STEVE INSKEEP, HOST:
During a visit to China, our team met a Chinese entrepreneur. His company is designing a computer chip that can be used for artificial intelligence. His intended market is companies that make self-driving cars. Over glasses of tea, this entrepreneur explained his business but said he did not want to be named or recorded. He doesn't want to risk attention from the United States, which keeps an entities list of companies it sanctions. When we got home to America, we put questions about this entrepreneur to Gregory C. Allen, who studies the competition between the U.S. and China.
He's not currently on the entities list, but he worries about it all the time. Should he be worried about it?
GREGORY C ALLEN: I think so. For autonomous driving, the types of computer chips that you put in those cars, these are really beefy processors. And what the United States government is, of course, wondering is, are the processors at the heart of those autonomous driving vehicles also going to be the same processors that are at the heart of a Chinese military autonomous tank in the not too distant future, if they're not already?
INSKEEP: This is one anecdote in the ever more complicated relationship between the U.S. and China. The U.S. says it wants to continue normal business, and at the same time, it wants to cut off technology it says China's military or spy agencies could use. Yesterday we heard from a Chinese company that is on the entities list. Today we hear about a different U.S. tactic. Since October 7, 2022, President Biden's administration has broadly banned the sale of the highest quality computer chips to all of China. China has responded by vowing to make its own. Gregory C. Allen has an American perspective on the chip race.
ALLEN: It's a little bit difficult to tell the sort of nuanced story here. But what the United States recognized was that when it comes to AI chips, China was using chips that it was buying ostensibly for commercial purposes, but those chips were being rerouted for military end uses, including nuclear weapons modeling and all manner of different kinds of banned activities. And the United States was essentially saying to its companies, we are in a race with China when it comes to the adoption of artificial intelligence technologies for military superiority. And this is a real risk to U.S. national security. And if we're in that race, we do not want your companies giving new shoes and Gatorade to the Chinese racer in this story. You should not be helping them improve their AI systems.
INSKEEP: Isn't China going to get there on its own given that it has lots of engineers, lots of money, and at this point, lots of experience?
ALLEN: This is a really complicated thing to actually answer. You have to come down to, what are the different parts of the AI value chain? There's the software, which companies like OpenAI do. Then there's chip design, which companies like Nvidia do, Nvidia which designs the AI chips that are the most popular in the world. And then there's companies like TSMC of Taiwan which manufacture the chips. And then there's the companies that make the equipment that goes in the factories that manufacture those chips.
So the October 7 policy was really interesting because it sought to choke off China's access to AI by choking off access to all parts of the chip value chain. So when you say, can't China just do it on their own because they have a lot of talented engineers? That's not just a question of replacing Nvidia, designing new chips. They have to also replace TSMC. They have to also replace ASML and Applied Materials, the companies that make these extraordinarily complicated machines. I mean, for your mental model, think James Webb Space Telescope hard. These are really, really, really complex machines that go inside these semiconductor fabs.
INSKEEP: So do you think this means China will never get there or that it will take them a long time to get there?
ALLEN: I think to replicate in particular extreme ultraviolet lithography machines - I mean, these machines cost $300 million a pop. And the research and development on it went on for 20-plus years. And unlike other types of lithography and other types of semiconductor manufacturing equipment which is already present in China, because it was sold during the period where it was legally allowed to be sold, there has never been an EUV machine inside China. So it's a lot harder to reverse engineer something that you've never gotten your hands on. And most of China's engineering ecosystem when it comes to semiconductor manufacturing equipment is just a straight reverse engineering story. They find a Western machine, they take it apart, they try and figure out, can they make a domestic alternative to every single individual part? It's really hard to do that if you've never gotten your hands on one of these machines.
INSKEEP: The United States exploded nuclear weapons in the 1940s, had a monopoly on the technology. Within five years, the Soviet Union had a bomb.
ALLEN: Well, they also had a spy inside the Manhattan Project, which helped a lot.
INSKEEP: (Laughter).
ALLEN: It was a really valuable spy. And I would just say, you know, nuclear weapons were incredibly complex and difficult in the 1940s, right? But today, regimes as crazy and backward and inefficient as North Korea are capable of doing that.
INSKEEP: Sure.
ALLEN: Extreme ultraviolet lithography is incredibly, astonishingly hard, even for the most technologically advanced organizations in the entire world. I'm not saying China can never get there. I just don't want anyone to be under the impression that this is going to be effortless or quick.
INSKEEP: Gregory C. Allen is at the Center for Strategic and International Studies, a policy research group here in Washington. He's one of many Americans tracking the tech competition between the U.S. and China. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
