Into Asia
Into Asia
Inside China's AI Boom
0:00
-39:03

Inside China's AI Boom

From Shanghai’s World AI Conference to start-up incubators across the city, the Chinese tech analyst Rui Ma shares what she’s seeing on the ground, and where China actually stands in the global AI race.

From Shanghai’s World AI Conference to start-up incubators across the city, the Chinese tech analyst Rui Ma shares what she’s seeing on the ground, and where China actually stands in the global AI race.

Follow Rui on X: https://x.com/ruima

Follow her newsletter via Tech Buzz China (Substack): https://techbuzzchina.substack.com

Related Links: 
https://techbuzzchina.substack.com/p/china-deep-tech-trip-what-we-did

Below is a transcript:


Chang:
Let’s start a little with your background. How would you summarize your work — what you’ve been spending most of your time thinking about over the past few years?

Rui:
I’ve worked pretty much in finance for my whole career, which is now — shockingly — over 20 years. Tech Buzz China was originally started as a hobby in 2018. It’s now, as you said, an investor-facing platform where we do bespoke research, organize trips to China, and more. I’m also working on a few personal projects, including a book on EVs.

Chang:
I’m interested in how you’d describe the development of AI in China. In the 2010s, there was a lot of talk about AI — the AlphaGo moment, when China became really excited about artificial intelligence. China has always been part of the conversation, given the quality of its engineers. Is there a difference between that phase and what we’re seeing now, with the rise of ChatGPT and large language models?

Rui:
It’s completely different. So I actually don’t remember exactly which year the book came out, but I think it was AI Superpowers or something by Kai-Fu Lee. So that was about machine learning, and Kai-Fu made the argument that the U.S. and China were the two superpowers to watch in terms of artificial intelligence, which at the time was more defined as machine learning — much more deterministic models — and computer vision, etc.

At that time China was actually leading in computer vision, and that was what the dialogue was primarily about. LLMs obviously completely changed all that, and generative AI as a discipline kind of caught China off guard. I remember very clearly, right when ChatGPT came out, there was a unicorn AI founder — his company was from a famous AI company from a decade ago — and he came to Silicon Valley and was like, “there were a few years where none of us in China were interested in Silicon Valley at all, because we felt like we had sort of surged past all the innovations that Silicon Valley had to deliver. And lo and behold, I’m here trying to learn from this place again, because it seems like while we were in China celebrating our successes in mobile internet, we completely missed the boat on generative AI.”

And that was very much the sentiment in 2022, 2023.

Now I think it’s a little bit less anxiety-ridden. People have said that generative AI stuff, we’re also beginning to figure it out. We’re also expanding into even broader aspects, which is one of the key points I’d probably like to bring up for this podcast: in China, AI is not seen as just purely software manipulation, but also manipulating the physical environment via robotics and what they’re calling embodied intelligence.

Ian:
And would you say that they’re ahead of the United States in that? Or are they still playing catch-up in that respect as well?

Rui:
So I think it’s really difficult to pinpoint AI as one measure. Even if you look at the LLM benchmarks out there — there are tons — because intelligence itself is quite multifaceted, right?

For AI, it’s becoming much more obvious that there are some tasks that Chinese companies are quite good at — maybe not yet world-leading, but quite good — and that typically revolves around image/video generation. And then there are some tasks where the U.S. still leads by a notable margin. And I think, like many other areas, we’re constantly sort of neck-and-neck in the narrow space we’re calling generative AI and LLMs. Overall, the gap has certainly narrowed — coming from the end of 2022, where Chinese researchers and entrepreneurs really felt ChatGPT came out of the blue and they were unaware of its power, to now, where there’s so much investment and talent devoted to this area that they’ve made significant strides and are actually doing really well in certain aspects of the technology.

