A city government that once measured success by how many algorithms it funded now asks a harder question: which real-world problems did those models actually solve? Beijing's answer was to stop scattering disconnected pilot projects and instead build a small number of flagship applications where the whole city becomes the testbed.
The instrument is the Beijing Municipal "AI+" Action Plan (2024–2025) (《北京市推动"人工智能+"行动计划(2024—2025年)》), released by the Beijing Municipal Development and Reform Commission (北京市发展和改革委员会) in July 2024. Its headline ambition is to "build an AI-native city" (建设 AI 原生城市) first, positioning Beijing as a globally influential source of AI innovation (人工智能创新策源地).
Key takeaways
- The plan is a formally adopted and implemented municipal action plan (生效实施, July 2024), with a stated run through the end of 2025.
- It organizes AI adoption into three tiers: 5 flagship application projects, 10 demonstration projects, and a broad set of commercialized applications.
- Concrete 2025 targets: 3–5 usable, self-controlled foundational large models, 100 industry large models, and 1,000 industry success cases.
- Beijing Economic-Technological Development Area (北京经济技术开发区, "E-Town") operationalizes the idea with vouchers and a dedicated "AI city" plan.
- The model is exportable: a city can coordinate supply (compute, data, talent) and demand (scenarios) rather than fund models in isolation.
- Stage caveat: the plan's named term has closed; treat it as a concluded policy whose design remains a reference template.
What the plan actually does
Most AI industrial policy hands money to model builders and hopes applications appear. Beijing inverted the sequence. The plan starts from scenarios—robotics, education, healthcare, culture and transport—and works backward to the data, compute and models those scenarios require.
The five flagship application projects (标杆型应用工程) are meant to be "benchmark" in the literal sense: each is a large, cross-sector engineering effort with measurable public value. The ten demonstration projects (示范性应用项目) cover research, government services, industrial intelligence, finance, spatial computing, digital marketing, justice, broadcasting, power and content safety. Beneath them sits a long tail of commercialized applications the city commits to promote.
At the launch press conference, officials from the Beijing Municipal Development and Reform Commission (北京市发展和改革委员会) framed the rollout around six safeguard areas—computing power, data, algorithms, capital, platforms and talent—rather than around any single technology. That framing matters: it treats AI adoption as an urban-operations problem, not a research-grant problem.
The benchmark application engineering model
The distinctive mechanism is what the plan calls joint R&D platforms (联合研发平台) for application scenarios. Instead of each agency building its own stack, the city aggregates high-quality industry datasets, opens scenario demand, and provides a shared development environment where model firms and users co-develop.
Consider healthcare. The plan proposes a "Beijing Doctor" medical large-model platform (北京医生医疗大模型平台) that links hospital, doctor and model-company incentives, builds a trusted training space for medical data, and explores doctor–agent collaboration for triage, online consultation and chronic-disease management. The point is not a cooler chatbot; it is a governed pathway for sensitive data to improve care without leaving a compliant boundary.
Education, transport and culture follow the same logic: the city supplies the coordinating layer, the data rules and the demand signal; vendors and institutions supply the models.
Beijing E-Town: a citywide AI testbed
The clearest concrete instantiation is in E-Town (亦庄), south of the city center. Its Implementation Plan for Building a Citywide AI City (2025) (《北京经济技术开发区关于加快建设全域人工智能之城的实施方案(2025)》) commits no less than ≈ US$141 million / HK$1.09 billion in special policy funding, delivered through three ≈ US$14 million / HK$109 million voucher pools—for computing power (算力券), models (模型券) and data (数据券).
E-Town also pledges to open 100 scenarios and 10,000 quality jobs, and set 2025 targets of at least 600 core AI enterprises and an industry scale above ≈ US$11.27 billion / HK$87.4 billion. The mechanism is deliberately hands-on: a "scenario innovation partner" program matches enterprises with real deployment sites.
Why this matters beyond China
For observers outside China, the interest is not the headline numbers but the governance choreography. Beijing shows one way a sub-national government can act as a demand-side orchestrator—pooling procurement, data access and compliance scaffolding so that smaller firms can reach production-grade deployments faster than they could alone.
It also reframes "AI city" away from surveillance tropes and toward public-service delivery: transport, healthcare, education, culture. Whether the model scales is a separate question (see limitations), but the template—flagship scenarios, shared platforms, voucher-style demand stimulation—is legible to any metropolitan government weighing how to turn compute investment into visible outcomes.
Honest limitations
- The plan's own timeline (2024–2025) has closed; this article analyzes its design, not verified end-of-term achievement figures, which the city has not consolidated in a single public scorecard we could confirm.
- Targets such as "100 industry large models" are plan commitments, not audited results. We did not find independent verification that every flagship project met its stated benchmark.
- The model depends heavily on a capable local administration and dense enterprise base; it may not transfer cleanly to regions without Beijing's research and capital concentration.
- Voucher and subsidy mechanisms create dependency risk if not paired with exit criteria, a point the plan itself does not address in detail.
What readers can do now
- Track the Beijing Municipal Development and Reform Commission (北京市发展和改革委员会) and E-Town releases for any 2026 successor framework before treating the 2024–2025 plan as current.
- If you advise a city government, map the three-tier (flagship / demonstration / commercial) structure onto your own procurement and data-access levers.
- Watch whether E-Town publishes audited outcomes for its voucher programs—those numbers will be the real test of the model.