Author | Zheng Senhong
For Chinese automakers, the first half of selling cars overseas has become relatively clear, but the truly difficult part has just begun. If going global in the past was more like a long voyage, then taking autonomous driving global is more like entering the "Odyssey Era": what everyone faces is no longer just a single market, but a completely different set of roads, rules, and physical worlds.
The easiest battle in the global expansion of Chinese cars might already be over. Shipping cars overseas, building distribution networks, and capturing local markets—Chinese automakers have become increasingly proficient in this playbook. In the first half of this year, Chery exported approximately 943,800 vehicles, accounting for about 69.5% of the group's total sales. During the same period, BYD sold 790,000 vehicles overseas, a year-on-year increase of nearly 70%, and Geely exported 474,200 vehicles, a year-on-year increase of 158%, making them major players in going global. The scale of China's auto exports is still growing rapidly. But the next battle is of a completely different difficulty level. Because while cars can be shipped out, technology cannot be simply replicated, especially autonomous driving.
In China, an autonomous driving system that has been fully trained and tested might face completely different right-of-way rules, traffic signs, driving habits, regulatory frameworks, and data requirements when it arrives in Germany. This means that the real problem Chinese automakers need to solve is no longer "how to sell cars overseas," but whether they can bring the intelligent capabilities trained in China into another physical world. This is the true second half of Chinese automakers going global.
01. Selling Cars Overseas: This Battle Has Reached Halftime
Chinese cars going global are divided into two strategies based on destination: incremental markets and mature markets. Mature markets such as Europe, Japan, and Australia present core challenges in branding, regulations, technology, and safety service systems; incremental markets such as the Middle East, Southeast Asia, Latin America, and Africa present core challenges in cost, channels, localization, and supply chains. The exploration of incremental markets has been relatively smooth. Early domestic cars entered with low prices, achieving a scale breakthrough from zero to one. Taking Saudi Arabia as an example, the Chery Tiggo 8 Pro Max was priced at 89,000 riyals, while the comparable Toyota Fortuner was 149,000 riyals. The Geely Monjaro was sold for 109,000 riyals, 26,000 riyals cheaper than the Toyota RAV4. For vehicles in the same class, Chinese brands were generally 15% to 40% cheaper.
With this price-killer strategy, domestic cars achieved impressive curves and growth figures in overseas markets. In the first year of going global (2001), sales were less than 20,000 units; four years later, it grew to 170,000 units, and in 2012, it broke the one million mark for the first time.
However, these dozen or so years were essentially fought on battlefields that others looked down upon. Chinese cars merely appeared on the world market and had not truly entered any mainstream market. Mature markets are the real test. In Asia, Africa, and Latin America, Chinese automakers attacked the flanks of their opponents, whereas in Europe, they needed to launch a frontal assault; the intensity of competition between the two is not on the same level. Take quality issues, for example. The quality standard in the European market is Euro NCAP. Volkswagen and Toyota's models on sale have all achieved the highest five-star safety rating, while domestic cars of the same period did not even have a record of participating in the tests. When Chinese cars first entered Europe, they even left a stereotypical impression of low quality and low price due to crash test results: "It's best not to sit in the driver's seat." Consecutive poor crash test results and shoddy workmanship created a stereotype of domestic cars in Europe for the next decade. It was not until the New Energy Vehicle (NEV) era that Chinese automakers began to redefine the overseas market's perception of Chinese manufacturing with safety, the three-electric system, and intelligence.
In 2022, Chinese brands appeared intensively in Euro NCAP tests. SAIC MG4, BYD Atto 3, NIO ET7, XPeng P7, G9, and G6 successively achieved five-star ratings. Public data shows that among the Chinese brand models participating in Euro NCAP tests in 2025, 33 models were tested and 30 received five-star ratings. Building cars with guaranteed quality and quantity is the foundation of going global. This not only requires massive R&D investment but also the accumulation of manpower and time for adjustment and integration. SAIC, Chery, and Great Wall have spent more than 20 years laying out the route of complete vehicle manufacturing plus the supply chain system. Today, whether it is complete vehicle exports or KD (Knocked Down) assembly exports, both are quite mature.
