Angela Huyue Zhang is Professor of Law at the University of Southern California, where she writes and teaches at the intersection of law, technology, and geopolitics. Her research examines the institutional dynamics of regulation and governance, with a current focus on the U.S.–China tech rivalry and the global regulation of artificial intelligence.
Zhang is the author of two award-winning books: Chinese Antitrust Exceptionalism: How the Rise of China Challenges Global Regulation (Oxford University Press, 2021) and High Wire: How China Regulates Big Tech and Governs Its Economy (Oxford University Press, 2024). Prior to joining USC, Zhang taught at the University of Hong Kong, NYU School of Law, and King’s College London. Earlier in her career, she practiced law for six years in Asia, the United States, and Europe. She received her LLB from Peking University and her LLM, JD, and JSD from the University of Chicago Law School.
Zhang, a speaker at the World Economic Forum’s Annual Meeting of the New Champions in Dalian, China, agreed to be interviewed about how the Chinese government is emulating platform companies. She is conducting this research project with Professor S. Alex Yang of London Business School.
Q: During a Forum dinner in Dalian, you said the Chinese state operates like a platform company. Please elaborate.
AZ: The platform business model has several defining features. Chief among them is that platform companies don’t own the assets. Alibaba and Taobao don’t own the shops. Airbnb doesn’t own the properties. Uber, for the most part, doesn’t own the cars. They are asset-light, which lets them scale quickly by facilitating transactions between participants on different sides of the platform.
The Chinese government governs the most innovative technologies — batteries, EVs, and now AI — in the same asset-light pattern. That’s not to say the government has no ownership. China’s government is huge: there’s the central government, local governments, ministries, sovereign wealth funds, and many species of state-owned company. But for the most part, the state is not the dominant owner. Ownership belongs mainly to private entrepreneurs, which creates a vibrant entrepreneurial culture and sparks the same kind of enthusiasm you see in Silicon Valley. This light-touch ownership approach avoids the misaligned incentives that plagued central planning in the past.
At the same time, control is embedded at every stage. The government’s hand is visible in how it grows the platform. It gives subsidies, but people tend to overlook that it does far more than offer money — it uses regulatory levers too, which is different from how a normal business platform operates.
Take electric vehicles. To boost demand, the government gave EVs privileges like priority access to busy roads and free parking. In some cities, when EVs first launched, it even organized test drives so ordinary citizens could try the technology. Now you see the same approach with humanoid robots, particularly in southern cities like Shenzhen, which has a dense physical-AI supply chain. The government becomes the first customer because people aren’t yet sure what humanoid robots are good for beyond entertainment — the use case isn’t clear, so the government steps in to try it out. It offers procurement guarantees to help grow the market, and it arranges public demonstrations: you see humanoid robots directing traffic in Shenzhen, and you see them in “6S shops.”
The 6S shop concept builds on the 4S auto dealership model — full service — but adds two things, including leasing, since many people don’t want to buy a humanoid robot outright even though it’s fairly cheap in China; they’d rather use one for a specific purpose. The government set up 6S shops to draw people in to see physical AI and try it for themselves. In this respect, the Chinese government does far more than democratic governments do to give these businesses a helping hand.
Q: Recently I was reporting on the battery industry, and one differentiator from a place like Europe is that the government ensures the whole innovation ecosystem is built out.
AZ: I completely agree. The Chinese government follows a pattern I call “grow, govern, and guard.” It imposes rules requiring that Chinese batteries be used by domestic EV customers — a restriction that was crucial to driving down battery costs, since the more they produce, the more they learn. That’s another tactic used to grow these ecosystems: what we’d call embedded control.
Think of it like running a platform: first you entice everyone to join. Then you have to govern them, because conflicts and tensions inevitably arise. The Chinese government is very active in governing its ecosystems to keep them healthy — governance is a core platform function. Law isn’t traditionally seen as important in China, but that’s changed in the platform era; the government has set clear rules so participants know what to expect.
