What Shenzhen Is Actually Doing About AI
A policy chain analysis from the ChinaGovernance corpus · 660 documents analyzed · 20 government sites
The question
When the State Council signals that artificial intelligence is a national priority, what happens next? Western coverage of Chinese AI policy tends to focus on central government announcements — the broad strokes. But in China's governance system, central directives are starting signals, not finishing lines. The real action is local: provinces, cities, and districts translating top-level ambitions into specific budgets, subsidies, and programs.
This analysis traces the AI policy chain in Shenzhen — from central directives down to district implementation — using 660 government documents from 16 departments and districts. The goal: make visible what a Chinese bureaucrat would intuit but the outside world cannot easily see.
The chain we found
The policy chain has four layers. At each layer, governments reference the layer above as justification and then add their own specifics:
The State Council and ministries set direction. Key documents include the Education Superpower Construction Plan (2024-2035) and the State Council's "AI+" action plan approved in August 2025.
Guangdong province channels central priorities into R&D funding. The provincial S&T department issues annual calls for "Next-Generation AI" research proposals, providing the funding pipeline that Shenzhen entities apply to.
Shenzhen has issued at least three major AI plans, each escalating in ambition:
- Next-Generation AI Development Action Plan (2019-2023)
- Accelerating High-Quality AI Development Action Plan (2023-2024) — cited by 2 district documents
- Accelerating the Building of an AI Pioneer City Action Plan — cited 3 times, the current flagship
- AI Pioneer City Action Plan (2025-2026) — the latest version
Key finding: None of these municipal plans are in our corpus. They're referenced by name from district documents but were published on a part of the Shenzhen portal we haven't crawled yet.
This is where policy becomes action. Districts translate municipal plans into budgets, subsidies, and specific programs. But the response is dramatically uneven.
The Longhua anomaly
Of the 16 departments and districts in our corpus, Longhua district produced 30 of the 89 AI-titled documents — more than a third. The next closest is the S&T Innovation Bureau (24, a municipal-level department), followed by the Development & Reform Commission (14). Most districts produced zero or one.
Longhua hasn't just talked about AI — it has issued three generations of formal policy:
| Year | Policy | Focus |
|---|---|---|
| 2021 | Measures to Promote AI Industry Development 深龙华府办规〔2021〕9号 | Public service platforms, incubators, R&D subsidies |
| 2024 (Feb) | AI Innovation Action Plan (2024-2025) | "Six-One" system: cluster, scenarios, platforms, policy, alliance, fund. |
| 2024 (Nov) | Measures to Promote AI and Robotics Industry Development 深龙华府办规〔2024〕20号 | Compute vouchers, algorithm registration rewards, embodied intelligence |
What the money reveals
The 2024 Longhua policy (深龙华府办规〔2024〕20号) contains specific subsidy amounts that reveal priorities:
- Compute vouchers: up to 3 million yuan per company (30% of procurement costs)
- Algorithm registration: up to 1 million yuan per algorithm registered with the Cyberspace Administration
- GenAI model filing: 500,000 yuan per model that passes national GenAI service filing
- Embodied intelligence R&D: 50% of R&D costs, up to 1 million yuan
- Application scenario matching: up to 4 million yuan for demand-side and 2 million for supply-side
- Public service platforms: up to 20 million yuan — the largest single item
- Standards development: 600,000 yuan for international standards, 400,000 for national
What's changed from 2021 to 2024
Comparing Longhua's 2021 and 2024 policies reveals how the AI landscape shifted:
- 2021 was about building foundations: incubators, service platforms, basic R&D support.
- 2024 is about deployment and governance: compute vouchers (GPU access is the bottleneck now), GenAI model registration, embodied intelligence.
- New in 2024: explicit references to "new quality productive forces" (新质生产力), Xi Jinping's signature economic phrase.
- The scope expanded from AI to AI + robotics — reflecting the national bet on embodied intelligence.
The silence of the other districts
If Longhua is the story of aggressive implementation, the other districts are a story of absence:
- Nanshan: 2 documents — one implementation plan from 2023
- Pingshan: 1 document — a draft subsidy measure under public comment in 2025
- Guangming: 1 document — a provincial leader visit mentioning AI in passing
- Other districts: zero AI-titled documents
What this tells us
- Chinese AI policy is not monolithic. The same central directive produces dramatically different local responses.
- The subsidy details reveal actual priorities. Central speeches say "AI is important." Longhua's 2024 policy says "we'll pay 500,000 yuan per GenAI model that passes national filing."
- Policy evolution is traceable. Comparing 2021-2024, you can see the shift from "build infrastructure" to "deploy and regulate." Local Chinese policy evolves in dialogue with global technology trends, not in isolation from them.