. a lot of people actually look at the process and they say. Well, okay, so Spick has this many lithography machines.
Okay, so based on these lithography machines, I think they can, uh, run, they can produce this many advanced seven nanometer chips. And I look at that and say, well, they also need other equipments also. And while, uh, while Spick itself is, probably has enough lithography machines to increase production. But it doesn't have enough of the etching machines, you know, many of the other kind of equipment that you need for the high volume production because they lost access to lamb and, uh, a match, you know, was in the past year, obviously.
So there is a question of if they can ramp up, the production and capacity as make, uh, sort of domestic equipment producers. right. And so that's an open question, right? Yeah.
Yes, I think, you know, I think if, if we go back to what I talked about before, if. we have been underestimating where, uh, SMIC's, uh, 7 nanometer, uh, manufacturing capability is. We may also be underestimating where the, uh, the different, uh, equipment manufacturers are at in terms of their, the quality of their equipment and such. Right, so this whole, this thing ties into all these issues, like, like, how far along is Amec, right? Uh, uh, so, one of the things is, you know, if the premise is the people at Huawei are not idiots, and clearly they're not.
And the calculations on, like, okay, what capacity do we actually have for all of these different aspects of the production chain? Surely, those guys can figure it out, right? They talk to their counterparts at SMIC. They talk to, you know, and I guess there's been talk of, like, mysterious buying of used DUV machines and all kinds of other stuff going on. you would guess that they have this lined up. Like, if they actually say they're going to, they're going to have tablets and, you know, uh, server chips and AI chips, then probably they sort of can see the path to actually producing them. Yeah, I, I think people should focus a little bit less on the lithography machine because I'll just tell you right now SMIC has enough lithography machines to increase the production by a lot.
So, so the. And, you need about, for 50, 000 wafers per month of 7 nanometer, you need about 20 advanced, uh, our, uh, DUV lithography machines. And, based on my estimates, they have more than that right now. right. So the bottleneck would might be in other stuff, but the other equipment, they would know if they have these guys, that calculation, these calculations are pretty straightforward. So they would know if, like, the bottlenecks going to be shifted.
Into some other aspect of the production process and they would know it well in advance, I think. Yeah. And I think they are actively validating the domestic tools right now. so a lot of the.
Some, I think, of all the equipment manufacturers from China, AMAC is probably the one that's most vocal about its progress, and it has shown some really remarkable improvements in the processes that it supports, uh, if you look at their, you know, the presentations for both, uh, memory chips and logic chips. So, My guess is that Huawei looked at it and then they based on their own estimation They think SMIC will be able to support all their, all their chip needs. Yeah, So when I talk about that I don't mean just by phones or computers and things like that. I'm talking about Earbuds, your smart watches, your TVs, smart home, everything. Like they're gonna need SMIC to produce all the stuff for them.
right. So again, there are certain assumptions made in this chain of reasoning that we're discussing, but it leads to a scenario in which I think you alluded to this in your comments across the whole production chain for semis. China is the gap is less than what we thought. And Huawei's behavior again, under these assumptions is suggesting that the domestic producers for some of these other categories, like, which compete with a matter, compete with, lamb or whatever that gap is closed faster than people expected. I would say so. And, I don't think it would have closed as quickly if there wasn't the October sanctions because it actually forced SMIC, for example, to work with, AMEC a lot more closely because they, they don't really have a choice.
This is a point that I make a lot to people, which I think if you're not, if you've never been a startup guy or a business guy, you don't understand this point. So, as a startup guy, I am confident that my team, because we're focused, we don't have bullshit, we don't have bureaucracy, we're focused and we recruit only the top talent. I feel confident that I can match up with the corporate team at a big company.
The thing that hurts me is nobody wants to go with a startup. Nobody wants to buy the machine from startup. There's so many risks. We could run out of money. We might not be able to scale up fast enough to meet their demand.
If they commit to our technology, there's so many reasons why, uh, startups fail generally tech is not one of them. It's not usually if it's a, if it's a mature product that I'm trying to match an existing large company, that's building that product, the technology I can catch up on. The problem is these other issues. But once you sanction the entire Chinese economy and you say, Hey, guys, you're not gonna be able to buy these things from the US or the Japanese or the Dutch vendors, then the question is purely a tech competition. Can the startups in China then close the gap on certain things like EUV, which are really hard? It's going to take a long time to close that gap.
