← The AI Policy Podcast
The AI Policy Podcast
China's AI Industrial Policy, New Controls on Huawei's Ascend Chips, and President Trump's Middle East Trip
00:28:15
We cover China’s AI policy, new US chip controls on Huawei, and Trump’s Middle East trip. Let me know if you'd like it to keep timestamps or specific names like Kyle Chan.
Loading summary
Transcript37 lines
- [00:00]
A
Foreign.
- [00:11]
B
Back with the A Policy podcast, my great colleague, Gregory C. Allen. As always, buddy, we got a lot to talk about today.
- [00:18]
A
There's a lot. A big, big week.
- [00:22]
B
Big week. Starting with, you know, I just want to go today. We're talking about Monday, May 19th. We're talking at midday. But there was an op ed this morning in the New York Times by Princeton researcher Kyle Chan, who published this piece called in the Future, China will be Dominant, the US Will be Irrelevant. Sort of shook me a little bit and I want to know your thoughts. What's the thrust of his argument and why should the policy community here in the United States care about it?
- [00:50]
A
Yeah, so Kyle has been writing a substack for a while and he's a, you know, a postgraduate student at Princeton. And you know, this, which I admire, by the way, I just want to say this is somebody who gets off on the strength of his ideas, not the strength of his credentials. Right. He's not somebody who had some highfalutin job, you know, before. And that's why people listen to him. To the extent people listen to him. It's because he makes interesting points and connects dots in an interesting way. And I just have to say, I think everybody should go read this article, like right now.
- [01:24]
B
So do I.
- [01:25]
A
Whether you, whether you, our producer will.
- [01:28]
B
Put it in the notes of this podcast so it's easy to find.
- [01:31]
A
Yeah. So whether you end up agreeing with this article, whether you end up disagreeing with this article, it's so provocative, it's so bluntly stated, and it really raises really important questions that I think just everybody wants to hear. So maybe I'll read a big quote from this. This OP said, and it's going to be long. I apologize for that. And then I'll sort of give give my take. So here's the, the quote from Chan. If each nation's current trajectory holds, China will likely end up completely dominating high end manufacturing, from cars and chips to MRI machines and commercial jets. The battle for AI supremacy will be fought not between the United States and China, but between high high tech Chinese cities like Shenzhen and Hangzhou. Chinese factories around the world will reconfigure supply chains with China at the center as the world's preeminent technological and economic superpower. America, by contrast, may end up as a profoundly diminished nation sheltered behind tariff walls. Its companies will sell almost exclusively to domestic consumers. The loss of international sales will degrade corporate earnings, leaving companies with less money to invest in their businesses. American consumers will be stuck with US Made goods that are of middling quality but more expensive than global products owing to higher US Manufacturing costs. And goes on. And I want to, just want to be honest, this, this piece is quite critical of recent Trump administration moves. Here's one of those criticisms. It doesn't matter that Washington and Beijing have reached an inconclusive and temporary truce in Mr. Trump's trade war. The US President immediately claimed it as a win. But that only underlines the fundamental problem for the Trump administration and America. A short sighted focus on inconsequential skirmishes as the larger war with China is being decisively lost.
- [03:15]
B
Okay, so that's strategic. That's interesting. Definitely worth talking about.
