Gasgoo Munich- Data from the China Association of Automobile Manufacturers shows that auto exports reached 4.059 million units in the first five months of this year, a 63% increase year-on-year. Focusing on new energy passenger vehicles, exports hit 1.792 million units over the same period—more than double the previous year. For the full year of 2025, China's auto exports had already shattered the 7 million mark, reaching 7.098 million units. China has now held the title of the world's largest auto exporter for three consecutive years.
Riding alongside the surge in export volumes is the penetration rate of Chinese smart driving. According to Gasgoo Automotive Industry Big Data Platform, the penetration rate of new passenger cars equipped with standard L2+ or higher features reached 28% in 2025. By April 2026, that figure had climbed past 41%.
Image Source: BYD
As "Chinese cars" become a common sight on global roads, can "Chinese smart driving" keep pace? With L2+ technology rapidly proliferating at home, is exporting smart driving becoming the new growth engine? The answer seems to be yes. But the real questions remain: How? At what pace? Should manufacturers sell the hardware first and upgrade software later, or deliver full capabilities from day one?
The First Hard Threshold for Exporting Smart Driving
Chen Weicheng, Senior Customer Success Manager for TomTom Greater China, put it bluntly at the 9th Intelligent Driving and Overseas Expansion Conference: "When Chinese smart driving goes global, the first thing it hits is the wall of compliance." It's not that the technology fails; it's that the rules won't allow it.
The EU's compliance system is a layered "pyramid." At the base is the framework established by the United Nations Economic Commission for Europe (UNECE), covering over 60 countries including the EU, South Korea, Australia, and Japan. On top of that, the EU imposes stricter General Safety Regulation (GSR) requirements. Above that lie the national laws of individual member states.
Image Source: Speaker Materials
On a technical level, four key regulations set the entry bar for the European market: UN R79 (steering systems, covering L1/L2 lane keeping); ELKS (Emergency Lane Keeping, mandatory in the EU since 2021); UN DCAS (Driver Control Assistance Systems, targeting L2+ scenarios and directly related to the NOA functions mass-produced by Chinese automakers); and UN R157 (Automated Lane Keeping Systems, the L3-level "hands-off, eyes-off" ticket to the highway).
Zhang Wei, Senior System Director at Knowhow, shared insights from the field: "In the EU, you have R79 for lateral functions, R152 for AEB, and R48 for headlights. The toughest part is the one requiring the most real-world verification: the Intelligent Speed Assist (ISA) requirements under the EU GSR framework, corresponding to UNECE R152 and related technical regulations." ISA demands vehicles accurately identify speed limit signs across different regions and respond accordingly. That sounds simple, but achieving high accuracy in Europe's multi-lingual traffic sign ecosystem requires massive on-road testing—which hits another wall: data compliance.
Regulations are evolving rapidly. Over the next three to five years, laws governing L4 autonomous driving will accelerate. The EU AI Act has incorporated AI into the automotive safety system—meaning algorithms themselves will face scrutiny, not just the final results. From 2021 to 2030, key regulatory milestones are taking effect almost every year.
Huang Luoyi, Director of Intelligent Driving Toolchains and Infrastructure at Bosch China, illustrated the complexity of regulations with a personal anecdote: spotting a traffic sign near Frankfurt Airport that he failed to recognize twice, nearly causing an accident. "Functional regulations aren't as simple as everyone thinks," he said. If a veteran driver of ten-plus years can misread a sign, what chance does an algorithm have?
Compliance is a wall that cannot be bypassed. For Chinese automakers, the question isn't whether they can climb over it, but how—and at what cost.
Localized Training Is a "Non-Negotiable Bottom Line"
If compliance is a "visible wall," then data is an "invisible abyss."
Huang Ziliang, an intelligent driving solution expert at Huawei, pointed out the crux of the data issue during a roundtable: "Data used to train smart driving large models certainly cannot leave the country where the cars are sold. In the future, if we go to Japan, Japanese traffic data must be trained in Japan; in Europe, it must be trained in Europe." This isn't a technical choice; it's a legal red line.
