Chips Avengers! CHIPS Act, AI, and Beijing’s Revenge
“Were the October sanctions put in place to limit China’s military capabilities, as the document says? Or is the intent to cripple China’s manufacturing capacity and their whole broader economy?”
The Chips Avengers assemble once again!
Reva Goujon of the Rhodium Group,
Dylan Patel, who writes SemiAnalysis,
and our latest addition: JP Kleinhans of the European think tank SNV.
They’re all here to talk about:
The US CHIPS Act vs. the Euro CHIPS Act;
If and how governments can be “reasonable technology pickers,” and what past industrial policy teaches;
A shoutout to the new “dollar-a-year” tech public servants in America;
The future of autonomous driving and generative AI;
The Tech Cold War and China’s retaliation.
ChinaTalk is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
First Impressions of the NOFO
So a few weeks ago, the CHIPS Act released its first request for funding: anyone who’s interested in getting money to manufacture in America now has their guidelines of what they’re going to have to submit in order to, hopefully, come away with millions (or even potentially billions) of dollars in support for building in the US.
What were everyone’s impressions of that document?
Dylan Patel: To me, it seemed as though the priorities were around the leading edge and advanced packaging, and maybe not so much about traditional packaging and trailing-edge process technology. I mean, there certainly are going to be efforts in that direction — but to me it just felt as though the administration was a lot more concerned about the leading edge (which is an interesting thread to go down).
Jordan Schneider: Yeah, when they say, “Two leading-edge networks or hubs” — that’s a lot of money, and getting two places in America up and running on the leading edge is committing a substantial portion of the $40 billion.
My sense was that they’re still leaving their options open, and that there weren’t a ton of choices that were made in this document. It listed everything that you could think of them ever wanting to do: you saw leading edge, you saw lagging edge, you saw packaging — the hierarchy of needs was just, “We want you to address national security and economic objectives.” We don’t really have the rubric of, for instance, “We’re giving 50% to leading edge and 50% to everything else.” We didn’t quite get all the way to that level of thinking.
Anyways — Jay, did anything strike you about the broad funding priorities?
Jay Goldberg: Yeah — to me it looked a lot like a policy document, as opposed to a technology document. It didn’t strike me as very strategic on which technologies really mattered.
Jordan Schneider: But I mean, there is an interesting negotiation dynamic thing here: when you say, “I want 100% leading edge and nothing else,” then all of a sudden the bargaining positions of the TSMCs and Samsungs of the world gets a lot better, right?
Jay Goldberg: But that speaks to a political dynamic — they clearly negotiated a bunch of deals ahead of it for TSMC and for some of the others, and that’s why those seem prominent. That dynamic is not from a “best practices” technology perspective — it’s more like, “Hey, we have this deal. Let’s make sure that there’s language in the policy that covers deals we’ve already closed,” and then everything else is like, “Here’s our priority list.”
Dylan Patel: I think they didn’t negotiate — because they say “two leading-edge fabs,” right? Well, Samsung’s got a huge announcement in Texas, Intel’s got a huge announcement in Ohio, and TSMC’s got a huge announcement in Arizona. And so all three of those could be significantly paired back — or actually go through, depending on how these negotiations pan out.
So I don’t think they negotiated these deals yet. I think they’re going to play them off of each other, and in the end, one of the three will maybe not get the money that they want.
Jay Goldberg: I think that’s true.
I also think some of them might not actually take the deal — I have to look at Samsung, who had already committed a lot of this before the CHIPS Act. And I’m not sure they’re going to be excited about it or want to take that money (which would be an interesting choice).
Dylan Patel: They’re leaking stuff to Reuters. On March 15, Samsung got Reuters to say that the cost went up by 80 percent (to $25 billion), versus what they had planned a couple of years ago. So I think that’s just more negotiation leverage that they’re trying to put out there.
‘Many, Many Strings Attached…’
Jordan Schneider: So Reva, we have money — but there are also many, many strings attached. What were some of the dramatic or potentially more costly asks that the CHIPS Act is going to make on companies considering taking this money?
Reva Goujon: Industrial policy provides a fantastic opportunity for governments to impose conditions on industry — and you can see that in the language of the CHIPS Act and in the provisions that we see coming out now in this implementation phase.
