Loading summary
Transcript60 lines
- [00:00]
A
Foreign.
- [00:08]
B
The American economy. Stock market gains are often due to the performance of a handful of companies in the tech sector. The build out of data centers has been a short term benefit for construction and employment. After a year of tariffs, imports of servers and other components going into those data centers are one of the few bright spots for international trade. As Samaya Keynes and I were working on our book, Semiconductors were the sector that everyone kept talking about. Chips are the foundation of what drives all that Artificial Intelligence. No chips, no AI. Semiconductors are also a really deep part of the US China trade war. That is why I am so excited about today's show where we tackle the the American government's policy odyssey into semiconductors. The book is titled how to Win a Trade War. If you pre order your copy today, you can receive a free bookplate signed by Semaya and me. While supplies last, just follow the instructions on the Simon and Schuster website. I will put a link to those instructions in the show notes. And with that, here's the show. You are listening to an episode of Trade Talks, a podcast about the economics of trade and policy. I'm your host, Chad Bowne, the Reginald Jones Senior Fellow at the Peterson Institute for International Economics in Washington. In this episode we're going to talk about semiconductors and the massive American policy interventions taking place in the chip sector today, from subsidies to tariffs to export controls. Why is the government doing it and how is the policy working out so far? To make sense of all this, I will be joined by a very special guest.
- [01:46]
A
I'm Dan Kim, Tech Insights Dan Kim
- [01:49]
B
is Chief Strategy Officer at Tech Insights. Dan is a PhD economist and has been my go to expert on chips for a long time. He's worked in senior roles at companies like Qualcomm and SK Hynix, and from 2022 to 2025, Dan Kim was the chief economist in the Chips Program Office at the US government's Department of Commerce, working on implementation of this massive American semiconductors industrial policy, the one which continues on to this day. Hi Dan.
- [02:17]
A
Hello.
- [02:18]
B
Today's conversation is going to be all about artificial intelligence. Why are semiconductors essential to AI?
- [02:27]
A
Well, there is no artificial intelligence without semiconductors. You know, when you use a chatbot, whether it's ChatGPT or Claude or any other models that you might be using, you're effectively talking to semiconductors. But advancements in semiconductors that have been advancing for the last 60 years, that's what's enabling artificial intelligence to happen. And that's what actually is going to drive it even Further. And so you can't talk about artificial intelligence without semiconductors.
- [02:52]
B
To understand what's happening today then for AI and chips in the United States, we really do need to go back about a decade. How did US Policymakers learn about America's vulnerabilities, as we might call it, in the semiconductor supply chain? What happened?
- [03:11]
A
Well, I think there was a couple of moments. I joined the International Trade Commission in 2015 and there were two people in the entire federal government that were focused on semiconductors. Two, there was one person at the ITC and there was one person at the US Commerce Department. And when I joined the itc, my division chief kind of pulled me aside and said, well, you know, you're covering semiconductors, but there's really nothing happening in that sector. US already dominates trade. There's no tariffs here. There's really not much trade disputes going on, so you probably want to find another thing to focus on while you're covering this. I said, well, I think it's going to be important, actually. Semiconductors, very important. And then later that year, China declared publicly its intention to be completely self sufficient in semiconductors. That set off alarm bells throughout the industry and they lobbied the government to say, hey, we need you to pay more attention to this because we know as an industry we've been telling you hands off, no government intervention, other than just making sure that free trade of goods and ideas flow. But now we might need you to intervene, but exactly how, we don't know yet was the tenor of it. Now this was during the Obama administration
- [04:25]
B
and this was the made in China 2025.
- [04:27]
A
Obviously this is made in China 2025. In the meantime, the manufacturing competitiveness of the United States was declining primarily because the foundry model that TSMC in Taiwan has really perfected with the introduction of the iPhone that really supercharged that and that relationship between Apple and tsmc, all of a sudden the United States found itself with a shrinking share, ever shrinking share of manufacturing capacity and manufacturing leadership. At the same time, and at the same time, China was declaring its intention to be a technology leader and to be self sufficient. Now contrast that to trade policy that's been happening to effectively enable the most globalized supply chain that I think of any industry we've ever seen. A chip travels many, many thousands of miles before it gets inserted into your phone or your PCs. The whole trading system were designed to make sure there was no frictions whatsoever that you could completely leverage your relative competitiveness in whatever your economy has to participate in that. And the US was a huge beneficiary of this. But some things happened along the way in which the manufacturing share had declined to close to 10%, from making all of it when semiconductors were invented in the United States, to China being a threat and all of a sudden no longer being the manufacturing leader. So that was all lurking in the background. Then Covid happens. There was a huge demand for compute work from home, school from home, and electronic devices were flying off the shelf, if you can find it in the stores. And it turns out that the same semiconductor factories that are making those chips for computers and others were also similarly making semiconductors for automotives, industrials, your refrigerators, dishwashers. When there was a huge demand for these goods, all of a sudden the semiconductor industry could not quite keep up. It is not designed to keep up with that kind of demand. So the shortages of semiconductors, it's a very visual, visceral thing when unfinished cars are sitting in parking lots in Michigan and Indiana and other places and your car prices are going up for no good reason and you can't find that laptop to send your kid to online school. And more importantly, price is going up because you couldn't buy these things. A lot of inflation was because of shortages of input goods like semiconductors. And for some reason I sort of had the front row seats in both government and in industry through this growing interest of semiconductors by the US government and quite frankly, all governments around the world.
