American politicians like to praise their actions as historic, but in early October America’s security policy actually recorded a first: The Biden administration imposed the strictest and most far-reaching export bans on computer chips against China to date. They concern state-of-the-art computer chips, which are needed to make the calculations required for artificial intelligence (AI) technologies. Washington’s goal is primarily to prevent the People’s Republic from further integrating AI into its military and thus becoming a military superpower.
The restrictions not only prohibit leading chip design companies from selling their most powerful products to China, they also affect companies that produce the software for chip design or the machinery to make chips – or even just machine components. In addition, Americans and green card holders are only allowed to work in the Chinese semiconductor industry to a very limited extent. China, so the goal, is to be shielded from technological progress in computer chips from now on. In doing so, the U.S. is hoping strongly for the support of its allies in Europe, Japan and other countries in East Asia.
The chip industry is one of the sectors in which China is enormously dependent on foreign countries. Every year, the country spends more money on the import of chips than on oil. The U.S., in turn, is in the driver’s seat. Because even though chip manufacturing has become an international process, the country is the world market leader in crucial areas of the computer chip production chain.
In addition to the export ban to China, the U.S. is investing heavily in the domestic chip industry. Congress passed the Chips and Science Act this summer, committing the government to spend nearly $53 billion on semiconductor research, development, manufacturing and skills training over the next few years. One of America’s experts on computer chips and AI is Greg Allen of the think tank Center for Strategic and International Studies (CSIS). He has studied the Biden administration’s new strategy in depth.
Mr. Allen, how effective will the combination of sanctions and subsidies be in curtailing China’s ambitions to become an AI superpower?
China is already an AI superpower. The Chinese put out great research and turn it into products that people want to buy. And they are extremely strong at incorporating AI technologies into their military and surveillance systems. The American export controls do not attempt to cripple China’s AI industry in the present, but to dramatically transform China’s prospects for the future. Without access to the leading edge AI chips, or to training large AI models in hyperscale data centers, slowly but surely, China’s AI research is going to become less and less relevant to the state of the art as it exists in the United States, Europe and Japan, for example.
On that score, these export controls are likely to be extremely effective. China is very good at the software side of AI. But all AI software has to run on semiconductor hardware somewhere. And the restrictions ensure that over the longer term, China is just going to be unable to access the state of the art hardware.
Obviously, the enforcement of the export controls plays a crucial role too. But they are designed to work, to materially affect the future prospects, especially of China’s military adoption of AI.
How are these export controls different from previous ones the U.S. has implemented for technology?
This regulation is really about targeting the future of the industry. These chips have two really important characteristics. One is the processing power: How many sorts of calculations per second can they do? The second is interconnect speed: How fast they can collaborate with other chips as part of a large scale supercomputer or data center? The new American regulations say you cannot sell chips above a certain performance threshold to China anymore – the one that represents the current state of the art for training large AI models
Historically, when the United States had technology performance- based export controls, they would revise them upwards every year, every two years. But in this policy, the Department of Commerce has written that these performance thresholds are never going to be revised upwards. They’re freezing the state of AI in the year 2022 for China. Soon, this will be very outdated technology.
How remarkable is it that an American administration is taking such steps to actively stop large segments of the Chinese technology industry from advancing?
When historians look back on the year 2022, I think two dates are going to echo in history. The first is February 24th when Russia invaded Ukraine, and the second is October 7th when the Biden administration announced these new export controls. This new policy is a significant reversal of 25 years of trade and technology policy towards China, in two ways. For the first time, chips above that performance threshold are prohibited from being exported to all of China — and not just to the military sector anymore. That’s a big, big change.
Secondly, historically the United States would sell these types of technologies to China, but only older models. Specifically with regards to semiconductor manufacturing equipment, they sold the equipment that would allow China to advance, but with a certain technological distance. Now, there is active degradation of Chinese technological capabilities taking place. One example is the semiconductor manufacturing facilities in China that can currently mass produce chips at the 14 nanometer process node. To those factories, you now can’t even sell the old stuff anymore. And every semiconductor manufacturing facility on Earth is dependent upon U.S. semiconductor manufacturing equipment. It´s not just a one- time dependency – you need spare parts, software updates, etc. This policy is designed to put those advanced facilities out of business.
Previously we tried to slow the pace of the advance. This is the first time we’re actively trying to reverse technological progress. That, again, is a big, big change.
