US-China Chip War with the Chip Avengers
"It takes a thousand steps to make a semiconductor and you're going to have to get them all right."
What will the Biden administration's new export controls mean for the US and Chinese semiconductor industries as well as the future of the US-China relationship?
The second-order implications of the Biden Administration's moves for industry;
What it would take for China to circumvent these controls;
How Beijing might strike back;
How the regulations could impact the risk of war.
If you haven't read my interview with Kevin Wolf, check that out first!
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber!
Were the Regulations “Right”?
Jordan Schneider: Was this the right thing for the Biden administration to do?
Jay Goldberg: I think “right” is a difficult word to use. My question is: right for what? One of the things that I've really puzzled over since these rules have come down is: what is the government really seeking to achieve? The scope is very broad. Was that the intent? What's the end goal? Some things are going to work very well; other parts of it, I think, are going to be very problematic.
Reva Goujon: One layer of this, of course, is the strategic competition element. As the US is scanning China's strengths and weaknesses, where does it have leverage over choke points? Where can it really hit hard? I think some of the real novelty in these controls showcases how it's concomitantly narrow and broad, and very deliberately so.
In what it prioritizes and where it goes, it is saying, “We want to know what inputs are going into China, period.” Companies are just going to be subject to US licensing as a result.
But there's also another layer here, which is petty politics. We look at political drivers trying to out-muscle one another on China policy and set the bounds for this debate. While China can be a very bipartisan issue on one level, on another level you can see a lot of variants in tactics and proposals of how a new export control regime should be structured. So here's the White House coming out ahead and saying, “Well, these are the technologies that matter. These are the thresholds that we're going to define, and these are the tools that we're either going to going to adapt or create anew.”
Martin Chorzempa: The signaling for this really goes back to Jake Sullivan's speech a few weeks ago, where he said,
We previously maintained a “sliding scale” approach that said we need to stay only a couple of generations ahead. That is not the strategic environment we are in today. Given the foundational nature of certain technologies, such as advanced logic and memory chips, we must maintain as large of a lead as possible.
If we're thinking about the goals of this, I think there's been a shift away from trying to freeze or hold back China's advances in chips. I see this as an attempt to roll back and degrade China's existing capabilities. China can produce 14-16nm logic chips. Biren can produce GPUs at these thresholds. YMTC can produce NAND at this threshold too. What's very interesting is that they did not set the thresholds at an aspirational level for China. They set them at levels that China already is able to do, and it's cutting off their ability to do it.
The fact that China's already at this level tells you that the Biden administration was in a bit of a rush. They felt like they had to do this now.
Jordan Schneider: If you've decided that the US is in a decade-long tussle with China over retaining a national power advantage, is this threshold the right one?
Doug O’Laughlin: I thought what Martin said about not trying to cut off their future, but actually calling back some steps, is a really good observation.
When you read it a little closer, the regs are actually extremely arbitrary. One of the first premises of the regulation is cutting off AI and HPC. They talk about [how] we're going to cut off their ability to buy from TSMC, etc. They're also going to cut off their ability to fab these kinds of chips in the future. But if you do a closer look, a lot of the same technologies that they're using are actually technologies that are used in 28nm and even 40nm. It's very backwards looking. For example, some of the metal deposition stuff is so broad and so encompassing that a very broad application of that would be able to put a stranglehold on even some lagging-edge technologies.
In my view, the zeitgeist has been doing this for a long time: from ZTE to Huawei to now this, it’s just the same story on a continuous escalation path. But now it seems different. It's much more punitive and broader.
Economic Competition and Timing
Jordan Schneider: A lot of the justifications within the regs were about supercomputers’ use for missile modeling and other WMD applications. I don't think you need fancy memory chips to do that?
Doug O’Laughlin: Memory is kind of a commodity, right? They're interchangeable: Apple can get it from YMTC, Micron, or Samsung. You could maybe make a logical assumption: because memory is one of the things that really matter for AI (especially DRAM), you go, “We're going to cut off better memory, because that's really important for the entire AI model computing power.” That's maybe a logical thrust, but how they actually implement it seems much more punitive.
Martin Chorzempa: The really big gun, which is the foreign direct product rule (FDPR), was not applied to memory, DRAM, or logic chips. It was only applied to AI chips and supercomputer controls. My understanding of the rules, as best I can gather, is that [China] can't use TSMC for the really advanced stuff, but stuff like memory and even 14nm are still going to be fine to fab at TSMC. It's only a restriction on exports from the United States to those places.