“Guard” means that once everyone is in the ecosystem and it has matured, it’s time to harvest the gains and prevent capability from leaking out to rival platforms. In China, the thing they worry about most is what’s called “Singapore washing.”
Q: What is Singapore washing?
AZ: It’s when Chinese companies move out of China and sell directly to Western customers. The government worries about technology transfer and brain drain — that’s why it reportedly holds back the passports of top AI researchers. I recently saw an article in the Financial Times about how the government discourages scientists in certain fields from publishing in top international journals, because the review process requires revealing too much sensitive data.
Data control is a major theme in Chinese regulation. So on the “guard” side, the government controls data, talent, and technology — and capital too. If you’re a new startup today, you don’t go public in America anymore; nobody is even thinking about a U.S. listing. And those who considered moving abroad are restructuring instead. This is China building its moat.
Control is the single most important platform strategy the Chinese government uses. But at the same time, it tries to nurture the ecosystem rather than simply pick winners. It lets companies fight it out first; whoever wins the market becomes the winner, and the government is willing to let the losers go. You see this pattern repeat: a wave of over-entry, a wave of bankruptcies, and the survivors — the fittest, strongest players — go on to scale.
Q: Would you say China has used essentially the same platform playbook all along — for solar, EVs, batteries, AI — just refining it along the way?
AZ: Yes. They started with solar, and once solar was established, batteries and EVs followed naturally. Now it’s humanoid robots. These sectors are all interrelated — once you have solar capacity, it’s a short step to renewables, EVs, and humanoid robots, because they’re integrated and interdependent. Throughout this process, the state’s capacity for control — its “state capacity,” as I call it — has grown too, which makes the platform strategy more effective over time. It’s a process of learning by doing.
Q: How do you see the strategy evolving in AI? I was struck that OpenAI was reportedly in discussions with the U.S. government about selling a stake — I never expected to see that in the U.S. How does that compare with what the Chinese government is doing?
AZ: The OpenAI case is interesting. Both OpenAI and Anthropic are competing against Chinese firms whose models are just a few months behind. And with the recent release of Kimi K3, that gap seems to be getting even smaller. That small advantage is what has supported their high valuations, which makes it a fairly risky business. If I put myself in Sam Altman’s shoes, aligning with the U.S. government makes a lot of sense: you become a national champion, and you’re not giving up much. Bernie Sanders, remember, was proposing something like a 50% stake; Altman is volunteering 5%, and in exchange he gets something like a soft budget constraint — if the valuation crashes, or an IPO goes badly, the government may be there to help. What if DeepSeek or Mootshot AI reaches 99% of frontier-model capability at a hundredth of the price? That risk is real.
So it’s a smart business strategy for OpenAI, particularly since it hasn’t been doing as well as Anthropic lately. You’ll recall Sam Altman talking about needing the U.S. government to help backstop the data centers required to keep building — a comment that was already controversial. Political alignment with the government makes sense from a business standpoint; this is American-style state capitalism. From the government’s side, it also addresses public unease about inequality — a way to be seen sharing in the wealth these AI firms are generating. It’s good politics for the White House too.
But it’s a completely different strategy from what the Chinese government is doing by investing directly in its top AI labs. I’ve written about China’s National AI Fund, whose predecessor was Phase III of the Semiconductor Fund. That fund recently invested less than 1% — under 0.3%, in fact — in DeepSeek’s first outside funding round, but it was the only investor with a voting stake. CATL and Tencent also invested, but neither holds voting rights the way the National AI Fund does. There’s very little public disclosure about what happens behind that.
