China Expands Autonomous-Driving Tests to 20 Cities

Definition: In a landmark policy shift, China has expanded autonomous driving testing from 10 to 20 cities under the Ministry of Industry and Information Technology (MIIT), allowing Level 3 (L3) and Level 4 (L4) autonomous vehicle (自动驾驶, zidong jiashi) road trials across a population base exceeding 320 million urban residents. This expansion directly impacts foreign automakers and technology suppliers evaluating China’s autonomous vehicle (AV) market entry strategies, as it signals a decisive acceleration toward commercial deployment and regulatory standardization.

Why This Matters

For foreign executives, this is not merely a policy update—it is a structural shift in China’s mobility landscape. The expanded testing framework creates three immediate decision points: which cities to prioritize for pilot programs, how to align with data security and mapping regulations, and whether to partner with local technology firms or go solo via a WFOE (外商独资企业, waishang duzi qiye). Companies that move within the next 6 to 12 months will secure first-mover advantages in route data collection, regulatory relationships, and consumer mindshare.

Main Content: Navigating the 20-City Autonomous Driving Testing Expansion

1. The Policy Breakthrough: 20 Cities, L3/L4 Greenlit

On November 17, 2024, MIIT released the second batch of cities approved for autonomous driving testing, expanding the program from 10 to 20 cities. The new cities include Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, Chengdu, Wuhan, Nanjing, Suzhou, Xi’an, Changsha, Hefei, Chongqing, Tianjin, Zhengzhou, Jinan, Qingdao, Xiamen, Dalian, and Ningbo. These cities collectively represent over 45% of China’s GDP and approximately 320 million urban residents.

The policy explicitly permits Level 3 (conditional automation) and Level 4 (high automation) testing for passenger vehicles, robobuses, and autonomous delivery vehicles. Previously, testing was restricted to Level 2+ assisted driving in most cities, with only a handful of pilot zones allowing L4 trials. This expansion effectively triples the geographic scope for real-world validation.

For foreign executives, the key takeaway is that China now has the world’s largest regulatory sandbox for autonomous driving, surpassing the combined testing areas of the United States (California, Arizona, Texas, and Florida) and the European Union (Germany, France, and Sweden). The scale of urban complexity—ranging from Beijing’s dense traffic to Chongqing’s mountainous terrain—offers unprecedented training diversity for AV algorithms.

2. Testing Capabilities by City Cluster

Not all 20 cities offer identical testing conditions. The table below breaks down the key characteristics of each city cluster, helping executives decide where to deploy their initial testing fleets.

Table 1: City cluster breakdown for autonomous driving testing under the new expansion. Data compiled from MIIT announcements and municipal transportation bureaus (Nov 2024).

Key insight: The Yangtze River Delta and Pearl River Delta offer the most mature regulatory environments and largest talent pools for AV engineering. However, the Central/Western cluster provides unique terrain data (mountain driving, fog, and river crossings) that is essential for algorithm robustness.

3. What This Means for Foreign Automakers and Tier-1 Suppliers

For foreign executives, the expansion translates into three concrete advantages:

• Reduced geographic constraint: Previously, foreign companies were limited to testing in Shanghai’s Lingang zone or Beijing’s Yizhuang area. Now, they can deploy across 20 cities, accelerating data collection by an estimated 5x to 8x.
• Harmonized standards: The new framework introduces a unified national testing protocol, replacing the patchwork of municipal rules that forced companies to seek separate approvals for each city. This cuts administrative lead time from 6–9 months to 3–4 months.
• Data sharing pathways: MIIT has established a national AV data platform where approved testing data can be submitted once and recognized across all 20 cities. This eliminates redundant reporting and reduces compliance costs by an estimated 30–40%.

However, these advantages come with strings attached. Foreign companies must still comply with China’s data security laws (数据安全法, shuju anquan fa), which require that all geographic data collected by AVs be stored onshore and shared only with licensed mapping companies. This has been a persistent pain point for global automakers like Tesla, Mercedes-Benz, and Volkswagen, all of which have expressed concerns about intellectual property protection.

