China's Automakers "Do it Themselves" Chip War in the Cockpit: China Builds its Own, Xiaomi is Now on Board

From Henrik Bork | Translated by AI 5 min Reading Time

Related Vendor

China's car manufacturers say they have "no other choice". They have to build their own car chips. Lei Jun, the founder of the electronics manufacturer Xiaomi, which is now also building electric cars, recently put it exactly this way.

In China, more and more car manufacturers are developing their own chips to circumvent US embargoes and supply chain uncertainties.(Image: Dall-E / AI-generated)
In China, more and more car manufacturers are developing their own chips to circumvent US embargoes and supply chain uncertainties.
(Image: Dall-E / AI-generated)

"As for the battle over chips, we have no other choice," Lei wrote last May in a post for his millions of followers on China's social media. At the same time, he announced that Xiaomi's first self-developed car chip would be unveiled "soon." He did not yet provide details on the architecture or its purpose—whether for autonomous driving or infotainment. However, this places Xiaomi among a growing group of automakers breaking away from Western chip manufacturers.

Nio, Xpeng, and Li Auto have been working for years on developing their own automotive semiconductors. "Go solo" is definitely the answer for some Chinese automakers, writes the Chinese specialist portal 36kr. That this trend is now accelerating further has two main reasons. In-house development is, on the one hand, the Chinese companies' response to the U.S. chip export controls.

In Washington, elected representatives have been busy for several years denying China access to high-end semiconductors, primarily those that are AI-capable. Nvidia is already prohibited from delivering many of its best chips to China, losing more and more market share there. And just at the beginning of this month, the latest tightening in a long list of export controls from the U.S. towards China came into effect. New restrictions on the delivery of EDA software for chip design to China have come into force, reports the Financial Times.

Here Again: Embargoes Do Not Ignite Properly

What the increasingly intense American attempt to cut China off from modern chip and AI technology has been causing for several years now is becoming very clear in the automotive industry. Jensen Huang, the CEO of American chip manufacturer Nvidia, has recently put it very vividly.

Washington's export controls have failed, Huang said during a visit to Taiwan in May. They have given Chinese companies "the spirit, energy, and government support" they need "to accelerate their own development," according to the Nvidia CEO. It is a misconception to think that Chinese industries can be kept away from modern chips. "If they don't have enough Nvidia, they will use their own," the New York Times quotes Huang from Taipei.

On the other hand, and this is the second reason why China's automakers believe they have "no other choice," delivery difficulties from Nvidia have also contributed to the Chinese do-it-yourself efforts. Technical difficulties had repeatedly delayed the delivery of Nvidia's Thor chip. "Thor is now so delayed that competitors' chips are already available," writes 36Kr. Nio and Xpeng have since removed Thor from their strategies; the risk is too great that planned models will falter due to chip supply issues.

The China Association of Automobile Manufacturers also recommended in a public statement "caution when purchasing American chips," reports the Chinese automotive portal Gasgoo.

Supply Chain Security for Market Dominance

The last thing Chinese automakers can afford in the face of very tough competition in the domestic and international car markets is uncertainties in supply chains, whether caused by political activism or manufacturers' delivery bottlenecks. China is therefore reluctantly relying on costly, yet promising a certain degree of autonomy over the product pipeline, self-development of car chips.

Nio has now installed its Shenji NX9031, based on a five-nanometer process, in several models such as the ET9, ES6, or EC6. The chip replaces the function "of four conventional smart-driving chips," said William Li, chairman and CEO of Nio, to CNEVPOST. This not only saves costs but also achieves significantly better integration into its own operating system, SkyOS. For the Nvidia founder, it must be painful to remember that Nio originally wanted to use Nvidia chips.

The same was true for Xpeng. The Chinese electric car startup had hoped for Nvidia, but then designed its own Turing chip, equipped with two specially developed neural cores. According to the company's own statements, the chip achieves a computing power of about 700 TOPS—comparable to the latest generation of Nvidia's Thor chips, which, after multiple delays, could finally be delivered in a slimmed-down version.

At the same time, Chinese automakers are practicing innovations regarding their car chips. They are striving to circumvent the limitations of edge computing and bring the power of LLM into the car. This aims to translate human-like "common sense" decisions into concrete driving instructions, such as a lane change or the use of brakes. "If we rely solely on vehicle computing, we reach limits," said Xpeng's autonomy chief Li Liyun at a technical briefing in April, reports 36Kr.

Subscribe to the newsletter now

Don't Miss out on Our Best Content

By clicking on „Subscribe to Newsletter“ I agree to the processing and use of my data according to the consent form (please expand for details) and accept the Terms of Use. For more information, please see our Privacy Policy. The consent declaration relates, among other things, to the sending of editorial newsletters by email and to data matching for marketing purposes with selected advertising partners (e.g., LinkedIn, Google, Meta)

Unfold for details of your consent

Distilled AI Models for the Car

Xpeng is now trying a hybrid strategy: large AI models with up to 72 billion parameters are trained in the cloud and then distilled down to smaller models for vehicle deployment. The Turing chip then processes these locally in the vehicle. Initial tests with this system architecture are said to have been "surprisingly powerful," according to the company.

Meanwhile, the newly joined "do-it-yourself" chip manufacturer in the Chinese automotive industry, Xiaomi, which has also just introduced its new smartphone chip "Xring O1", relies on the Taiwanese contract manufacturer TSMC for chip production. This follows the same path as Apple and Nvidia. However, according to a report by TrendForce, Xiaomi uses U.S. EDA software for its internal design processes, which now fall under export controls. Access to these tools is uncertain in the future, which could also affect Xiaomi's car chip plans, according to the specialist publication.

Nevertheless, Lei Jun is showing a fighting spirit. Xiaomi plans to invest at least 50 billion yuan (around 6.5 billion US dollars) in chip development over the next ten years, he announced. This year alone, it will be over six billion yuan, about 790 million US dollars. Xiaomi has a R&D team with 2,500 employees.

The Government Supports the Ambitions

The Chinese government supports these efforts. The state broadcaster CCTV referred to Xiaomi's announcement of its latest smartphone chip as a "breakthrough," showing that private companies in China are also able to achieve technological sovereignty. Of course, by using their own chips, Chinese automakers not only increase their response speed but also their margin. Nio, for example, was able to replace the use of four Nvidia Orin-X chips with two of its own, as its founder claims, aiming to achieve cost reduction noticeable in the price competition on the domestic market.

The shift towards Chinese chip autonomy in the automotive industry is happening rapidly. "Decoupling begins: Xpeng introduces Turing AI chip in the second quarter and reduces dependence on Nvidia" was the headline recently on the CarNewsChina portal. Tesla demonstrated to Chinese automakers with its "Full Self Driving" architecture how far vertical integration can go. Now, out of necessity, the Chinese are following suit with their own operating systems, data platforms, and AI models. One consequence is that they will need to order less from the Western abroad in the future. (sb)