中国科技的走钢丝：电力、监管与 Angela Zhang 眼中的 AI 竞赛

Today we're continuing to build foundational context for understanding China as we enter the AI era. My guest is Angela Zhang, professor at the USC Gould School of Law and author of High Wire: How China Regulates Big Tech and Governs Its Economy. Having grown up in China, lived in Europe, and become a law school professor in the United States, Professor Zhang combines an insider view of China with a global outlook to create a unique perspective on Chinese technology governance and international technology competition.

After first discussing the difficulty of studying China, and how much better China collectively understands the United States compared with how well we in the United States understand China, we unpack the dynamic pyramid model that she developed in her book to explain China's regulatory approach, as well as the prevailing characteristics of hierarchy, volatility, and fragility that it exhibits. We then examine the stark power imbalance between government and business in China, where the state clearly has the upper hand, and compare and contrast the Chinese approach with its European and American counterparts.

We also discuss several notable moments in history, including the Chinese government's decision to exclude American tech companies, thus creating the market opportunity in which its own tech sector could develop; the famous 2020 crackdown on Big Tech, which involved canceled IPOs, major destruction of the market capitalization of Chinese tech firms, and the sidelining of Jack Ma, who had been one of China's leading technologists and global ambassadors up until that time; as well as the more recent apparent thaw between the Chinese Communist Party and leading Chinese tech companies. Some see that thaw as evidence that the government recognizes it needs the most vibrant tech sector possible in order to effectively compete with the United States going forward.

Along the way, as you'll hear, I ask a bunch of basic questions about the practicalities of Chinese society, including how concentrated power really is and how internet censorship works at the operational level. On the topic of AI specifically, I was also very interested to get Professor Zhang's takes on the light touch that the Chinese government has taken on AI so far; how China understands the U.S. export controls on chips and semiconductor manufacturing equipment; how to understand the DeepSeek moment; whether we should expect another crackdown coming soon; and even whether Xi Jinping himself might be, as some in the AI safety community have speculated, a doomer.

Please note that I am well aware that a number of the opinions Professor Zhang expresses in this conversation are highly controversial. To highlight just a couple: while I'm not sold on the strategic value of export controls, I am also not convinced that the DeepSeek moment proves that export controls won't matter going forward. On the contrary, I think there's a very good chance that, held in place as they are, they will. More specifically, with respect to the Chinese tech market, I know that Jeff Ding's work on technology diffusion suggests that this is not, in fact, a great strength but rather a weakness for China.

That said, because my time with Professor Zhang was limited, and because I'm interested in hearing different perspectives on such important questions, I chose not to push back on each of these topics in real time. I think it's worth taking her self-description as a neutral analyst seriously and genuinely trying to consider her analysis from a neutral point of view. Of course, this won't be the final word, and you can expect multiple strongly contrasting points of view on this feed coming up.

Professor Zhang, from the Gould School of Law at USC and author of High Wire, welcome to The Cognitive Revolution.

I think it's okay to pronounce it as “Gould” because it's a good school.

Yeah, okay, good. I like it. So, quick preface for this conversation: I don't know a lot about China, and I'm trying to catch up quickly because I think the assumptions that people are making about China, its technology development, what it's going to do in AI, and its attitudes toward the United States or the West are fundamental to a lot of the AI discourse happening in the United States right now. I have a sneaking suspicion that a lot of it is not super well founded.

I'm trying to get a better understanding myself and hopefully be able to have better conversations with people and contribute to a healthier discourse. You've written this book, High Wire, about how China manages, oversees, and regulates its technology sector, and exactly to what degree those different words properly apply. I would love to go through it, but maybe the first question I have to just get started is this: I don't think people have a very good intuition for this. What are your methods for doing this kind of research? What sources do you go to, how do you understand them, and what tools do you use? Is AI helpful?

I'm interested in that because I think you probably need a lot more people looking closely at China, and I think a lot of people might be interested. I have probably no idea where to start.

Yeah, that's a good question. It's increasingly more and more difficult for researchers to do field studies in China, but fortunately I was born and raised in China. I grew up there, I did my undergraduate work in China, and I also practiced law there for a bit. When I was working overseas, I also worked on China-related projects.

Over the years, I built up all my connections with Chinese industries, and I also worked for 7 years at the University of Hong Kong. Hong Kong has very close proximity to China, so all these things add up. It's incredibly helpful for me to accumulate firsthand experience understanding what's going on.

I have to say that, unfortunately, it is impossible to truly understand China without being there. To really understand things, you don't get it from an interview. It's more like you have to be absorbed in the environment and have that immersive experience to really understand what's going on.

I rely a lot on firsthand understanding and my knowledge and conversations with numerous contacts that I have. But I also do a lot of empirical studies, looking at court cases. I try everything—I also use secondary resources. I'm just kind of like an AI, in a sense: I do the best I can, gather all the sources, and try to answer the questions that I want to understand.

Yeah, I'm actually thinking of trying to arrange a trip to China because I've never been. I feel like it would just help my intuition a little bit, at least, and hopefully set a positive example for general engagement.

I'm sure it would be fun. There would be a lot of great food. I would definitely welcome you and all your podcast listeners to come and see China.

Now that I'm living in the United States, I think there's a lot of misconceptions about China. A lot of people worry about traveling to China, but I do think that being there in the first place is incredibly important. The vast majority of people have a really incredible experience after traveling there. It's not as scary as the media has portrayed it to be. It's just like another country, and in fact, people will be surprised by the technology when they arrive in China.

Several directions I want to go in and follow up there. How helpful—I mean, you don't need them, but do you have a sense of how helpful translation tools are in today's world, in various contexts? If I want to go read content on the Chinese internet and translate it with DeepSeek or whatever, how much does that actually help me read in a way that would foster real understanding, versus being too superficial?

I guess maybe the same question applies to street conversation. Can I take a real-time translator at this point and have meaningful conversations? What would you expect in that regard?

I'm not a big expert in that, but I would say that China is a leader in voice-recognition technology, and that kind of technology has been built into a lot of software and is readily available. When I was living in Hong Kong, my domestic helper only spoke English, and my mother-in-law spoke Chinese, but they communicated flawlessly with each other using a cellphone. So I don't think that's a problem at all.

Yeah, cool. What would you say are some of the technology surprises that the average American would be most taken aback by when they visit China?