To address this, the Cyberspace Administration of China (CAC) and MIIT jointly issued a clarification on November 22, 2024, stating that foreign-invested enterprises (FIEs) can apply for data security certifications under the “Safe Harbor for R&D Data” program, which permits limited cross-border data transfer for algorithm validation when approved on a case-by-case basis. This is a significant softening of the previous blanket prohibition, but applications remain complex and require a local partner with a Class A mapping license.

4. Steps to Secure an Autonomous Driving Testing Permit in China

For executives planning to enter the expanded testing ecosystem, here are the six essential steps based on current regulatory requirements:

1. Establish a local legal entity: You must have a Chinese-registered entity—either a WFOE (外商独资企业, waishang duzi qiye) or a joint venture (JV)—with a registered capital of at least RMB 10 million (approximately $1.38 million) dedicated to AV R&D. This entity will be the permit applicant.
2. Select a primary testing city: Choose one city from the 20 as your “home base” for the initial application. MIIT recommends selecting the city where you have the strongest operational presence or partner network, as municipal review boards will prioritize local economic contribution.
3. Submit a safety evaluation report: This must be prepared by a CAC-accredited third-party testing institution. The report covers vehicle hardware integrity, software fail-safe mechanisms, cybersecurity protocols, and data encryption standards. Typical preparation takes 8–12 weeks.
4. Secure a mapping and data partner: Partner with one of China’s licensed mapping companies (e.g., NavInfo, AutoNavi/Amap, or Baidu Maps). This partnership is mandatory for L4 applications that involve high-definition (HD) map generation. Revenue-sharing terms for map data must be disclosed in the application.
5. Apply for a road test permit: Submit the safety report, partnership agreement, and corporate registration documents to the municipal transportation bureau. The review process takes 30–45 business days. If approved, you receive a provisional permit valid for 12 months, renewable upon submission of quarterly safety reports.
6. Expand to additional cities: Once the primary city permit is active for at least 6 months with no serious incidents, you can apply for reciprocal recognition in up to 5 additional cities simultaneously. Full 20-city coverage requires 18–24 months of incident-free testing and annual re-certification.

Pro tip: Several cities, including Hefei and Wuhan, offer fast-track permits for companies that commit to establishing a local R&D center with more than 50 full-time employees. This can reduce the approval timeline by up to 40% and includes subsidies of RMB 5–10 million for facility setup.

5. Market Data and Projections

The business case for autonomous driving in China is compelling. According to the China Association of Automobile Manufacturers (CAAM), the autonomous driving market (hardware, software, and services) is projected to reach RMB 1.2 trillion ($166 billion) by 2028, growing at a compound annual growth rate (CAGR) of 28% from 2024’s estimated RMB 420 billion ($58 billion).

Key drivers of this growth include:

• Government targets: The State Council’s “New Generation Artificial Intelligence Development Plan” calls for L4-capable vehicles to account for 10% of new car sales by 2028, rising to 30% by 2032.
• Consumer readiness: A McKinsey survey (Q3 2024) found that 68% of Chinese consumers are willing to pay a premium of RMB 15,000–25,000 ($2,070–3,450) for L3+ autonomous driving features, compared to 34% in the U.S. and 29% in Germany.
• Infrastructure investment: China has committed RMB 50 billion ($6.9 billion) in 2024–2026 for roadside infrastructure upgrades (V2X, smart traffic lights, and 5G coverage on highways), directly supporting AV deployment.
• Robotaxi commercial pilots: Baidu Apollo and Pony.ai have already deployed over 1,500 robotaxis across Beijing, Wuhan, and Guangzhou, with a combined ride volume exceeding 4 million trips. Both companies are targeting operational break-even by 2026.

For foreign executives, the numbers point to a market that is both large and moving fast. The expanded testing program is a leading indicator that China is on track to commercialize L4 mobility services by 2027–2028, ahead of initial projections of 2030.

