The Guangdong Paradox: A Province Full of AI Founders, With No Star AI Company to Call Its Own

Guangdong province keeps producing the founders of China's most ambitious AI labs, and keeps losing their companies. The political-incentive structure that grades Chinese local officials on visible wins rewards the measurable work of recruiting headquarters, not the decade-long work of building the research gravity that makes founders stay.

Consider four names behind three of China's most closely watched labs. DeepSeek CEO Liang Wenfeng grew up in Zhanjiang, a port city in western Guangdong. Moonshot AI co-founder Yang Zhilin is a Guangdong native. Two lead authors on Moonshot's Kimi K3 technical report, Jianlin Su and Guangyu Chen, are also from Guangdong. None of their companies are headquartered in the province. DeepSeek sits in Hangzhou. Moonshot AI runs out of Beijing.

That is the paradox the Sinpers.ai analyst Caixian Chen pressed in a WeChat essay translated into English by the ChinAI newsletter: Guangdong exports the people. Beijing and Shanghai keep the companies. The geography is not an accident of taste. It is the predictable output of how Chinese local officials are evaluated, and what they are evaluated on.

"Frontier AI" in this context means the state-of-the-art, large-scale foundation models these labs are racing to build. The Six Little Tigers of China's LLM (large language model) start-up wave, Zhipu AI, Moonshot AI, MiniMax, Baichuan Intelligence, StepFun, and 01.AI, are all headquartered outside Guangdong.

Chinese officials climb on visible metrics: GDP growth, fiscal revenue, and headline projects. A ribbon-cutting on a tower occupied by a name-brand AI lab moves all three at once. Cultivating a frontier-research cluster, by contrast, takes a decade or more of university investment, faculty recruitment, and tolerance for failures that will not show up in the next five-year plan. A Shenzhen official who hands a free office building to an existing lab gets a measurable win this quarter. An official who tries to grow the same lab from a research-group seed gets nothing this quarter, and may not be in office when the seeds mature.

Per Chen, Shenzhen authorities tried to recruit two of the Six Little Tigers with a free office building. The deal fell through, with the source framing the courtship as rumor. Chen's reported reason: one of the two companies later dropped out of the race to build the most capable models, and the other was unlikely ever to leave Beijing. The recruitment pitch did not change the underlying geography. It just confirmed it.

That confirmation is itself a story. Two Sina Finance pieces in February 2025 pressed the same "AI three heroes all from Guangdong, none developed in Guangdong" argument as a critique of Guangdong industrial policy. The complaint was not really about AI. It was a way of asking why a province with the deepest hardware supply chain in the country, anchored by Shenzhen and a constellation of component makers, could not convert founder pipelines into headquarters pipelines.

The economic logic belongs to the Chinese economist Zhang Jun, cited in the same conversation: city-level competition, sustained by central-government incentives for local officials, has been a driving force in China's rapid industrialization. The Guangdong paradox is the same mechanism with the steering wheel in a different place. Local officials are not irrational when they chase existing headquarters. They are optimizing for the metrics they are graded on. The Beijing and Shanghai clusters that already exist are the cumulative result of decades of decisions like that one, compounded.

That is why the headquarters-recruitment playbook is a symptom, not a fix. The labs already sitting in Beijing and Shanghai benefit from a deep bench of cofounders, ex-ByteDance and ex-Baidu researchers, university pipelines, and a venture capital ecosystem that has co-located around them. A free tower in Nanshan does not undo that. The cost of switching headquarters is the network, not the rent.

The same pattern shows up in other Chinese industries built on the same incentive structure: biotech, EV batteries, semiconductors. A province can recruit the headquarters, or it can build the research gravity, but it cannot do both on a five-year timeline. Which one it picks tells you what its officials are actually being graded on.

The Guangdong paradox is not a Guangdong problem. It is a Chinese local-governance problem, wearing a province's name. Until the metrics that grade local officials include research-cluster cultivation, the next wave of frontier founders will follow the same route the last one did. The founders will leave Guangdong. The companies will open in Beijing or Shanghai. Shenzhen's headquarters-recruitment budget will keep producing ribbon-cuttings, not labs.