One line that really stood out is the clawback provision, which said that Commerce can clawback funding if a recipient “knowingly engages in any joint research or technology licensing effort with a foreign entity of concern that relates to a technology or product that raises national security concerns.”
That’s obviously very broad, and the Secretary of Commerce has the remit to interpret or adjust that definition over time. And that, of course, should make companies nervous.
Samsung and SK hynix are already in a really uncomfortable position, just being beholden to BIS licenses to continue operating their fabs in China. The signal has already been sent: “Steer your leading-edge memory chip production out of China. You’re on a timeline. Go.”
I think part of the reticence you’re seeing from companies like Samsung is, “How tight are these conditions going to be? How will the US interpretation evolve over time?” And then you see TSMC is like, “This is all very reasonable. We’re good. We’re going to go for it.”
So it’s really interesting to see middle-power balancing manifested through these companies’ policies as they’re reacting to what’s coming out.
Dylan Patel: So there’s a provision in the CHIPS Act that basically says, “Any excess profits would be shared with the government” — which is very interesting, because I don’t think that has happened a lot to companies in industries that the US government has subsidized.
So what’s up with that?
Jay Goldberg: I read that as: they added that in the last minute because it emerged that the European Chips Act had those provisions in it — and I think some US policymakers thought, “Oh yeah, we should have that, too. That sounds like a good idea.”
JP Kleinhans: Actually, depending on your reading, some of the provisions in Europe are a little bit more drastic. In one subsidy vehicle, it is negotiated that companies have to make the financial gap analysis and tell the government, “This is the delta that we are looking at, and this is why we need subsidies.” And then some governments argue that, “Well, obviously you have a lot of leeway with this financial gap analysis: if you calculate extremely conservatively how profits will develop in your market, then your financial gap analysis might be extremely big, even though the market is extremely profitable for you.”
Over the seven-year period that the subsidies are going out, if their profits are above a certain percent, they have to pay back the entirety of the subsidy. So suddenly companies have to put aside money because, depending on how the market develops, they might have to pay back all of the subsidies they have received.
This is definitely an interesting situation for companies to receive subsidies but, at the same time, they have to put aside money — because if their investment is too profitable or if the market develops too nicely, they might have to pay back all of those subsidies.
Reva Goujon: How have companies responded to that on the European side so far?
JP Kleinhans: Interestingly, there’s not a huge discussion about it in Europe on the public level. But there’s a very active conversation among member states because — as always in Europe — member states differ on their perception of how subsidies should be used, how stringent the rules should be, and so on.
To use a cliche, you can imagine that the more frugal Scandinavian countries are more opposed to the idea of giving away subsidies freely — whereas other countries, including Germany, have a different idea of how to give subsidies to companies.
It’s a little bit of a mess, but it’s interesting to see that, in this particular regard, apparently (as Jay put it) the US government maybe got some inspiration from Europe.
A 21st-Century Test of American Industrial Policy
As a side note to the previous conversation: Jay, you mentioned before that, to you, the NOFO read more like a policy document and less like a technology document — for me, the underlying question is, “To what extent can governments be reasonable technology pickers?” Because ultimately, if you look at industrial policy, governments want to be the winning technology pickers.
I look at, for example, the financial sector, and how tricky it is to pick a winning technology for VC — so I’m wondering, “Is government in the right place to make those rather intricate and long-term decisions on the technology level, when the sector itself already struggles to identify winning technologies in the long term?”
Jay Goldberg: My kneejerk reaction to that is, “No.” But if you think historically, there have certainly been many instances in which the US government has had a technologically savvy policy toward tech.
Usually, it’s rooted in some basic need of the government, say, around defense and military. And so I’m wondering to what extent some of these priorities are reflected in the CHIPS Act. Certainly the chip sanctions back in October were heavily influenced by the Defense Department and adjacent agencies.
I don’t get the sense from the NOFO that it’s saying, “This sector of the US economy or the US government needs such-and-such technology, and we’re going to favor it.” It feels broader to me. And I think that’s reflective of its origins in Commerce, which doesn’t have a big need for advanced chips the way that the Defense Department does.
Reva Goujon: But it can evolve over time, right? Part of its design is to stay ahead of that innovation curve. We’ll see how it evolves — but the trailing edge policy debate still seems to be very much in play. Yes, the emphasis is on leading edge, but you’ve still got $10 billion for legacy chip manufacturing.
