_________

Published on Sep. 28, 2026

China's auto industry has largely completed its first and second transformations. Electric drivetrains, batteries, and charging networks are established or emerging. Intelligent driving features are fast becoming standard in entry-level cars as well.

Now, companies and the government in Beijing alike are getting serious about tackling the third transformation with artificial intelligence. For the first time ever, a high-level plan explicitely calls “AI + Automobiles” a priority.

On September 9, 2026, the Ministry of Industry and Information Technology (MIIT) and eight other government departments published the “15th Five-Year Plan for the Development of the Intelligent Connected New Energy Vehicle Industry.” It has a dedicated section titled “Action Plan for the Development of ‘AI + Automobiles’” (in Chinese).

The whole NEV plan targets a 70 percent share of e-vehicles and plug-in hybrids among passenger cars and 40 percent among commercial vehicles by 2030, with widespread autonomous-driving deployment. In other words, it treats the transformation to smart electric vehicles as “mission accomplished”.

China's policy focus now shifts from the rapid adoption of new energy vehicles, or NEVs, to new core technologies led by AI and to topics like the global expansion of Chinese car manufacturing and domestic supply-chain resilience.

The “AI + Automobiles” chapter reads like an industry checklist for the coming years. Three main to-dos for Chinese carmakers stand out.

First, MIIT wants to see innovation in AI models and intelligent agents, prioritizing their automotive deployment with pilot bases and computing facilities from public and proprietary resources.

Second, the planners call for deeper AI deployment within vehicles, focusing on energy management, motion control, multimodal human-machine interaction, and fault prediction, calling for “cockpit-driving integration” through cross-domain coordination.

Third, the MIIT plan addresses progress in the area of automotive data, from trusted data spaces and better industry datasets to large models for the industry and around 50 typical application scenarios.

This is, mind you, a “plan”, a blueprint for the future. AI in China's automotive industry is currently at an early stage of experimentation which can easily be mocked.

More than 60 automakers and suppliers in China have already localized the popular DeepSeek AI model. “However, most remain at the stage of ‘technology rebranding’,” the Chinese industry newspaper Zhongguo Qiche Bao (China Auto News) recently quoted from a white paper by Baidu Intelligent Cloud and China EV100 (in Chinese).

While a chatbot in the cockpit is a far cry from the era of AI-defined driving that some marketing copy is already announcing, it would be equally wrong to dismiss the real and widespread experimentation with AI in the Chinese automotive industry that goes far beyond the cockpit. That was on display at the recent fourth “AI-Defined Vehicle Forum”.

Regarding AI agents in engineering, the first item we highlighted on the MIIT wish list, a German company is active in China. Siemens' Xcelerator platform, launched in China three years ago, has more than 600 local partners, 900 applications, and 660,000 users, the Chinese auto portal Gasgoo quoted Qian Shiming of Siemens Digital Technology from the Shanghai forum.

The variety of use cases gives a first glimpse of the scale of transformation underway. In the automotive sector, Siemens offers more than 100 AI applications across eight scenarios. In PLC programming, the Eigen agent integrates with TIA Portal, generating code, HMI screens, and hardware configurations in one step. In drawing management, the number of manual 3D-to-2D drawing operations has fallen by 90 percent, while drawing output has risen fivefold, Siemens says.

In production, the German company claims that its MES platform helped Yanfeng Visteon cut quality costs by 15 to 20 percent. And the quality agent “Zhi Xiao Q” improved quality engineers' efficiency at Amphenol by 66 percent, also according to its maker.

In product design, NX MCD reduces on-site commissioning by 30 percent and trial-and-error costs by 20 percent, reports Siemens.

Such productivity gains are difficult to verify independently. The direction of travel is clear, however. This is part of the engineering upgrade the MIIT plan seeks, delivered by a European supplier.

For the second field, cockpit and driving integration, the Chinese carmaker Seres serves as a good example both for what has been achieved already and of the future roadmap regarding AI integration.