If you look at what the National AI Fund’s predecessor, the Semiconductor Fund, was built to do — its whole mission was to build chip self-sufficiency — you can see the logic: invest across the AI stack to get there, starting with infrastructure. DeepSeek is treated as infrastructure because it’s widely used and open source. My interpretation is that the more DeepSeek’s models are optimized for domestic chips — Huawei’s or others’ — for training or inference, the more that increases demand for those chips. Chinese firms, for the most part, aren’t making money selling open-source models; API access is very cheap. The real end goal is to sell the chips themselves, the most profitable part of the AI supply chain — and that’s China’s key bottleneck.
The government’s small investment in DeepSeek is really a way to understand what the company is doing, and to nudge it toward greater interoperability with domestic chips — which is happening. DeepSeek’s latest model is optimized for inference on domestic chips, and Meituan, which is a bit like a Chinese Groupon, recently trained its AI model entirely on domestic chips. From the government’s standpoint, that state investment is less about capital support and more about nudging the ecosystem in the right direction.
That’s the endgame: AI sovereignty. It’s why Jensen Huang is so worried about being locked out of the Chinese market — China has tightly controlled the purchase of Nvidia’s H200 chip , even though there’s enormous domestic demand. But this follows the same playbook as batteries: The more they produce, the more they learn, the more costs fall, and eventually the gap narrows. That’s exactly what happened with batteries — because of the domestic content requirements for EV subsidies, within about five years, at a critical moment, Chinese battery costs fell below Japanese and Korean costs. There’s good research showing this. That’s how they took off.
Q: What could Europe or the U.S. learn from the Chinese strategy?
AZ: It’s easy to talk about the strategy itself, but there’s a precondition that matters just as much: state capacity. We can list the playbook — asset-light ownership with embedded control, nurturing the ecosystem instead of just picking winners, scale first and harvest later — and hand that list to Vietnam or Cambodia, but it won’t work the same way there. The “platform state” is bigger than industrial policy, because every state is endowed with different conditions.
In China’s case, you see very strong state capacity, which I break into three parts. First, coordination within the bureaucracy — the ability to consolidate regulatory power quickly, which requires a tightly coupled, hierarchical bureaucracy that democratic societies don’t have, because democracies are built on decentralized checks and balances. That’s actually part of why the Trump administration has been pushing for a more unitary executive — closer to the Chinese model of control.
Second, low institutional resistance. This is a classic feature of an authoritarian state: when it wants to get something done, there’s little pushback. Compare that with data centers in the U.S., where a large majority of Americans oppose new data centers and many cities have suspended projects because of the federal system and the institutional resistance that comes with it. In China, when the government wants to build something like the “Eastern Data Western Compute” project, it happens — they recently announced $300 billion for interconnected data center clusters across the country, and it will get built. Media is state-controlled, unlike in America, where a driveless Waymo robotaxi hitting a cat would make the New York Times that sparked backlash against autonomous driving.
Third, control of state resources. The Chinese government still controls a great deal: state-owned farms, land, information, finance — the very interfaces of the economy — in ways no free-market economy does.
I was talking to someone from Vietnam in Davos who ran an industrial park there, and I asked how his country handled state capacity. He said Vietnam can move quickly and get things done — fairly top-down. But what Vietnam lacks, which China has, is scale: a large domestic market for experimentation, a large customer base, supplier base, and talent base, and enough local governments and agencies competing with each other to allow many different experiments to run in parallel, with room for trial and error. You simply need scale, and Vietnam doesn’t have it.
This is something to think through: how might elements of China’s model be replicated elsewhere? Maybe applied to a specific technology where a country has both strong state capacity and workable scale. But you can’t just lift the Chinese playbook and drop it into America or Europe— it wouldn’t work at all. That specific combination of conditions is what makes China’s platform state fairly unique.
What other countries can learn, though, is the ecosystem approach. Industrial policy elsewhere tends to be narrow — one sector, one industry at a time. China instead treats technology development as a single ecosystem spanning multiple sectors: batteries, renewables, EVs, large-language models, humanoid robots, brain-computer interface, biotech and more. Each sector cross-fertilizes and reinforces the others — and that, more than any individual policy, is the real lesson.