I would say the first is high-speed trains. It's appalling that the United States—correct me if I'm wrong—still doesn't have high-speed trains. I think California has been trying to build this high-speed train for a long time, and now it has finished one-third of the project. Who knows when it will be completed?

Roughly a third, as I understand it.

Look, when you think about that, I don't think China is 10 feet tall. But in terms of execution capability, you can just look at the capacity to build high-speed trains. A project like this, between San Francisco and Los Angeles, would take China 1 or 2 years to complete, while it's taking the state ages, and we still don't know when it will be completed. There are numerous obstacles along the way.

I was talking to my student the other day about the U.S.-China tech divide. This is just one example that shows you where China is good: China speed, Chinese execution, and efficiency in doing things.

You mentioned that there are a lot of misconceptions. I wonder if the highest-level one, which it sounds like you're already alluding to, is that China is this invincible, monolithic, super-powerful entity. When you say China isn't 10 feet tall, what other misconceptions do you think are most prevalent or problematic in the United States with respect to China?

I'm also interested in the reverse. Does China understand the United States better than the United States understands China? If so, is that just because of our general history of cultural exports? What big misconceptions do you think China collectively has about the United States?

This is an excellent question. I don't want to paint a rosy picture of China. The Chinese economy is at rock bottom at the moment, and young people have a very pessimistic outlook on the future. My book, for the most part, is very critical of China's governance model.

That said, I want to be fair. Because I was born and grew up there, I understand that there are certain advantages to an authoritarian regime. It can be very efficient, it can get things done very quickly, and it places more emphasis on the interests of the majority.

Even for a lot of controversies that you've seen in Western jurisdictions about China, the vast majority of Chinese people actually like these things. Take surveillance technology as an example. In the West, when we talk about Chinese surveillance using all these cameras everywhere and facial recognition, it's viewed as incredibly negative. But if you go to China and ask an ordinary Chinese grandmother or grandfather whether they like it, they don't care. In fact, some people will tell you that they feel safer because there are cameras everywhere, and they don't need to worry about being robbed. The crime rate is extremely low.

I think the Chinese Communist Party knows how to get the endorsement of the majority of the people. That's why its popularity rating was pretty high until the mishandling of COVID-19.

As for the United States, does China have misconceptions about it? I do think it does. One of the biggest misconceptions is that the United States is a very democratic government, which this new administration clearly shows it is not.

For the most part, though, I think China understands the United States much better than the United States understands China. That's understandable because China's governance is very opaque. The bureaucratic regime is extremely efficient, but at the same time, the government deliberately keeps things very ambiguous and opaque. Also, because of internet censorship and information control, it's much more difficult to understand what's really going on inside China's bubble.

Every day when I read the news in China, I get to know almost everything going on in the United States. But the government always tells people bad news about other countries. Depending on where I live, the information I get is always biased one way or the other.

I was going to ask just briefly before turning to the framework of the book: when you said approval was high up until the COVID-19 mishandling, is that a felt sense, or are there data sources that you trust? Are there data sources that have actually shown a drop in satisfaction?

I'm not aware of that, but it's really interesting if that's out there.

Yeah, there is. There are a bunch of scholars who track public-opinion ratings of the Chinese government. They show that the Communist Party gains legitimacy, a lot of the time, through performance. It wasn't democratically elected, so the government needs to ensure economic prosperity and social stability for the vast majority of the people. Otherwise, there will be a lot of discontent. That distinguishes China from a lot of other authoritarian countries that tend to be extremely repressive.

By the way, during COVID, when Xi Jinping was able to get things under control—particularly from 2020 to late 2022—his approval rating was extremely high. If you talked to ordinary Chinese people, they felt great about the country because, when they looked around the world, everybody was living in chaos. China was able to achieve zero-COVID, was the first country to emerge from the COVID chaos, and was the first to resume production.

There was talk at that time about the rise of the East and the decline of the West. But things quickly turned sour after Omicron emerged, when the rest of the world started to open up while China was still under severe lockdown.

Yeah, that was a really interesting moment. Maybe we’ll come back to that in a little bit. Let’s go to the book.

The core framework of the book—and again, the subject is how the Chinese government oversees, manages, and regulates its technology industry—is what you call the dynamic pyramid model. You describe a couple of different dimensions. One is the traits that it has: hierarchy, volatility, and fragility. Then there are also the different actors that play in that system.

Let’s start with an overview of the hierarchy. We’ve got the top leaders, the regulators, the firms, and the platform participants, which I take to be individuals, small businesses, and anybody who’s doing stuff on these internet platforms. It strikes me that we have basically the same model here in some important sense, but maybe that provides a good jumping-off point to compare and contrast how the different actors are positioned differently, have different levels of power, or have different interests in China versus what we’re familiar with.

Great question. The purpose of having this model is to help our readers understand how China governs more generally. In fact, this model is not only helpful for us to predict how China regulates consumer tech businesses, but almost every major policy that you see coming out of the Chinese government. I also applied this framework to analyze how China had regulated COVID-19, the property crisis, and so on.

The book focuses mostly on the tech sector because it gives us a very good case study—a massive case study—to analyze this model in depth. When we talk about models, the first thing we need to look at is the players.

The hierarchy describes the structure of decision-making in China’s regulatory governance system. There are 4 major actors. There is the top leadership, which mostly refers to President Xi and the other members of the Politburo. Then we have the bureaucrats and regulators, who are in charge of the day-to-day implementation of the top leadership’s initiatives. These are the people who really do the hard work of turning out regulations and conducting enforcement.

Then we have the firms, which are the main subjects of the regulation. At the bottom are the platform participants who interact with the firms.

We obviously have the same 4 major players in other democratic settings, like the United States and the European Union. However, the way that they’re organized is entirely different. The major distinction is the power imbalance.

If we see it as a spectrum, China would be at one extreme, where the government clearly has the upper hand over businesses. When we talk about the pyramid, at the apex of the pyramid in China are the top leaders, and then come the regulators.

The regulators are subordinate to the top leadership. They’re not independent. They’re held accountable to the top, not necessarily to the bottom. They’re not so accountable to the public, because in China that’s the first and foremost priority for them.

I just had to search for how many members are on the Politburo. I got 24 back as my answer from Google. Are we literally talking about the top leadership—not a general elite class, or even a 500-member, U.S. Congress-sized group? Are we literally talking about just a couple of dozen people?