But a lot of things which are nevertheless just standard components of semiconductor production process. Thank you very much. I am confident that small companies could close those gaps if given the opportunity. So the way I would say it is the U. S.
sanctions solved a kind of alignment problem or coordination problem within China which basically forced all these companies to work together and it's to the benefit of all the startups that are trying to catch up. So that's how I explain this at least. I'm not sure that's actually the situation and that description might only apply to certain subsets of the production chain but as a general remark The U. S. did, in a sense, a huge favor, at least to the startups in China that are trying to catch up on technology.
Yeah, so Steve, I'll, I'll use the example of this, of something I read, from someone, you know, who works at, uh, SMIC. And, uh, it's regarding to their new fab in, what they call the Jingcheng fab. It's actually, it's called Jingcheng because it's in Beijing.
And, one of the things about that fab is when SMIC opened it up, they wanted to make it, a de Americanized production line and people that are in charge of, uh, getting the yield up in SMIC was really not happy about it because they said we prefer to use AMAT equipments. Because we can get 90 percent yield for 28 nanometer process using AMAT material, AMAT equipments. But using these Chinese equipments, we can only get 75 percent and I'm going to get penalized on this. So, but you know, uh, the company itself actually, uh, pressured, you know, said we have to go this direction because We could get sanctioned. It turned out that, you know, the entire industry in China did get sanctioned. So over time, obviously the, the domestic producers will improve their, in their quality and they'll work with the, the major fabs in China to actually improve their yields until at least for things like 28 or 40 nanometer or in 14 nanometers.
They catch up to the same quality as, uh, or some, something close to where AMAT and long is that. You know, a lot of the sentiment of Americans who, you know, are hawkish on China, they're like, we want, we got to smash these Chinese guys, we can't let them catch up with us technologically, blah, blah, blah. Even if you have that sentiment, uh, very, you have a very strong desire to keep them down. You fear them, you loathe them, et cetera. This still might be the wrong strategy.
The, the current sanctions, which are put in by people, I think, who fundamentally don't understand technological innovation or semiconductors, it could be a, a bad move for the U. S. even if you had as your goal, just crushing the Chinese, you know, techno, tech innovation landscape. So I think that's something people have to consider. You can't just say like, Oh, I'm doing something mean to Chinese companies, so it must be good.
But the world is not that simple. Yeah. I would say you're completely right about that. I mean, China is not Russia or Iran. It has a lot of resources at hands and a lot of money at hands to, uh, throw out these problems. Yeah, it if you calculate the number of Engineering type graduates from chinese universities.
It's you're talking about easily something like 5x And if you, you know, you could even be 10 X times the U. S. So if you, there's another point I sometimes make, which is that if you look across many areas of technology, whether it's space, satellite communication, uh, semiconductors, EVs, uh, AI, quantum computing, material science, chemistry, you name it, they have enough human capital that if the other conditions are right, they can catch up and catching up is easier than innovating. So, you know, it's a separate question of whether once they get to the innovation frontier whether they can push it forward as fast as the Americans can catch up is really constrained by capital, human capital and financial capital.
And, uh, everything I see across all of those different sub sectors shows that they're catching up extremely fast and in some cases, surpassing the West. So I think people are just not aware of the numbers, the scale of how much human capital they have to draw on. Yep.
Definitely agreed. So, let's now, we were talking about fabrication, let's now shift to the modem and, uh, RF chips. So, the point that you made, Which is that, you know, first of all, 5G modems are pretty hard. Right? Right now, everybody is basically except Huawei now dependent on Qualcomm, right? Apple was trying to actually develop its own 5G modem for its phones at the moment, as far as I understand, hasn't succeeded. And so they put in large orders, I think, even for the next iPhone.
Is it iPhone 15? 14 or 15? they put in orders. It's going to be Qualcomm 5G modems in those phones, right? So, I think it's, they're, they're getting until 2026. so, It's a tough technology. Anybody who says like, Oh, these Chinese guys can't innovate. It's too hard, you know, but nevertheless, within Huawei, it seems they've managed to actually build a 5g modem, uh, chip or chipset, which is comparable to Qualcomm. As far as I can tell from all the independent tests.