Of course, policymakers want to steer sourcing away from China (which is hard to do, just given market dynamics). But it’s interesting to see where some emerging policy areas are cropping up to try to address that — for example, in demanding more transparency on subsidy disclosure. That’s where the EU is already a couple of steps ahead of the US in throwing up procurement investment restrictions around subsidy-driven market distortions.
There are still ways for the US to try to steer some of that policy and diversification, although, again, the economics of this is really hard for US companies to reconcile with.
Jay Goldberg: But there’s flexibility built into these latest policies — and my question is, “Do we think that’s a feature or a bug?”
The “glass half full” version is like what you said: they’re going to adapt over time, depending on needs.
The “glass half empty” view is: they don’t really know what they want, and it’s going to come down to backroom deals and last-minute horse trading, as opposed to a coherent policy objective.
Jordan Schneider: You got to have a sense that there’s been more strategy thinking within the CHIPS Act than just what we saw in the vision document and the NOFO.
JP Kleinhans: In Europe, from the beginning the conversation about subsidizing chips was focused on cutting-edge frontend, because that’s the stuff we don’t have at all. But then, over the past eighteen months or so, policymakers understood, “Oh, there’s this thing called packaging; and oh, there’s backend; and packaging is advancing, and then you call it ‘advanced packaging’ — and maybe we need that.”
I remember in one of the Commerce Department’s requests for comment, a couple of US printed circuit board (PCB) manufacturers replied, “If you’re worried about national security, don’t forget about us.” And with PCBs, the interesting thing is that China is actually not lagging behind — the most cutting-edge PCBs are in China.
So my question here: hypothetically, if I am a really cool PCB startup or a small PCB company, would I get anything out of the CHIPS Act? Because I wouldn’t get anything out of the EU one.
Jay Goldberg: I don’t know the answer to that — but I will say: being an advanced PCB manufacturer is not as exciting as it sounds. It’s a pretty miserable business: very low margin, fairly labor intensive. I don’t know how much standard packaging is going to come back to the US now.
Advanced packaging, on the other hand, is important. Some of the advanced packaging stuff is super hard to do, and it’s going to be done by the foundries. But the bulk of the business is not coming to the US. If it’s not in China, it’s going to go to Malaysia.
JP Kleinhans: But at the same time, the weird thing is that, since we are all doubling down on national security, PCBs should be of concern to us, right? And maybe even more so than the 15 billion transistors in your cutting-edge processors: compromising a chip that then ends up on a PCB is much more vulnerable to an attack than, say, trying to infiltrate Intel’s or Nvidia’s next cutting-edge processor and hiding a back door among the billions of transistors.
So I find it interesting that, for some reason with PCBs, the US government forgot to think about the national security angle.
Dylan Patel: The question is whether the policymakers believe that there is a certain level of minimum production that is in the US or in Europe for various technologies. Should we be able to make an advanced Nvidia AI server and the networking equipment to connect them all together — even if it’s a small amount?
And to Jay’s point — that’s not economically viable. There’s too much labor involved; a frontend fab requires very few people, and labor is less than twenty percent of the cost of a final chip. It’s all capital. So that makes more sense, at least from an economic standpoint.
But then are policymakers going to subsidize: “Hey, actually we want to be able to have three percent, or five percent, or ten percent of the world’s PCB manufacturing here.”
Reva Goujon: But remember, there are multiple policy tools in play. Your industrial policy is your incentive — but then the other huge whopper of a policy that’s emerging now is the RESTRICT Act.
Senator Mark Warner is behind this. He’s a very serious player in Congress, and it’s really interesting because this is something that focuses on information and communications technology (ICT) supply chains. Remember, Commerce already has ICTS rules in effect — these were introduced on the last day of the Trump administration, and then went into effect under the Biden administration. We’ve been waiting to see, “Will this be fully utilized?” — and the TikTok debate has just catalyzed this.
But if you look at the text of the RESTRICT Act, it is deliberately wide-ranging, and it’s meant to cover hardware, software, services, even the provision of those services — anything around ICT supply chains that has a nexus with a foreign entity of concern. So to your point about printed circuit boards and the national security risk — well, the RESTRICT Act is the catchall for something like that.