- [03:20]
A
Yes, 100%. And I think it's, it's really just worth pointing out. And he says this in the op ed, which I think is interesting. You know, even when we talk, when Donald Trump talks about, you know, the consequences of tariffs might be that Americans have to buy, you know, two dolls instead of 30 dolls. That's really portraying China as it was, which was a low skill manufacturer of low end goods. And the reality is, is that China is now a very powerful manufacturer in a bunch of high tech indust. Electric vehicles and solar and robotics are all well known case studies. But it goes beyond that. I mean, China is just really good at making really complicated goods. And one of the things that I like to explain in terms of this story is just the difference between the Chinese economic model and the Soviet economic model and like what China is optimizing for in my mind. So if you think about the Soviet economic system, that was a centrally planned system, you know, the, the bureaucrats in the Kremlin had to decide to match supply and demand. So they picked all the winners and losers in the economy. And that's basically just impossible, you know, for any bureaucracy, however competent, to pull that off. So they made a ton of terrible mistakes and their economy was really technologically backward. Contrast that with China, it's still a centrally planned system to a certain extent. But most of what they're trying to do is not match supply and demand. It's trying to have supply radically exceedingly demand. And the reason for that is that they're not optimizing for GDP growth per se, at least not in the Xi Jinping era. They're optimizing for competitiveness, industry competitiveness. So they're saying, okay, you know, we know that our bureaucrats are not smart enough to just pick, you know, which company should win in the following markets, especially not in the early days when China is not necessarily all that strong in this market. So instead of trying to match supply and demand, we're going to create, you know, through subsidies, through favorable, you know, industrial policy, we're going to create supply that like 5x exceeds demand. And then the survival of the fittest that washes out in those in that industry where there's a bunch of companies who are all struggling to survive when there's not enough demand to go around. Whatever companies survive that culling are going to be absolutely world class because they were sort of grown in this ultra competitive jungle environment. And what they've done in electric cars, in solar panels, in so many other industries is use overcapacity to get around the problem of the government having to pick winners and losers. Right. It's still the market that's picking winners and losers at the end of the day, but China is essentially guaranteeing that Chinese firms will be the ultra competitive winners at the end of the day. So rather than picking companies, they're picking industries. And the thing is, while it has a ton of costs for Chinese consumers, while it has a ton of costs for Chinese citizens, it still is fundamentally effective at creating these ultra competitive companies like Huawei, like others who can then be national champions after they've established this incredibly competitive corporate culture and this world leading technology.
- [06:36]
B
All right, Greg, so you know, to what extent do you agree with Kyle Chan's conclusions about the future of US and China's AI competition? I mean, we generally think the opposite. That, and you and I have talked about this on the podcast, that, you know, they are good at making a lot of things, but they're not good at innovating, as least as good as we are. You've just not.
- [06:57]
A
I think they are. I mean, at this stage, I think they are. Yeah, what, what I, what I agree with and what I have said is what Chan concludes with. Right. He concludes with this line. Whether this century will be Chinese or American is up to us, but the time to change course is quickly running. But you know, that, that, that final line almost like doesn't seem supported by the rest of his argument. The best of the rest of his argument seems to say that basically China's going to win and we're doomed. And I would say offer a lot.
- [07:25]
B
Of solutions either does it.
- [07:26]
A
Right. You know, which you only get the space you get in the Times op Ed. But I think, I think it is not settled that the United States wins and it is not settled that the United States loses. We could totally blow. This is what I would say. And that's why I appreciated this op ed as a wakeup call saying, you know, if you think that this resembles the. The rosy memory of the Cold War, which is we had a good system, they had a bad system. All we had to do was wait for their system to collapse under its own failures. That is just not the situation we face here. Right. And I actually would argue it was not the. The situation we faced in the Cold War either. I think there's plenty of ways that we could have either nuked each other or lost the Cold War if we hadn't, you know, done certain things right. But the situation facing the US And China is definitely of that nature. Right. Which is if we play our cards wrong, we could totally lose. I mean, to put it in like this is overly simplistic terms, but China has four times the population of the United States. The United States, I'm, you know, roughly here, has four times the GDP per capita of China. So if they were to match us in GDP per capita, we should expect, you know, their economy to be four times larger than that of the United States. And you kind of never learn about somebody's character, right, when they're in a position of, you know, greatly reduced power. You kind of learn who they truly are when they're the top dog. And when the United States was the top dog, we let the rest of the world, you know, get wealthy. We spent a ton of money trying to make the rest of the world healthy. I would say, all in all, we were pretty nice hegemon. I would say, just from the data points that we have in terms of how they treat their neighbors, like the Philippines, like others, I think China is likely to be a pretty rough hegemon for the world to have to deal with. And I think as much as people love to say that, you know, the United States's advantage is in innovation, I would say that that is not a birthright. Right? That is a contest that you have to fight every single day. That is a, you know, a position of leadership that you have to earn every single day. And I think the reason why I'm encouraging everybody to read this Chan piece is to just sort of wake up that. That this could be lost. This could go bad.