You can just think about it as 1 person: the president. He basically controls the top leadership. The regulators then report directly to that high level, so you feel like their direction is coming through very few links in a chain of command from the very top.

Yes, very few links in a chain of command from the very top.

That’s a good way to put it. The regulators are the government—the ministries, the departments, and the local governments. There’s a huge bureaucracy spread across 31 provincial entities, as well as numerous central ministries, departments, and agencies.

All these bureaucrats have the expertise and knowledge to implement the law. However, when they do so, they need to closely align their work with the top leadership’s initiatives. They always look upstairs to see what the shifting policy winds are, or what the policy signals are.

In China, that signal is incredibly important. In some ways, the signal is policy, and the policy is signal. When bureaucrats carry out their work, the first and foremost thing is to understand the policy signal, because the policy wind can shift all the time. That’s incredibly important.

When things happen in the United States, it’s often the case that this is not what the president wanted. We’ve got—although this is, of course, up for contest—independent agencies that are historically understood and meant to be insulated from the executive’s current mood. We’ve also got state-level regulators.

In general, it seems like you just cannot assume that there’s a president or anybody else you can point to. You can only go up so high before you find the executive decision-maker in that particular domain.

It sounds like the Chinese model involves a lot of pull, in the sense that there are local executives who have a very clear understanding that they don’t have an independent power base. They’re not elected or appointed by some other group that they’re more beholden to. It’s all the discretion, or the pleasure, of the top leadership, and their goal is to understand what that top leadership would want them to do.

They’re not necessarily always told super-explicitly, but they’re trying to map these high-level notions onto their domains in a way that they hope will ultimately be pleasing to the central leadership. Is that how I should understand it?

Yes. Everybody answers to 1 boss upstairs. Depending on what level they’re at, they may have other bosses, but they know who the big boss is.

That’s why Xi is also called “Xi Dada,” which means he’s big—he’s a really big boss in the country. But even at his level, he may not have the time, expertise, or capacity to handle everything. He has to delegate to government officials and bureaucrats to implement these policies.

The bureaucrats do have discretion, but at the same time they need to closely align what they do with the top leadership’s initiatives, or frame it in a way that aligns with those initiatives. That also works for their own interests, because we need to take into account the agencies’ own bureaucratic incentives when they carry out enforcement mandates.

Going back to what we were talking about, the key issue is the power imbalance. The key actors here are the businesses. I put them at a lower level of the hierarchy than the agencies because, unlike in other jurisdictions, where businesses go to war with agencies, businesses in China don’t do that very often.

Whether it’s a public company, a private company, a foreign multinational, or a Chinese company, they almost never directly confront the government. Think about Elon Musk. When he went to China, he behaved like a sheep. He was trying to please the Chinese government, unlike in the United States, where he’s sending a poop emoji to government agencies.

Can you imagine him ever daring to do that in China? He understands that in China you play by the Chinese rules and show respect to the government agencies. Businesses understand that the government clearly has the upper hand and is the most powerful actor here.

At the bottom are the people—the public. Think about the online merchants, delivery workers, and consumers who interact day to day with the platforms. Just like people in the West, they’re very weak in bargaining with a powerful platform, and there are often a lot of complaints.

But those voices tend to be muted in China because of censorship and information control, which make it very difficult for people to have their voices heard. That said, the Chinese government is very strategic. If it wants to have complaints heard, it can amplify those complaints.

At that point, you do hear a lot of public opinion, because that public opinion helps the agencies gain more legitimacy for their actions. The Chinese government is extremely media-savvy. It knows how to gain more public approval and support.

If I were to place Europe at the center, I’d say that the government is still powerful, but businesses have the opportunity to challenge it. In the European Union, where I used to live—I used to work in Brussels and also taught in London—U.S. tech businesses are investigated all the time by the European Commission.

They almost always challenge the Commission in Luxembourg, in the General Court or the Court of Justice. The courts have been very deferential in Europe, possibly because of their legal culture, but occasionally businesses do win. They continue to challenge and fight their cases, and from the standpoint of European officials, they don’t think that’s abnormal. They won’t hold a grudge because a business fights.

They respect that this is a rule-of-law system. Moving to the United States, I would see it as another extreme, the polar opposite of China, where businesses seem to have the upper hand.

Just think about the past 5 years. Despite a cascade of congressional hearings, interviews, and testimony with all these big tech companies—Meta and the others—has anything happened? Nothing. Congress hasn’t passed a single law to discipline the big tech firms.

The only law it adopted is the law against TikTok, to ban or divest TikTok. The United States still doesn’t have a federal data privacy law. The only law that restricts data flows is targeted at TikTok. The Biden administration signed an executive order to restrict the transfer of sensitive data to countries of concern, including China.

The only platform regulation we have in the United States is almost exclusively targeted at companies with a Chinese identity. With respect to its own big tech giants, the United States has done almost nothing.

You can ask whether that’s regulatory capture or a scandal. When you see all the big tech CEOs surrounding the president on Inauguration Day, rubbing shoulders with the president—Trump is surrounded by all these big tech moguls these days, and even assisted by Elon Musk—you worry about another form of crony capitalism in the United States.

Going back for 1 second to what you mentioned about the government knowing how to turn up complaints selectively when it wants to, could you describe a little bit more how the staffing, or actual management, of the censorship apparatus works?

I would imagine this is something the tech companies had to build. When you were saying that complaints about the tech platforms themselves are, by default, suppressed, that made sense to me because there’s already this censorship regime, and they’re implementing whatever censorship guidelines they’re getting.

Maybe they just salt in a little of their own brand protection as they do it. But when you said the government could actually tinker with that in a more granular way, I was wondering what the shape of that is—who has access, who staffs what, and whether that’s even known or clear.

This isn’t a very transparent system. Obviously, both the big tech firms and the Chinese government invest hugely in content moderation. In the West, companies like Facebook and Meta also employ an army of people to do content moderation, although they’re getting rid of them these days.

In China, this is a very important part of the compliance obligations of all the big tech platforms that operate social media businesses. One of the first and foremost priorities of government regulation is to ensure that the content is politically aligned with the government, or with the socialist values of the Chinese Communist Party.

The platforms have to receive instructions from the Cyberspace Administration of China about what kind of content is allowed or not allowed. They receive very timely updates as circumstances change.