On the other hand, technology is a powerful tool to meet demands and create new markets. The involution in the domestic NEV sector is all-encompassing. From industry to technology, and then to products, the competition in every link has pieced together a full-dimensional capability system. For example, the upstream battery industry driven by electric vehicles: in the early days, the equipment and materials segments of the lithium battery industry were almost entirely dominated by Japanese and South Korean companies, but today, Chinese companies occupy core positions in every segment. This has made overseas consumers realize anew that Chinese cars also possess high-value technology, thereby increasing their favorability towards Chinese car brands. Price advantages are responsible for opening the market, while technological advantages are beginning to change the perception of Chinese brands overseas. In 2025, China's auto exports reached 7.098 million vehicles, nearly twice as many as Japan, the second-place country. Among them, NEV exports reached 2.615 million units, a year-on-year increase of 103.7%, while the export volume of fuel vehicles decreased by 2% year-on-year. In this round of going global, domestic automakers have finally attacked the hinterland of their opponents relying on the technological generation gap.
02. Why Do Domestic Cars Lose Their Intelligence After Leaving the Country?
Regarding domestic cars going global, there might be a misconception: overseas consumers are indifferent to Chinese-style intelligence. This is not the case. In the European volume of the "Global Automotive Consumer Study" released by Deloitte in 2024, 56% of German consumers were interested in navigation functions with congestion alerts and route planning, and 62% were interested in OTA (Over-The-Air) software upgrades. The acceptance of novel things like intelligence among overseas consumers is no less than that in the Chinese market. However, the possibility of Chinese automakers fully meeting the user needs of every country is zero. In the past, a seemingly simple navigation system could not actually perfectly adapt to the usage in every country. First is language. Languages are not unified among European countries, and there are multiple dialects within the same country. Even if the scope is narrowed down to the European Union, the 24 official languages plus several regional languages are enough to keep the software R&D teams of automakers busy for months.
A French owner of a BYD Yuan PLUS complained that he basically had no other complaints about the car, except that his French accent often failed to wake up the voice assistant. Second is the map. Chinese suppliers are generally bound to AutoNavi and Baidu, while the mainstream map sources in Europe are HERE and Google Maps, which do not match the expectations of local consumers. A survey by TechInsights of over 5,000 car owners in North America and Western Europe showed that 54% use CarPlay navigation directly, 23.5% use the built-in navigation of the car machine, and 15.5% mount their phones on the dashboard. It sounds a bit like a castrated version of a smart car, but in the past, the models going global were indeed vastly different from those in China. The full-scenario voice interaction, ecosystem applications, and online navigation on domestic models are mostly reduced to basic Bluetooth and FM radio in overseas versions. The most stable functions on the car machine are ironically the ones with the least "intelligent content."
At the autonomous driving level, it is even more direct. In China, almost everyone has full-scenario navigation autonomous driving, but when going global, it is first downgraded to lane keeping plus adaptive cruise control, which is still mandatory standard equipment in the EU market. Facts have proved that the more fiercely domestic intelligent features compete, the more are cut when going global. After all, the performance of intelligence highly depends on the degree of localization. This localization dilemma is an unsolvable problem for domestic automakers. If more localization needs are to be considered, the workload for software adaptation is virtually limitless. If there is not enough determination within the automaker, this kind of localization is difficult to promote. For example, when XPeng first entered the German market, it only laid out a sales network and used TomTom for navigation. Later, XPeng officially launched its Munich R&D center, extending the investment in going global from the channel end to the R&D end.
In July this year, XPeng launched the MONA L03 in Munich, officially announcing the integration of the Google Maps Auto SDK, becoming the first automaker in the Asia-Pacific region to link Google Maps' underlying data with the vehicle's overall intelligent data. Specifically, XPeng is responsible for the map's interface, rendering, and interaction, while the underlying real-time traffic conditions, place search, and charging planning are all replaced with Google's data. This model not only retains XPeng's self-developed car machine system and voice assistant but also significantly shortens the R&D time. Originally, Google estimated the workload to take more than six months, but the XPeng team finally completed the Demo in just one month. XPeng's sample sufficiently illustrates that what blocks the globalization of intelligence is how the automaker's engineering system can completely move the functions that have long been mature in China out of the country. But cockpit localization is only the first hurdle; the truly difficult part is autonomous driving.