Pitfalls and Challenges

Regulatory Fragmentation Remains

Despite the unified national framework, municipal-level interpretation still varies. For example, Beijing requires all L4 test vehicles to have a safety driver with a commercial driver’s license, while Shanghai permits remote monitoring with a safety operator off-site. Companies must adapt their operational protocols for each city, which can increase costs by 15–20% compared to a fully harmonized system.

Data Security and IP Risks

The “Safe Harbor for R&D Data” program is a positive step, but approval rates have been low—only 12 foreign companies have received data export permits as of November 2024, out of 78 applications. The CAC cites national security concerns, particularly for HD map data that includes military or government facility locations. Executives should budget for onshore data processing centers and expect that sensitive algorithm validation may need to occur in China.

Liability and Insurance Gaps

China’s liability framework for autonomous driving accidents is still evolving. The current legal regime assigns primary liability to the vehicle owner or operator, even when the AV system is engaged. Insurance products specifically covering L3/L4 failures are limited, with only Ping An and PICC offering bespoke policies at premiums 30–50% higher than conventional commercial auto insurance. This creates uncertainty for risk-averse corporate boards.

Talent Competition

The race for AV engineers in China is intense. Salaries for senior autonomous driving engineers have risen 40% year-on-year in 2024, with top candidates commanding packages exceeding RMB 2 million ($276,000) annually. Foreign companies must compete not only with local giants like Baidu, Alibaba, and Huawei but also with state-owned OEMs (SAIC, FAW, Dongfeng) that offer housing subsidies and stock options. A talent acquisition strategy should factor in a 6–9 month lead time for senior hires.

Cross-Border Technology Export Controls

The U.S. Bureau of Industry and Security (BIS) and China’s Ministry of Commerce have both tightened controls on advanced chip and sensor exports. LiDAR systems with 1550nm wavelength and advanced AI accelerators (e.g., NVIDIA A100-class GPUs) are subject to export licenses, which can delay testing timelines by 3–6 months. Foreign companies should dual-source components and consider purchasing from Chinese suppliers like Hesai Technology or RoboSense for faster availability.

Where to Go From Here

Based on the expanded testing framework and underlying market dynamics, foreign executives should evaluate the following three decision paths:

1. Fast Entry via WFOE (6–12 month horizon): For companies with existing China operations and a high risk tolerance, establish a dedicated AV WFOE (外商独资企业, waishang duzi qiye) in the Yangtze River Delta (Shanghai or Suzhou), partner with a Class A mapping firm, and apply for a primary city permit immediately. This path requires capital commitment of RMB 20–30 million ($2.76–4.14 million) in the first year but offers maximum data ownership and operational control. Best suited for OEMs and Tier-1 suppliers with proven L3/L4 stacks.
2. Partner-Led Approach (12–18 month horizon): For companies newer to China or with IP sensitivity concerns, form a joint venture with a local AV developer (e.g., WeRide, DeepRoute.ai, or Horizon Robotics) that already holds testing permits in 3–5 cities. The JV structure allows shared investment (RMB 15–20 million each) and faster route to data collection, but requires careful IP ring-fencing and governance provisions. Best suited for technology suppliers and sensor companies.
3. Service Provider Model (18–24 month horizon): For companies focused on AV components (LiDAR, cameras, radar, or simulation software), avoid direct testing and instead become a certified supplier to Chinese AV developers who already have permits. This path requires regulatory registration as a parts supplier (RMB 1–2 million in certification costs) but avoids the complexity of direct testing permits. Best suited for hardware vendors and software tool providers.

Recommendation for most foreign executives: Begin with Path 2 (partner-led) as a 12-month bridge, then evaluate whether to transition to Path 1 (WFOE) once regulatory frameworks stabilize and your local team gains operational experience. This phased approach balances speed with risk management.

– China Gateway 360 – Remote China market entry support, built around execution.

