How Are Chip Companies Navigating US-China Tensions?

The title of your presentation is really very telling and very timely here, AI competition and the future of semiconductors. Can you just lay out kind of the state of play right now amidst this kind of global competition that we're seeing? Well, I think you see global competition, but it remains a competition between companies. What's very interesting is that while geopolitics matters, risk reduction matters, supply chain resilience matters, this remains a product and performance industry. And this remains an industry where if you have the best product, if you have the best technology, your customers are going to buy it. So even though we're five years in, 10 years in, 15 years into the chip war, the industry hasn't really changed that much. And the statistic I mentioned the most is that even though everyone is trying to de-risk from Taiwan, Taiwan's share of global semiconductor manufacturing has actually gone up in the last five years. And the level of supply chain change is actually quite small. Where are we in the competition right now? Everyone's doubling down on AI. We may not see a bump from new smartphone sales, but from a technologist's perspective, what's going into these smartphones is actually quite, we are moving to AI phones, which means you have an NPU inside the processor. We're going to see an upgrade of SOC capabilities. We're going to see real-time transition. We're going to see lots of things of AI, not just in the data center, but at a device level, which hopefully expands the pie of the AI market. Part of your survey also includes asking Chinese companies how they feel about what's going on. How do they feel? Well, so I think the number one thing is that everyone in the industry is optimistic. And again, it goes back to, we say we have a chip war, we say we have all these geopolitical problems. These last five years have maybe been the best five years in the entire history of the industry for everyone. So AI, so Chinese companies believe the future is bright. They're going to see growth. They don't see margins collapsing. They actually see margins going up. But they do have, they do see some fundamental differences with how, when you ask actually Chinese companies, where will the Chinese semiconductor industry be the most successful? They don't say AI. They say automotive electrification, which are markets where China has more downstream industry control and a larger percentage of the global TAM. In addition, while Chinese companies and Chinese respondents are optimistic about the Chinese market, they actually still have a tremendous amount of respect for the United States ecosystem. And the one I always pick out, about one third of our participants sit in China. And we ask them, where will the best talent in the industry go? What place is the most attractive for talent? 85% of people working in China said the United States. And 60% of the people sitting in China said the United States would be the most attractive for capital investment in the semiconductor industry. So I think they're very clear on where the competitive balance lies. Yeah, I think you were just discussing there that sort of shifted, or that dominance more in like the consumer electronics space, automotives, and they're sort of the lower end chips, but advanced chip making capabilities. What do you see from China? And what is that gap that you see in the race to catch up to the U.S. right now in terms of years? Well, I think you have to describe what is the gap. So if you just narrow down to a very specific thing about, for instance, the nanometers of a semiconductor manufacturing process and making actually the most complicated parts of the chip, it's still probably three to four to five years. And there's a big debate about how close you can get to it without adding EUV. And you are able to get closer doing double patterning, triple patterning, which means you mean an older technology, and you just run the machines more often. But it's more expensive, the costs are higher, and the yields are lower. And you can get the performance closer. So how far the performance gap is depends, maybe 20%, 30%. But then you have to take a step back and say, what does that actually mean? Because we don't care about the speed of the transistor, we care about the capability of the AI cluster. And to bring up a statistic, NVIDIA said over the last 10 years, their inference of their chips has improved 1,000 times. Only 4x of that 1,000 was specifically from semiconductor manufacturing. The rest is about how you build the models, how you do the error correction, how you design the chip. Those are all areas where there's much lower barriers. So Chinese companies, to be particular, can get closer even though they don't have the latest performance. You also have to look at the entire AI cluster, the backplane, the model setup, how you run out the data center, all of these things. And the other thing that Chinese companies are doing above and beyond trying to get closer on the most advanced and difficult models, more than two-thirds of the models that have been launched by Chinese companies are actually smaller models with fewer data points that don't need the most advanced processing. So they're actually going in a different direction to try to get into the market without competing at the threshold, the bleeding edge of performance. You mentioned about EUV, and we have seen those restrictions around not only the sales of equipment from ASML and Tokyo Electron, but it's the servicing of those components that are already in China. So how much is China being restricted again at the more advanced end, and where do you see that going from here? Where do we see that going from here? So I can't predict where government policies go. I think one thing that we've seen with some other restrictions around servicing is that the fabs in China, they've moved from being technicians where they just run a machine that someone else is servicing, to being the engineers themselves. So they actually figure out how to do the servicing themselves and build up those capabilities. This may, you know, this is one of those things that in the short run it may slow down the technology cadence, but in the long run as Chinese companies invest to do that themselves, it brings up the capabilities of the Chinese market. Where does it go from here? I think the real question is, is will the industry have a right-hand turn where all the existing technology is not as important and you have a new leapfrog that can jump over it? Now we talk about quantum, we talk about photonics, those things are not ready for primetime yet. But the critical thing is there's so much installed capital in this industry. And so it's very doubtful in my mind that you can create a whole new innovation that is purely a leapfrog and all that old capital doesn't matter. It's going to be incremental on top of the existing base, and it's going to require that existing base to be used. So this is an industry where having an installed base, having a lead is very difficult to conquer. And so it will be difficult again at the bleeding edge for Chinese companies to catch up. And you mentioned that in the end, these Chinese companies have to go global because despite all these terror threats and sanctions and the like, I'm just wondering how possible is that? And really, you mentioned that we're not seeing that much de-risking really. How far can decoupling actually go then? Yeah, those are interesting questions. So there's two ways that Chinese companies can go global, Chinese semiconductor companies can go global. One is a Chinese box or a Chinese car that's made with Chinese chips can go be sold overseas. And you see that in the automotive space, you see that in the PC space and other things. The other one is where Chinese chip makers sell chips to overseas companies. They would sell them to Apple or to Dell or to a European OEM. And that's going to be really important to do, because at the end of the day, 75 percent of the chip making market, the people who decide which chips to be bought, are the United States and its allies, U.S., Japan, Korea, Europe, etc. So unless China only wants to compete in that 25 percent, where the scale is simply too small to drive the R&D in the capital, they have to sell to those other countries. Now you say, well, we have geopolitics, we have de-risking, people aren't going to buy. But the critical thing the survey showed us is that for every single type of company in every region, performance and price matters more than the nationality of the supplier. Because this is a product industry. You know, I would love to have a fully de-risked supply chain, but it costs money, it costs time, it builds risk. Who's going to pay for that? And people very clearly said, if you had the choice between investing in innovation or de-risking your supply chain, I'm going to invest in innovation. So actually, the industry hasn't changed that much in the last 10 years. How far can decoupling go? It can never fully decouple because there isn't enough money and time and capital in the world to create two separate supply chains. It's just not possible. But then how does it work with the restrictions, though? Because you're seeing even more restrictions that are coming through. Just last week, it was that focus on quantum computing. And even though you're saying that's not ready for the primetime, obviously regulators already have it in their sights. Yeah, so what you see companies doing today is they are preparing and executing some level of a de-risking strategy, but relatively low hanging fruit and doing things that are specifically responding to government restrictions. They're not being proactive and saying, hey, there could be a government restriction in the future. I better stop using that supplier now. And in what our survey told us is, what's very interesting, I'll just bring up one thing. One third of companies said over the long term, we need to split ourselves up into two separate companies, one focused on the U.S. market, one focused on the Chinese market. Let's say there's about a thousand interesting semiconductor companies. If one third of them do that, that means we should have seen 300 to 350 JVs spin. How many have we seen? It hasn't happened yet. Even the companies under the most pressure, there's still thousands of Nvidia chips flowing through China. ASML is still selling into China. So companies are very nimble at keeping business going despite all this political pressure.