Chang:
I wanted to get back to how China’s AI systems evolved. You mentioned computer vision — an earlier paradigm of AI that has to do with recognizing things in the visual space. A lot of these systems are associated with surveillance technology, right? You mentioned SenseTime as one of the leading companies supplying surveillance technology. That was how many people thought about AI in the 2010s. Then after ChatGPT we moved into what Rui said was this generative-AI paradigm. It also involves images and vision, but it’s more closely associated with ChatGPT — models that can spit out information, can hallucinate, are closely associated with “intelligence,” and now we can actually communicate with them with language. That has become the new paradigm.

When we talk about foundation models, I think we’re referring to these big companies — OpenAI, Meta, Google — who have used basically the entire internet to produce these language machines that sound very convincingly like humans. And China has created their own versions, the most famous being DeepSeek.

Tell us a little bit about the recent trip you did with your community in July. You went to the World AI Conference. First of all, what was that like? What were your impressions from being there? I know there were some pretty high-profile people there — U.S./AI space — and just share a little bit about some of the other conversations you had.

Rui:
The trip was amazingly informative. We actually did not intend initially for the World AI Conference to be a super big part of the trip — that’s why we only spent, I think, a little over half a day. But this year it really blew up, partly because there was support from the central government in the form of, like you said, the Premier going and making remarks and getting Nobel Prize winners to talk about global AI safety and collaboration.

And this year was really big for robotics — specifically humanoid robotics. That’s actually where I ended up spending most of my time at the venue. So there were panels (none of which we attended because we had a lot of private meetings set up), and then there was the actual showroom floor, which we did go to. There was a very, very large building completely filled with booths of various AI and robotics companies. I spent most of my time on the robotics floor. It was probably the first event I’ve been to where there were robots running all around you — they weren’t even staying in their booths; they were just walking willy-nilly everywhere.

The conference — I think I read — had 300,000 or maybe 350,000 people attend over four days. I know it sounds unbelievable, but having been there for a few hours over two days, it certainly felt like that. It was extremely well attended and very, very high energy. And it really goes to show how much enthusiasm there is — not just from the government and businesses, but also from the general public — about AI.

Ian:
Since you bring up Chinese government involvement, what is the effect of politics in the United States and in China on the development of AI? In the U.S., there’s pressure to deregulate the AI industry and not export certain things to China; the government is getting involved now in companies like Intel. In China, especially under Xi Jinping, there are big efforts to regulate opinion — what can be said and what can’t be said. Is this irrelevant, or does it affect AI development in both countries?

Rui:
I’m sure it does, but at the business level we were interacting with, it came up very little. One of the people on our trip, who was at the conference as a speaker — Paul Triolo — spoke on AI safety, and he said that was a big part of the conference. However, in everyday investor-facing business conversations, it never really came up.

So I think regulations, export controls — all these things — definitely affect the industry’s development and are on certain people’s minds. But for the most part, as a business person, the major growth areas and the major areas of interest for most AI entrepreneurs and investors don’t involve that just yet. It’s such a new and burgeoning industry. Regulations and hardware-stack limitations — sure, important — but people are trying to figure out how to get their first customers, get to their first million dollars of revenue.

Ian:
Right. But safety is one thing, and people have different interpretations of what “internet safety” means. What about robotics, which almost inevitably will have a political side because it’s going to take people’s jobs? Do you think the Chinese will have less trouble developing robotics than Americans, because there would be less concern about people losing jobs — or possibly more concern?

Rui:
That’s really interesting. We talked to a few robotics companies. Right now, the level of robotics development is definitely as an augmentation to human labor and not really displacing human labor. The few types of jobs they can displace are largely jobs that are either very dangerous or very uncomfortable for humans to do.

For example, we saw a vertical-climbing robot that can climb caves and bridges and the outside of buildings to do inspections and light maintenance. Obviously cleaning — those are things currently very dangerous or inconvenient for humans. So no one’s going to complain about these robots “taking away” human jobs because that’s not the reality. I’d say most robotics is at that level of “job annihilation.”

