Dwarkesh Hirani

Today, I'm talking with Victor Shi, who is the director of the 21st Century China Center at UC San Diego.

China is obviously the most important economic and geopolitical issue over time, and doubly so if you believe what I believe about AI.

I was especially keen to talk to you because you have deep expertise not only in Chinese elite politics, but also its economic system, its fiscal banking policies.

We'll get into all of that before we get into the AI topics.

Is China actually a more decentralized system than America?

If you look at government spending by provincial governments versus the national government in China, it's like 85% local versus provincial, 15% national.

In the U.S., it's actually, if you look at state and local, it's 50%.

Federal government is 50%.

When you think of an authoritarian system, you often think of it as very top-down.

The center controls everything.

But if you look at these numbers, it just seems like it's quite decentralized.

Or is that the wrong way to look at the numbers?

Let's start at the very top.

If you look at the members, first I want to understand the personnel.

If you look at the members of the Politburo and you look at their past, a lot of them will be provincial leaders and then some local thing before that.

But in many cases, they'll have stints, as you said, as they were the head of this chemical engineering SOE or they studied.

Xi Jinping himself has a PhD in chemical engineering, right?

Something.

Some kind of.

Well, but I... Yeah, that's what you want to ask about.

If you just look through the list, a lot of them are like PhD economists, PhD industrial engineer, et cetera, et cetera.

I mean, some of them that you look, it's like, oh, they have a PhD in Marxist theory or something.

Which that seems fake.

But...

Are they actual technocrats?

If you look at their degrees, they seem quite impressive.

Or is this just fake, made-up credentialism?

If there's a PhD economist in the Politburo, should I be impressed and be like, oh, they probably really understand economics?

Or should I just think this is kind of a fake degree?

Yeah.

I still feel confused about how to model the competence of the party because on the one hand, they are

super educated in STEM and are selected at least partly on merit, et cetera.

On the other hand, we just see them make... A lot of zero COVID was just genuinely stupid, scrubbing down airport runways and so forth.

And it just lasted for too long.

And why was nobody steering the ship then?

So...

Well, no.

What's the way to put together this picture?

But if you're an economist, okay, so here's a question.

Like, traditionally, the way I understand it, if you study economics in at least an American university, you'll learn about, like...

supply-demand, here's why certain regulations can be bad, here's why tariffs are bad.

Just the basic 101 teaches you that.

So if a lot of the members at the highest levels of the Communist Party have this basic understanding, then what's the reason that they make mistakes that many economists say they shouldn't do?

Like, why...

Obviously, people talk about rebalancing a lot.

Why don't they think that increasing consumption is a good idea?

At some point, we'll talk about leading small groups and the way they work.

But the picture you get from the fact that

Xi is monitoring what all these different Politburo members are doing.

He's noticing whether they're dragging their feet.

He's leading all these small teams.

You'll explain this in a second, how these work.

But they organize different leaders who are the heads of the ministries or commissions that are relevant to a specific project working.

And I guess every week they meet and they hammer out who's a bottleneck, what do we need to do to get a project moving.

So Xi has ratcheted up the amount of such meetings that – and he's – the picture you get is somebody who is micromanaging details across all these policy areas.

If you read Stalin biographies, there's a big thing.

Like Stalin actually was a person like this, obviously not to great effect.

But he was interested in micromanaging every single detail across even the theater productions in the country to like steel production, everything.

Is that who Xi is?

So the study sessions with the Politburo, these meetings, etc.

Are they like, are they substantive?

Are they real?

Is he like actually learning the intricacies of like, here's the, you know, here's how zero COVID has impacted this province and here's why we should whatever, whatever.

Or are they just faux?

Because if you read the announcements from the party or departments, they're just like so vapid, like typical communist speak.

How real are these meetings?

Yeah, and it's stupendous to think about.

We understand that you can tell somebody how big is China, the population, et cetera, besides the economy.

I think visiting China gives you a tangible sense of just how big.

You have provinces, each of which have as many people as a normal nation state.

Provinces which have hundreds of millions of people, 40 million people, you know, so forth.

And then one person...

will say something at the end of a meeting that determines a policy for basically like everyone.

Yeah.

For like the population of like North and South America or something.

Does he say things at the end of meetings, during his speeches, et cetera, which indicate that he actually has a good understanding of these different policy areas?