- [07:19]
B
So US officials see these vulnerabilities and decide to act. They've ultimately used a lot of different policies to intervene in the semiconductor supply chain ecosystem over the last few years. But let's start with subsidies. Let's talk about the CHIPS Act. What was the CHIPS act designed to do and why?
- [07:40]
A
Well, the theory was when the question was asked, why has the US lost so much of its share of manufacturing of semiconductors? The answer that was given primarily by the industry was the cost of production is just too high in the US compared to others like in East Asia, particularly Taiwan, Korea, China. The argument they made was that a lot of the cost difference was because of government support in those countries of which the US offered none. Therefore, if we want to get in the game again, then we would need to offer the same. That was the argument. And there was a bipartisan congressional agreement that that was the case. So I, I remember actually being in the industry when this was being debated. I was a chief economist of a company called SK Hynix. I remember there was a conversation within the industry as to what to lobby for. Do we lobby for tax credit, do we lobby for a grant program? There are advantages and disadvantages of both. They didn't think they would get both, so they had to make a choice. So they lobbied for a grant program. The theory was that if there was sufficient supplied side support, that if you can just lower the cost of production enough to have the companies justify building in the US then you don't really have to worry about the demand. They'll just build it, subsidize, and they will come. I think it was the theory. To everyone's surprise, Congress offered both. They offered the tax credit and the grant program. So 25% fully refundable tax credit, meaning if your tax liability is zero for whatever reason, then 25% of your eligible capital expenditures you would get back in cash, which is an incredibly powerful tool. And 25% of a $20 billion fab is quite a bit of incentive. And the tax credit will be a non competitive one, meaning there's no limit that the treasury has. If there is a trillion dollars worth of investments all qualified for, then there would be $250 billion worth of tax credit that the treasury would give. And on top of that, there was a competitive grant program, direct funding, meaning there was a finite number of dollars out of the $52 billion of CHIPS Act, $39 billion would be allocated. And so companies would compete for those funds for the next four or five years.
- [10:13]
B
You've been on both sides of this. But before you came into government, as you said, you had just started to work at SK Hynix, a company headquartered in South Korea. How much subsidies had the Korean government been giving to semiconductor firms to build factories there?
- [10:28]
A
So I was the chief economist of Hynix, and I asked our team that was building a FAB in Korea. They said, how much subsidy are we getting from the South Korean government? They came back and said, I don't think they thought that they had misunderstood the questions. The government doesn't really support us here. Immediately I realized actually government subsidies was really not the driving factor here, at least for this company deciding to build in Korea.
- [10:55]
B
Okay, but what factors would then drive the decision of where to locate one of these semiconductor fabs? If there were no subsidies, why did SK Hynix build there?
- [11:05]
A
The easiest answer is that they already have fabs in Korea, so they know the ecosystem works. There is a supplier base already built in. A fab uses something like 3,000 suppliers to keep the factory running 24, 7. They know that the workforce is there. There are engineers, there are factory workers that could run that fab 24 7. They know that the infrastructure is there and then obviously the cost competitiveness has to be there. And it's a mix of all those things. If you're imagining that, oh, if you just have low cost of labor, you're going to build fabs there. If that was the case, then every fab would be in Southeast Asia and there are almost no fabs in Southeast Asia. If you're imagining that it's only high skilled labor that's going to drive this, well, then every fab will be in Silicon Valley and there's no fabs in Silicon Valley. It's the sweet spot of where everything has to be present. And so given all of that, I think I intuitively understood that the financial support that a government can give is just one piece of the picture.
- [12:07]
B
When you were in the private sector, you did site selection for SK Hynix. What did you look for?
- [12:13]
A
So I actually had to make a checklist of all the must haves and all the nice to have. All the must haves have to be there in order for any site to compete. The nice to haves then become a point of what can the government do to make up for some of the deficiencies that a place might have? Whether it's cost, workforce, infrastructure, whatever it may be the must haves, I'll give you a couple of examples. Is it going to be near a place where there's sufficient people and workforce? Surprisingly, that's not always the case. Another must have is absolutely reliable energy grid. Another one is absolutely reliable water supply. Those are just some examples that a site selection requires you to consider. It also requires you to consider what kind of vibration there is in the ground or in the air. Even so, it can't be too close to an airport, can't be too close to a railway, because when you're working at nanometer scale of the semiconductor manufacturing, then any kind of vibration could actually make your chips production really difficult. The nice to haves then become about costs and in some cases political buy in. Do you have local, state, federal partners that will work with you to get the right permitting to make sure that it's done on time? Because if you're sinking 20 plus billion dollars into a factory, if there is a six month delay because of some permitting issue, that six month delay is extremely expensive because you are building a fabric and every second that you're producing you make money. Every second that you're not producing, you're losing money. And so a delay of a day is actually multimillion dollar cost to you. Another nice to have, I think is universities nearby you who's willing to work with you to do R and D. You can do R and D separately from your fab. And so if you have a university partnership that you really like, that's actually really nice to have, some might consider that an essential, but some might consider that nice to have. If you combine the nice to haves and the must haves, the easiest sometimes decision you can make is you look at an area and say, is there already a semiconductor fab somewhere in that area? If that is, then you know that there's a supplier base, you know that the workforce is kind of there, you know that universities are there, you know that the government officials are used to doing permitting for, you know, large scale fab. And so the easiest thing might be to say, where in the United States are there already fabs at the scale that I'm looking to build and if there is, I'll just go near there. So I don't think it's an accident that TSMC looked at Arizona and said, you know, there are intel fabs there. They've been operating there for many, many years. So it's an obvious site for them to try. Samsung too. They've been operating fabs in Texas for many, many years. There are other fabs in Texas in that Austin area. So choosing another site near there makes a lot of sense. It becomes a bit riskier if you're going to an arena in an area in which there's never been a fab before. So you have to take a chance and think about can a new ecosystem be developed around you? You're the center of it. Will there be 3,000 suppliers around you? Will there be universities supporting you? Will there be government officials supporting you? All those questions has to be asked. I think in those cases there's probably more fierce competition due to government funding. So in those cases where government officials know this is a risk that a company has to take, then you have to offer something to them to sweeten the offer. I don't know if any company will go out there and say whoever gives us the highest dollar of incentives at a state level will go there. I can't see that happening because you have to meet all these requirements.