I can give you a concrete example of how it works in practice. In the book, I talk about China’s largest ride-hailing platform, Didi, which is the Uber of China.

When Didi wanted to go public in the United States, it encountered opposition from within the government. One of the agencies we just talked about, the Cyberspace Administration of China, asked Didi to postpone its IPO filing to ensure that it complied with the data security regulations.

Didi was a huge ride-hailing firm. It had a lot of sensitive geographical data and sensitive personal data from the Chinese population. From the standpoint of the CAC, the online-listing process, including the submission of information to U.S. regulators, could potentially pose a national security concern.

However, Didi pressed ahead with its IPO filing in the United States, disregarding the CAC’s warning. Obviously, the CAC was very upset. Two days after Didi’s listing in the United States, the CAC conducted a cybersecurity review of the firm.

That caused the firm’s stock to tumble more than 10%, wiping out a lot of money from its market capitalization. That was really a moment that made international investors upset and led them to conclude that China was uninvestable. This kind of regulation seemed totally unpredictable and arbitrary.

When the CAC conducted that investigation, it said it wanted to conduct a cybersecurity review of the firm. But that kind of review, according to the law, is usually about procurement practices—whether the firm is buying equipment that makes it insecure and poses a national security threat.

It had never been used in a situation where a firm was going public and merely submitting documents to the SEC. It was a dubious act by the government. However, in order to enhance the legitimacy of the action, you started to hear a lot of public opinion in the Chinese media saying that Didi was a traitor for turning over its sensitive data to the U.S. government.

Didi’s president, Liu Qing, is a former banker and a very successful businesswoman. She was also labeled a traitor who was divulging Chinese secrets and selling them to the United States.

You can see the Chinese government drum up these nationalistic sentiments in the country. There were also posts circulating that showed Didi had collected a lot of sensitive government data because it could reveal the traffic patterns around different Chinese ministries—what the traffic was like at certain times, which ministries were busy, and when they were busy.

When these media posts circulated, they created the impression among the Chinese public that this company had the ability to reveal information that could pose a national security threat, and therefore the government’s intervention and cybersecurity review were a good thing.

Of course, the general public had no idea what a cybersecurity review meant or whether there was a clear legal basis for it. You can see that in this example.

Another example is when Jack Ma’s Ant Group IPO was suspended. There was a flood of media reports on the Chinese internet accusing him of being an extremely greedy businessman who was trying to start up a huge fintech company that would rip off the Chinese people.

There were a lot of reports about the dishonest practices of Ant Group, the fintech company that was supposed to go public and was allegedly trying to exploit consumers. The Chinese media machinery is also part of the Chinese bureaucracy. The different departments—the Propaganda Department, the media, and the regulatory authorities—all work together in tandem to increase the legitimacy of government actions.

That is quite interesting, and there are a lot of different directions we could go. Could we spend just a minute on the rise of China’s tech giants?

I don’t know a lot about this, and one question I’m particularly interested in is to what degree these companies were winners from a relatively free and open market-competition process, versus at what stage they became favorites of the party or were otherwise designated winners.

In the book, I discuss how Chinese government censorship and the Great Firewall worked as trade barriers to prevent foreign, particularly U.S., social media businesses from coming to operate in China.

Companies like Baidu, which operates China’s largest search engine, faced much less competition because the only other U.S.-operated search engine is Bing, which now has only 1% of the market share. Google exited years ago.

Now let’s consider the counterfactual that you proposed. Had China not imposed these kinds of restraints on the Chinese internet sector, would China still have had these domestic national champions?

I think the domestic firms still would have won. Just think about what’s happening nowadays. Why is TikTok conquering the world? Why are the majority of the 10 most-downloaded apps in the U.S. app stores Chinese firms?

The AI firms are also leading and doing well. For the most part, I think China’s capability of adoption and diffusion is much better than that of its U.S. peers. China is not good at going from 0 to 1. That’s why the major breakthroughs in technology almost never occur in China.

I don’t see DeepSeek as a technological breakthrough, either. DeepSeek wasn’t the company that discovered the transformer model, or that figured out how to use neural networks to model an AI architecture. That is what I see as a true scientific breakthrough.

When we talk about Joseph Schumpeter’s innovation cycle, we have innovation, and then we have adoption and diffusion. China is more advanced at the second and third stages. It’s extremely good at adoption and particularly good at diffusion.

DeepSeek is very good at diffusion. Diffusion is about how we make things more efficient, how we lower costs, and how we make things more accessible to consumers. In those areas, China can go from 1 to 10 extremely well.

That’s why, with all the consumer tech businesses I’m talking about, I have no doubt that, a few years from now, unless the United States bans all Chinese apps—which would be difficult—we will see more and more Chinese firms flooding into the U.S. market.

You already see Temu. I ordered from Temu the other day. Prices were 30% cheaper than on Amazon if you can wait, and people can deliver very quickly these days. Temu also has local warehouses.

Then there’s Shein, the fast-fashion company, which is coming into the U.S. market—not to mention DeepSeek and other companies.

That’s really helpful context. It sounds like the tech giants we know—Tencent, Alibaba, and so on—started as organic companies and succeeded on their market merits.

There was this exclusion of U.S. companies, which created a less competitive environment than they would otherwise have faced. But in terms of why it was Alibaba or Tencent, versus a different Chinese company competing for that niche, these leaders won their markets fairly and squarely, so to speak. Is that right?

Yes. It’s an extremely competitive market. Before everybody knew DeepSeek, DeepSeek had already been known in China for more than a year. I was just so surprised that the Western community had no interest in China and totally underestimated DeepSeek.

Around the middle of 2024, DeepSeek engaged in a price war in the AI community. It was the company that significantly slashed prices and started this price war, in a way, and then everybody had to follow, including the leading AI companies.

That’s why, in China, DeepSeek was called the Pinduoduo of AI. Pinduoduo is the parent company of Temu and is very good at huge discounts and deep-discount sales.

I was surprised that the Western community didn’t know about the firm’s existence until 8 or 9 months later. All of a sudden, everybody woke up and said, “This firm is doing this.”

It shows that there’s a misconception we were talking about. The misconception in the United States is polarized. One school of thought thinks China is 10 feet tall, that China is ahead of us, and that China is going to beat us, so we have to control it.

