On 18 December 2024 Shenzhen’s industry and information technology bureau issued 18 measures to build Shenzhen into an AI pioneer city, carried as Shen Gong Xin Gui [2024] No. 13. The package combines computing supply, model development, application support and ecosystem building — a full industrial-policy stack rather than a single incentive.
What the 18 measures do
The measures work on several layers at once. They address computing resources, since access to training and inference capacity remains the binding constraint for most Chinese AI companies. They support model development and open-source work. They promote applications across industry. And they build out the surrounding ecosystem — platforms, services and talent.
That layering is deliberate. A city that subsidises only models, without computing or demand, produces demos rather than businesses. The later subsidy rules and the government AI scenario fund show how the package is operationalised once the direction is set.
The competition between cities
Shenzhen’s move should be read alongside what other cities are doing. Shanghai’s AI application subsidy targets the same demand-side logic, and the national AI application service provider action plan sets the framework that local programmes elaborate. For AI companies, the practical consequence is that location decisions now have measurable financial consequences, and the differences between cities are in the detail of eligibility, not in the general direction. A founder comparing Shenzhen and Shanghai is really comparing two application menus, two computing arrangements and two sets of paperwork.
Practical actions for AI companies
First, match your business to the measures before choosing a site. Subsidy programmes reward specific activities — training runs, application projects, platform building — and a company whose work falls outside those categories gains little from the headline.
Second, document eligibility as you go. Local programmes typically require evidence of local operation, project milestones and measurable outcomes; assembling this retrospectively is slow and often incomplete.
Third, think about computing and data access, not only cash. The measures’ supply-side elements may matter more to a growing model company than the grant itself.
Finally, remember that incentives do not displace compliance. Data handling, security review and filing obligations apply in Shenzhen exactly as elsewhere in China, and a subsidy that funds a project which later stalls on a filing requirement is not a bargain. Grant money is not a substitute for a launch plan that already clears the regulatory baseline.
This article is based on Shenzhen’s 18 measures for building an AI pioneer city (Shen Gong Xin Gui [2024] No. 13) and public information and is for reference only; it does not constitute legal advice.