So if you go on Chinese internet. You can see guys, uh, in places in China where they have a real 5G network and they're downloading almost gigabit per second speeds, uh, to their phones from the wireless network. Uh, and they, they compare the, the, the Mate 60 against the latest iPhone and seems like Mate 60 is at least as good or comparable. So it does seem that there's evidence that they've closed this 5G modem gap.
Do you, do you want to comment on that? Yeah, I think the ongoing joke online on China is just how bad the iPhone signals are in China compared to everyone else. And I think one of the rationale is that iPhone doesn't have an integrated modem to their SoC because it uses the Qualcomm modem. Right, But I was actually very impressed, not just with the modem itself, but its integration with, all this. Front end analog chips, chips that we talked about that you just never know when a new player is coming into industry, they can say they have, we have a product that's comparable to a Skyworks or a Broadcom products, but you, until you test it out and verify with your system, you don't, you don't know.
So, so what we actually saw is, Huawei, not only doing an advanced SOC and a 5G modem to it. They also cultivated a domestic, RF supply chain from almost nothing. right. So, so RF is radio frequency. And you're talking about the thing that allows this tiny antenna in your phone to connect to a base station, or even potentially to a satellite, which we'll talk about in a moment.
So RF technology is, it's, it's very tricky technology. That's real electrical engineering, right? It's like you're, you're talking about, uh, signal processing, pretty hard stuff that, you know, honestly, like, as a professor in the, U. 30 years, The kids who study that, the kids who go to grad school to study that stuff nowadays, chip design, analog, chip design, it's generally not Americans, right? Americans who are closer to the money figured out, Oh, I can make more money more easily in software, or social networking startups, blah, blah, blah. Those kids are not the ones who go into chip design or these really hard, E. E.
Electrical engineering type things. It's basically kids from China. And so, to say you're surprised that the Chinese can catch up in these areas is kind of, to me, like, very stupid. It just shows, like, you don't really understand who's doing this work, even in America. yeah, I agree a hundred percent.
Like, obviously, we know what the, you know, professions in America that you can make the most money and that, that kind of draws people in. yeah, and you, you were mentioning, like, okay, the leading, one of the leading companies in this space is Broadcom. Like, big U. S.
chip company, right? If I said, like, well, who's working at Broadcom? What new engineers do they hire? Or even team leaders who are designing their chips? It's probably not people born and raised in the United States. Or if they are born and raised in the United States, they're immigrant kids. so, to say you're surprised that they can't, that they could do this in China, is like, really, kind of, like, just shows, I think, your ignorance of all this.
Steve, can I just raise another point I think hasn't been talked about enough at all, yeah, which is, which is this. So we talked about how important this radio frequency area is, and we talk about these existing engineering talent. You're having this, these amazing American companies in Qualcomm, you know, Skyworks, Broadcom, and Quervo, and I keep mentioning those four companies because they have existing engineering talent who have years of experience in this field. Now they, a lot of them, especially Qualcomm, but the other guys also, they get a huge percentage of their revenue actually from Chinese OEMs who use their stuff for their phones, for their tablets, for the, any kind of Wi Fi front end modules, any kind of smart, smart home modules right now, they use that. So if you think about it, uh, China is the largest, you know, international market for sure for them, but also it could be the, Largest market overall for, for Qualcomm.
And then if Huawei's domestic suppliers can ramp up their production to the point where they can supply all the Chinese OEMs, then these companies in America, they're going to suffer huge revenue losses. And we know what happens when companies suffer revenue losses in America. Wall Street will tell them to, you need to cut back your expenses. You need to, you know, reduce your headcount. And then when we reduce headcount, what happens in the process is that the best trained engineers we have, they don't stay in this industry.
They go somewhere else. So then you lose all this talent over time. Yeah, I, you, you were very perceptive again, I think, uh, in pointing out, I think, in your tweets or comments. Posts that, you know, if the OEMs who are using these chips to make phones, whether it's Oppo or Xiaomi or Huawei, or even, uh, the Apple phones are made in China, it's much easier for a Chinese supplier once they've shown that their chips are comparable quality to Qualcomm, and they can produce enough of them for them to get into the production, into the supply chain, goods.