I think [the BIS], on purpose, applied a different level of control based on how concerned they are about the military implications of these chips. But then, a sense of economic competition is also mixed in. I think Kevin Wolf alluded to this in an earlier ChinaTalk: it seems to be a move back to Cold War-style thinking. I know this is something that China insults the United States with, but it was a longstanding US policy that we wanted to keep, in general, a technological advantage on foundational technologies ahead of the Soviet Union and China. Now, [the regulators] are arguing that you can't really differentiate anymore between what's important for commercial and what's important for the military side, and they have to control it all.
Jordan Schneider: I remember that just a few years ago, people rolled their eyes at military-civil fusion being the be-all-end-all conversation winner. To have it put front-and-center by a centrist Democrat gives me the sense that this policy and outlook towards China, whether or not it's accurate, is something that is going to be locked in for a very long time to come in American discourse and policies.
Jay Goldberg: If that's all true, then we have to come to grips with the fact that this is essentially economic Cold War, right? If you start talking about broad limitations to military-use civilian technology, you're talking about cutting off large parts of China's economy and some of its most successful sectors: the internet, the Super Seven, the hyperscalers, Alibaba, Baidu, Tencent… There’s going to be potential for big impacts on automotive as well. Hurting those industries is a whole different ballgame than cutting off missiles and supercomputers.
Reva Goujon: Which is why both the implicit and explicit objectives are really important to parse out here. What's striking about these rules is: where it talks about advanced semiconductor thresholds, it basically creates a leap that defines advanced semiconductor thresholds as below 16 or 14 nanometers, non-planar, transistor architecture, etc. These could be used in a WMD application. They could've set that threshold at sub-10nm or sub-5nm, but they chose a pretty hard line there [based on] where China's already developing. That seems very intentional, and that WMD justification can be applied quite liberally in the way that these rules are written.
Martin, to your point on [the] foreign direct product rule, I think there's a window that the US is signaling here. [It’s] to say, “This is where we're defining those thresholds. Now, tech partners, CHIP Four, DTC, etc.: follow suit or else.” There's political will here to resort to extraterritorial measures, as we're seeing in other aspects of how the US persons activities rule is being applied here in terms of where FDPR was used.
I think this is really a test of how other partners will choose to align. This is not the way other countries like to design their export controls. This is a much more creative, novel approach to it.
Doug O’Laughlin: The thing that's so brilliant, taking away the normative side, is the timing. The economy of China is falling off a cliff, so the historical levers they have to pull back against the United States are a lot weaker. If Chinese low-end smartphones are falling off and automotive sales are going down, how much more can they cripple you?
Jay Goldberg: Did they know that they were going to release the rules right ahead of the Party Congress?
Doug O’Laughlin: If you read the regs, they feel rushed. Also, [the US government] did a press release on the Monday right after the Party Congress about national security.
Reva Goujon: It seems like there was a big lead-up to this, though. We had a lot of clues dropped over the summer, [like] the “is-informed” letter sent to Nvidia. Sullivan really set out the sea change in policy. Of course, there's the midterms [and] political timing. Lots of different drivers there to factor [in].
Jordan Schneider: What folks in the Administration have told me is that [they] didn't plan on this being a giant middle finger to the Party Congress, but look, this is how it landed.
Retaliation Options for China
Jordan Schneider: Let’s talk about potential retaliatory measures. I wanted to come to this paragraph that Jay wrote a little while back:
We also have to be cognizant that China is starting to develop industries where it occupies higher strategic ground, albeit not in semis. These include areas such as industrial systems (machine tools, lasers, robots, etc.), electrical power systems, batteries and more. These are not as glamorous as semis, and Chinese companies still have some ways to go, but somewhere down the road they may become strategic enough to provide China with a little bit of the leverage it does not have now.
Jay Goldberg: I follow venture investing in China as closely as anybody can from this far removed. There's a lot of activity in some important: machine tools, robots, lasers, etc. Those are all important industrial tools, and China's actually done really well in developing those. CATL is investing in batteries all over the place. In this country, we're talking more about electric vehicles and the electrification of everything. China has a lot of battery capacity, and I don't think we're seeing the same level of investment in the US or anywhere else.
What happens five years down the road, when China's been investing heavily in dirty, polluting battery factories that are then going into clean vehicles? They're going to have that capacity; we may not. Taking a step back, my bigger concern than the restrictions is what kinds of steps the US government is taking to encourage all these new technologies.
Jordan Schneider: Doug, say you're sitting in MIIT and you want to screw over American microelectronics, What's in your playbook, and how are you weighing the advantages and disadvantages of various paths?