Because obviously the official speeches are very vague and –

It's always something like we will make sure that we are robustly pursuing Chinese growth while hewing to the tenets of Mao's thought.

So never anything specific or interesting.

But do we have an indication that behind closed doors he's – Well, you don't think there's anything specific or interesting.

I'm interested in what it reveals about him personally.

Does he display any analytical ability, any sort of like empirical understanding of different policies, etc.

?

Yeah.

The advantage of the American system, of course, is that at least in theory, it can survive a leader who is not up to the snuff.

publicly available data is running out.

So major AI labs like Meta, Google DeepMind, and OpenAI all partner with Scale to push the boundaries of what's possible.

Through Scale's data foundry, major labs get access to high-quality data to fuel post-training, including advanced reasoning capabilities.

Scale's research team, SEAL, is creating the foundations for integrating advanced AI into society through practical AI safety frameworks and public leaderboards around safety and alignment.

Their latest leaderboards include Humanities Last Exam, Enigma Eval, Multi-Challenge, and Vista, which test a range of capabilities from expert-level reasoning to multimodal puzzle solving to performance on multi-churn conversations.

Scale also just released Scale Evaluation, which helps diagnose model limitations.

Leading frontier model developers rely on Scale Evaluation to improve the reasoning capabilities of their best models.

If you're an AI researcher or engineer and you want to learn more about how Scale's data foundry and research lab can help you go beyond the current frontier of capabilities, go to scale.com slash thwarkash.

Okay, let's go back to Ding Zhaozhang.

How do you pronounce his name?

Ding Zhaozhang.

Okay, sorry.

So Politburo has, what, 25 members?

20, 25?

Yes, 25 members.

Standing Committee has seven members.

This is the key group inside the Politburo.

And this person is one of those seven.

He runs the Center for Science... Sorry, the...

The Central Science and Technology Commission, the reason I'm especially interested in him is because any large-scale AI effort that China would launch would be under his purview.

Do we understand what he wants, who he is, what his relationship with Xi is, what he would do if AGI is around the corner, etc.

?

What are the different organizational milestones we should be watching for to understand how they're thinking about AI when things have escalated?

So to add more color to this question, what I expect to happen, let's say next year, in terms of raw AI capabilities,

we might have computer use agents.

So a thing which is not just a chatbot but can actually do real work for you.

You put it loose on your computer, it'll go through email and compile your taxes, et cetera.

And then dot, dot, dot.

In five years, I expect it to be able to do all white collar work or most white collar work, which is like 40% of the economy potentially.

Eventually, we'll have full AGI, even robotics and so forth.

As this is happening, how should we be tracking how the...

party is thinking about what is happening how seriously they take it what they should do about it should be I think last week or a couple weeks ago there was a Paula Bureau study session on AI but I don't know if there was like AGI AI or what kind of AI they were discussing AI yeah

Should we be looking for a leading group on AGI in particular?

Should we be trying to just read more speeches?

What should we be paying attention to?

But so far, after the deep sea became a big deal, the immediate reaction, it seems, in China was all the different big tech companies, WeChat, Tencent, everybody was encouraged to adopt it.

And they did adopt it as fast as possible.

Xi himself met with Lin Wenfeng and did a whole meeting where all the industrial heads were there and discussed.

And Xi said we have to accelerate, I don't know, whatever, technology in China, et cetera.

So it seems like so far the response has been the greater the capabilities of AI, the more they're excited about it, the more they think this is the beginning of Chinese greatness.

But you think that changes at some point?

Suppose it turns out you were saying, look, they can help High Flyer source GPUs and et cetera.

What is the mechanism by which, suppose it turns out that for High Flyer to continue making, DeepSeq to continue making progress, they need all the GPUs that Huawei can produce and they need a whole bunch of other things.

They need energy.

They need data centers.

What would it actually look like for the government to say every single Huawei GPU has to go to DeepSeek?

They're going to get access to all this spare land to build data centers.

Who would coordinate that?

Would that be the leading cybersecurity group?

Would that be somebody else?

Delta and Road View too.

Yeah, exactly.

Does it have any end implication to the progression of AI?

Who actually ends up controlling these leading groups?

Whose purview it ends up under?

Or is it just basically a detail?

But in terms of...

what happens with AI in China, it doesn't have that big an implication.

That ironically sounds like a great tourist site.