03. Autonomous Driving Going Global: The Endgame Is Selling Technology
The most important aspects of taking autonomous driving global lie in two points:
First, deep localization is required, which demands a sufficient understanding of local laws, regulations, and road characteristics to conduct in-depth development and adaptation. For example, cyclists represent the most sensitive scenario. In Germany, bicycle lanes are physically separated from motor vehicle lanes, and cyclists are road users with priority heavily protected by law. There are strict statutory requirements for following distances, and the lateral clearance for overtaking must not be less than 1.5 meters. For the same type of target, it is a vague obstacle in China, but a party with strong right-of-way in Germany, which determines different following distances and overtaking timings. Traffic signs are another lesson. The German StVO regulations, including variants, have more than 670 types, whereas the main signs in the Chinese national standard only have more than 200 types. Diamond priority signs and circular speed limit variants are rare in China, and the number of variants in subdivided scenarios is far greater than that in China.
Second, the autonomous driving needs in different markets vary. In China, the competition is on speed and the upper limit of technology, while in Europe, more emphasis is placed on compliance, safety, and the lower limit of technology. Any minor decision-making overstep is unacceptable; only by truly following the rules can it be considered a high-level capability. Industry insiders revealed that the priorities of German users are very clear: compliance, safety, and predictability are all the primary prerequisites. Only on this basis will they accept reasonable decisiveness. When rules are ambiguous, they prefer the system to be conservative rather than accept risk-taking aggression. From another perspective, German users recognize agile decision-making within a compliant framework, rather than the extremes of conservatism or aggression.
What capabilities are really needed for Chinese autonomous driving to go global? XPeng is providing a referenceable sample, which can be broken down into three capabilities:
First, general model capability. Can the models trained in China be transferred? Following this standard, XPeng has carved out a set of autonomous driving globalization models that can be reused globally: the foundation model is the hull, and the local team is the helmsman. The hull undertakes generalization and transfer capabilities. XPeng's second-generation VLA (Vision-Language-Action) large model is mainly trained based on Chinese road conditions. Without large-scale localization adaptation, the actual experience on European city roads is highly close to that in China. Behind this is the fact that the second-generation VLA directly gives driving decisions based on perceived scenarios, and the training signals do not rely on manual annotation either. The driver's actual operations are the supervision samples, and what the model learns is how humans drive. Facing the scarcity of real road condition data overseas, XPeng makes up for it with synthetic data generated by the world model that conforms to physical laws. For example, for Germany-specific signs and right-of-way, only a small amount of local data needs to be supplemented.
Second, ontological engineering capability. When the model arrives in Germany, what constitutes "correct driving"? The helmsman undertakes the judgment capability. Local engineers and road safety experts define scenario standards, study the road rules, driver behavioral habits, and traffic sign systems of Germany and major European countries, and then iterate the model in combination with the feedback from local test drivers. The hull is responsible for general capabilities, and the helmsman decides how to quickly adapt to different sea areas.
It is understood that currently, a single post-training run by XPeng to produce a European version has its time compressed to a few hours. Selling cars overseas and exporting technology are two tasks of different difficulty levels. The former only requires shipping cars to overseas markets to sell them, while the latter, when entering mature overseas markets, faces extremely strict safety supervision and requires a very high compliance threshold. Taking the European market as an example, the compliance threshold is superimposed by three layers: first, the mandatory access requirements of the GSR (General Safety Regulation) regulations set the lower limit; second, the non-mandatory E-NCAP evaluation procedures determine the upper limit; and finally, the GDPR draws the data red line, covering the entire data closed-loop of collection, transmission, storage, use, sharing, and destruction, which is more difficult than in China. Currently, XPeng is promoting the preparation for overseas VLA regulations and localization testing, planning to take the lead in obtaining European regulatory approval in the first half of 2027, and simultaneously delivering the second-generation VLA.
Third, commercialization capability. What exactly is being sold in the end? XPeng has set up a dedicated business team responsible for turning autonomous driving capabilities into externally licensed technology cooperation projects and is currently in talks with multiple partners. Over the past two decades, the globalization of Chinese cars has solved one problem: Can cars made in China be sold globally?
Now, this problem is turning into another: Can the intelligent vehicle technology accumulated in China be replicated globally? The former problem tests manufacturing, cost, and channels, while the latter tests model, data, engineering, regulatory, localization, and global operation capabilities. Landing Chinese technology overseas is far more challenging than selling Chinese cars overseas. This is why selling cars overseas is only the first half of the globalization of Chinese cars. What truly determines the competitive landscape of the next stage is that Chinese cars are not just selling vehicles globally, but are beginning to sell intelligent capabilities globally. If this step can be successfully taken, what Chinese cars export when going global will no longer be just a car, but a complete set of intelligent vehicle technology systems from China.