Experts say China is behind the US in AI tools like ChatGPT due to censorship, geopolitical tensions, and the government's growing control of the private sector
New York TimesLi Yuan
Context & Ripple Effects
This February 2023 diagnosis of a China-US generative AI gap landed just as Beijing was closing doors rather than opening them: within days, sources reported that regulators had told Tencent, Ant Group, and others not to offer public ChatGPT services after state media criticized chatbots over misinformation. The NYT piece frames that as structural — censorship, geopolitics, and growing state control of the private sector — not a temporary engineering deficit.
What followed validates the framing while complicating it: engineers later pegged the lag at [[a:849587|at least one year in generative AI, pushing Chinese companies onto US models like Meta's LLaMA]], and by mid-2024 technologists at Chinese firms credited open-source technologies with narrowing the gap. By 2025, the relationship had inverted into physical risk, with China reportedly telling its top AI leaders to avoid US travel over detention fears.
First-order effects
Chinese consumer internet giants — Tencent, Ant Group, and peers named in the regulatory guidance — cannot ship public ChatGPT-style services even where they have the models, ceding the domestic chatbot market's first-mover moment.
Censorship requirements mean any compliant Chinese chatbot must be trained or filtered to state standards, adding cost and latency relative to unconstrained US rivals like OpenAI.
Second-order effects
Barred from building freely at home and blocked from US services, Chinese developers route around both limits through Meta's openly licensed LLaMA and other open-weight models — making Western open-source releases an unintended supply line for Chinese AI capability.
US model providers gain leverage they did not seek: access decisions on weights and APIs become de facto policy instruments, which cuts both ways once Washington weighs whether openness aids an adversary.
Third-order effects
If the pattern holds — state control suppressing domestic experimentation, then open-source closing the gap — China's AI industry consolidates around labs that are state-compatible by design rather than the most commercially aggressive ones.
The 2025 travel advisories for Chinese AI leaders point toward talent flows and research collaboration hardening along geopolitical lines, with model access itself becoming a managed channel between blocs.
The trend: Generative AI competition between the US and China is settling into a structure where state control shapes who builds domestically, open-source weights leak across borders, and each side treats model access as a geopolitical lever.
“The AI model underlying [Baidu's] bot has been in development since 2019 with its newest generation trained on 260bn parameters — comparable to GPT3, the technology underpinning ChatGPT, in terms of size, although trained on a much smaller data set.” https://www.ft.com/...
Why China Didn't Invent ChatGPT. “The state's hardening censorship and heavier hand have held back its tech industry; so has entrepreneurs' reluctance to invest for the long term. It wasn't always that way.” https://www.nytimes.com/...
I remember when it was commonly accepted that China will surpass the West in machine learning and AI because it is not constrained by rule of law, privacy regulations and consumer rights. https://twitter.com/...
“The development of any significant technological product is inseparable from the system and environment in which it operates...” https://www.nytimes.com/...
“Once the open environment is gone, it will be challenging to create such products,” he said. If a decade ago China was the wild, wild East for tech entrepreneurship and innovation, it's a very different country now. https://www.nytimes.com/...
“'Why ChatGPT wasn't invented in China?' they asked...They're also asking more fundamental questions...: Have censorship, geopolitical tensions and the government's growing control of the private sector made China less friendly to innovation?” @LiYuan6 https://www.nytimes.com/...
Given the interest in ChatGPT, a fascinating essay by @LiYuan6 about how China led in AI technology but then fumbled because of political controls. Chinese techies joke, “We need to teach machines not only how to speak, but also how not to speak.” https://www.nytimes.com/...
“But the [Chinese] govt has been the biggest barrier to A.I. — its obsession w/ censorship perhaps its heaviest club....a wide range of data is crucial to developing tech like ChatGPT, and that is increasingly harder to come by in a censored ..environment.” https://www.nytimes.co…
If a decade ago China was the wild, wild East for tech entrepreneurship & innovation,it's a very different country now under Xi.ChatGPT serves as both a shock&a revelation:China lags far behind the U.S. in A.I. & gov't control stifles innovation. My latest https://www.nytimes.com…
Why Chatbots like ChatGPT Weren't Invented in China Today, jokes circulate that capture the dark mood among tech people. A popular one: “We need to teach machines not only how to speak, but also how not to speak.” https://www.nytimes.com/...