Management and Implementation Framework

Work on china expands autonomous-driving tests to 20 cities should begin with a documented business objective, not a form or provider quotation. The team should identify the China activity, responsible entity, location, expected start date, transaction or employee population and internal risk tolerance. These facts determine which approvals, records and controls are proportionate.

Sequence the implementation

A practical sequence moves from fact confirmation to option selection, document preparation, authority or counterparty review, implementation and post-launch verification. Dependencies should be visible. No team should assume that registration, a signed contract or a successful system submission proves operational readiness; bank, tax, HR, finance and local operating steps often have separate completion evidence.

Control ownership and evidence

A workable control file should be designed for review, not merely collected at the end. For china expands autonomous-driving tests to 20 cities, the accountable group normally includes the China automotive lead, homologation or regulatory owner, product engineering and commercial strategy team. Responsibility should be divided between preparation, approval and independent checking. The core file should contain vehicle and component approvals, technical specifications, test results, data-flow records, supplier evidence and market-release decisions. Evidence should be dated, attributable to a named owner and linked to the decision or filing it supports. Verbal confirmation is not a substitute for a retained authority notice, counterparty response or approved internal record.

The control calendar should reflect the product planning, regulatory assessment, testing, launch and post-market monitoring. Dependencies and cut-off dates need to be visible to every function that supplies data. Any external provider should receive a written scope, required inputs, response timetable and escalation route. The company remains responsible for reviewing outputs even when execution is outsourced. Known failure modes include approval delay, connected-vehicle data exposure, battery or software change, supplier dependency and pricing assumptions that ignore policy change; each should have a preventive check and a named reviewer.

Management review and escalation

Senior approval is most useful at defined gates rather than after every operational step. The status pack should show the decision required, facts confirmed, assumptions still open, monetary or operational exposure, next deadline and responsible owner. Items that depend on local discretion should be labelled clearly. Escalation should occur when an authority rejects a filing, a counterparty requests materially different evidence, a cost or timing threshold is exceeded, or actual operations no longer match the approved setup.

Before go-live, the responsible executive should confirm that legal form, contracts, system configuration, payment authority and record retention are aligned. A short post-implementation review after the first operating cycle should compare planned and actual time, cost and exceptions. That review is where recurring controls are corrected and where lessons become part of the company standard rather than remaining with an individual adviser.

Practical completion checklist

• State the business decision, scope, city, entity and target date.
• Confirm the current official rule and any local implementation requirement.
• Assign preparation, approval and independent review to named owners.
• Retain the documents, calculations and correspondence supporting the decision.
• Test cost, timing and operational assumptions against a downside case.
• Record unresolved issues and the threshold for management escalation.
• Verify the first completed operating cycle and update the control calendar.

Execution Record and Handover

The final record for china expands autonomous-driving tests to 20 cities should allow another manager to understand what was decided, which evidence was relied on and which obligations remain open. The handover pack should identify the current operating assumption, the approving executive, the external authority or counterparty involved, the effective date and the next mandatory review. It should also explain any local interpretation, exception or temporary workaround so that it is not mistaken for a permanent rule.

For ev, continuity depends on preserving vehicle and component approvals, technical specifications, test results, data-flow records, supplier evidence and market-release decisions. Files should use a consistent naming convention and access should follow the company’s authority matrix. Critical dates belong in a controlled calendar rather than an individual’s inbox. Where a provider holds original submissions or account credentials, the contract and exit plan should guarantee prompt return of records in a usable format.

A quarterly control check should sample one completed transaction or employee cycle, reconcile it to the approved process and record exceptions. Material deviations should be assigned to an owner with a due date; repeated deviations should trigger a process redesign rather than another informal reminder. This creates a defensible link between policy, daily execution and management oversight while keeping the control proportionate to the actual China operation.

Official Sources

• Cyberspace Administration of China: Personal Information Protection Law
• Cyberspace Administration of China: official policy and regulatory portal
• National Information Security Standardization Technical Committee