- [15:59]
B
In 2022, after Congress passes the Chips act, you're still working at SK Hynix. The US government ultimately hired you to be the chief economist in the Brand New CHIPS office that the Commerce Department had to set up from scratch to implement this new law. How did that Come about.
- [16:16]
A
I was actually called into an interview with Secretary Raimondo with the premise that we would just share some ideas. I did not know that it was going to be a recruiting meeting. And I sat down. Secretary Raimondo, any cabinet secretary, they don't have a lot of time for you. We sat down. I'm sitting across the table from her in her office, and she asked me about my family. And I'm an immigrant to this country. My grandfather was a refugee in the Korean War. The U.S. marines helped save our family. And so I owe a lot to the service members there. So I kind of told the story because she asked and we talked a little bit about semiconductors. And she pauses at the end and she says, dan, you owe a lot to this country. I need you to serve. And she appealed to my sense of duty to this country. And now, mind you, I had already served in the government before, so this wouldn't be my first time doing it. But she emphasized that this is a national economic security program, it's not a political program. And she told me at the time that she would do her very, very best to shield the team from politics to enable us to do the right things. And quite frankly, didn't really give me a choice other than to join. And so I had to fly back to Korea to tell my CEO that this had come about. And so I had no choice to accept and do it.
- [17:34]
B
So tell us now about what does a chief economist of the CHIPS office do? Why is it important to have a high level economist involved in the strategy here?
- [17:45]
A
Yeah, that's a good question. I like to joke that I was chief economist of a team of one economist, but my team actually consisted of what we call the economic security team. So I had three titles. I was the chief economist, I was the Director of Strategic Planning, and then I was director of Economic Security. We created a team of industry experts, technical experts, data scientists. I was the economist. They could actually formulate a strategy around what does economic security mean in this sector? What does the country need most? And actually come up with a metric of success and evaluation criteria for companies that would come to us in their applications. We would be setting requirements as to information they would be giving us. And then we would effectively have a rubric of deciding how companies ranked depending on the answers.
- [18:40]
B
What were the priorities for economic security? How would you characterize those?
- [18:46]
A
Yeah, and that was the first assignment that I was given when I joined CHIPS is to actually formulate what we call a vision for success paper which articulated what we thought success Looked like, what do we need to accomplish, what do we need to have in this country to say chips succeeded? So we outlined certain types of production that we thought was necessary in this country to have and that we were willing to incentivize. So we outlined, for example, the need to have the most advanced logic foundry production in the United States. And by most advanced, I mean at nanometer scale, the finest, the techniques that actually can cram the most transistors onto a chip to have the most functionality. That's really important because it's a bit of a winner take all situation where in order to enable your smartphone or your PC or now your data centers, that's enabling your AI, you need the most advanced manufacturing techniques. That's what we would call leading edge. So we needed to have that because we didn't have any at the time that could serve multiple customers. At a foundry model, like a contract manufacturing model, we signaled very clearly that we wanted not only one, but multiple manufacturing clusters of that type of factories. And by clusters, what we meant was we wanted a growing set of factories within a site that would drive down the cost of manufacturing sufficiently to compete globally. We wanted and emphasized the need for memory production, particularly dram, of which there is a very severe shortage in the world right now. But we also signaled that we wanted advanced packaging, which has become increasingly important to enable advanced functionalities such as AI chips. If you open up an AI chip, it's got a GPU in the middle, but it's got multiple high bandwidth memory DRAM modules around it that is packaged together in a way that is very advanced so that there's quicker communication between those chips.
- [20:44]
B
So GPUs, those are the AI chips, say that Nvidia and AMD and those types of companies design and that TSMC will make in high bandwidth memory. That's the fancy memory chips that companies like sk, Hynix or Micron will make. And this advanced packaging is basically taking all those chips and putting them together in the right way so that they work.