Seres built the Aito brand in collaboration with Huawei and has recently taken control of it. Current vehicles have separate domains for intelligent driving, the cockpit, and the chassis, says Liu Xiaofei of Seres Technology. In the future, they will integrate as “embodied intelligent agents.”

Seres proposes a three-tier hardware architecture of “brain, cerebellum, and zone control.” A central computing platform merges cockpit, driving, and vehicle control resources, becoming the car's perception and decision hub. A motion domain controller, the cerebellum, coordinates drive, braking, steering, and suspension. Zone controllers reduce wiring and cut weight and cost.

The computing is hybrid, with different tasks performed either in the cloud or in the car. Deep inference occurs in the cloud, with millisecond responses on the vehicle, according to Seres. Cross-domain coordination, another MIIT item, is embedded as a concrete electronic architecture right in the car.

The Chinese company QCraft is developing AI models for driving intelligence. Its cloud-based world model generates controllable video, zero-shot scenarios, and low-cost simulations. A vehicle-side model combines vision-language-action with reinforcement learning, covering perception to action, Qian Xiangjun of QCraft said at the AI forum in Shanghai.

These are no longer small pilots. Qcraft's assisted-driving software runs in more than one million vehicles across more than 40 models and 10 brands, according to the company.

On-device computing power, currently one of AI's biggest constraints, is growing fast. The company's Pro platform was the first to deploy urban navigation assistance on the 128 TOPS Horizon J600M chip, while the Max platform already works with more than 500 TOPS.

Assisted driving and full autonomy share the same platform. QCraft is not only testing this technology in Chinese cities, but also in Munich and Paris. Autonomous driving is the “first large-scale commercial application of physical AI”, says Qian Xiangjun.

There is a growing realization in the Chinese automobile industry that no single company can master the coming transformation alone. The costs for R&D are too high and the iteration cycles are too short. To acknowledge this is new for proud brands used to closed-door development.

Yet this is another trend clearly visible in China. More and more companies are actively exploring partnerships, even with their competitors. “OEM + OEM + key partners + institutions” is what Wang Xiaojuan of Geely's Institute for External Cooperation and Research calls this new collaboration model.

She was speaking at the World New Energy Vehicle Congress in Haikou on September 22, 2026. The reasons the expert gave for the need to team up were technical.

A vehicle model used to last five years, she said, while AI-based driving algorithms need updates every few months. Developing a chip like Nvidia's 254 TOPS Orin from scratch would cost at least several billion yuan before ecosystem adaptation.

Competitors, suppliers, and research institutes should co-create from the product definition stage, Wang suggested, while vehicle engineering control and safety systems remain under the carmaker's control.

The Baidu white paper quoted earlier provided an even longer list of obstacles facing every industrial company attempting to integrate AI in its processes and products. The current reality for most is that AI applications are still limited to pilot projects. Legacy business data are hard to connect. Even departments within a single company coordinate poorly.

Try to introduce the smartest large model into this environment, and the results will disappoint. “The challenge is that general-purpose large models are trained on publicly available data, while automakers’ core data is stored in systems such as ERP, MES and PLM. The two therefore suffer from a ‘comprehension gap’,” writes the industry newspaper Zhongguo Qiche Bao.

When understanding is superficial, agents may “guess the meaning based on field names” and turn deviations into business risks. Data silos and connecting legacy and new systems are systemic issues.

The focus in the industry is therefore shifting from “pursuing stronger AI” to “truly understanding business data,” the Baidu white paper concludes. This is what the Chinese industry ministry means by “trusted data spaces” for automotive datasets.

While such challenges aren't unique to this country, the environment for tackling them is. China's large market acts like a testing ground for new technologies across a number of different scenarios. Once an innovation is proven, a complete industrial chain allows it to move quickly from testing to mass production.

This is why the mockery of carmakers that deploy DeepSeek in the cockpit and boast about their AI skills will be justified only for a fleeting moment. The third transformation of the Chinese car industry is moving at an even faster speed than the first and the second.