Not because it's real Manhattan, but because it's big.

We went to this, in Amisham, we went to this huge, huge Buddhist temple.

By huge, I just mean like there would be this structure, the shrine, and you'd go through it and you'd realize behind it was an even bigger shrine that was hiding from view.

And then you'd go through that one, there'd be an even bigger one.

This happened like five times.

If you were to drive through this, it would take you like,

But it's new, right?

Not historical, right?

Oh, yeah, it's new.

And then I asked the head... By the way, there was nobody else there.

It was me, three white guys, and nobody else.

And I asked the head monk, how did you guys finance this?

And he says, we've got a lot of supporters.

A lot of donations.

But it's still been compatible with them innovating and having leading companies in even content, right?

Like Red Note and TikTok and so forth, not to mention WeChat.

So if AI ends up – and AI might be even more robust to these sensitive content because you can just teach the AI to not say certain kinds of things.

As somebody who is trying to observe what the party is doing, have you found that using LLMs has been helpful?

And if so, which model gives you the best insight into what the CCP is thinking?

Do you think it's just because they have more Chinese language data and so it's just better at understanding the party communications?

Look, coffee isn't the reason that you're going to stay at the Ritz, but it could be the reason that you leave.

Because when you offer a truly premium product, your customers require excellence in all the parts, not just the core.

So, the Ritz works with a specialty roaster to ensure that they can deliver great coffee.

That's what WorkOS does for your product, but for enterprise features, like single sign-on, access controls, and audit logs.

These features are not the reason that people will love your product.

But if you don't have them at all, you simply won't get a seat at the table with your biggest, most important customers.

Companies like OpenAI, Anthropic, Vercel, and Perplexity all rely on WorkOS to serve their enterprise customers.

In fact, NVIDIA might be the only other vendor that all of these AI companies share.

And that's because not only are these features critical, they require excellence.

WorkOS makes it super easy to plug in these features into your product and get back to delivering the premium experience that your customers expect.

If you want to learn more, go to WorkOS.com and tell them that I sent you.

All right.

Back to Victor.

Okay.

And then this is going to sound like a naive question, but we're talking about how the party in the abstract wants to maintain control over every aspect of Chinese life and Chinese economy.

But of course, the party is made up of individuals.

And many of these individuals, Xi himself, went through periods where the party had extraordinary control over people's life during the Cultural Revolution and so forth.

And they personally and their families suffered as a result of this.

Yeah.

and they're educated people.

Many of them had been in industry, right?

These are not naive people.

So maybe I'm just projecting too much of my Western bias here, but what is the reason that they think it's so crucial that, uh,

You know, the banks are run by the party and the AI is run by the party and nothing escapes the party state.

What is the main difference between him and Stalin in the sort of political maneuvering sense?

Because all of this sounds very similar to—

And if you read the Stalin biography, so many of the – in the early part –

Half the people in the Politburo are later going to end up in the gulags, and then many of the people in the Politburo later have been to the gulag.

I wonder if there's a Trotsky-analogous person in the story, somebody who was a flamboyant speechmaker.

Yeah, yeah.

I think that would be really interesting, actually.

Let's move on to another topic I know you study deeply, which is the –

local government debt situation.

Sure.

What is the newest there?

What is your sense of the situation now?

That's a top of 60% that is owed by?

Yeah.

And the high level way to understand what this debt was for, the debt in Western countries and other OECD countries is often because you owe money basically to pensioners.

In this case, it was just that they built for

Right.

For startups, is the fact that they can get cheap credit from banks more relevant?

Is the fact that they can get loans from provincial funds more important?

There's also these central government funds, the big fund and so forth.

Which is the bigger piece here?

And the crucial thing here is that the banking sector is totally controlled by the state.

So you're not getting a true rate of return based on what the productive value of investment is.

It's literally like 1%.

Yes.

The deposit rate is 1%.

Yeah.

Which is, I mean, it's like an insanely low return.

Yeah.

There's a couple of mixed up questions I want to ask here, basically.

So you said, well, if there were no capital controls, these people might want to take their money and go invest in Silicon Valley.

Very basic development economics would show you that if a country is less developed, they'll have higher rates of return.

So people would want to invest.

The Silicon Valley would want to invest in China.

Why would getting rid of capital controls make capital go the other way?