- [09:40]
B
Okay? So wake up. We gotta get on it. Important op ed. I'm glad we talked about it this morning again. It will be in the notes for this podcast. Sarah, our excellent producer, will make sure of it. Craig, I want to move on on May 13th. This is first, I want to ask you, what is the Bureau of Industry and Security? It is one of your favorite things. We've talked about it many times on this podcast. But this Bureau of Industry and security on May 13 announced the use of Huawei Ascend chips in anywhere. That announced that the use of Huawei Ascend ships anywhere in the world is likely a violation of US Export controls. First, tell us what is the BIS Bureau of Industry and Security, and why did it come to this conclusion?
- [10:24]
A
Yeah, so bis, just to refresh everybody's memory, this is the shop in the Department of Commerce that is responsible for licensing, administering those licenses and enforcing US Dual use export controls. So for everything related to AI chips, for everything related to stuff that's not just out and out weapons, the Department of Commerce, Bureau of Industry and Security, they're the folks who are in charge of this. And this move that they made, to announce that the use of Huawei Ascend chips, and this is a quote, anywhere in the world is a violation of US Export controls is a really interesting move. Now, on the one hand, let's just talk about, like the legal grounds for this. What they're saying is that because those Huawei Ascend chips were manufactured using US Semiconductor manufacturing equipment, then they are subject to, you know, the foreign Direct Product Rule and the other mechanisms whereby goods can have export controls applied on an extraterritorial basis. So what they're saying is, hey, you're making these chips in China, we admit that, but you're making them in China with US Equipment. And that means that our export controls governing the use of that equipment apply. And that means that you cannot be a customer for the purchase of these illicitly made chips. This is the very, very similar to the legal justification that was used to say that Huawei in 2020 cannot buy American chips when they're made at Taiwanese semiconductor manufacturing facilities. Because we say, hey, just because they're made in Taiwan does not mean that they're not subject to US Export control law, because when Taiwan makes those chips, they're using US Machines. So what America here is saying is that, hey, you know, companies in the Middle east, hey, companies in Africa, hey, companies in, you know, even China, if you were thinking about buy a lot of these Huawei Ascend chips, you need to understand that the force of US law, you know, which could be financial sanctions, could be export controls, etc. Etc. We view ourselves as having the legal justification to pursue penalties for using those chips. So for folks like Nvidia who were saying, hey, we think Huawei is a significant competitive threat to us and not only just in China, but potentially you know, in a lot of the global south, like Africa, for example, this is basically saying the force of the US Law is going to be used as a tool for preventing Huawei from getting that sort of global adoption.
- [12:51]
B
Greg, so what was Beijing's reaction to our measure?
- [12:56]
A
So, unsurprisingly, they're opposed to this, just as they're opposed to almost all uses of U.S. export control law and U.S. sanctions law. You know, their language in terms of what the Chinese Ministry of Commerce noted was, you know, relatively sedate, honestly, by Chinese standards. But they said, quote, the U.S. announcement is a typical non market and unilateral bullying practice. You know, the US Move is not conducive to the long term, mutually beneficial and sustainable cooperation and development of companies on both sides. This measure fully exposes its unilateralist protectionist nature. So they hate it. They want everybody to know they hate it. But they haven't said, you know, specifically because you're doing this, we're going to do Y. And I think that's kind of an interesting move. You know, one thing I just want to point out here is that logic that was used to say that the Huawei Ascend chips are illegal in principle would also apply to Huawei's phone chips, right, that they're using for manufacturing their smartphones. But there's no sort of additional, you know, guidance on how the United States is going to enforce this. So I think, you know, while the United States has the legal right, at least you know, under US Law to go into these things, there's still a bit of a strategic calculus on the part of the United States thinking about how it's negotiating with third party countries, how it's even negotiating with China, honestly, when it talks about what it's going to go after and what it has clearly communicated across both the Biden administration and the Trump administration is that AI chips are a priority and they want to keep those out of China, whether they're made by Huawei, whether they're made by Nvidia, whether they're made by anybody.