Doug O’Laughlin: First and foremost, I would look at everything that's done in Hong Kong. There's a lot of distribution done there. Some [Chinese] companies already end up doing a lot of distribution and logistics there.
Packaging is probably the place [where] China has a relative advantage compared to the US. There's effectively no packaging done in the United States, and I would love to see that changed. This includes advanced packaging and not-so-advanced packaging: at the end of the day, whenever you make a semiconductor, you still have to put it into a PCV or whatever form factor you want to then make the device. Obviously, that intermediate step is a lot lower value than actually fabbing the chip, but there is a huge choke point there and [China] could probably flex that.
The problem with that is: if you pull that lever really hard, OSAT is not a crazy business. It is a little human capital-intensive, but [companies can] move that to Malaysia or Southeast Asia really quickly. If [China pushes] too hard on that lever too quickly, that just further encourages companies to leave ASAP, which is already happening to a certain extent. People are diversifying their supply chains away from China, mostly through other Southeast Asian countries which are lower cost anyway. Maybe that will hurt in the extremely short run, but that will be a Pyrrhic victory in a three- or five-year period.
Jay Goldberg: The bottom line is: China doesn't have a lot of great levers to respond with anytime soon.
Reva Goujon: It just underscores the unreliability factor for China at a point of extreme economic weakness, which probably went into the timing as well for the US in unleashing this as a barrage of controls — it wasn't incrementally trickled out by any means.
Another area, maybe less obvious or explicit, is the anti-competition element. When we look at China's longarm tools and the arsenal that it's built, there was an article recently where China was basically congratulating itself on building up these extraterritorial measures. Obviously, it wants to mirror the US in that kind of leverage, but it doesn't have the same punch — not even close. Anti-foreign sanctions laws have just been symbolically applied to entities that don't have a whole lot of exposure in China. Now that the US is moving to next-level tech controls on China, if China does take the gloves off and target companies that are more exposed in China and face more domestic competition, that would be interesting and a real risk.
In the anti-competition realm, SAMR has had real leverage to disrupt M&As. If you look at that in parallel to the US-led friend-shoring current, where, inevitably, this is going to involve tie-ups and all kinds of deals to coordinate supply chains and get more vertical integration, this is where you could see some movement from SAMR as some big deals come on the radar.
Problems of Expertise
Jay Goldberg: I would love to hear from the people who wrote the regulations: in crafting these rules, which agencies showed up and really did their homework, and who phoned it in? Because it feels like some people did more work than others. Some things are incredibly specific, like the rules around AI performance. [In] other areas, it's just like, “supercomputers.”
Doug O’Laughlin: They go ham on metals deposition. They talk about 14 different types of metals, different types of processes, ALD… And then they don't talk about etch once. Whoever did the etch part mailed it in, and whoever did the deposition became an industry expert. The metals deposition part was mind-blowingly specific, and then they talk about ion implantation, which is something that I know a little bit about, and it’s so vague it could mean the entire set of tools that have ever been made.
Jordan Schneider: There is limited expertise in the US government on this stuff, and I think it's probably pretty spiky for certain things and not others.
Jay Goldberg: In last week’s show, Kevin was really clear that the timing between [the] announcement and implementation of the rules was deliberately very, very short. I understand they had valid reasons for keeping that window tight, but it opens up all sorts of unintended consequences. In the rush to get the rules out, there was very little or no consultation with industry. My second question is: when the authors are putting this together, to what extent did they wargame or think through some scenarios of what's going to happen next?
Reva Goujon: It raises a question: how much quality industry input was coming from the toolmakers vs. the chip makers in designing a policy like this? You definitely see some gaps there in terms of that two-way communication, and not a whole lot of voice coming from the SME front.
Jay Goldberg: I've heard that it was deliberate: they wanted to keep certain industry bodies out of the process because they have a lot of lobbying weight.
Doug O’Laughlin: I follow earnings calls. The Lam Research call was despondent: they were like, “We had no say in this. This is going to hurt the crap out of our company; we can't even quantify [it]. No one really gave us a heads-up at all.” At the same time, Pat Gelsinger is out here doing somersaults, a press release, interviews… I'm sure Pat was in on it.
Reva Goujon: Obviously, BIS is creating a lot of work for itself: it's ballooning regulations., [and] it's severely understaffed already. You could take that and infer that there's no way that this can be enforced adequately. On the other hand, when you have limited resources, you make an example out of companies to draw broader compliance.
