On 30 July 2024 the General Office of the Shenzhen Municipal Party Committee and the General Office of the Municipal Government issued the Action Plan for Accelerating the Building of an AI Pioneer City, containing 22 measures. The plan sets out five fronts — full-stack innovation, intelligent products, cross-border data, scenario applications and intelligent driving — and a target for intelligent compute capacity in 2024. For AI companies in Shenzhen, it places industrial policy, factor supply and compliance arrangements in one document.
Five fronts, five parts of the value chain
Full-stack innovation points at the chain from chips and frameworks to models and applications, with an emphasis on self-reliance and coordination. Intelligent products point at devices and hardware, where volume and market access matter. Cross-border data addresses factor flows: how the data needed for training and operations moves in and out. Scenario applications address sector deployment. Intelligent driving addresses intelligent connected vehicles as a distinct category.
The five are not parallel slogans; they are different combinations of policy tools. Full-stack innovation relies on research spending and ecosystem coordination. Intelligent products rely on supply chains and channels. Cross-border data relies on rule-based pilots. Scenario applications rely on scenarios opened by government and state-owned enterprises. Intelligent driving relies on testing and admission. A company should first locate itself on that chain, then decide which department to approach.
The limits of the cross-border data front
“Pioneer” is easily misread as deregulation. Policy support for cross-border data means facilitation mechanisms and pilot arrangements, not an exemption from the existing outbound-transfer regime. Security assessment, standard contracts and certification remain available routes, and the choice depends on the nature and volume of data, the recipient and the purpose.
The first task is to draw the data flow: where training corpora come from, where inference data is processed, where logs and model outputs are stored, and whether overseas customers can access them. Only with that map can a company tell which steps are outbound transfers and which are merely remote access, and whether anonymisation, pseudonymisation or local deployment can reduce the burden.
Intelligent driving and admission
Intelligent driving carries the densest regulation of the five fronts. Road testing, demonstration application, product admission and road operation each have requirements touching vehicles, maps, data recording and cybersecurity. Policy support expands testing scope and room for innovation; it does not replace admission conditions.
Features involving data captured outside the vehicle and personal information require the collection scope and retention period to be fixed at the design stage. High-precision maps and positioning bring separate surveying and mapping requirements. Treating intelligent driving as ordinary software usually means discovering qualification and data problems only at the testing stage.
From a compute target to company action
The plan’s 2024 target for intelligent compute capacity concerns supply. More compute lowers training and inference costs, but it does not mean a company automatically obtains resources; procurement, energy, facilities and supply chains still need separate arrangements.
When applying for policy support, project ledgers, contracts and invoices, IP ownership and security-assessment materials usually need to corroborate one another. Using the policy document in fundraising also requires converting “aligned with the direction” into “meets the conditions”, and that conversion rests on evidence rather than wording.
For data-flow mapping and outbound-transfer routes, see our data compliance service; for intelligent driving and product admission, see business licensing; for filing before launching a generative service, see algorithm filing.
This article is based on the Action Plan for Accelerating the Building of an AI Pioneer City issued by the General Office of the Shenzhen Municipal Party Committee and the General Office of the Municipal Government on 30 July 2024, and on public information. It is for reference only and does not constitute legal advice.