And this goes into other questions of why is it the case that the Chinese stock market has performed so badly, even though the economy has grown a lot?

And then why is it the case that...

they have a current account surplus and they're basically accumulating T-bills.

The pattern you should see, obviously, is that they can earn much higher rates of return building productive things in China than accumulating 4% yield T-bills.

But the companies aren't socialist, right?

The companies are profit-seeking.

What is the sort of outer loop feedback like?

The thing which keeps the system disciplined, here there's private investors who care about actually earning a rate of return.

And so they're only going to invest in companies which they think have a viable shot of becoming huge companies that can actually be at the frontier in technology.

If these banks are not actually competing against real investors.

And it's not their money.

Exactly.

So they don't care.

But it seems like this is compatible with many of the world's leading companies being developed in the system.

So how is it working?

But doesn't this just belie the whole idea?

And CATL and DJI.

Yeah, exactly.

And it's all financed, as you say, with this financial repression, which is basically a tax on savers.

And so even though there's no meaningful income tax and just value-added tax and corporate income tax, if you factor in the currency devaluation, which is a tax on consumers, and then this financial repression, which is a tax on savers, it actually might be a meaningful tax.

decrease in the quality of life of an average person.

And then just to close the loop on this, so –

I think sometimes we get impressed when we see, oh, they've got the Made in China 2025 and the NDRC and the MIIT and so forth have these...

very specific, you know, LIDAR needs to get to this point and we need the EUV machines at this point and so forth.

But your claim is that that part of the system is actually dysfunctional.

And in fact, the part that actually works is the 5% that is totally private.

Going back to the local government debt burden, there's a world where AI isn't a big deal and there's a world where AI is a big deal.

If we live in the world where AI truly causes huge uplift to economic growth and so forth, is it possible that this problem doesn't meaningfully harm China's ability to compete at the frontier?

And here's why that might be the case.

It seems like this problem is especially affecting poorer provinces, whereas you have the actual numbers, but are Shanghai and Guangdong in a good position with respect to their fiscal situation?

And could they keep funding

top-end semiconductors, AI firms, data centers.

And so China could still keep competing at the frontier in AI, even if the poorer provinces don't have the money to fund basic services and so forth.

And in the long run, they get advanced AIs, and that's a big uplift to economic growth.

And they can just grow their way out of this debt problem.

And what would a solution at this point look like?

Because if you've already spent all this money –

First of all, actually, why is it the case that these investments aren't productive?

Because it's not like it's entitlement spending where you're doing it for, I don't know, old people.

But then at the end of the day, you're not getting any return for it.

Theoretically, these are like you're building real bridges.

You're building airports.

So why aren't they productive?

And then what does a solution at this point look like?

Can I ask about this?

Yeah.

If the reason that the provincial leaders kept doing this construction was because they want to get promoted and they want to get, you know, GDP numbers on the book and government spending now is as the GDP, even if the ultimate thing you're building is not that productive.

But then the central government has had a problem with this for a while.

And then they think this is fake.

They don't want to keep this fake growth going.

And they see the local debt numbers.

Were they still promoted nonetheless?

Why were people doing this?

Who was the person who created?

And also, a lot of these people got arrested in the aftermath, right?

So they weren't promoted.

They were actually arrested.

So what was incentivizing them to keep?

I find it fascinating the fact that –

whether a corrupt political equilibrium leads to more construction or less construction, because the political equilibrium in America today is that, I don't know, California isn't the least corrupt state in the world.

But the political equilibrium is that

The fact that there's so many different factions who get their hand into the pocket means that it's not that everybody's like, we need California Rail to happen yesterday because I need my share of it.

It's that, oh, I'm going to get a little consulting fee to slow this down by five years.

So it's quite interesting.

And if you read the...

Cairo biography of Robert Moses, and you look at the strategies he employed, it actually is very similar to what you're mentioning here, where it's in every single faction's interest that a bridge that Robert Moses is working on gets done through some of the mechanisms you mentioned.

For example, he would give the banks these discounted bonds

For this construction.

So the the bank lobbyists were encouraged to keep the construction going and he would give Obviously the unions wanted employment.

So they'd want the construction to have every single person at every stage was incentivized So I find it interesting that China has maintained to me keeping political equilibrium which encourages too much building whereas our political equilibrium is That the guy it's discouraged and

Yeah.

Okay.