The one near-term big question mark is autonomous driving. When we were there during the World AI Conference, they made a big announcement that a few select, high-profile companies were granted licenses to do truly driverless robotaxis in China, and that was really exciting — but it’s still a relatively small area they’re allowed to operate in. I think it will be a while before they really take over drivers’ jobs. You do see some media mentions of drivers expressing worry, because that’s a very big gig-economy job in China right now.

Ian:
What about industrial jobs? Is it having an effect there?

Rui:
I would say industrial jobs — really not yet. A lot of the very labor-intensive things — it’s not economical yet for robots to take over. I’m talking about honestly low-value labor, like sewing clothing, etc. The robots that are commercializing tend to be much larger industrial robotic arms.

For example, we do an EV investor trip, and the EV factories are often “dark factories” now, where you have huge robots moving parts from one place to another and putting them together. These are jobs that wouldn’t have been done that much by a human anyway. So I don’t see them net-replacing human workers, but I do see manufacturing lines that, in the past decade, have been redesigned from the ground up because of these machines.

Ian:
And do you detect — because you spend a lot of time in China — differences in anxiety or optimism between Chinese attitudes and American attitudes? Or is it pretty similar?

Rui:
I think everywhere there’s a mix, right? I live in Silicon Valley — lots of optimists, lots of doom-sayers. In China, it’s the same, except I think the mix tilts more optimistic. There’s more of a sense of, “Hey, AI is a great business opportunity — let’s master it and use it to our advantage,” versus “AI is going to render me useless,” which in the U.S. is a little more prevalent. But, like I said, it’s a mix.

I do think acceptance of new technologies tends to be a little easier in China, partly because in the past 30 years there’s been so much change in the economy and daily life that people are used to that kind of change. Whereas the U.S. has been more developed and therefore stable for longer, so people are less at ease with change.

Chang:
Rui, you had mentioned on a tweet after you came back that “the energy problem is not a problem,” or something like that. Could you elaborate — maybe start with what you’ve been noticing in the United States in terms of energy as a potential bottleneck for AI development?

Rui:
I’m not an energy expert — I want to preface this — but I went a little bit viral for saying that in China, electricity generation for AI data centers (which are predicted to be huge power consumers) is considered a “solved problem.” Multiple people said that. We spoke to one company that actually runs AI data centers and went into more detail. That’s very different from what you hear in daily conversation in Silicon Valley, where I live — a lot more hand-wringing about not having enough power to run all the AI capacity we want to bring online.

And it’s true: in the U.S., in the past one or two decades, electricity generation has been largely flat. In China, it’s almost doubled in the last 10 years and is continuing to become cleaner and more efficient. Electricity prices are lower than the average here — at least pure generation. I think that’s a big difference in how people think about opportunities and challenges to make society fully intelligent and connected.

Chang:
You mentioned wanting to focus on this podcast about the diversity in conceptions of AI in China. One thing I thought back to is something Kai-Fu Lee popularized — in the U.S., the tech industry often goes for big moonshots, trying for an AI takeoff (basically what OpenAI is going for). And Kai-Fu has always said China’s strength is in applications — using major technologies and diffusing them through society. In this iteration of the AI arms race, do you see what he said holding up? What are you noticing?

Rui:
Any time you look at what businesses are doing — you could include academia too — it’s about who is funding these activities (whether pure research or commercializable work) and what their expectations are.

That explains a lot of Chinese entrepreneurs’ behavior. Their incentives were tied to the fact they couldn’t get funding without showing really good cash flow, tied to the exits available to them and capital-market conditions.

Whereas for U.S. entrepreneurs, you tended to have more buyers if your company failed (talent acquisitions), more patient capital, and a stock market willing to evaluate you based on your story and future prospects versus only what you’ve already accomplished.

All that led to what Kai-Fu rightly noted — more U.S. companies willing to go for bigger moonshots. That said, as the Chinese ecosystem has matured, you see diversity. The most notable example is DeepSeek: they’re clearly not trying to be a $100–300B valuation company like OpenAI. They’re highly focused on R&D. They have very little staff on go-to-market, sales, or even customer support — bare-bones people doing that, just enough to understand if their technology is any good. That’s very much a moonshot company, whereas OpenAI has become more of a commercialization company — putting out products, building an ecosystem, lots of sales and GTM.