Jordan Schneider: So coming up from PCBs for a second — every single company at any point in the value chain is going to hire a smart lobbyist to help them frame their thing from a national security perspective. And Commerce is going to have to sift through potentially thousands of companies asking for support, saying, “Look, maybe you weren’t aware of this, but our thing is really important to national security. So give me a slice of the pie.”
And I’m sure there’s actually going to be a lot of learning and processing. And interestingly, in the NOFO as well, there was a lot of room for back and forth — a company would send an initial twenty pages, and then the Commerce Department would say, “Oh, we want to learn more about this, that, and the other thing.”
But I imagine that, over this iterative process, there will be priorities that the government will be sold on which nonetheless aren’t necessarily the ones that policymakers have already internalized around, say, leading-edge logic.
Should we talk about childcare for a second?
Reva Goujon: The interesting thing about that: the provisions that got thrown in on childcare have reopened some of the partisan vibes. The Republicans are saying, “Wait, you can’t just throw that in there. I thought there was some consensus around this — how many standards do you want to pile on?” Then there’s, of course, a legitimate argument about what kind of labor standards the US wants to lead with here.
But then you’ve got Morris Chang of TSMC saying, “The labor cost is so expensive in China and in the US; and globalization is dead.”
But this is just the new world we live in. I don’t think childcare stops this in its tracks or anything — it’s just a revealing political element to the policy evolution.
Jordan Schneider: Yeah. You spend money, and you’re going to push as far as you can, and you’re going to get some sort of return on your investment — from economic security, national security, securing the future of the industry, creating an ecosystem, or whatever. And there’s a give and take with all of this.
Maybe it was surprising to me that childcare was the one thing that the national media picked up on — because we’ve been talking for thirty minutes about lots of other things that, I think, are going to be far more consequential than whether a thousand employees in the US end up having more subsidized childcare than they would otherwise. Childcare is important, and I think everyone should have childcare in America — but I’m not sure that the future of America’s semiconductor ecosystem is going to hinge on whether or not TSMC’s Arizona fab has C+ or A- preschool.
An Encomium for Public Servants
JP Kleinhans: What is striking to me is that the Department of Commerce is on a hiring spree for the implementation of the US CHIPS Act — and some of those hires are really impressive. Among others, Commerce got the chief economist from one of the largest semiconductor companies in the world to now be the chief economist for the implementation of the CHIPS Act. [Ed. This hire is Dan Kim, formerly at SK hynix, the world’s second-largest memory chip producer.] These are people who know what they’re talking about when they talk about chips.
I find this spree pretty impressive — not just in terms of talking the talk (ie. writing regulations) but also walking the walk, and trying to exactly figure out, “How do we spend this money most meaningfully, and how do we sift through probably hundreds of applications, where some of them might be sketchy?”
I don’t see the same thing happening in Europe, even though the EU Chips Act talks about almost the same amount of money — theoretically €43 billion. I don’t see a hiring spree in the European Commission or in member states to step up their industry-knowledge game.
So to me, if we compare the US and EU, it is really increasingly to see who puts the money behind government resources to actually implement all this stuff.
Jordan Schneider: So during the Great Depression and the New Deal, there was this new class of “dollar-a-year men” (they were all men back then), who basically came in from industry to spend all the money that FDR got appropriated to rebuild America.
And I think it’s worth acknowledging on this podcast that the pay cuts all these people from industry are taking by doing these jobs are factors of five or ten.
The CIO spent thirty years or something at KKR, and he probably isn’t making more than $200k to do this job. [Ed. This hire is Todd Fisher; Kohlberg Kravis Roberts & Co. is a global investment company.] This is a real public service.
So it has been really encouraging to see the type of folks that the administration has been able to persuade to take these opportunities, because they’re not doing it to get rich. We’ve talked for thirty minutes — these are incredibly difficult things to weigh and tradeoffs to make that you can’t figure out just by looking at these things for a few months. Because if you do that, then the lobbyists will beat you, and you won’t be able to get the best bang for your buck.
Paid subscribers get early access to the second half of our conversation — in 3,000 words, we get into:
The US/China relative economic impacts of generative AI;
What China’s wave of retaliation on US firms portends for foreign firms’ future China prospects;
The diverging interests of the US and EU, and strategies to engage with China;
How Singapore could turn into the “Switzerland of Cloud.”
Keep reading with a 7-day free trial
Subscribe to ChinaTalk to keep reading this post and get 7 days of free access to the full post archives.