The other school thinks China is 3 feet tall and totally underestimates China’s innovation capabilities. I’m just trying to be neutral and present what China is.

We haven’t officially touched on volatility and fragility, but maybe you could illustrate those with a little bit of history from the 2020 moment, when you alluded to Jack Ma being put in his place, through to the present.

I know the story in the roughest terms. He said a couple of critical things, then all of a sudden he wasn’t heard from very much for a while. The IPO was canceled, and vast amounts of wealth seemingly came off the books.

My sense is that the company still carried on without much disruption to its operations. It was just clear that there was a limit to your upside. That can affect the stock price even if today’s operations are still running pretty smoothly.

That’s my sense of what happened, but why did it happen? What were they really trying to accomplish? Did they accomplish their goals, or did it turn out differently from what the decision-makers had hoped?

I also have a vague sense, from reading the news, that very recently—perhaps since the DeepSeek moment—there’s been a sense that China might need to bring its technology companies back a little bit. What exactly does that mean? What are they bringing them back from? What are they trying to accomplish?

By volatility, I mean that the regulatory pendulum swing is very dramatic. The Chinese government is either very lax with tech regulation or becomes extremely intense and strict with it.

For the most part, from the early 2010s until late 2020, until the IPO debacle, Chinese consumer tech businesses enjoyed about a decade-long honeymoon with the Chinese government. The government took a very light-touch approach with these firms.

From the IPO until early 2022, there was an 18-month period of very intense crackdown. After that, we reverted back to normal. If you think about how China regulates AI these days, I also think it’s really lax at the moment.

Exactly what happened is very complicated. During the first decade-long honeymoon period, the government wasn’t doing much. It was actually doing very little, and that also created a lot of problems. Had these problems been addressed earlier, they would not have become what they are today.

This led to an extremely concentrated tech sector. We’re mostly talking about 2 major players: Alibaba and Tencent. There are also a few other companies in China, whereas in the United States you have 7 major tech companies. The U.S. market is still a little less concentrated in that sense.

That concentration created a lot of issues and a lot of regulatory problems. The last straw was the IPO of Ant Group, China’s largest fintech company and, at the time, the world’s largest fintech company.

It was going to be the largest IPO in history in late 2020, and the IPO was suspended at the 11th hour. That stunned investors. That’s the kind of volatility I was talking about.

Why was the IPO called off at the last minute? Obviously, Jack Ma’s speech was very helpful in contributing to the debacle, but fundamentally, the issue was the regulatory tension between China’s financial regulators and Ant Group.

Alibaba’s empire mostly started from e-commerce. But when Jack Ma spun off Alipay, China’s payment app, to start Ant Group as a financial business, he began to cross the line into the financial industry. That touched a sensitive nerve of the Chinese Communist Party, which is obsessed with control.

When you’re selling clothes, cabbages, toys, or video games, it doesn’t pose a threat to social stability or political stability. But as soon as you venture into finance, and as soon as you grow to such an enormous scale, that puts the Chinese top leadership on high alert.

The last thing they want is for a single private entrepreneur—just imagine Jack Ma, who used to control more than 50% of the voting rights in Ant Group—to amass that much fortune and control the largest financial institution in China.

Not to mention, what happens if the firm collapses? Its ownership structure was also problematic because there was a dominant controlling shareholder.

We probably won’t have time to go into all the details, but I have a lot of sympathy for why the Chinese government wanted to intervene in the affairs of this firm. To be fair, Ant Group provided all sorts of financial services, but it argued that it wasn’t a financial institution.

It changed its name from Ant Financial to Ant Group to avoid being identified as a financial firm. From the IPO perspective, it didn’t even want to be seen as a fintech business. It called itself a tech-fin business.

Eventually, it received a very high IPO valuation of $320 billion, which was higher than JPMorgan’s valuation at the time. That stunned Chinese regulators. They thought, “This firm was spun off from Alibaba less than 6 years ago, and now it has grown to be the largest financial institution in the world. That looks like a huge bubble in the making.”

The government thought it needed to intervene before it was too late, because the firm now posed systemic risk to the Chinese financial industry. I think this was the major reason the government realized that these firms weren’t just selling clothes or video games. They were venturing into the most important parts of Chinese business.

The government needed to reduce their influence and make sure they listened to the government. It needed to remind them who the big boss was.

Should we understand that as Xi personally? When something like this happens, should I be thinking that Xi gathered the cabinet, got reports from different directions, and ultimately made a personal decision that this was not going forward?

I don’t think Jack Ma was put under house arrest. In fact, I saw Jack Ma the other day. We were at an event together, and everything was fine.

So what’s the new turn, then? They put him in his place for a while. I hear the argument that this is now a national public good, and it’s simply not the Chinese style to let random entrepreneurs control something so important. You’re no exception.

Are we swinging back because of competitive reasons? How do you account for the most recent apparent vibe shift?

I don’t think it’s swinging back in the sense that the government is encouraging this kind of financial service. I don’t think that’s happening.

From Xi’s standpoint, it remains clear what the priority industries are that can really help China become strong and prosper. You can see that from the entrepreneurs he invited to the symposium held a couple of weeks ago and who were seated at Xi’s table.

These are entrepreneurs who work on AI. Even DeepSeek’s founder, whom nobody knew about a year ago, was at the table. There were also a couple of consumer tech businesses, but for the most part they were firms in AI, electric vehicles, clean energy, and robotics.

The most important guest was the boss of Huawei. These are the people the Chinese government really relies on. These are the most important priority strategic industries.

The government understands that it should let companies continue to sell video games and operate consumer businesses as usual, because China still needs employment. Young people desperately need jobs, and the unemployment rate is too low.

But those businesses aren’t going to have a transformative impact on the Chinese economy. China has gone through a very painful structural transformation over the past few years under Xi. Xi is very ambitious in trying to transform China from the old, dirty economy into a new economy.

You have to say that when you look back at what China has been trying to do, it has achieved most of what it wanted to do. Even with the controversial Made in China 2025 project, Bloomberg ran a report the other day saying that China had achieved most of its goals.

China has now become a leader in most of the technologies for which it set goals earlier. China also went through very painful property reforms because it didn’t want its economy to be built on a property bubble.

Now you can see the economy has shifted toward high-end manufacturing and clean-energy exports. We have to wait and see what the trajectory of the Chinese economy looks like in a few years, but the government’s emphasis has clearly shifted from tech in general to hard tech—particularly hardcore technology that can give China a competitive edge.