- [21:06]
A
That's right. And I think most people understand what GPUs are because it's in the news all the time. Most people probably have not heard what a high bandwidth memory is. It is a specialized type of memory of DRAM that is stacked. So if you imagine stacking eight different dies of chips together, and then you have to drill precise horizontal holes in these chips so they communicate with each other. Something like 2,000 of these, what we call three silicon vias, and then you have to control for the heat in between each of these dies in a way that doesn't warp. This is extremely difficult to do. Probably the most difficult manufacturing technique out there other than logic. Now, mind you, when I was writing this strategy paper and I say we, there were multiple people writing it with me. When I joined chips, it was November 2022. That's the same year that ChatGPT was launched to the public. And so the whole conception of AI was really just at the very, very beginning we did not see coming. For example, Nvidia being the behemoth that it is in terms of both market size and the importance of the chips that they make, or AMD for that matter, Leading Edge. For many years, we effectively equated with your personal computers and your smartphones in order for you to have the next iteration of your smartphone. What does it take to make your next smartphone smarter and faster? It is effectively the advancement of the chip. We did also want that. But as we thought about trade offs, we made very clear that we wanted to have the ability to make chips that would go into our compute infrastructure. And I think now if I were to write it again, I would be very specific that it's more AI related. But back then again, AI really wasn't in the jargon as much as it is now. But we made it very clear that compute infrastructure was a critical end use that we wanted to have some production capacity to supply.
- [22:59]
B
I want to ask you about Taiwan. TSMC in Taiwan is the concentrated source of production of the most advanced chips. That concentration means supply chain risks. And then there is the geopolitical risk of President Xi Jinping's stated desires to reunify Taiwan with the mainland. What sorts of questions did your office face on Taiwan?
- [23:23]
A
One of the questions was, can we use the $39 billion to completely offset our exposure and risk to Taiwan? And my response was, no, that is not possible with the amount of money you're talking about. I think within that question is this concern that more than 90% of leading edge foundry manufacturing comes from Taiwan. And one company, tsmc, practically all of your iPhone chips that are most advanced comes from there. Practically all of your Nvidia AI chips, at least the GPU part of it, comes from there. Same with amd. Pretty much anything that you could conceive of that is driving some kind of compute power in your life have been manufactured by tsmc. The concern for some was that's a very geopolitically sensitive place. What can we do to offset that risk? So the question comes to me, can we use this to completely offset that risk. And then the economics has to come into play, which is the answer is no. There is too much volume that's already coming out of Taiwan, meaning there's so much volume there that to completely off risk it, you would have to build so much more in the United States or at least away from Taiwan, that it would be completely impractical. At least with the money you're giving me. If you're giving me $3 trillion, maybe, right? But not 39 billion each. Fab costs somewhere around $20 billion. And by the way, Taiwan does an amazing job. They've enabled so much compute to happen. They've enabled pretty much all of our semiconductor design companies like Nvidia, like Qualcomm, like amd, even intel to thrive in their respective businesses. They've been an amazing partner for us. And so I understood the question, but I had to gently push back and say, no, that's not going to be the driving factor here. Although it is a background that is important for us to understand. But at the same time, if that's what you wanted, then this is not the tool to get that done. And secondly, I'm not sure if I agree with the premise that we need to completely zero out our risk there. I'm not sure if that's the right move either. Now, that wasn't my job to dictate that, but I could understand the fear of it. So that obviously was lurking in the background.
- [25:48]
B
Let's turn now to the impact of the CHIPS Act. Unlike the Inflation Reduction Act. So that was a US industrial policy passed right around the same time in 2022, but for electric vehicles, batteries and other green tech. President Trump did not kill the CHIPS Act. It lives on and has been enhanced in some ways. We'll come back to that. Now we're nearly four years on since the legislation passed. What can we say so far? Maybe let's start with positives. Have parts of the CHIPS act worked?
- [26:17]
A
The parts that worked is that now people are talking about it. Remember when I came into government 2015, there was almost no attention on semiconductors. Now there almost seems to be almost too much. There's so much attention on it that people talk about it. So I think that's a success. The fact that there is a team, multiple teams, one at Commerce, there's other folks at State Department, there's other folks that Defense and others that are so focused on semiconductors now, I think is a good thing. I think there's a recognition of the criticality of this industry that's A good thing in terms of share investments, I think it's a bit hard to conceptualize because when we left we had announced that there were about $400 billion worth of announced investments out of the result of the $39 billion worth of committed funds, that's effectively 10 to 1. I think that's a pretty good return on taxpayer money, especially if you consider the multiplying effects not only in economic security, but the jobs they would create and the investments, you know, downstream that they would create. I think the administration has announced now it's north of $600 billion. So there's more committed funds to that.
- [27:21]
B
What about leading edge and American manufacturing of the most advanced chips?