So you can make generalizations, but I don’t think they’re “cultural truths.” They’re indicative of the funding conditions under which these businesses survive.

Chang:
We didn’t really get to open source. I remember that being a pretty big difference between Chinese AI and the American one. Could you share how you think about open source in China? And you mentioned with that company using DeepSeek — why do all the apps I use in China have DeepSeek built in, whereas Siri on my iPhone is still as awful as it was several years ago? Is that related to open source, or what’s going on?

Rui:
I think a couple of things are going on.

The most important is DeepSeek — a company that philosophically believes AI should be open source, because they’re not trying to compete on commercialization. They’re primarily a true research company and sell services at basically a hairline profit. That said, there are companies like Alibaba that are also open-sourcing — and Alibaba has been a big proponent of open source for almost two decades. That’s also a philosophical position. Not all Chinese AI companies are open source — quite a few are not, or they only open-source their least advanced models. So it’s not a cultural universal; it’s specific actors.

It’s helped by the fact that the government is a big proponent of open source, and Chinese businesses trust open source more because they want more control over their own implementation.

Why does the Chinese government like open source? Because from their perspective, they don’t care about creating one super-giant profitable AI company (or a handful). They care about unleashing the power of AI throughout the entire economy — that’s a much bigger impact and what they’re going for, versus just creating trillion-dollar companies. That’s why the government likes it. It also helps that open source establishes credibility globally and helps diffuse the technology across borders.

Chang:
We’ve talked a lot about the U.S. and China tech sectors, but can I ask about the social component — how AI is impacting daily life? There’ve been reports about AI girlfriends in China. Have you noticed any impact at the social level, or is it too early?

Rui:
From usage numbers, you see a good portion — Tencent, ByteDance, DeepSeek’s models — tens of millions of users. Like in the U.S., it’s not always obvious: few people talk about how they’re using AI; they’re just using it. For consumers, and for businesses, they’re using AI to market products more.

Basically every consumer-electronics company is trying to become an AI company. I don’t know if you saw this, but Midea, the appliance company, was promoting an air conditioner with DeepSeek. I don’t know why you need DeepSeek in your air conditioner, but clearly they think it differentiates them and maybe gets a premium.

You see things like that. Day-to-day life doesn’t look that different from here: people share AI-generated pictures or videos, or use AI search to find where to eat or which tourist spots to visit — sharing within WeChat. Not really different from here. I don’t see any “higher-level” use of AI or closer integration yet.

Ian:
China has one of the oldest — perhaps the oldest — examination systems in the world. In the U.S., where I teach part-time, professors are more worried about students using AI to do exams or write papers. Is that something that worries people in Chinese education?

Rui:
Far less so. So much of Chinese education is testing, and you obviously can’t use AI for in-class testing. Sure, you can use AI to do your homework, but it doesn’t affect your standing much because the bulk is based on tests.

I see more use of AI to study — also to do some homework, but as a study aid. All the big ed-tech companies in China have integrated devices — basically a tablet — with AI curriculum and proprietary adaptive material. I’m personally looking into that because I want to educate my kids using AI. It’s less about cheating and more about studying.

And the government is putting some AI instruction into elementary schools starting this fall. I didn’t look into the details, but I think it’s basic concepts behind AI — explained to a fifth-grader.

Ian:
Can I ask you to generalize once more? Different philosophies, attitudes, strategies — which of the two countries is best placed to “win” the AI competition, if it is a competition: the United States or China?

Rui:
It depends on what you mean by “win.” A lot of people define it as having the most advanced model — but even defining that is tricky because these models do many things, and depending on benchmarks, datasets, languages, results differ. If you define it as the single most advanced technology, I think the U.S. system tends to incentivize that — it’s less about diffusion and more about the entity that can get the super-advanced tech and sell it.