That is clearly the main focus of the Chinese government, also because of pressure from the United States. China views this as an urgent signal that it has to become independent and self-sufficient.

How about AI? We’re maybe in the early innings of the great AI competition, which I’m not particularly enthusiastic about, because I feel like the technology is unwieldy enough without putting it at the center of a geopolitical struggle. Nevertheless, that’s clearly the way the winds are blowing at the moment.

One thing that has surprised me is that you mentioned AI is still in its honeymoon phase. I’ve heard a little bit about this before. There are some standards and expectations, but it’s not as though the government is conducting pre-launch reviews of all these products.

My sense is that the government is more or less letting people do things, go to market, and see what happens. If there’s a problem, you’re probably going to get smacked for it, but you’re given freedom to operate until something goes wrong.

Still, I was surprised by how free and almost whimsical DeepSeek’s releases seemed to be. I’m not sure whether anybody knows this, but did DeepSeek run those releases by the top leadership?

Was it something that would rise to the level of senior decision-makers—“This company is planning to launch, or wants approval to launch, an open-source frontier model that’s on the level of the best stuff out there”? I could imagine a world where Xi would definitely be looking at that and signing off on it.

I could also imagine a world where it didn’t take the leadership totally by surprise, but they heard about it on Twitter with the rest of us. What do you think the level of alertness to the AI moment is right now in the Chinese government?

I definitely think DeepSeek is quite independent in turning out its own products. Obviously, any business, whether foreign or domestic, operates under the authority of the Chinese government. You need to play by the Chinese rules.

However, the true disruptive technologies coming out of China are all the result of private-business independence and innovation. I almost have to be too harsh and say that anything the Chinese government touches isn’t doing that well, to be honest.

Whenever a government tries to allocate resources, it leads to corruption, a lot of inefficiencies, and losses. We’ve seen and observed that in the semiconductor industry. There have been a lot of corruption scandals, and so far China isn’t achieving that much in semiconductor capabilities.

That’s why China needs to rely on Huawei. Huawei is not a state-owned firm; it’s a private business. The true innovative capabilities still come from private entrepreneurs—from the animal spirits of business leaders in China.

DeepSeek is a maverick in the sense that it’s not even a big tech firm. It’s a kind of random quantitative hedge fund which, because of regulatory restrictions, was limited in trading certain financial products. It devoted more resources to training AI models and became an overnight success.

It’s a very distinctive success story. At the same time, you see all the major big tech companies—Alibaba, Tencent, Baidu, and many others—as well as a lot of AI startups, making tremendous progress on AI efficiency. It’s just that DeepSeek is the best. Everybody else is not far behind.

I don’t think the Chinese government had anything to do with that. I do think the Chinese government is very supportive of AI development, but if you ask me what the first and foremost priority is, I would still say it’s hardware and semiconductors. That is really China’s bottleneck at the moment.

Is that how we should understand the recent announcement of this mega-funding program? I’ve seen it called $100 billion-plus, and even a quarter-trillion dollars in purchasing-power-parity terms.

Is that primarily a hardware buildout, or is there anything else we should infer from it? We’re hearing here in the United States—not from super-credible sources just yet—that the Overton window now extends to a Manhattan Project for superintelligence.

Part of the reason people are willing to entertain that is the idea that China is going to do its own version of a Manhattan Project for superintelligence. Does this large investment tell us anything about how centralized the effort might be?

We also hear a lot about civil-military fusion in China, and it sounds like we may have media-government fusion from the earlier generation of social media platforms. Where is this money going, and what does it imply?

To what degree do you expect that, even if all these developments came up independently, there are government monitors at the DeepSeek offices now, looking at the developments on an ongoing basis?

A lot of people seem to think that’s what’s happening. If that’s laughable, please disabuse me of the notion. I’m genuinely ignorant about what’s going on in reality.

I do think DeepSeek needs to communicate closely with the Chinese government, given that it’s such a major player in the game right now. But for the most part, I think its intentions are benign, because it’s open-source and providing a public good for the rest of the world.

From that standpoint, it also makes sense why the company wants to be open-source. It makes the company more credible and helps China become an industry standard-setter in a way.

China has gained a reputation as the second-most-important AI superpower. Because DeepSeek is widely accessible around the world, China can potentially shape the standards of AI technology with its open-source AI models.

The Chinese government is playing a long game. On the one hand, it understands that China is not best positioned to create or run the technology. On the other hand, it wants to guide the technology in a way that maximizes national interests.

Why do I think the U.S. semiconductor project, particularly all these export controls, might fail? Why might China experience another DeepSeek moment with a semiconductor breakthrough?

The U.S. export controls have created alignment between the private sector and the government’s interests. In Washington, because of the emergence of DeepSeek, there are a lot of industry leaders and policy analysts advocating for very strict export controls on AI chips to China for national security reasons.

Dario Amodei was one of the most vocal proponents of that idea. He even wrote an essay saying that we should double down on export controls and develop a smarter export-control strategy.

I think this school of thought disregards the unintended consequences—not just of DeepSeek’s innovation, but also of the alignment of interests between the private sector and the government.

Once the private sector knows it can’t access advanced foreign technology, it has to innovate. It has to work with the government and think about solutions. You’re incentivizing strategic alignment between the private and public sectors.

That is one of the unintended consequences of all these U.S. restrictions. I don’t think it’s a remote possibility that China could achieve a semiconductor breakthrough.

A lot of industry people say that China is a decade behind in semiconductor technology. I’m more optimistic about China’s capabilities. This is a country with 1.4 billion people and a lot of highly capable engineers.

Many of those engineers came back from the United States because of the U.S. anti-espionage investigation. The United States lost a lot of talent to China in the past few years, partly because of Donald Trump’s old China Initiative program.

When you have people, you have the capacity to innovate. That’s exactly what Ren Zhengfei, the Huawei boss, said during that symposium with Xi Jinping. The most important thing is that we have people. Even if we lose everything else, as long as we have people, we can innovate.

I think that a few years from now we may experience another DeepSeek moment, and people will say, “Holy cow, how did the Chinese crack this really difficult problem?” At that point, I think the U.S.-China tech war will end, because there will be no other important obstacle along the way.

When you say that the tech war will end, what does that mean? Does it mean we enter a new period of friendship?