In China is going to be that much easier, right? Uh, it's a small world, right? So, Qualcomm and these other RF chip manufacturers, I think their future revenues are definitely at risk now. Yeah, I did some, calculation. I, I would say what, what we know is that for each phone, you have about 35 off our front end. So for, uh, the 700 million phones that Chinese OEMs produce every year, uh, you know, that works out to be. Over 20 billion, and then if we add in all the other stuff and, you know, the Chinese suppliers could also supply Samsung, for example, uh, you could very well look at a 30 billion pie that gets cut down to much less than that for American companies.
And that really cuts down to, cuts down on R and D. Do these RF chips also have to be made using a 7 nanometer processor or is it more of a trailing, uh, edge feature size? Uh, they don't. So like, this is another interesting topic. So the two main ones that people care about are the, uh, BAW filters.
which gets, made in what's called a MEMS fab, a micro, microelectronic, uh, sensor, I think, something like that, and, uh, the port amplifiers use what's, uh, use, they get fabbed in what's known as, uh, gallium arsenide. Uh, fabs. So none of these need the latest seven nanometer process or lithography machines, but, uh, they're also very technical in their own ways, but they just, you just don't need the latest lithography machines for them. So it's, it's plausible there that the capacity might exist in China for these local manufacturers, local designers of RF chips to actually, uh, ship a lot of products. so one, the, the main, gallium arsenide foundries are actually in Taiwan. And, uh, what they've noticed is that in the past year, uh, the Chinese, uh.
Foundry, which is called sand has basically overnight seeing it's a market share climb from 5 percent of the market to 15 percent of the market, whereas the Taiwanese, uh, gallium arsenide foundries are sitting extremely underutilized. So, and this is a process that's likely to. Keep going forward and depending on what happens with the gallium sanctions that China placed on the rest of the world We'll see what happens to these foundries. Yeah, it's super interesting.
I, you know, there's gotta be some, uh, analysts who, you know, track semis, semiconductor stocks and companies that. You know, are trying to drill down into this and track this more carefully. And they have to have people on their teams that are fluent in Mandarin, et cetera, so they can do some boots on the ground research. I, I think there are huge trading opportunities. Not, not that I run a hedge fund that invests in semiconductor stocks, but, but it just seems like an obvious thing. That's.
Uh, certain amount of research could pay off. you, you and I should start a consultancy for, for a semiconductor analysis. But anyway, let, let's talk a little bit about the satellite capability of this phone, which is, I believe, unprecedented, right? So it allows for voice anywhere with coverage to, uh, I forgot the name of the, Chinese satellite systems at Tiantong. Yes, Tiantong.
So that was actually one of the most surprising part of this is incorporation of satellite technology So beforehand actually was made 50 and p60 while we had already demonstrated Uh, two way SMS capabilities using the Baidu, Beidou navigation system. So that wasn't new. but this, uh, satellite phone capability is actually new. And if you see some of the map coverage, the Tiantong satellites can cover a good chunk of, uh, Southeast, like all Southeast Asia, Japan, Korea, even all the way to like, uh, United Arab Emirates. It's amazing. Now, I don't know if it's actually used that far in practice.
I think right now, China Mobile, no, China Telecom is, is saying it's only operational on the ground in, within the border of China. But people have already used this on airplanes, so that's kind of interesting. but the more, the more interesting engineering part of this is they manage to somehow make the chip, uh, small enough. And still work inside of smartphones. So the old satellite phone we have all look extremely clunky with huge antennas. And these, the Mate 60 clearly are just, just a regular smartphone.
And the chip is strong enough to go all the way up to geo satellite, which is really far away. And, and receive signals from that. So, and we're going to see more phones.
From the Chinese market with this capability is going forward because, uh, you know, other OEMs are going to want to catch up to this capability. Yeah, I don't know to what extent those capabilities are available from other vendors than Huawei. It could be a unique leapfrog for them that other people can't currently reproduce. In a future podcast, just to advertise, I, I, there is a, uh, a friend of mine whom I've known since we were both students at Caltech who is, The founder and CEO of a company called Lynk, L-Y-N-K, which is a satellite company, which, uh, now their satellites are, I think, low Earth orbit.