- [27:26]
A
There's about five companies that are able to do leading edge manufacturing of semiconductors. In the logic side there's tsmc, Samsung and Intel. On the DRAM memory side there is Samsung, sk, Hynix and Micron. All of those companies have committed to a large scale manufacturing in the United States in one form or the other. Whether it's in front end fabs or advanced packaging, no country in the world has all of those companies. This is an industry in which has national champions. You go to Taiwan, you got tsmc, obviously you go to Korea, you have sk, Hynix and Samsung. And in the US it was Intel. But now we have manufacturing presence. We're committed manufacturing presence by all those companies. And I think they then pull in R and D, they pull in the skills training. I think that kind of ecosystem being created to me is the biggest success. And I don't also want to ignore that. Companies like Texas Instruments and Global Foundries and others are also building out in large scale manufacturing. There were many other smaller scale manufacturing also very important for end uses like critical infrastructure and defense goods. MEMS production of which we didn't have too much before. We have some committed funds to that. There's a lot of attention now on critical minerals. I think having the chips act there for there to be a conduit of conversation, to have those conversations I think is an excellent thing too. I think those are good things that have happened. When I take a look back and I look at it, I had really high ambitions. And would I have done things differently? Some things absolutely. But I have to be realistic in setting expectations because the US manufacturing ecosystem, particularly in semiconductors was quite atrophied at the time that we started this. And the fact that these big successful companies with unbelievable manufacturing precision and discipline would say we're going to take a chance in the United States again. I think, to me, is just a huge, unqualified success.
- [29:30]
B
Those are successes. How about examples of things that didn't quite work? Where did the CHIPS act go wrong?
- [29:36]
A
I think what didn't go well, and I'm not sure how much it could have been helped, is that we had to stand up a new team, new process, new procedures to receive applications, go through them and give the grants out. I sense that there was frustration that it was going too slow, even though I could assure you we were working day and night. We were all very tired, we were all very overworked in the background. I think we could have taken a little bit more risk, pulled triggers a bit earlier in some cases. I think sometimes we did have a tendency to live in our heads a little bit. And I think that showed.
- [30:10]
B
When you were engaging with the private sector, with these companies, what did they complain about most? When it came to CHIPS act, we
- [30:18]
A
weren't giving enough money. A lot of the complaint that we heard was that they had underestimated how much the American manufacturing in the semiconductor ecosystem had atrophied and therefore how much more costly it was to get the fabs up and running. And they had underestimated the cost that it would do it. And we didn't have the flexibility to sort of plus up those funding if it was. And so that's one complaint. The other complaint, obviously, was the amount of information that was required and the time that it was taking for the applications to go through. But I can understand why the bureaucracy can be really frustrating from a company's perspective, especially if they're global companies. And if they had experiences like in East Asia, where you go to South Korea or Japan, they're so used to these permits and other things. And even when there are some financial incentives involved, like they would often comment that they would never have to jump through as many hoops as they had to jump through with us. And so that is something to consider is how to streamline all of that.
- [31:21]
B
When I were interviewing semiconductor executives for our book, one of them said to us that this was the grants part of the CHIPS Act. They said it was the most expensive free money they'd ever received. It seems like the companies really did prefer the tax credits to the grants. Was there anything particular with the grants that really helped move the needle? Things that wouldn't have happened but for that particular part of the Chips Act.
- [31:52]
A
So some types of manufacturing, you know, 25% tax credit, simply would not have been enough to push the investment decision over the edge. So in which cases like you had to have the grant put push that. So simply the math of it would not work without a grant program. I think also for the companies when they would receive a grant, it was a signal to the market too that the government had given a stamp of kind of faith that this technology was important, that it would be supported by the U.S. government. And they've never really had that before. And so we would sometimes get feedback. Like when we mention to our potential customers that we are in line to receive CHIPS grants, they're all very interested because the customers are evaluating you not only in the technology front but also on the financial front. And so having a government backer pushes down that risk quite a bit because the government grant is a competitive program that is not easy to get, that there is some sort of due diligence built into it that your customers would validate at the end of it. So we've heard that example as well.
- [33:05]
B
The CHIPS act is all about supply side tools. You can incentivize TSMC and Intel and Samsung maybe to all build leading edge capacity in the United States. But there's still the question of demand for Nvidia and amd. Who do they take their designs to to actually do the manufacturing of their AI chips. They all still seem to want to go to tsmc. One problem with the CHIPS act of course is that there were no demand side tools in the legislation. You had nothing to work with. All you could do was give out money. The legislation did not give you authority to do offtake agreements or to otherwise incentivize AI chip design companies to use this new capacity. So let's brainstorm a little bit. Are there creative ways policymakers to could incentivize these AI design companies to diversify away from TSMC a little bit to maybe use an intel or a Samsung fab?
- [34:01]
A
One example I could give you is there is a cost, an actual cost of porting your design from one fab to another fab. So if you're already at TSMC going to a Samsung or an intel, there is actually significant hundreds of millions of dollars in some cases to design your chips specific to this other fab. Could the government play a role in effectively zeroing out that cost risk so that you can try it for free or even give you something on top of that, something that effectively says we're going to really incentivize you to try these other sources or R and D together or whatever it might be. I think if we were to really put our brains to it, we could come up with a range of options that we may not have thought of before rather than you must use this other fab that you don't want to use. Otherwise it may be tempting to think, well, you got to force these companies to use a fab they don't want to use. That's the only way to get them to use it. I'm not sure if that's the case. These are long decisions that a company has to take. When you choose a fab, you are effectively saying you are my technology partner going forward for the next decade or so. That is not a light decision to take. And we've had experiences, Chad of other countries doing local content requirements and others to try to force these kinds to happen with varying degrees of success. And secondly, I think you need to understand that from a company's perspective because there are limited clean room space at whatever fab you want a fab at TSMC or Samsung, wherever. If you don't fill that space, your competitor might. And if there is a fab and or clean room space somewhere in the world, that's going to give you the most bang for your buck to use that facility unless you use it, your competitors might. And so how could you tell them not to use it is a really difficult question, right? Unless perhaps you force everyone to not use it or to use something else. And so I could totally understand why like demand side questions are difficult. And so I think we need to be very careful in thinking about how to design those.