One question I often ask, and that I think gets glossed over too quickly in many of these conversations, is why the 2 countries are at odds right now. I know there’s a long list of grievances that both countries will cite, but in your view, is there a fundamental, interest-oriented analysis that leads to unavoidable tension or conflict?

Or would you describe this more as a malfunction of leadership on both sides that has created conflict, perhaps for domestic political reasons, without a more fundamental reason for it to exist in the long term?

Nathan, you should invite me back for another episode. This is exactly what I’m writing about right now. It’s called “America’s Legal Gambit Against China.”

To give a preview and answer your question, just think about the deterioration of the U.S.-China relationship. When did it start? It started in 2016, after Trump was elected. Particularly after 2018, approval of China among the American public dropped dramatically and continued to decrease over time.

We’re now in the worst period of this relationship. How did that happen? The outcome is driven by both demand and supply.

From the demand side, 2 things are happening. The first is the trade imbalance that The Economist was talking about. After the United States liberalized its trade relationship with China, there was an influx of Chinese imports into the United States.

That took away a lot of U.S. jobs, caused manufacturing decline, and generated a lot of discontent in American society. That’s the part of the story we’ve all heard. Trump’s major campaign promise was to bring jobs back to the United States. China was taking and stealing U.S. jobs.

The other factor is China’s rise. China’s geopolitical rise disrupted the existing great-power order and challenged U.S. hegemony. Those are the demand-side reasons why the relationship turned sour.

What I’m writing about is the supply side. We need to think not only about why the United States needs to get tougher on China, but also about what institutions are shaping the policy response to those demands.

I think Trump played a pivotal role in contributing to the deterioration of the U.S.-China relationship. When we talk about foreign relations, the president and the executive branch have tremendous discretion.

Even though the United States is a democratic country, foreign policy is different. Power is concentrated in a way that makes the United States look like China. Power is highly concentrated in the executive branch.

When Trump came into power, one of his major cards was that he was against China. Being tough on China was his brand. Trump broke a lot of presidential norms by taking action against China.

He was the first president to invoke Section 301 to impose tariffs on China, accusing China of forced technology transfer. That broke norms, because Section 301 was a violation of WTO commitments and had rarely been invoked.

No president before Trump had used tariffs in that way to deal with these issues. Trump started the game. He was the first to restrict capital flows into China and to toughen sanctions and restrictions on Chinese firms like Huawei and ZTE.

He was the first to propose banning TikTok and restricting information flows. Trump was also the first to start the China Initiative program, which led to a massive anti-espionage investigation into thousands of ethnic Chinese scientists working in the United States.

Many of those people left and returned to China. Trump broke with all these norms. As he did so, just like the Chinese government, he also drummed up demand. He talked very toughly about China, so public opinion became very negative.

He would say that every Chinese student who came to live in the United States was a spy. He called COVID-19 the “China virus.” He was hitting all the negative notes about China.

By responding to the demand, he also increased it. He inflated the demand. Even after Trump left office, Joe Biden inherited almost all of his policies and doubled down in some areas, particularly on export controls.

That’s the part most people don’t talk about. It’s important to make an institutional comparison with China. When I finished the article, I was thinking that, in its efforts to contain China, the United States is becoming more like China in many ways.

You see the concentration of power, and you see volatility in its practices. TikTok is a good example. Now Trump is TikTok’s biggest savior, by the way.

You also see fragility in the regulatory outcomes. The emergence of DeepSeek is a clear example of how U.S. export controls generated a huge backlash and wiped out over $1 trillion in market capitalization from U.S. tech firms.

What has that achieved? People in Washington are still arguing for tougher export controls. There are rumors today that Trump is going to appoint somebody to the Bureau of Industry and Security, and that export controls are going to get tougher.

Once these institutions are set up, they’re there, and they have long-lasting impacts.

How do you think the Chinese government understands the United States’ goals with respect to China right now?

Do they see this in terms of increasing volatility and think, “Xi is probably going to be gone in a couple of years, so maybe we can wait this guy out. We’ll develop our own domestic chip industry, but hopefully, other than that, we can get back to more reasonable leadership”?

Or do you think this is starting to take on its own hard-to-control momentum, where China may really be starting to see the United States as a long-term enemy in a way that will be hard to undo or get over?

I think China’s biggest enemy is not the United States. It’s China messing itself up.

The title of my book is High Wire: How China Regulates Big Tech and Governs Its Economy, but you could also say it’s about how China has messed itself up. The crackdowns wiped trillions of dollars of market capitalization from China’s big tech firms, hardly injected any competition into the tech sector, and discouraged new entrants.

Private investors left the Chinese market because they believed China was uninvestable. That totally sabotaged the entrepreneurial spirit. It wasn’t the United States.

Yes, there are tariffs, export controls, and other restrictions creating obstacles for Chinese entrepreneurs, but those aren’t the biggest worries about operating in China. The biggest obstacle is the lack of a truly great vision and a good rule-of-law system that gives people the confidence to invest and prosper.

At the end of the day, it comes down to good governance. The same story applies to the United States. The biggest enemy of the United States is not China. It’s its own democracy and the constitutional crisis people are talking about at the moment.

The United States is messing itself up. China isn’t outcompeting the United States. Even if you look at the DeepSeek story, it’s a story about geopolitics only in the sense that DeepSeek created this technology out of necessity.

You also have to ask yourself why none of the big U.S. tech companies came up with a similarly efficient solution to training AI models when they had abundant resources and were the clear technology leaders. They knew very well what China had done, because this is the internet age. For more than a year, everybody in Silicon Valley knew that Chinese firms were very good at improving the efficiency of AI models.

None of the big tech firms showed interest in doing that or following similar strategies. People say that this is a good thing for Silicon Valley. I don’t think so. At the end of the day, it was a deliberate business choice.

I just wrote an article in The Independent and Project Syndicate called “How DeepSeek Upended U.S. Tech Dominance.” It argues that DeepSeek succeeded because the U.S. tech industry is too concentrated.

Think about the AI industry. At the infrastructure level, it’s controlled by a few monopolies. At the very top, you have Nvidia, which is a virtual monopoly, and then you have the 4 big cloud providers.

Those companies also invest downstream. In a way, they’re vertically integrated. The money is all upstream. That’s where the profit is, and that’s where the profit margin is.

