The AI race is in full swing. China is chasing digital sovereignty, but there’s a glaring hole in that plan…it doesn’t have the advanced manufacturing capabilities to achieve it. Here’s the real kicker: nobody does.
China might be a bit more flexible with its ethics and data manipulation, but that means nothing without the advanced chipmaking equipment. China can produce less efficient chips, but has incredibly low yields on leading-edge chips. To produce the highest-level chips, there are roughly 65 countries involved in a complex supply chain.
Which means that not even the U.S. could go out and do this on its own (unless it underwent a 20+ year tech ecosystem rebuild). America’s real advantage in the AI race comes from the broad coalition of allied countries all working together to produce high-end semiconductors.
Hey everybody. Peter Zeihan here, coming to you from the top of little Castella, taking advantage of a break in the wind to answer a question from the Patreon crowd. Oh, by the way, that’s an ash mountain over there. And all of that is the Turner Ranch. Ted Turner, I guess, is a state, is a single largest landowner in the country.
And that’s where he used to party. Anyway, the question is, is how is China’s artificial intelligence program coming along? They clearly don’t have any of the ethical restrictions that the United States has. And is that given them an edge? Is this something we need to worry about? Number one, I would argue that if you’re comparing the tech bro culture of Silicon Valley to China and the idea that one of them is morally superior, you need to do a little research.
Silicon Valley, and especially the folks that are involved in AI, really don’t give a rat’s ass about your personal privacy and are perfectly willing to pillage your data in any possible form. In fact. Hey, I got a little visitor here. Hey, buddy. Yep. Never mind. He’s gone. Anyway, point being that if you are on Facebook or what is it now? Meta, whatever, any sort of social media whatsoever, all of your data is being mined actively in a way that the Chinese honestly sometimes have trouble keeping up with. And while the Chinese certainly do it with a little bit more flair, the AI Kings in the United States, the data kings in the United States have no problem getting all of the information that they need from you already.
So the info influx really is not much of an issue. There used to be a moment when things had to be manually entered in, and the Chinese had an advantage of that for the numbers, but now AI has reached the point that it can self process. So again, that’s gone away too. So the structural issues really are not relevant.
It really comes down to hardware. Now the Chinese are under export restrictions, so they cannot import the hardware that is necessary to make high end semiconductors, specifically the extreme ultraviolet machines that come from the Dutch company ASML. And without those, there’s no way they can make anything that’s really under seven nanometers reliably. That means that the reliant on importing chips and their export restrictions on anything coming out of the United States or Taiwan or Japan or Ireland, or any of the countries that make high end semiconductors, that limits them to approximately a 4 to 5 nanometer ceiling. Now, it’s not that those are bad chips at six nanometers.
I mean, they’re fine, but if you want to do AI, it’s all about the number of computations you can crunch per chip and the power that takes per computation. And right now, in the rest of the world, the high end chips are in the 2 to 4 nanometer range, and we are now starting to get ships that were designed with coolant topography in mind. let me spell that out. What that means for an operational if you have a one gigawatt hyperscale data center, you probably have about a half a million GPUs on top of that. Lots of CPUs, lots of Dram, lots of Nand, different sorts of processing capacity, lots and lots of hardware.
But the 500 GPUs or the core of it because large models. And when we’re talking about AI today, that’s what most people are discussing. Use those chips almost exclusively. And the better they are, the more computations, the more energy efficient per computation. The Chinese don’t have access to those 2 to 4 nanometer chips. Sometimes they can import some five and especially six nanometer chips.
But they have made the decision that for sovereign control reasons, they don’t want to depend upon anyone’s imported chips. If they can make anything that’s even partially functional themselves really cool in. But there’s a middle ground. There’s a gray area because while the Chinese can’t import those high end chips in that high end equipment, they can still import a lot of equipment and a lot of gear. And they do that. They imported deep ultraviolet, DUV machines that are still pretty good and that are the basis for a lot more mature nodes.
And they have used these to successfully make nodes as small as seven nanometers. Now I have to emphasize this. These chips kind of suck. They you generate about one third the computations per chip compared to what comes out of Asme and Taiwan. They generate about twice as much power, probably closer to three times as much power. Once you consider that the rest of the engineering of Chinese data centers, is that just that much worse?
So you’re looking at overall one ninth efficiency compared to the leading edge stuff. On top of that, the yields for these seven nanometer chips is awful. It’s typically less than 30%. And as a rule, anywhere else in the world, any yield rate below 95% is something that wouldn’t be considered because it’s just such a massive waste of resources.
China does the math differently, and they’re concerned with digital sovereignty and keeping them with the United States. But what it means is they need to build nine data centers for everyone in the United States in order to get an equal number of compute, and that is assuming the hardware can run the software that is required to do things like large language models.
Bottom line, the Chinese are doing everything they can. They are sparing no horses, and they’re not even remotely close. They have tried over and over and over and over again for the last four years, to try to push past that seven nanometer limit, and they have succeeded in making a few emphasis on the word few chips, the meter level using technology. But it’s highly inefficient. And you’re talking about yields that are below 5%, nowhere near enough to supply a single data center, much less a raft of them. But before I laugh in their face, I have to say that the technology of AI is changing day by day, week by week, one by month, and the hardware is always two years behind.
So I don’t want to rule it out there. Certainly motivated, if that’s the right phrase. But when it comes to the domestic manufacturing of the hardware that is required to do this, no, they’re just not even in the same game. Now, before anyone goes and toots America’s horn, the United States can’t make these chips either. Neither can Japan or Korea or Taiwan takes a team effort with 65 countries and 100,000 manufacturing supply chain steps and almost 10,000 companies.
It really does take everyone. And if anyone is actually trying for real digital sovereignty, they’re going to fail unless they rebuild the entire tech ecosystem. And that is not something you do in less than 20 years. So are the Chinese going to fail at this by the rules of the game, as I understand them? Yeah. Are the Americans going to fail at this by my understanding of the rules of the game?
Yeah, but the United States is part of a coalition of like minded countries that implement these technologies together. That’s how we do AI. And if the US decides to go it alone, it’ll be at best in China’s position.