- [36:07]
B
Now for leading edge manufacturing, we've got the relative success you mentioned about getting the most advanced makers of these chips to do more in the United States. What about the older chips, the less fancy mature nodes? You spoke earlier about how shortages of these less fancy chips during the pandemic meant consumers were hurt when companies not getting these chips couldn't make cars, washing machines, refrigerators. How successful has the chips act been at getting more mature node semiconductor production in the United States?
- [36:42]
A
One regret that I have is that despite our efforts to focus on mature node semiconductors, we did not have huge success in attracting really high volume semiconductor production and mature node. And that's actually purely due to economics. The economics are just really, really tough. And because we had a universal set of tax credits, despite what kind of types of chips you were making or the economics of that particular node of manufacturing, it's just 25% now, it's 35% with the big beautiful bill in place. But the economics actually is quite different and we didn't account for that at least in the tax credit part. And I don't think we also appreciated that China's overproduction of mature chips would effectively make so much hesitance in investing in new factories. In that the fact that we didn't really have a renaissance of mature node production is a regret that I have. And I frequently think about what else we could have done differently to do that.
- [37:45]
B
Why are the economics different for mature node semiconductor manufacturing?
- [37:50]
A
You know, at Tech Insights we actually do what's called economic models for semiconductor manufacturing. So we could actually tell you at a given wafer for a semiconductor what the component of the cost was. And if you look at a leading edge, most advanced node manufacturing wafer or chip, somewhere between 60 and 80% of that cost is in depreciation of new equipment. Because the depreciation schedule usually is five years or so for equipment. And so because the equipment is so expensive and because constructing a new fab is so expensive for the first five years or so, by far the largest cost is the equipment. And that's how it's priced in, that's how you price in the chip. It's quite expensive. It's expensive to design it, it's expensive to get it. And so once you have the scale, then you're off and running. But for mature node production, imagine competing against a fully depreciated fab in East Asia by building a new fab in the United States. The fully depreciated fab in East Asia are no longer having to pay off that equipment in the five year depreciation schedule. Whereas you are in the US all of a sudden you might be two to one disadvantaged economically in that scenario. If you want to help companies overcome that disadvantage by inviting greenfield investments, then you have to do something about the economics. And 25% just isn't going to do it in a lot of cases. So we leaned into modernization of pretty much every semiconductor fab or site in the United States. I think we've received modernization applications from pretty much all of them, if not most of them. And we like those because all of a sudden now then it's about modernization. So it might be some new tools, but there are the depreciation aspect of it. There's less pressure on that. And so I think that's a really good bang for the buck. But if you really want to move the needle, then you have to think a little bit bigger. That gets even more difficult when you consider that China is really just building out most of the mature node capacity that is being built out into the world right now. They seem to not care about the economics of it. If you have a production place where the economics don't matter to them and overproduction does not matter to them, and that's the case not only in semiconductors, but in other parts of the economy that we've seen, then how do you compete with that? You can consider trade restrictions. Those are tough to pull off. But on the supply side of it, it's the most predictable thing that a government can give to a producer. To say if you build that thing, then we can give you a financial Runway here. Unless that's present, I think it's just very difficult for you to have commitments to build large scale manufacturing and mature no production in the U.S. let's talk about import restrictions.
- [40:38]
B
We've talked a lot about Chips act and subsidies, but that's certainly not the only policy the US government has been using in the chip space in the last six to eight years. Start with trade restrictions for mature node. What's the story there?
- [40:52]
A
I don't see a universal support for tariff in the industry right now, even for matured node, partly because the users of semiconductors, your automotive manufacturers, your medical device, any kind of tariff increases their bill of materials costs directly. And so I can understand why they're not a fan of it. But my prediction is that in the next five years or so, when the Chinese economy, their downstream economy from their semiconductor production has absorbed all of the production capacity that China has to offer, it's not quite there yet, but I think it's starting to sort of reach saturation in the next five years. When that excess capacity then gets exported out for very cheap around the world, including here in the United States, then I think that tune is going to change quite a bit. That is my prediction of it. In the next five years there will be a clamoring of protectionism against flooding of mature node production from China, I think we'll look back at that time to say maybe we should have done something about it a bit earlier. Just as we say the same thing about solar panels, about printed circuit board manufacturing and other things that we say the best time would have been to fix it when we actually had some manufacturing back then to protect. Now you protect it. Who's going to come back? I think you're going to have similar discussions. You can't rewind the clock and there's only so much effectiveness that those measures can have once an industry has completely fled. It just takes time. And then you have to commit to protecting that industry indefinitely in order for that to come back. Especially if you think it's really critical for your national security. And mature node production of semiconductors is absolutely critical for national security. Most of the stuff that's in our missile systems, in our defense systems, they're all mature pretty much. And so unless we have a robust manufacturing ecosystem around that, it's just really hard to keep it going.