The GDPR (General Data Protection Regulation) is the most typical representative. Huang Luoyi noted that GDPR clearly stipulates: fines for general violations can reach 10 million euros or 2% of global turnover (whichever is higher); for serious violations, 20 million euros or 4%. "Small companies might not care, but for Bosch, we can't afford to lose."
An even bigger challenge comes from the "opt-in" rules of data privacy—some countries require explicit consent from subjects before collecting or processing personal data. Huang Luoyi asked rhetorically: "I go out to collect data—how do I get consent from pedestrians and car owners on the road? I can't do that."
This leads directly to a widespread phenomenon. Chen Yichi, CTO of KPIT China, shared a key observation during the roundtable: "Last year we received many projects for collecting data overseas, which were anonymized and compliantly brought back. But in the end, the data portion was often cut. The priority became selling the cars first, with smart driving and ADAS features either disabled overseas or only partially activated."
Chen Yichi's conclusion: "Technically, there's nothing that difficult; you just do it bit by bit. It's mainly that the cost is indeed not low." With domestic smart driving yet to achieve a commercial closed loop, spending heavily on overseas expansion "doesn't seem particularly wise at this stage."
Huang Ziliang proposed a solution: a supplier-shared data center model. A supplier builds a data center in Japan, and the trained base model can serve multiple automakers. Du Jianning, Head of JetBrains' Smart Vehicle Business in China, suggested "abstracting a public layer" while processing upper-layer sensitive data locally.
In February 2026, eight departments including the Ministry of Industry and Information Technology jointly issued the "Guidelines for the Security of Cross-Border Transfer of Automotive Data (2026 Edition)," clarifying management methods, rules for determining important data, and security protection requirements. Policy is opening up, but "blockages" in overseas compliance remain.
Data localization is not an option; it is a mandatory question. And the answer to that question directly determines how far Chinese smart driving can go abroad.
When Chinese Algorithms Meet Overseas Roads
Compliance is the threshold, and data is the foundation, but what truly determines user experience is the algorithm's ability to adapt to local scenarios.
Zhang Wei offered a vivid comparison: "Parking spaces in China are very standard, but overseas, many spots might not be recognized, or the car simply can't park in them." Overseas parking spots come in diverse forms—horizontal spaces are common and often narrow, brick pavement lacks clear markings, and spots for the elderly or disabled feature ground icons. Knowhow has built a database covering dozens of countries and over a thousand parking scenarios.
Image Source: Speaker Materials
The differences in driving scenarios are equally massive. Zhang Wei noted: "In Europe, lane lines might be narrower, while in the Middle East, you have ceramic nail lane markers." In Germany, unrestricted highways place higher demands on control algorithms. Meanwhile, the variety of unusual vehicles overseas—cars towing boats, towing small cars, oddly shaped bicycles—all pose tests for perception models.
Huang Luoyi used the example of German construction zones to highlight a deeper issue: "Without construction zones, the standard lane width is 3.75 meters. In construction zones, the yellow line is only about two-thirds the width of the white line, so the car is basically driving inside the yellow line with very little room." This imposes entirely different requirements on perception and control algorithms.
Even trickier are the differences in driving habits. Zhang Wei mentioned: "We did a project in Eastern Europe and found some of our functions were overly sensitive, constantly alarming. We later realized locals follow the car in front very closely and drive very fast." Bosch's Huang Luoyi pointed out that German traffic laws explicitly require vehicles to pull slightly to the left or right when temporarily stopping—depending on their lane—to leave a clear path for ambulances and police. This kind of "rule-based behavior" is difficult for imitation learning based on Chinese data to pick up.
Huang Yinhua, Senior Solution Architect at HERE Technologies, offered a sharp assessment: "We can treat maps as beyond-line-of-sight sensors, completing environmental understanding before on-board sensors arrive, providing prior information for the algorithm." Maps aren't being marginalized; they are upgrading from navigation tools to the environmental foundation of smart driving systems.
