Chip War Escalates: Xpeng Executive Says Single-Chip Computing Power Misleads the Public

"Some cars have inferior configurations, unreasonable prices, and insufficient computing power, and cannot be modified. What can they do? They focus on local facts and deliberately use various methods to release data that obscures the truth." On February 10th, Yu Tong, head of smart experience at Xpeng Motors, made a sharp statement on social media. Although he didn't name names, it was clear to discerning people that his target was Li Auto.

Yu Tong accused Li Auto of selectively presenting facts in its chip computing power advertising, deliberately releasing misleading data in various ways, focusing solely on single-chip computing power while avoiding discussion of total computing power, thereby intentionally misleading the public by discussing "single points" and not the "whole." The fundamental reason for this is that "total computing power is insufficient."

The direct trigger of this storm is a recent "mainstream autonomous driving chip computing power ranking" circulating online. The chart shows that Ideal's self-developed Mach 100 chip ranks first with 1280TOPS per chip; followed by NIO's Shenji chip with 1000TOPS; and then Tesla's Ai4; while XPeng's Turing chip is tied with NVIDIA's Thor-U, with a claimed computing power of only 700TOPS.

Regarding this, Yu Tong pointed out that the actual single-chip computing power of the XPENG Turing chip is 750TOPS, while the figure mistakenly wrote the Turing chip's computing power as 700TOPS, deliberately obscuring the truth for selfish purposes. Based on the 750TOPS data for re-ranking, the XPENG Turing chip should rank third, only behind Li Auto's MACH 100 chip and NIO's Shenji chip.

"If we're going to compare, let's do it openly, objectively, fairly, and comprehensively, instead of trying to be opportunistic with localized clashes," Yu Tong said, also previewing Xpeng's upcoming second-generation technology platform, which will bring a "leapfrog experience driven by high computing power." While seemingly a response to the image, it is actually a direct response to some views previously expressed by Li Xiang, Chairman of Li Auto. This is already the third exchange between Li Auto and Xpeng within a week.

Let's go back to the beginning.

On February 6th, Li Auto teased the new generation Li L9, claiming that its two self-developed Mach 100 chips achieved 2560TOPS of effective computing power, making it the "world's most powerful brain." However, unexpectedly, just three hours later, XPeng Motors released the XPeng GX, equipped with four Turing chips, with a local effective computing power of 3000TOPS. This led to Li Auto being ridiculed for its "world's most powerful brain" claim lasting only three hours. This is considered a "sniper strike" from XPeng at the time of Li Auto's new car release – it is unclear whether this was unintentional or deliberate.

Following this, on February 9th, Li Xiang, the founder of Li Auto, responded on social media to the "Most Powerful Brain" lasting only three hours. He stated that the new L9 will be equipped with a self-developed Mach 100 dual-chip, with a total computing power of 2560 TOPS. The key is not the total computing power, but the effective computing power of a single chip. Li Xiang said that the effective computing power of a single Mach 100 chip is as high as 1280 TOPS, adopting a data flow architecture, and its utilization rate far exceeds that of traditional GPUs. The effective computing power of a single Mach 100 is 3 times that of NVIDIA Thor U, and the dual-chip combination reaches 5-6 times that of it.

From the perspective of overall computing power, Xpeng GX ranks first with 4 Turing chips and an effective computing power of 3000 TOPS, while Li Auto L9 Livis ranks second with an effective computing power of 2560 TOPS. However, from the perspective of single-chip computing power, Li Auto's Mach 100 indeed ranks first with an effective computing power of 1280 TOPS. Aside from the aforementioned erroneous chip computing power ranking, it's understandable for both companies to promote their products from angles that favor them.

However, this heated war of words throws a key question to consumers: In the era of intelligent driving, how should we understand "computing power"? From a hardware perspective, the total computing power of a vehicle indeed represents the "ceiling" of its intelligent hardware. Higher total computing power means greater potential for future OTA upgrades and the ability to support more complex algorithms and functions.

However, high computing power does not directly equate to an excellent intelligent driving experience. The quality of the experience is the result of the deep integration of hardware, software, and data. For example, in the past two years, many car companies have been competing on the number of lidar sensors. But in terms of actual intelligent driving experience, the performance of some models equipped with multiple lidar sensors is even inferior to some pure vision-based models, which serves as a good illustration.

Therefore, for consumers, simply comparing parameters without actual experience has limited significance. Whether a single chip is powerful or multiple chips combine for high computing power, it ultimately comes down to practical scenarios, observing whether the vehicle can provide a safer, smoother, and more "experienced driver"-like driving experience across all scenarios such as highways, urban areas, and parking.

Furthermore, the current computing power competition between Li Auto and XPeng is not simply a battle of technical parameters, but a pre-launch skirmish between their new models, the next-generation L9 and the XPeng GX, to seize the "technical high ground" label and influence consumer perceptions. Both new cars are large six-seater extended-range SUVs targeting the premium 400,000 RMB market, making them direct competitors and two of the most important products for each company this year. Currently, both vehicles have been declared to the Ministry of Industry and Information Technology, and their launch dates are not far off. The proof of the pudding is in the eating.