Uh, their network allows you, using an existing cell phone, the existing antenna on your cell phone, which is meant to connect to, like, a local, uh, uh, base station, like, in your neighborhood. that same phone can talk to these Link satellites that are in orbit. And so he is selling, uh, emergency calling from an ordinary existing cell phone through his network to big carriers, uh, uh, telco carriers around the world. So just as an emergency feature, mainly, so I'm going to have him on the podcast and we're going to discuss part of what we're going to discuss is what these Huawei capabilities, but it does seem to leapfrog anything that I had heard about from other.
Uh, companies or vendors, Yeah, I think like this technology itself from what I read, it's basically going to be a very important part of the 6G network where you're going to have your lower earth orbital satellites, which provide additional transmission data transmission points. So that no matter where you are, if there's no ground stations, you can still get streaming, you can still, you know, download as much data as you want, provided that obviously your, you know, your chips can actually handle this much power without using up all the battery life. Right. So, but, you know, back to again, like, uh, this is something that's kind of popped into my head from reading your analysis. Even if you are, say you are behind TSMC in, you know, chip fabrication. And so, you know, your SOS, your, your, your main CPU or GPUs in the phone are a little bit slower, less power efficient, and you maybe have to pay a little bit more.
You know, per unit for them, these other capabilities in your phone, like the sat capability or, some other aspect of the phone, you know, might make the consumer willing to pay more for it. So you can trade those things off against each other. It's not the simplistic analysis that I hear from a lot of American analysts is like, oh, 7 nanometer. That's like 5 years old and, uh, they're still way behind and Apple's next chip is going to. Beyond a three nanometer process, so surely these guys are way behind. Phone competition, you know, competition between handsets isn't that simplistic.
It's like a bundle of lots of different features that determine, uh, which handset is preferable to the other. And, you know, I wouldn't actually, it's, I think there's already pretty ample evidence that nationalist sentiment in China is going to buoy Huawei sales. People in China understand that the U.
S. government unfairly targeted Huawei and just out of spite, they're going to buy a Huawei phone, even if it's slightly less good than a foreign phone. And so I wouldn't, I wouldn't underestimate that, uh, dynamic. Yeah, I think like most people when they use their phone, they're not going to know what their power consumption data on their phones. I have no idea any what that is for any of my phones. I don't know what the ISOC is on my.
You know, Google Pixel phone. I don't really care. Uh, things that people, you know, that are like really good advantage for this phone, Mate 60, apparently its camera is really good.
Its speaker is really good. Obviously, the 5G signals are really good. And it has, uses very advanced fast charging technology. And its, uh, heat dissipation technology is good.
So even though... With an older process, you generate a lot more heat because the heat dissipation technology is so good that the phone, the surface of phone doesn't really get that much hotter than your four or five nanometer S. O.
C. So, yeah, So, so you, you could argue that Huawei's strength in what I call, like, real engineering, like physical engineering or applied physics, you know, allows them, like, say their phone charges faster, right, than an iPhone. It allows them to compensate for, you know, being somewhat behind in the fabrication process for the SoC. And as you also pointed out, people don't notice, people don't really know, unless you're a hardcore gamer on your phone, you are actually not going to notice if you have like a main CPU, which is a couple generations behind, uh, the leading edge, bleeding edge. in fact, one of the things that I often complain about on Twitter is that the Analysts and strategists in Washington seem to think that they're crippling the Chinese military by cutting off nanometer process.
But no one uses anything like that bleeding edge fab process for weapons, first of all. And secondly, even the AI chips that people, uh, are using to train LLMs and things like this. Those are seven nanometers.
So, so really all you're doing is like crippling your ability to make bleeding edge flagship phone chips, right? Uh, when you, when you cut off, uh, China from three nanometer, four nanometer process. I did some, you know, brainstorming of applications that would absolutely need, something that's, uh, T. S. M. C.