We're in an interesting phase right now where we're coming off [the] CHIPS Act’s funding passage, and all of that is now in the implementation phase.
The government probably feels like it has considerable leverage over the major recipients of those funds to comply: these are the conditions attached, [and] there is intent to steer production and high-performance computing chips away from Chinese access.
There's an acknowledgement in these controls that this is hard. It’s really hard to get the technical thresholds right. Civil-military fusion is just really, really complicated for regulators to wrap their hands around. So guess what? We won't. The US is basically saying that the onus is on Chinese entities to demonstrate their bona fides, or you're on the unverified list, which is a shortcut to the BIS entity list (or worse, if you end up getting FDPR-ed).
Jay Goldberg: I think that CHIPS Act leverage is a really, really good point. If you look at all the stuff that's happened over the last four years, starting with Huawei and ZTE, the biggest force working against all the government's rules are US companies themselves. I think the authors of these latest restrictions were very cognizant of that fact. Now, we're going to see how it actually gets implemented, because we've already had a bunch of waivers go out.
We've been talking before about how the devil's going to be in the details of implementation, and a big part of that will be how we treat the US companies who want to continue selling different products. I mean, here we are four years later, and Huawei is still shipping base stations and phones; not a lot of them, but they're still able to do it and they're sourcing US parts somehow.
The Village of American IP
Martin Chorzempa: The firms that have agreed, through receiving funding [from] the CHIPS Act, [to set] up more semiconductor-related activity in the United States have already agreed that they're not going to expand, and in some cases upgrade, their semiconductor activities in China. These restrictions could, in a sense, supercharge that by making those decisions even more clear for them, because whether they accept the CHIPS ACT funding or not, they won't be able to make those investments in China. They probably will not be able to get the most updated advanced equipment going forward if they only have these potentially shaky waivers in the short term.
If you're a firm that has a lot of business in China, you might decide that you would rather excise as much US technology and US components from your equipment, supply chains, or other processes as possible so that the US can't have any of this longarm jurisdiction on you.
Doug O’Laughlin: Something that's really weird about semiconductors is that in a lot of ways, the way that they have been made is almost identical to when the industry was birthed 50 years ago. It's almost all the same steps but just miniaturized.
That's one of the reasons why America has so much long-term IP: they invented it and it's almost the same, just a billion times smaller. Semiconductors are a very cumulative effort. It's taken so long to get here, and it's been a village. ASML, on their earnings call recently, said, “We are not going to be as impacted by this because we're a European company. We're shipping European goods.” But I'm almost positive that the BIS has a lever to go after them, because one of the meaningful subsystem parts is made in the United States — it's a San Diego-based company. It's really, really hard because of how much entrenchment there is.
I think that there are ways to eventually get there in some of the commoditized processes of the semiconductor fabrication stack. CMP is not the craziest thing ever, and I think that if you give anyone a sufficient amount of time and resources, they could figure it out. But it's going to take a lot of time and resources, and that's before we even have a working product. There are some places where I think there's potential, but it takes a thousand steps to make a semiconductor and you're going to have to get them all right.
Jay Goldberg: SMIC’s Shanghai foundry has been poaching staff from TSMC since its inception and they still have not been able to close the gap. I think there is an important distinction here between technical capability and commercial capabilities. I'm sure that there is some research lab in China somewhere that has a small EUV system and can produce 5nm chips, but they can't do it at scale. They can do a wafer a day as opposed to 30,000 a month.
Doug O’Laughlin: It's one thing to make one wafer, but it's another thing to make thousands of wafers with 99.999% accuracy over and over and over again. That's the real problem: this is a cumulative benefit. If this is all you've ever been doing since the 1980s, all that cumulative knowledge and experience comes with you.
Jordan Schneider: If you can do a wafer a day in a lab, you can probably make enough for your missile systems, no? And if so, then it really is just an economic competition.
Doug O’Laughlin: There was a really good article I read the other day about how Russia is forced to import Chinese chips in order to support [its] military operations. The effective yield of that is like 60%, meaning 40% of the chips that they're [importing] have defects. This is lagging-edge technology: it’s not rocket science at all. Maybe you can make some artisanal handmade rockets, but we're looking for scale.
The entire output of the entire [Chinese] academic system making batch wafers would be equal to three hours at TSMC.
Jay Goldberg: But to your broader point, Jordan, I do think that to really restrict the PLA from getting advanced chips is going to be a big struggle. The supply chain is so porous. They're not going to need to go to academic institutes; they're going to get them on the open market through dozens of shell companies and complex distribution channels.