- [14:34]
B
Greg, I want to put this in a little bit of context, though. So these Ascend chips, how do they compare to our very best chips?
- [14:42]
A
Yeah, so the best chips that we have right now, the sort of industry standard are of course, these Nvidia chips. The last generation of Nvidia chip, the best one was the H100, you know, with the more current generation is the GB200. And so if you think about H100 as kind of the yardstick here, a GB200, which now is on the market, you can go buy that from Nvidia, is about 2 and a half times better than an H100A Huawei Ascend 910C, which is still pretty hard to get just because they haven't had the amount of advanced manufacturing capacity to package all these chips. You have to make the logic chip die. You have to get all the high bandwidth memory, and then you have to integrate that into a finished unit. And that integration into a finished unit sort of requires its own complex manufacturing process. So there's not a ton of 910Cs on the market right now, but Huawei is ferociously committed to getting a lot of these on the market in the not too distant future future. And according to Tom's Hardware, which has talked to folks who are customers of Huawei, the Ascend 910C seemingly delivers about 60% of the competing Nvidia H100's inference performance. So that means they're only 60% as good as the last generation of Nvidia chips. And then ultimately, Huawei is going to try and come out with another product called the 920. I think they're still going to have manufacturing rates challenges with that chip as well. But the basic point here, I think, is that they're going to continue to try and make technological progress. They're still going to face challenges mostly thrown up by these export controls. And this is an instance in which the United States is now trying to threaten the customers of Huawei, not just the suppliers of Huawei and Huawei itself. All right, one other thing I think that's worth say, okay, yeah. There was a piece that came out in the South China Morning post on May 16th that I think is really just worthwhile. So to refresh everybody's memory, you know, the CEO of Deep Seq, that Chinese AI model, he had said in multiple interviews that the greatest challenge his company faced was US Export controls and the challenges of getting, you know, enough of these chips. Now we have another quote from another senior Chinese executive at a different firm, Tencent, saying very similar things this year. Right, because those Deep Seat quotes were from July of 2024. But now we have a May 2025 in which Chi, who is the vice president at Tencent Cloud Computing Unit, he said, quote, the most severe problem is the limited resources of graphics cards and computing resources and that he expected the Titan trade controls to, quote, widen the gap regarding AI adoption between China and the US in the short term. So the point is, these export controls, as you know, Complicated. And as frustrating as all their implementation has been, Chinese AI executives still perceive this as the number one challenge that they're facing. And if you think about Huawei as the potential savior of all this, as we just talked about, Huawei comes with very serious performance limitations compared to what Nvidia offers. And Deepseek points out both where that China can still continue to make progress, which is in training AI models that have almost as good performance, but where the real pain point comes out. And the tencent executive talks about this is on the inference side, which is using those models. Right. OpenAI now serves ChatGPT to I think it's 500 million users every single week. And that requires an awful lot of computing power. That's specifically the flavor of computing power that we have denied China with these export controls. And I think it's totally plausible that in the absence of these export controls, when you think about some of the largest computing clusters that exist today, like the 200,000 H100 cluster that Elon Musk built in Memphis, I think it's totally plausible that in the absence of these export controls, there would be a 300,000 chip cluster in China already. And that's kind of what these executives are getting at when they talk about all the pain and suffering they've endured as a result of these export controls, for all of their failings.