You said earlier that China is already an AI superpower. Should the American government have acted sooner to stop that?
It took a really long time for the Biden administration to get to this point because we’re talking about a reversal of 25 years of trade technology policy. There is a very high threshold for what the justification had to be to do this type of policy: decades of the Chinese government ratcheting up provocations such as state-sponsored corporate espionage, forced technology transfer, human rights atrocities, the development of AI-enabled autonomous weapons, etc.
A lot needed to have happened to justify that kind of reversal. I think this is absolutely justified. Yes, it would have been more effective if it had happened five or ten years ago. But in order to be ultimately effective, other countries will to need to get on board and pass similar types of restrictions. And those other countries wanted to see the United States try diplomacy first and try diplomacy themselves first. We’ve come to this policy change at the end of trying so many other things that didn’t work.
However, the U.S. moved forwards unilaterally and is now trying to bring allies on board retroactively. Isn’t this approach really risky, given what’s at stake?
Time was of utmost priority to the administration in this case. There was some prior consultation with allies too. We asked them to tell us right now if they viewed it as unacceptable. The United States has risked a tremendous amount of its future national security on the idea that our allies are going to come along with us in this regard. But we think it is urgently important to do this – and that many of our allies share this view.
How do you expect China to react or retaliate?
There’s a few options for public responses. First, they just filed a complaint with the WTO, but I don’ t think China has a real hope of winning that. Second, the Chinese government recently announced an investment fund of the equivalent of $143 billion for the domestic semiconductor industry. It sounds like an astonishing amount of money, but they have had a $100 billion investment fund before, and all the people in charge of it have been arrested for corruption. Third, China is, of course, trying to divide the United States from its allies. China is trying to persuade Japan, Netherlands and other countries leading in chip technology to fill the void left by the United States. They offer to pay them a lot of money to figure out how to replace the U.S. equipment that is currently in their facilities.
But there are also non-public, that is, illegal responses that I think are absolutely part of the Chinese toolbox. One is export control evasions — trying to route purchases through third countries, trying to disguise your identity as a purchaser. The other is intellectual property theft and technology espionage. The Chinese government has a long history of deploying their intelligence services in support of their commercial industries or technological espionage. I have to expect that’s a priority for the Chinese government right now.
Looking at the subsidies, not only the U.S., but the E.U., Japan and many other nations are pumping money into their domestic chip production. Is there a risk of going from shortage to oversupply?
This is absolutely a risk. The semiconductor industry has a history as a boom and bust industry. When there’s a shortage and you invest to increase production capacity, it takes a long time for that capacity to come online, so eventually, there’s an overproduction. That’s why I’m so strongly in favor of allies coming together and figuring out a way to coordinate these investments. There’s so many different parts of the semiconductor market. It will be a disaster if all the allies simultaneously invest in the same market niches, right? There’s certainly an appetite in the United States to find a way to coordinate these investments.
What does it mean for Taiwan’s national security that other countries are now investing so heavily in their own products and in their own fabrication infrastructure?
It shouldn’t be overstated what the Taiwanese semiconductor giants are preparing to do abroad versus what they expect to continue doing in Taiwan. Even in this future where Taiwan Semiconductor Manufacturing Company is diversifying worldwide and opening some facilities in the United States and potentially Japan and Europe as well, a massive share of the advanced chip production is still taking place in Taiwan. Maybe it’s no longer 100% or 90%, but it is very possible it will continue to be 50%. TSMC has more than 64,000 employees in Taiwan, and the facilities abroad will ultimately employ maybe 2,000 to 3,000 people.
All these countries are trying to figure out a way to be not so entirely dependent upon Taiwan. But the most likely future is still a significant degree of dependance upon Taiwan.
So the fact that the U.S. is investing so heavily in its own chip fabrication capacity doesn’t diminish its commitment towards Taiwan’s protection?
How many times does Joe Biden have to say that we would defend Taiwan for people to believe it? I think he’s made himself very clear. If China invades Taiwan, the United States is going to be there.
Latest NZZ in English articles
The Neue Zürcher Zeitung (or NZZ for short) is one of the preeminent news sources in the German-speaking world, with a tradition of independent, high-quality journalism reaching back over 240 years.
With its curated selection of English-language articles on Swiss and international news, politics, business, technology and society, NZZ in English provides readers with a fresh and independent perspective. Learn more about the project here.