Nvidia’s operating profit margin is over 60%. Downstream, the AI firms—including OpenAI—are not profitable. They’re losing money. The monopoly profits are upstream, not downstream.

Those companies aren’t interested in competing downstream. They want to increase demand for AI chips and computing resources. They continue to project a rosy picture of demand and scarcity for computing resources. That’s why they have to keep scaling things up.

That’s the business strategy. That’s how Silicon Valley works. When we think, “How did DeepSeek succeed?” we also need to reflect on ourselves and ask why we failed.

We didn’t fail because we lacked innovative capability. We clearly have that capability. We failed because the concentration problem in the United States blinded us to the direction of innovation.

How do you think about your own position in this crazy world?

Having grown up in China, my understanding is that the Chinese state may still take an interest in what you’re saying even while you live abroad. I don’t know whether you feel that way, but I’ve certainly heard people say that this is something the Chinese state does.

You’re clearly willing to be critical of both governments. Do you feel totally comfortable doing that? What is your strategy or goal in the work you’re doing and in all this public communication?

Think about it this way: If I see myself as an entrepreneur, I’m producing a product, which is my scholarship, and I need to have my brand.

My brand is that I’m very neutral. I grew up in China and worked extensively overseas in Europe. I know Europe pretty well, and I know the United States. I think I can offer a unique perspective by being neutral and honest.

When people read my work, they’ll see how critical I am of the Chinese government. I’m equally critical of what I’m writing about with respect to the U.S. government. I think we need this kind of niche product at the moment.

Who can really offer fair and honest opinions about what’s going on? I think there’s a demand for that. Thanks to your podcast, my product is getting distributed and my ideas are being disseminated, which is very important—something scholars don’t usually do.

What about this question, which may be too far outside what’s knowable to have a real opinion on?

In AI safety circles, there’s the notion that Xi might be a doomer. What exactly that would mean is unclear. In American AI safety circles, it means taking seriously the possibility that AI could lead to human extinction or the collapse of society.

As a result, even if you’re historically very pro-technology and libertarian in your politics, you’re inclined to have a special level of caution and support for regulation for this particular wave of technology, especially frontier AI.

Is there any way to read what Xi might really be thinking with respect to these extreme-risk scenarios? Is there any reason to think he can or can’t be engaged or reasoned with on that level?

I have no idea what President Xi is thinking. But from what I observe, both the United States and China are pressing ahead with AI development at full speed.

If you’re doing that, then if you’re rational, you shouldn’t be an AI doomer. You should be an optimist who thinks AI is going to do more good than harm.

Unfortunately, I think this trajectory is set. We’ve already seen that AI can have dangerous capabilities alongside its great promise.

I wrote an article called “The Promise and Perils of China’s Regulation of Artificial Intelligence.” Taking this very lax approach could sow the seeds for potential disasters and catastrophes down the road.

That’s particularly true when we think about China’s dynamic-pyramid model of governance. Problems don’t receive attention from the very top until it’s too late. That’s a recurring theme in Chinese regulatory governance.

Think about COVID-19 pandemic control, tech regulation, and the property crisis. One followed another. Will AI be the next disaster? We’ll have to wait and see.

At the same time, the United States is following China’s lead by relaxing regulation and going to another extreme. That’s truly worrying, because the United States is still the clear leader in this technology.

Certainly at the frontier, I agree with that.

As a last question, how much value do you see in Track 2 dialogues, particularly those motivated by the idea that this AI could become really powerful one day? We might need to step back from the brink of all-out competition and have some kind of arms control or coordination.

How much value is there in Track 2 work? Do you have any advice for people who want to build these kinds of bridges or relationships that might one day be useful in that scenario?

I think what the world needs at the moment is a disaster. Let’s hope the disaster isn’t so disastrous that it kills us all, but that it’s relatively contained.

We need to show the world’s leaders the dangerous capabilities of AI and the potential misuse of AI. AI could turn into a huge accident that threatens social and political stability. Then our leaders and politicians around the world will wake up and take proper action.

In my opinion, the biggest obstacle to U.S.-China AI cooperation isn’t geopolitics. It’s the fact that we haven’t seen a disaster. We don’t understand the capabilities of AI, and we don’t know how to regulate it.

This is not just my view. Geoffrey Hinton has said that we don’t understand how these things work because we don’t understand how the brain works. This is a more powerful brain than ours, and we simply feed it computing power, data, and algorithms. It works, but we don’t know how it works or what capabilities will emerge.

There’s a lot of uncertainty and unknowns when it comes to regulating AI. That’s why regulators aren’t serious. During the Cold War, the Soviet Union and the United States signed the Nuclear Non-Proliferation Treaty even at the brink of nuclear war, because they knew nuclear war could kill us.

We don’t know what will happen with AI.

That’s a more optimistic take than some, but I honestly don’t disagree with the assessment.

In the meantime, a lot of people feel like they want to do something useful now—something that might help in a crisis later. I want to give you 1 more chance to advise me.

If I want to do something that would marginally improve the situation, what could I do that might build social capital, communication bandwidth, or something else that could be useful if that disaster hypothetically hits?

There’s a proposal to create an independent agency modeled after the IPCC, the Intergovernmental Panel on Climate Change. The idea would be to collect intelligence about AI and potential disasters.

If we had an independent international organization staffed with good scientists from around the world, it could share intelligence about AI capabilities. We already have a lot of evidence.

Those scientists could then update their governments about what’s really going on and what the potential consequences are. At least we could make sure we’re on the same page, understand the risks together, and take precautions before it’s too late.

If there’s even a 1% or 0.1% chance that AI could get out of control and the consequences could be disastrous, is it worth investing a little bit of resources in preparing for that? Clearly it is. That’s the whole point of buying insurance.

To abandon AI regulation, walk away, and overlook the problem until it’s too late is not a wise regulatory approach.

I’m with you on that for sure. I’ll look more into those proposals.

I think this has been a fascinating 90 minutes, so I really appreciate it. Do you have any closing thoughts, or do you want to share where people can find you online before we break?

Harvard Law professor William Alford said that I have the best-made personal website of any legal scholar in the United States. If you search for my name, Angela Zhang, you’ll find me online.

He said it was the most elaborate personal website of any law professor in the country.

We’ll have to check it out. For now, thank you again for doing this, Professor Angela Zhang.

Thank you for having me. It was a great pleasure.