- [18:59]
B
Lots of pain and suffering. Indeed. And let's talk about one more big story over the last week, and that, of course, is President Trump's visit to the Middle East. Last week, President Trump visited the Middle east in his first major overseas trip of the second term. And AI, of course, was at the top of the agenda. The administration announced major deals in both Saudi Arabia and UAE. I want to start first with the UAE deal. On May 15, the administration announced the US UAE AI acceleration partnership. What does this partnership entail?
- [19:33]
A
So, according to a fact sheet that was published by the Department of Commerce on May 15, one of the big things that's going to come out of this is a massive data center sort of campus that UAE companies are going to operate on, American companies are going to operate on, and it's going to be a 5 gigawatt center. So when we're talking gigawatts, right.
- [19:56]
B
You lots of energy to run it.
- [19:58]
A
Yes. I mean, you can convert that into Hoover dams. I want to say that each hoover dam is 1 gigawatt, if I'm not mistaken. So it's going to be a massive, massive amount of energy production. And all of that energy production is going to go into producing AI and cooling, you know, AI, because it generates an awful lot of waste heat that you have to dump when you do.
- [20:19]
B
All of this in the middle of the desert.
- [20:21]
A
In the middle of the desert. And you know, unfortunately, like, actually sometimes being in the desert can be helpful if it's super, super dry. But the UAE desert can be pretty humid too, sometimes. So it's a problems here to overcome. Yeah, I mean, it's going to be, it's going to be a challenge to cool all of this. But that is a challenge that can easily be overcome with money. Because if you just throw enough energy at it, if you just throw enough money at it, it's all going to work. And I think you know what US Companies like about this deal, and a lot of US companies were with Donald Trump on this trip. You know, Nvidia obviously loves it for the massive number of chips that they're going to sell to countries in the Middle East. But other companies, what they like is the lower construction cost of data centers. I mean, it can be, depending on the circumstances, anywhere between 10 and 50% cheaper to build this kind of stuff in the Middle east just because it's so much easier to get permits, so much easier to construct new energy generation capacity. So this is a big, big deal for everyone involved. And understandably it's been a bit controversial because now we're kind of seeing, when you talk about the scale of this deal, you know, when you, when you talk about a 5 gigawatt data center, that's equivalent to 2.5 million Nvidia B200 chips, recall that the biggest data center that's in operation today is the Colossus cluster, which I think is about 200,000 Nvidia H1 hundreds, or at least it was a couple months ago. So we're not only upgrading the generation of chip, we're going to build something that's 10 times bigger in terms of the aggregate number of chips. This is going to be massive. At least according to some analysis by, by Leonard Heim out at Rand. This is a bigger development just in the UAE. Just this one development is bigger than everything that OpenAI is talking about with Stargate, which was announced in, you know, Donald Trump's first week in office. So we're talking about.
- [22:17]
B
Really.
- [22:17]
A
Yes, we're talking about that big. Yeah, that was in the ballpark of 4.5 gigawatts. So this is going to be a really, really big development. According to Reuters reporting on May 14, the agreement with the UAE will, quote, allow it to import 500,000 of Nvidia's most advanced AI chips per year starting in 2025. So the draft deal calls for 20% of the chips, or 100,000 of them per year, to go to the UAE tech firm G42, while the rest would be split among U.S. companies with massive AI operations in the UAE, like Microsoft. So American companies still going to be a big part of this story, but unambiguously, you know, a massive win for the uae. And you kind of learn when, when the Biden administration was having negotiations with the UAE in all of this, one of the things that was not really out there in the open reporting was just what exactly is the UAE asking for? Right? Like, how much would they actually be willing to build and buy under unrestricted circumstances? And here we're kind of seeing this is just how gargantuan these deals could be. You know, we're talking about a pretty massive share of global computing capacity that in the future will likely be based in the Middle East.
- [23:35]
B
All right, let's talk about Saudi Arabia. If it's big in uae, how big is it in Saudi Arabia?
