Surge in New Energy Vehicle Exports, Intelligent Driving Overseas Remains an Untapped Area

The lack of core product power means intelligent driving overseas needs to accelerate implementation.

With the announcement of October sales for automakers, unsurprisingly, exports are becoming the core sales volume for domestic automakers, and Chinese car exports are showing an explosive growth trend.

According to China Association of Automobile Manufacturers data, total Chinese car exports from January to September 2026 reached 8.63 million vehicles, up 52% year-on-year. In October, the proportion of new energy passenger car exports alone reached 44.2%.

At the same time, domestic automakers are becoming the main force at overseas auto shows. Selling points like cabins and intelligent driving, which are commonplace at domestic auto shows, continue to attract overseas consumers.

Before the Paris Auto Show this year, He Xiaopeng brought Xpeng's NGP to the European market and had a direct showdown with Tesla's FSD.

Recently, He Xiaopeng proposed a new testing ground for domestic automakers and intelligent driving companies, stating that if an automaker wants to prove its intelligent driving technology is truly robust, the most direct verification method is to drive the product to overseas markets and undergo actual tests in different road conditions and traffic rule scenarios around the world.

Setting aside how Xpeng's intelligent driving actually performs, behind He Xiaopeng's viewpoint is the common status quo faced by Chinese new energy vehicle overseas expansion: cars go out, but intelligent driving cannot be used.

Intelligent Driving Untapped Market

With the intensification of price wars and the manifestation of supply chain scale advantages, the prices of domestic new energy vehicle intelligent assisted driving solutions are gradually decreasing, and penetration rates are also climbing. Ministry of Industry and Information Technology data shows that in 2026, the penetration rate of new passenger cars equipped with L2-level integrated driving assistance functions reached 70.5%, and the penetration rate of models equipped with Pilot Driving Assistance functions (NOA) also broke through 34.2%.

Looking at Europe, although it is the birthplace of the automotive industry and the region most actively promoting assisted driving, the penetration rate is still not high.

At the L2 level, under the EU GSR2 regulation framework, the L2 installation rate was forcibly pushed to 100% in 2026. However, this 100% mainly corresponds to basic ADAS functions such as intelligent speed assist and emergency lane keeping, which are not entirely equivalent to the L2 that domestic consumers perceive as simultaneously controlling steering and acceleration/deceleration.

Data from JATO Dynamics statistics of five European countries (Germany, France, Italy, Spain, UK) shows that the penetration rate of new L2 cars that truly possess the ability to control steering and acceleration/deceleration simultaneously is about 35%. In 2022, this figure was 22%, an increase of only 13 percentage points over five years.

As for the penetration rate of high-order urban NOA in Europe, it is less than 1%. The EU's UN R171.02 regulation requires system behavior to be 100% consistent with the written declaration thresholds submitted by the manufacturer to the regulatory authorities in advance, and does not allow any autonomous decision-making behavior beyond the declared scope.

As for the closer Southeast Asian market, although the growth rate has accelerated, it is still lagging in infrastructure, regulations, and after-sales verification systems.

Data shows that the overall ADAS penetration rate in Southeast Asia jumped from 18% in 2025 to 24% in the first quarter of 2026. The rate of pre-installation in pure electric vehicles has reached 42%, and 29% in plug-in hybrid models.

Behind this, the gradual improvement of regulations in Southeast Asian countries. Thailand took the lead in stipulating that new registered electric passenger cars must be equipped with at least three basic ADAS functions starting from January 2026. Indonesia plans to follow up in 2027. Regulations are the primary driver for current penetration rate increases.

However, high-order functions still remain at the verification stage, and test cars are mainly concentrated in closed testing parks in Thailand and Malaysia. Mass-produced urban NOA for consumers basically does not exist.

As for the Middle East and Latin America regions, they are still in the popularization stage of L2 basic functions. There is still a long way to go before reaching high-order intelligent driving.

As for the only region that truly achieved NOA penetration, it is the North American market. Driven by Tesla, the penetration rate of new cars equipped with North American NOA is expected to break through 45% in 2026. Q2 data shows that 55% of users in newly delivered Tesla vehicles in North America activated FSD subscriptions at the time of pickup. As for data on truly realizable urban NOA, it is also less than 5%, contributed entirely by Tesla.