And then what would a solution to this problem look like?

Yeah.

Why is it the case that economists – so when they're talking about rebalancing more towards consumption, they'll often suggest that you should do increasing amount of welfare.

But it seems like the more straightforward thing to do would just be to –

get rid of financial repression so that by default savers would have more purchasing power, get rid of currency devaluation.

Again, the same effect.

Okay, so you mentioned a second ago, well, why do they need this big military?

Obviously, the reason is not just self-defense, but a potential invasion of Taiwan.

So given your reading of elite politics, what's your sense on the probability that they would actually invade Taiwan this decade?

Well, zero COVID was kind of reckless.

Yeah.

I mean, it depends on how you – but you could say the same thing about Taiwan.

It's like it's conservative because in the long run, maybe they'll pose a national security threat, right?

But it is a change from the norm.

I guess you could argue that he has been trying to build up enough self-sufficiency that in the aftermath, China would not be left in the lurch with food or power.

And that's why they're doing clean tech.

That's why they're doing investments in semis and so forth.

So that over the next few years, if they build up enough self-sufficiency, they're in a position where they actually could without it being a huge – he couldn't have done the same thing in 2015.

This gets to the question of how much information flow is there

How much true information is getting up to Xi Jinping?

In authoritarian systems, obviously, there's a danger that the leader is not getting told what they don't want to know.

But you were mentioning a second ago that you think that the top leadership is in communication with experts all the time, is in meetings all day, where they are learning about what's happening.

Is that your sense –

But then you read about zero COVID, and it seems like by year three, it was very obviously a mistake.

So to the extent that it wasn't stopped at that point, was that because information wasn't getting up to the top leadership?

Similar story with – I think there was a very famous case in 2024 where Xi said, why don't we have a billion-dollar unicorn tech companies in China?

And people were responding –

Does he not realize that the 2021 tech crackdown is the reason that's the case?

Did somebody not tell him?

So what's your sense on how much he is getting told things which might make him uncomfortable, for example, that we might not succeed in Taiwan?

And how did that manifest in Zero COVID?

Yeah, I guess I'm just trying to get a sense of where does this net out in the competence of the political system?

It seems like they have a track record of making some smart calls, but then occasionally making a call that is so bad that it wipes out all the smart calls and zero code being an example of that or obviously one child policy before that, not to mention all the things that Mao did.

Yeah.

And so I'm getting the sense that they will consult with economists or scientists or whatever on some things.

But then I guess I still don't understand.

Well, then why did they – okay, we're not going to buy the American vaccine because of this –

This is the time of all times to be from making sure the pharmaceutical industry in China is in a good position.

Sure.

But then did they not know that people were locked in their apartments starving and that this was not a long term viable strategy and so forth?

Like, how is this compatible with this very technocratic picture we have?

What is the succession plan after Xi is gone?

What's your sense of what would happen?

Tomorrow he drops dead.

What happens next?

Oh, God.

That's interesting.

Yeah.

But they can stop it.

And suppose it doesn't happen tomorrow, but in 2028, 2030, he just goes senile.

And over the course of months, it's clear that there needs to be a successor.

What does then the succession process look like?

What are the important factions in Chinese politics today?

What is the equivalent of liberal or conservative in American politics?

Historically, there's things like the Shanghai clique or whatever, but do those still exist?

And if they do, what do they stand for?

I mean, they're all beholden to Xi Jinping in one way or another.

But do they have different ideas about what should happen with China or what should happen with the world?

What's your forecast on –

Chinese growth or the Chinese economy relative to the U.S., if AI is not a thing, let's just forecast out everything else.

In the year 2040, is it 2x as big as U.S.

economy?

PPP, is it 3, 4x?

Is it 1x?

Okay.

So this is quite a bearish forecast because usually people expect it to be quite dominant by 2040.

Final question.

Sure.

What advice do you have for me on how to land, if not somebody on the Politburo, like one step away from the Politburo?

Do you think it could be an interesting interview?

Okay, this was super useful to get a very much more tangible sense of what is actually how the Chinese political system actually works.

We talk about China so much and I just, most of us have such little understanding of even the basic like Congress president level Chinese political system.

So, and obviously a lot of it is obfuscated on purpose.

So I appreciate you adding a lot more color to it.

Great.

Thank you for having me.

Thank you for tuning in.

I'll see you on the next one.