- [23:41]
A
Yeah. So the ua, the, the, the Saudi thing does not include anything comparable, at least not in the public literature yet, in terms of, like, one campus with a massive, massive build. Which doesn't mean that they might not announce that at some point in the future. I think the UAE was just further along in planning, you know, these kinds of things. So what the Saudi deal talks about is, number one, they're committing to Data Vault, you know, which is one of the Saudi Arabian companies that's a big part of the story, is theoretically going to invest $20 billion in AI data centers and energy infrastructure in the United States. And on the same, you know, on the same flip side, a bunch of Saudi companies, companies like Humane, which is a subsidiary of Saudi Arabia's public investment fund, are going to be buying, at least at first, 18,000 Blackwell Nvidia chips. And they're going to be doing big things, too. I mean, it's just. It's just tough to say all this. There's, there's so many different parts of this deal. AMD is a part of this deal. Google is a part of this deal. A bunch of other companies are a part of this deal, and Amazon Web Services is part of this deal. So there's going to be a big AI zone in Saudi Arabia. It's going to involve billions of dollars of investment. There's at least, you know, one discussion about 500 megawatts of computing Capacity, I think that number is very, very likely to go up in the not too dist. So, as I said, right, this is a significant shift in the global balance of computing capability towards the Middle East. And it's because there's a lot of money on the table to do all of that. I mean, these are, these are economies that are looking to diversify away from oil. They see AI as an incredibly attractive opportunity for that kind of diversification. And, you know, if you're Europe, you kind of got to just be looking at all of these deals with your jaw hanging off the floor because where are you in this story, right? Is anybody talking about building this kind of infrastructure in Europe? No. No.
- [25:35]
B
Well, so, you know, it reminds me of the Godfather in the Middle East. Here everybody wants to and is going to get to wet their beak. Maybe not so much.
- [25:44]
A
Yeah. And what's interesting is, you know, one of the biggest announcements from the French AI Action summit, not that long ago back in. Gosh, was it January or February, I forget. But one of the biggest.
- [25:56]
B
2025.
- [25:57]
A
Yes. One of the biggest announcements was Macron scored a deal with the UAE for them to invest in France. So the fact that, you know, there's just so much money available to invest is something that, you know, the United States and even European allies are just sort of thinking, how can we get a piece of this? And understandably, right, because the thing that was holding all of this up previously was national security concern turns, and whether or not we can have some confidence that the uae, Saudi Arabia, other countries are actually going to be aligned with us at the end of the story versus aligned with China at the end of the story, you know, that's kind of where the debate has been, is a national security concern. I think one other thing that just kind of has to be asked here, right? And, and if I was, you know, a senator looking to have a hearing on this topic, I would basically be going to OpenAI and say, hey, you were saying you were going to Invest, you know, $500 billion in Stargate. You know, that was based on an amount of compute that you wanted to build and an assumption of how much demand was going to be. Now that, you know that the UAE is going to be coming online with all of this capacity, are you still going to do Stargate? You know, what share of Stargate is going to be in the United States, how big is it going to be? Now this is, this gets to the risk that some, like Jimmy Goodrich also it ran, have raised, which is the idea of cannibalization Right. Which is how much of this investment that's going to happen in the uae, in Saudi Arabia, how much of that is in addition to investment that was going to happen in the United States, and how much of that is instead of investment that was going to happen in the United States?
- [27:39]
B
Greg, as always, great discussion. I know we're going to be continuing to talk about these issues, but I look forward to this every other week. Really appreciate your time.
- [27:47]
A
Hey, thanks, Andrew. Always good to talk to you. Thanks for listening to this week's episode of the AI Policy Podcast. If you like what you heard, there's an easy way for you to help us. Please give us a five star review on your favorite podcast platform and subscribe and tell your friends. It really helps spread the word. This podcast was produced by Sarah Baker, Isaac Goldston and Sadie McCullum. See you next time.