- [42:43]
B
So how would you strike the right balance? This is going to be unpopular with the using industry. Suppose we did something right now, but how would you structure it to ease the burden on them, but also to sustain a matured node manufacturing base in the United States?
- [43:00]
A
Yeah, first I think I would ensure or at least encourage transparency. I think BIS did a survey not too long ago asking companies of their awareness of Chinese matured node chips in their systems. And these are companies that are fairly close to the semiconductor industry, if not chipmakers themselves. A surprising number of companies were just simply not aware of how much was in there. These are 10 cent dollar pieces really that are often distributed not directly from the manufacturer but through distributors. You just get it often tier 2, tier 3 suppliers are in charge of it, not you. And so the awareness of it just was very shallow. And so even if you were to sort of ask companies what would be the impact, I don't think they could really tell you. And so gradually increasing the awareness of it I think is the first step. I have a unique view now because Tech Insights, we have the most extensive building materials database in the market. So we actually can tell you how much chips like Chinese designed, Chinese made chips are in our systems, in our smartphones and whatever else there is. Quite a bit. Right. And if you give enough lead time, supply chain managers can adjust. And so I don't think it's something that you announce and say this needs to be implemented tomorrow, but I think it is something you need to signal. And I know that at the moment it might seem alien for government to say let's have a measured phased in approach when it comes to tariff, really consider the impacts and implement it over time with the support of our allies, with consultation with the industry, with a recognized problem and a stick by it. We're sort of in a mode of, I guess tariffs go up when there is a post from the president era. So it can't be like that. But I think understanding the exposure, gradually increasing the transparency of it, either through encouragement or through requirements and then going from there, I think is probably the best way to do it.
- [44:57]
B
So that would be a rational argument for a U.S. tariff to incentivize supply chain managers to think about sourcing their mature node chips out of China in ways that might sustain production elsewhere, including potentially in the United States. So in September, October of 2025, there was this moment where the United States decided to expand the coverage of its export control regime to now include subsidiaries of companies that were already on the entity list. This caught up a company called Nysperia. And it turns out that Nixperia makes a lot of mature node chips, especially ones that are important for the automobile sector. The Dutch government ousted the CEO of the Dutch subsidiary of this company and then the Chinese government got unhappy and banned Nuxperia's chips to the world. Suddenly, the automobile sector's production was almost nearly shut down worldwide because they couldn't access these chips anymore. So was that a wake up call for users of these chips like these car companies? Have you seen any change in their chip sourcing behavior since that experience?
- [46:07]
A
I think it was a wake up call, but I have not seen any change in behavior. And I think maybe we'll look back at this era and say there were a lot of risks, but because there are so many risks that companies didn't quite know what to do with those risks. Yes, automotive is very important for automotive manufacturing, but there are substitutes. If you give it enough time, there are other substitutes that could take place in experience. This is not a comparison to TSMC making the most advanced node that no one else could possibly do at the scale or the cost or the yield that they do. That's not the case here. This is no, not highly sophisticated manufacturing. The economics is different. If you give enough economic signals for their competitors, they could come in and fill it with enough lead time and supply chain managers can deal with it. And it's not just automotive manufacturing, by the way. Actually, we went back at Tech Insights and found nextperia chips in smartphones. In AI data center servers, for example, we found a super micro data center server that obviously the Nvidia and the HBMs are highest value. That made up for most of the bill of materials costing there. We counted I think about 15 Experia chips in that server, all adding up to about a dollar. Right? But the server does not function without those Nuxperia chips. So it turns out like we're talking about asymmetric impacts. This is it again. If you give it enough time, supply chain managers can figure it out, but China can threaten to say no. And then the impact is immediate because there's not enough time to adjust. I think economics is about substitution and trade offs and time costs, and that's what we're seeing. So it is a waco call. I am not personally seeing a huge amount of change as a result of it yet.
- [47:51]
B
But if we don't see any behavioral change happening on its own, that's where policy comes in. And policymakers need to create the financial and economic incentives for supply chain managers to make alternative decisions if they're not willing to make it because the underlying economics in the absence of that don't cause them to. Let's talk a bit more about export controls and the semiconductor odyssey. There's US export controls in the chip sector really got going back in 2019 with restrictions on chips sent to Huawei. At the time, the US government was concerned with Huawei's 5G telecommunications equipment, so base stations and things like that. And the government restricted exports of high end chips to the company. But Huawei also made smartphones. You worked at Qualcomm at the time. And American chip companies suddenly were at risk of not being able to sell to Chinese smartphone companies. What do policymakers need to better understand about American companies when they design these sorts of export controls?