Reva Goujon: There's an interesting underlying policy debate: should the US maintain and grow a strategic one-way dependency with China?
China remains highly dependent on the US for critical inputs, especially for mature technologies that go into more commercial spaces. Even in the CHIPS Act guardrails, there was a provision in there that basically said, “Don't increase production in China (it set the threshold at that point at 28nm), but if you're going to be producing in China for the China market at that more mature threshold, that's okay.” That gets undermined if you go too far with your controls, because once you get to an FDPR-level control you're getting to a level of blacklisting that undermines the strategic one-way dependency argument. You're not going to be able to sell to those Chinese entities. You're not going to be able to grow that dependency.
Why Hasn’t China Succeeded?
Jay Goldberg: I think it's very interesting that there is this presumption of success on behalf of China's semiconductor complex. We're assuming that they have this incredible capability. In some senses, the opposite is true, right?
Over the past decade, they've spent a hundred billion dollars plus, and they have not actually advanced their ability to manufacture chips at all. The gap between SMIC and TSMC is just as wide as it was ten years ago. The domestic wafer fabrication equipment industry is not capable. They don't have EDA tools. They've had a lot of success on the Fabless side, [with] thousands of companies doing chip designs, but they're all as dependent on TSMC as everyone else.
There is a reasonable argument to be made that if your adversary is making mistakes, don't stop them, right? If they spent a hundred billion dollars last decade, let's see if we could just let them keep going and get them to spend 200 billion this decade.
Doug O’Laughlin: I do agree that it's been very hard for them, but I want to push back a little bit on the struggles. YMTC is a perfect counterexample of that: they really do have a technology roadmap that's as good as the Western companies. At the same time, I also think that SMIC has had some grenades thrown in their process along the way. I really wonder: if SMIC had no export controls, the Huawei thing never happened, [and] they were never stopped, would that gap be shorter? I think yes.
The book Tiger Technology talks about each decade of semiconductor countries essentially catching up, and the playbook is the same. The government heavily subsidizes loans and props up a semiconductor industry; they make tons of losses [and] are meaningfully behind, but they close that gap and then eventually become a leader. This happened in Japan, obviously, then Korea, then Taiwan.
China literally just found the playbook. I don't think that given infinite time and infinite resources, they wouldn't have done everything else that history has done as well.
Jay Goldberg: I agree with that economic framework in Tiger Technology, but there's a key step there: to close the gap and transition to being a leader is much harder. The number of economies that have actually been founded on that isn’t very [big]. That transition from a middle-income trap to becoming a world leader is very difficult.
Doug O’Laughlin: That's the thing that makes it so interesting: right now, there are actually a lot of pushbacks. People are getting arrested essentially because of their high-profile failures — being given billions of dollars and having nothing to speak of it. If we're talking pure capitalism, I don't think we ever thought that China would [get] the greatest ROI or ROE of that capital. If you got a 0% ROE on a hundred billion dollars for a Chinese semiconductor industry, I think the state would be okay with that.
Jordan Schneider: As long as you can deliver the goods, get your global market share, and help out the PLA when they need it. That's not quite where we're at, and getting there is going to be more buckets of tens of billions of dollars with a definitely uncertain future of whether or not this is even possible.
Risks of Conflict and the Future
Reva Goujon: The tech-control escalation ladder and the Taiwan escalation ladder cross over in interesting ways. In a beyond-2024 scenario [with] potent rules under the national security umbrella or under this new package of BIS expert controls, how those are applied really, really matters.
You can take a precision approach, or you can take a really blunt approach. If you push China toward an existential crisis, that could have some big effects. I think that's something that weighs heavily on the minds of US partners. Japan has experience with the US: if you look at history, what do you resort to when pushed into a corner? Europe, looking at the war on its own continent, is really, really hesitant to push China this far when it's already dealing with 20th-century problems creeping into this century.
Jay Goldberg: Reva raises the key point in all this: what comes next? If I were to pick a second order priority: now that we've just bought ourselves a ten-year window, what are we going to do with it? Are we going to spend all this money pushing lagging-edge technologies, or are we going to actually try and invest in things that matter?
The drawback of a planned economy is you're building last century’s railways and hydro plants. What are we going to build for the 21st and 22nd century — power systems, quantum computing, and new models of compute?
How are we going to spend the next few years: focused on beating China, or focused on building what's important, what's new, and what's coming next?
Outro song by Taiwanese rapper and pop star SHOU 婁峻碩: