AI in China Update: National AI Development Plan Updated for 2026-2030 — Key Takeaways
China has released an updated National AI Development Plan (国家人工智能发展规划第二阶段, guojia rengong zhineng fazhan guihua di er jieduan) covering 2026–2030, which targets over 400 billion RMB ($55 billion) in core AI industry output by 2030, up from approximately 150 billion RMB in 2023. The plan outlines 14 priority action areas, including foundational models, autonomous systems, and AI governance, as China moves from catch-up to global leadership in artificial intelligence.
Why This Matters
For foreign executives, the updated plan signals where Beijing will direct the next wave of regulatory support, subsidies, and infrastructure spend—directly shaping market access, partnership viability, and technology transfer rules. Understanding the plan’s priorities helps you calibrate your China AI entry or expansion strategy for the second half of this decade.
China’s AI sector attracted $12.8 billion in venture funding in 2024, second only to the United States. The new plan aims to increase the number of AI-holding enterprises to 35,000 by 2030, up from roughly 20,000 at end-2024. The plan also calls for tripling the number of AI patents granted annually, reaching 8,000 by 2030, compared with 2,800 in 2023.
Key Updates in the 2026–2030 Plan
The updated plan pivots from earlier broad-based support to targeted domain dominance. Below are the six major shifts that foreign businesses must track.
| Area | 2021–2025 Plan | 2026–2030 Update | Implication for Foreign Firms |
|---|---|---|---|
| Foundational Models | Develop domestic LLMs (large language models) like Ernie and Tongyi | Require “world-class” LLM performance on 12 benchmark categories; mandate domestic training data for sensitive sectors | Data localization rules tighten; foreign LLM licensing subject to stricter scrutiny |
| Autonomous Systems | Pilot autonomous driving in 10 cities | Scale to 50 cities by 2028; target 70% “safe autonomy” in new industrial AGVs by 2030 | Opportunity for sensor/software suppliers; full vehicle autonomy still restricted for foreign OEMs |
| AI Chips | Encourage domestic substitutes for NVIDIA GPUs | Mandate “AI chip national certification” for government procurement; goal of 80% domestic chip use in new data centers by 2030 | Foreign chipmakers face reduced public-sector access; partnerships with domestic fabs become essential |
| Governance & Ethics | Voluntary guidelines on AI safety | Binding AI Safety Law expected by 2027; mandatory stress testing for “high-impact” AI applications | All AI products sold in China will need certification—compliance costs rise, but clear rules reduce regulatory uncertainty |
| Healthcare AI | Funding for diagnostic algorithms (radiology, pathology) | Expand to “full clinical decision support” with 500 approved AI medical devices by 2030 | Foreign device makers need NMPA registration; algorithm sharing will require onshore servers and Chinese data controllers |
| International Cooperation | Encourage joint research with Belt & Road countries | Emphasize “reciprocal AI cooperation” with OECD-aligned nations; “restricted AI technology export list” updated | Technology transfer approvals become more bilateral—joint ventures may be the only route for core AI IP |
7 Priority Action Areas for Foreign Executives
Based on the plan’s detailed annex, these seven areas will receive the most state-level resources.
- Multimodal foundational models—China aims to launch three “GPT-5 level” domestic models by 2027; expect heavy compute subsidies for local AI labs
- Autonomous manufacturing AGVs—target of 1.5 million connected AGVs by 2030, creating demand for edge AI, sensors, and fleet management software
- AI governance software—mandatory “AI explainability” and “adversarial robustness” testing creates a new compliance software market projected at ¥18 billion ($2.5 billion) by 2028
- Domain-specific AI datasets—auto, pharma, and finance are prioritized for national data-sharing platforms; foreign firms can join only via Chinese JV partners with data security clearance
- AI for energy grid optimization—target 15% efficiency gain in national grid by 2029; smart energy AI solutions will be fast-tracked for procurement
- Edge AI for consumer devices—smartphones, wearables, and home appliances must meet new “AI-embedded” standards by 2030; foreign OEMs must adapt designs
- AI talent development—target to train 50,000 “AI engineers with domain expertise” annually by 2028; foreign talent recruitment eased for registered WFOEs (外商独资企业, waishang duzi qiye) in designated AI zones
Pitfalls to Navigate
Data Localization Deepens
The updated plan explicitly ties AI model training to “secure data circulation” within China. Any AI product sold in China—whether software, hardware, or service—must be trained or fine-tuned on data that stays within Chinese borders. Cross-border data flows require a new “AI data cross-border security assessment” that adds 6–9 months to approval timelines. Foreign firms that rely on global model architectures should prepare to deploy fully onshore training pipelines, including domestic GPU clusters, to comply.
IP and Technology Transfer Risks Escalate
With the updated “restricted AI technology export list,” Beijing can now control the transfer of algorithms for foundational models, autonomous driving, and AI chip design. Joint ventures may face clauses that require transferring core AI IP to the Chinese partner after a certain period. Foreign executives should negotiate exit provisions and IP valuation caps upfront, and consider using a WFOE structure for AI R&D to retain control over the technology stack under PRC law.
Certification Bottlenecks
The plan’s mandatory AI safety certification, expected to be enforced from 2027, will apply to any “high-impact” AI system—defined as those used in healthcare, autonomous driving, finance, and critical infrastructure. Certification requires testing by state-accredited labs, which currently have limited capacity. Early estimates suggest a backlog of 1,200–1,500 applications in the first year. Foreign companies must plan for a 9- to 12-month certification cycle and factor that into product launch timelines.
Subsidy Access Conditioned on Localization
While the plan offers generous subsidies—up to 40% of R&D costs for priority AI areas—these are available only to enterprises registered in China. Foreign WFOEs may qualify, but only if they meet “local content” requirements: at least 70% of the AI supply chain must be domestic by value. This includes chips, data storage, and even some software libraries. Firms that rely heavily on non-Chinese AI components (e.g., NVIDIA GPUs, AWS cloud, certain open-source frameworks) may find themselves excluded from subsidies unless they pivot to domestic sourcing.
Opportunities Worth Pursuing
AI Safety and Compliance Consulting
The binding AI Safety Law will create a compliance gap. Fewer than 50 domestic firms are currently accredited to perform AI stress testing, and nearly all state-owned enterprise (SOE) clients prefer vendors that have both domestic and international experience. Foreign consultants with expertise in AI auditing (e.g., ISO/IEC 42001, EU AI Act alignment) can command premium rates in China—if they partner with a local testing lab. This is a low-capital, high-margin entry point.
Edge AI Components for Consumer Devices
The “AI-embedded” standard for consumer electronics, due by 2030, will require tens of millions of edge AI processors, sensors, and middleware stacks annually. Foreign semiconductor and sensor companies that can offer low-power, high-efficiency edge AI modules—and that localize design and assembly in China—can capture significant market share. The plan designates five special economic zones (Shanghai Lingang, Shenzhen, Hefei, Chengdu, Beijing Zhongguancun) as fast-track hubs for such component certification, reducing time-to-market from 24 to 12 months.
Smart Factory AGV and Fleet Management Software
China’s plan to deploy 1.5 million connected AGVs by 2030 creates a software-defined automation market worth an estimated ¥120 billion ($16.5 billion). Foreign software firms specializing in fleet orchestration, predictive maintenance, and V2X communication can enter via a WFOE and license to domestic AGV manufacturers. The key is to offer modular software that integrates with major Chinese AGV platforms (Geek+, Quicktron, Hai Robotics) and to host all data onshore. This avoids the IP transfer risk of hardware joint ventures while still tapping state subsidies for smart manufacturing.
Timeline to Watch
The plan includes specific milestones that will affect your decision-making.
- 2026 Q1—Detailed budgets for each priority action area released to provincial governments; first round of AI Safety Law draft for public comment
- 2026 Q3—Updated restricted AI technology export list takes effect; AI data cross-border security assessment regime begins
- 2027 Q2—Binding AI Safety Law enacted; mandatory certification for high-impact AI systems begins (phased in over two years)
- 2028—50-city autonomous driving expansion complete; AI governance software market reaches ¥18 billion
- 2030—Core AI industry output hits ¥400 billion; 80% domestic chip use in new AI data centers; 500 approved AI medical devices
Where to Go From Here
Decision-path 1: Enter via AI compliance consulting or edge AI components. If your firm has expertise in AI testing, certification, or edge hardware, the lower capital requirement and faster certification pathways make these the lowest-risk entry points. Form a WFOE in one of the five designated AI hubs, partner with a domestic testing lab, and target SOE clients or consumer electronics OEMs.
Decision-path 2: Form a joint venture for foundational models or healthcare AI—with strict IP safeguards. For large AI players, the plan’s subsidies and data access effectively require a Chinese partner. Negotiate a minority-equity JV (49% or less) with explicit IP exit provisions, and keep your foundational model training in a separate onshore entity under WFOE control. Invest in domestic GPU capacity to qualify for the 40% R&D subsidy, but avoid transferring core algorithm code to the JV.
Decision-path 3: Wait and monitor for autonomous systems or smart energy—but prepare a local sandbox unit now. If your technology targets autonomous driving, AGVs, or energy grid AI, the plan’s full impact won’t materialize until 2028–2030. However, early access to the 50-city pilot program requires a registered entity in China by end of 2026. Set up a small WFOE focused on field trials and data collection, and submit joint proposals with a Chinese partner to gain pilot slots. This keeps your options open while deferring large-scale investment until certification timelines become clearer.