- [49:06]
A
Well, that the costs are real and they should be taken seriously. I've had more than one conversations where when the companies present, hey, these are the kind of revenue hits that we're taking because of the import export bans or whatever. And these are the kind of costs on R and D and things like that presented to them. Government officials will listen. And then after the meeting they would turn to me or others and say, but they had a really good quarter though. Look at the record amount of revenue they're making in China or whatever. And I understand that, but it's an actual serious thing for companies. That's an actual loss and they have to plan around it. So when I was in Qualcomm, for example, I mean if there was ever a national champion in like communications technologies, it's Qualcomm, at least one of them, absolute pioneer in inventing new things that we use every day and take for granted. When Huawei was being put on the entities list and there were considerations of other handsets makers, other smartphone makers being placed on the entities list, like Oppo, Vivo, Xiaomi, these other Chinese handset makers with the idea that they're somehow all connected to the Chinese military. And I respect all those opinions, but like to the reality for the companies that are selling these chipset was if half the handsets being sold into the world are by Chinese smartphone makers, then that's an extremely big risk that you're facing now for the revenue that you use for your next set of inventions. I think that cannot be underestimated, that these revenue loss impacts are real and should not just be dismissed. Now there are some times when you see, you know, evasion of some of these rules and that's not what I mean by that. So that was really interesting to observe as to how to like wrap your minds around what are the risks here and the fact that when there is a foreign competitor that are not under those restrictions, then all of a sudden they now have market access, preferred market access in the Chinese market that you no longer have. And that's a real hit to you as well. And that's real. I mean, I wouldn't dismiss that.
- [51:08]
B
In October of 2022, the U.S. government expanded its export controls to cover equipment that companies in China needed to manufacture chips. The concern was Chinese companies getting the best equipment to manufacture AI chips used by the Chinese military. This was the start of the so called small yard, high fence policy. You were based in Washington for SK Hynix at that time and your company operated facilities in China to make other kinds of chips. Your company also imported this equipment and needed to send workers from your suppliers in to constantly service it. With this sudden new October export controls, your company needed a legal exception, a license. Did you meet with US policymakers? What happened?
- [51:57]
A
I remember being pulled into, and I was at the company at the time and because I was located in Washington, I was kind of got pulled into this meeting and they had assured us that this would not be impactful for us. And then as I was playing with my daughter in the playground that weekend, I got a call from one of our suppliers saying, hey, do you realize that you need a license for us to continue operating here? And I said no. I mean, they told us, they showed us this does not apply to us and that we were exempted. And they said no, our legal guidance is unless you show us the license, we have to walk out. And so the company was all sudden facing this situation where if they walk out, we might have to shut down production and that's potential billions of dollars of lost revenue. What do we do? And so that led to frantic calls and credit to the government. They got put into action and they granted us the license or the exemption right away because it was not intended as such. Right. The license requirement was about Chinese companies operating in China, not multinationals at the time. It is worthwhile noting that, like there are obviously efforts to this and there's always like discussions about like these loopholes, how they're exploited. But there are also very earnest companies that are trying to do the right thing, that sometimes are impacted negatively by it. You know, you don't see them really complaining about it, and rightly so, because they are companies that are genuinely interested in being aligned to U.S. national security positions and they want to help, not get in the way in a lot of cases. But my goodness, it was sometimes very messy to be in the middle of that from time to time. From the corporate side, one argument that
- [53:33]
B
you often hear is that for every round of these export restrictions that the US government has imposed, all it's done is incentivize China to do more, to try to break the dominance that the American and the Western tech ecosystem has, especially for chips that are needed for AI. What's your sense of that?
- [53:54]
A
So I buy it, but I also don't buy it. To some degree. China has made it very clear that they want to be self sufficient in this sector and they're plowing so much money into it and they're putting all of its resources possible to do that towards a self sufficiency drive that measures that the United States or others can take to curb that I don't think is really deterring them to not have that desire on principle. At the same time it is the case that they have to force themselves to innovate in ways that otherwise would not. So we found for example in the latest Huawei smartphone smics manufacturing capacity at say about 6nm. This is their N3 node, really innovative multi patterning of lithography that otherwise they would never have to do if they had access to EUV equipment, for example. So it is actually forcing them to innovate in ways that they otherwise would not. So maybe they get better at some things than the US because they're forced to. Maybe they're forced to come up with more efficient AI models because they're constrained in compute and we're sort of seeing that happen in real time. But you do also see examples though of where if something is available in China that is not Chinese, there is actually huge preference for non Chinese technology. So for example, if you crack open a Huawei phone, which is not allowed to have American technology in them, you actually find very, very little American technology there. But if you open up a Chinese made smartphone that is allowed to have American technology, then it ends up being mostly American technology because they also have to compete in the global market space. Chinese goods are not as competitive and so they would prefer to have a better one. Or a cheaper one. It's important to keep in mind that the government wants to be self sufficient, but the Chinese industry may not and they vote with their feet as well to the degree that they're allowed to. So I think we just have to always keep that in mind.
- [55:51]
B
Dan, this has been fantastic. Thank you very much.
- [55:53]
A
Thank you. Really appreciate it, Chad.
- [55:58]
B
And that is all for Trade Talks. A huge thanks to Dan Kim at Tech Insights and former chief economist at the CHIPS Program office. A big thanks as well to Isabel Robertson, our audio producer. Thanks to Melina Kalb, our supervising producer. And thanks to Sam El Boise and Sarah Allen on Digital. Please subscribe to Trade Talks on Apple, Podcasts, on Spotify or wherever you get your podcasts. Even if you're a longtime listener. Please take two minutes to leave a review. That is how new listeners will find the show. Please order a copy of my new book titled how to Win a Trade War. It's out in May. That is how to Win a Trade War. See you next week, everybody.