Overall, currently the European and Southeast Asian markets where new energy vehicles mainly export have gaps in the field of high-order assisted driving. For domestic automakers, this is undoubtedly another area of technical dividends. There is no stock competition; once successful, it will become an absolute technical barrier.

But the reality is that intelligent driving overseas faces dual pressure from regulations and infrastructure, and there is still a long way to go before landing.

Three Ways to Seek Breakthroughs

Currently, major domestic automakers exporting overseas have already laid out landing points for intelligent driving solutions overseas to achieve breakthroughs locally. The main paths are divided into three categories, corresponding to these different solution models.

First is the self-developed + localization adaptation solution, which is also one of the current mainstream implementation paths. Tesla's global layout also follows the same solution. A set of core models combined with local data can theoretically achieve adaptation quickly.

Taking Xpeng mentioned at the beginning of intelligent driving overseas as an example, it adopts the "one model eats all globally" strategy. The second-generation VLA model uses world models and reinforcement learning to support L2 assisted driving and L4 autonomous driving with the same architecture.

At the same time, Xpeng has established a local data center in Europe to meet GDPR compliance requirements and has passed the German TÜV Rheinland UN R171 certification, becoming the first Chinese intelligent driving system to receive this certification. It plans to deliver intelligent driving systems adapted to the local environment to European users in 2027, and is also the automaker domestically that has provided a definite timeline so far.

Traditional automaker Changan Automobile also follows the same solution. In details, Changan Automobile adopts a "layered layout" strategy in Europe. The European version uses the Tiandu System + Horizon chip adapted to UN R171 regulations. It plans to launch highway NOA in Europe in 2027, while urban NOA is suspended due to insufficient local verification.

At the same time, Changan established a R&D center in Munich and cooperates with Qualcomm to develop the European data training system, which will accelerate the verification and landing of urban NOA.

The second path is starting from scratch, re-walking the domestic training route in overseas markets. Currently, BYD mainly adopts this solution. Currently, all of BYD's export models are downgraded to basic L2-level functions, and Sky Eye System has not landed overseas.

Intelligent driving technology adapts to the market by cooperating with local companies and building data training centers locally. It uses physical isolation and full-stack localization strategies to cope with GDPR compliance requirements, ensuring European user data is 100% stored in EU data centers and cutting off the data return link.

As for the third route, it is the easiest to implement, which is to directly adopt locally compliant intelligent driving solutions. Just like the Aito export version. Aito models in China are equipped with Huawei Qiankun solutions, but due to regulatory limitations, in order to land overseas as soon as possible, the European version of Aito models plans to adopt the Bosch intelligent driving solution. It plans to mass-produce highway NOA functions in September 2027.

This third path also triggered the rhythm of solution suppliers accelerating landing overseas. Horizon entered the European market through CARIZON, a joint venture with Volkswagen. At the same time, Horizon is accelerating the landing of HSD. It has already secured design wins for 11 automakers and more than 40 export models.

Zhuoyu Technology activated its European headquarters located in Braunschweig, Germany in May 2026, promoting the "In Europe, For Europe" localization strategy. It used ID.Buzz for highway NOA, urban NOA, and automatic parking road trials, planning to complete verification by the end of 2026.

Momenta, one of the four giants in intelligent driving, is the fastest to place its pieces. On the one hand, it has obtained L4 testing permits in the Europe and Middle East region. On the other hand, with cooperation with local European automakers, Momenta's high-order solutions are also accelerating landing. After signing a global strategic cooperation agreement with Dongfeng Peugeot-Citroën Technology, it directly entered Stellantis's global supply chain.

As for true landing, it actually lies in the periodicity of local data training and multiple regulatory certifications. Facing the EU's GDPR compliance requirements, whether it is automakers or solution suppliers, they can only choose to build local data processing and training capabilities locally, slowing down the landing speed.

On the other hand, under the multi-country road conditions overseas, various regulations and certification systems are the issues that currently most affect landing. The EU's WVTA certification takes an average of 12-18 months. Under such a long period of time, intelligent driving solutions have already iterated several versions, making it difficult to truly achieve global synchronization.

The competition of intelligent driving going overseas is no longer about whose technology is more advanced, but about who can complete a mass-producible system-level closed loop under the multiple constraints of the target market, with acceptable costs and cycles. The proof of intelligent driving going overseas is not only technical strength, but also the efficiency competition in local layout under globalization strategies.