Alternatively, “winning” could mean the general intelligence unleashed in your society: how smart is everything, and how much can everyone — even the poorest — benefit from AI? In that sense, I think the Chinese government has more incentives built around that. That is literally what they try to do with state capitalism. It doesn’t always work, but that’s the KPI they talk about.

To be great at AI, you need both: the most advanced tech and the widest access. The two societies have different strengths in how that happens. It’s like a 2×2 matrix — they’re in different boxes.

Chang:
I feel like in the U.S. everyone thinks about the AI race based on the frontier — there’s this idea of a runaway moment where AI can do everything. That’s not really as prominent in China, would you say?

Rui:
People talk about it, but Hollywood popularized that moment — AI becomes fully sentient, exponentially smarter, manipulates matter, hacks our systems, and we’re all prisoners of the superintelligence. I can see that happening, but I feel like you’d have a pretty long warning before that.

It’s like climate change — you can see where the tipping point will happen. It confuses me why people don’t believe in the climate tipping point (which we can forecast and feel with extreme weather) but are terrified of AI takeover, which is nowhere near as predictable. It’s irrational — thanks to Hollywood and sci-fi — and because we anthropomorphize robots/AI.

Chang:
Yeah, that’s a great point.

Ian:
There may be a cultural or religious element. I once heard a talk by the economic historian Robert Skidelsky, who wrote the biography on Keynes, who is a devout Christian. He denounced AI on religious grounds — a form of hubris, humans “playing God.” That’s an old Christian taboo — the Tower of Babel and so on. I wonder if the Chinese have anything quite that strong.

Chang:
I also feel like part of the reason people like Peter Thiel are against that kind of thinking is because “climate” has a negative connotation — tipping points are apocalyptic. Whereas a lot of people in that orbit in Silicon Valley are excited about a tipping point for AI — they’re racing toward it. That valence matters.

Rui:
Yeah, they probably think that if there’s superintelligence, robotics — subsets like exoskeletons — could extend quality of life as you age. And maybe superintelligence can figure out how to delay death — which all the billionaires here are super obsessed about.

Ian:
That’s absolutely true. But think of the Frankenstein story, which again has roots in the Christian tradition. There’s a deep suspicion in Western culture of people playing God — creating humans in ways not natural. That’s Frankenstein. It’s why there’s perhaps more resistance to robots in most Western societies compared to Japan and China — the “replacing God” idea is deep in Judeo-Christian thought, perhaps not as profound in East Asia.

Rui:
Exactly. It’s not. Modern Chinese society is quite atheist and non-religious, but even in pervasive mythologies, the concept of life is circular, with much less of a hierarchy — less binary, more circular. “Superintelligence” implies “more intelligent,” but it’s just a different form of consciousness or life. People believe you can be reincarnated into trees, rats, cats, whatever. There’s no concept like that in Judeo-Christian religions, where humans are humans and closest to God.

Chang:
There’s also the element that a lot of U.S. AI people literally went into the industry because they watched Star Trek or got inspired by those ideas. That’s not the case in China — AI engineers there aren’t like “I saw Frankenstein and wanted to repeat that.” Different cultural influences; a lot of them just want to do something that makes money — and right now AI makes money.

Ian:
But if any of them did this inspired by Frankenstein and thought “I want to do the same thing,” that would be deeply creepy.

Chang:
Yeah — motivations really matter. In Silicon Valley, there’s self-selection with heavy ideas coming in. For example, Yann LeCun — one of the godfathers of AI — I think he went into AI after watching a debate with Chomsky (and others) back in the day. You can trace influences. People are inspired by ideas within Western culture that frame AI as transformative — and people want to join that. There’s an evangelical dimension. In China the motivating drivers are completely different.

Rui:
Yeah. It’s much more just an engineering problem.

Chang:
I think that’s a great place to end it. Thanks so much, Rui.

Ian:
Thank you.

Rui:
Yeah no problem! This was a fun conversation.

Thanks for reading Recent Works by Chang Che! Subscribe for free to receive new posts and support my work.

Discussion about this episode

User's avatar