In five process or better. And, Aside from smartphones, I really can't think of something that you really can't use something else for because the main, the main downside of having a less advanced process, like a 7 nanometer versus a 3 nanometer is that the power consumption numbers are worse, you need a bigger chip to get similar level performances or, you know, those are the kind of things where if you have a phone, it really matters because of space and power constraints. But outside of, uh, even for laptops, You know, it's not that big of a deal and definitely for data centers, you can, you can compensate for that with better liquid cooling of your data centers. You can use better gallium nitride inverters or something or things like that to save on power consumptions. Yeah, I think this point is totally lost on the people in Washington who thought they were crippling the Chinese military with these sanctions, which is totally wrong.
Crippling the consumer who wants a phone, right? Who wants a flagship phone. But, so now that we've gotten onto that, let's talk about the consequences of, suppose it's true that SMIC has plenty of capacity in its seven nanometer N plus two process to make AI chips. So, last time I checked, there were companies, design firms like BRN and HiSilicon that made AI chips that are actually comparable.
I'm not saying they're better or exactly equal to NVIDIA, latest and greatest from NVIDIA, A100, uh, or whatever. But it does seem like they're pretty competitive. And if it is true that there's enough capacity, then we should see a lot of Chinese, uh, domestically made and designed chips which are competing for, with NVIDIA chips, uh, to be used in AI model training.
Do you want to comment on that? Yeah. So I think in terms of, uh, just in terms of, uh, NVIDIA and AI, obviously NVIDIA has played a really big role in this early stages of. LLMs and AI development in general, but I think a lot of what got NVIDIA to this point is, uh, they're the only game in town, basically. And, everyone basically wrote their back in the days, you know, PyTorch and things like that. Obviously, I'm not an expert in this area, but my understanding is that, there's a lot of CUDA developers and also the, these, these AI software or architectures platforms were optimized to work with CUDA.
That's a different situation inside China where, uh, Huawei itself is a behemoth when it comes to AI. And it has its own stack of ascent platforms and also what it calls MindSpore. So that's their version of PyPorch.
And then that you have, uh, uh, Baidu was paddle paddle and, they have, you know, various platforms domestically for this. And the, the part that I found interesting about Huawei is that it has been training its own developer pool all this time. It has all these Huawei schools to teach people to program in China program to ascend to the point where I think, I think I read there was something like 2 million ascent developers in China, you know, to put things in perspective. I think the last time I read it was, there was about four to five million, maybe five million CUDA developers. So there's still more CUDA developers, but there's a lot of Ascent developers also that are, that can write stuff for the, uh, Ascent GPU. So while there's actually a lot of, uh, quite a few, you know, AI chip, uh, make chip designers in China, like, like Byron and, more threads, MetaX.
There's quite a few of them, but, from a usage point of view, it seems to me that Ascent is, it is the most popular one that's being used in China by different large language models. in, in fact, in Huawei's own, you know, Ascend, uh, press releases and events, they would say that half the LLMs in China are trained using Ascend platform, which is, you know, you wouldn't know that if you were in America. But, so people are suggesting to cut off NVIDIA from China or restricting, Make the, even restrict the a h eight 800 from China. that will obviously, you know, hurt certain people like, uh, Tencent, who, uh, you know, wrote their entire AI stack around Nvidia platform, but they're, it's not gonna stop the other people in that are using the, you know, ascend platform and, and other platforms. So, So, so let me just summarize that for the audience. So CUDA is an acronym for Compute Unified Device Architecture, and it is a platform or a software framework that AI developers use.
And the whole framework was developed by NVIDIA. It's specifically for their chips. And so a big part of the lock in that NVIDIA has is not necessarily just the quality of their chips or GPUs or NPUs, whatever you want to call them. Basically chips that are good at huge matrix operations that are necessary for AI. the lock in is Both that their chips are good, but also that they built the software environment or framework that everybody uses. And so it's difficult to retool your code so that it runs well on somebody else's chip set.
So that's part of the lock in that NVIDIA has, but what TP has just pointed out is that that isn't completely the case in China. So there are some companies that are probably locked into NVIDIA hardware, but there are a lot of developers in China that could use the software framework that's built around Huawei's. NPUs or GP
2023-09-26 01:51