HERE has launched a full-stack, layered map product portfolio covering L0 to L3, matching different driving levels and regulatory needs across countries. TomTom emphasized that "mature autonomous driving shouldn't just be about perception, but anticipating road conditions like an experienced driver"—the essence being "not just seeing the world, but understanding it in advance."
Chinese algorithms excel in data scale and iteration speed, but overseas roads, signs, and driving habits constitute a brand-new "exam system." In this system, Chinese smart driving needs to start learning from scratch.
"Go First, Talk Later" or "Ready Before You Go"?
Facing these layered challenges, the paths for Chinese automakers exporting smart driving are clearly diverging.
Path 1: Sell the Car First, Add Features Later.
This is the pragmatic choice for most Chinese automakers right now. Chen Yichi's observation is representative: "With domestic smart driving yet to achieve a commercial closed loop, spending a huge sum on overseas expansion doesn't seem particularly wise at this stage." Consequently, many domestic automakers choose to sell vehicles first and upgrade smart driving features later.
Bosch's Huang Luoyi also acknowledged this reality: "Some friends ask, can't we just have one algorithm strong enough to generalize globally? At least from our current understanding, we haven't found such an algorithm." Since one algorithm can't conquer the world, the strategy is to land first, then iterate.
Path 2: Systematic Layout, One-Step Delivery.
Chen Long, Senior Expert in Intelligent Driving Products at Great Wall Motor, represents a different mindset: "We should do our duty from the user's perspective, serve them well, avoid gimmicks, and create value for the user—only then does the enterprise have value." He emphasized that Great Wall Motor's entire technical layout is always centered on user experience.
Huang Ziliang used a clever metaphor to express this approach: "In China, we must first cultivate the large model based on domestic computing power—including Huawei, Horizon Robotics, and Black Sesame—let it bloom and bear fruit, and then promote it globally. We shouldn't plant the seeds of Chinese models directly onto foreign computing power. It's like the lychees of Chang'an: if you transport them, you bring the soil and lychees from Lingnan to Chang'an, not just the seed. Planting just the seed in Chang'an's soil won't work, or won't work well."
TomTom's Chen Weicheng offered a framework from another dimension: "Compliance decides entry, scale decides breadth, and trust decides how long you last." He advocates entering markets in phases—first breaking into regions with looser regulations like the Middle East and Southeast Asia, before tackling high-barrier markets like Europe.
The Third Path: Leverage Local Strengths.
Du Jianning suggested learning from the "joint venture model traditional foreign companies used to enter China." Bosch showcased the technical path of "federated learning"—where data doesn't cross borders, only model parameters are exchanged. HERE collaborates with over 30 Chinese OEMs, providing global capabilities and ecosystem support.
Knowhow's Zhang Wei offered the most pragmatic summary: "On one hand, we follow automakers overseas; on the other, we've established connections with many overseas host plants to apply our technology to their vehicles." Following along, while also walking independently.
Conclusion
Returning to the initial question: Exporting smart driving—just go and figure it out later?
The answer is: Go first, but you can't just "figure it out later."
"Going first" is the right move. The scale of China's auto exports—4.059 million units from January to May 2026—means staying put means missing a historic window. Europe has surpassed the Middle East to become China's largest auto export region, accounting for about 25% of the total. The competitive logic of China's auto exports is shifting from "competing on batteries" to "competing on AI." At this turning point, any hesitation could mean falling behind.
But "not just figuring it out later" is equally critical. Compliance isn't something to deal with later; the GSR, GDPR, and UN R series regulations constitute "veto power" thresholds. Data localization isn't a later issue; it determines whether algorithms can iterate locally and if experience can improve. Scenario adaptation isn't a later consideration; it decides whether local users are willing to pay for your smart driving.
As Chen Weicheng put it: "True globalization is not just entering the world, but having the ability to operate continuously within it." Going first takes courage; the ability to keep running is the real skill.
