Summary of the News in Plain Language
In essence, this is a self-staged, cost-free marketing campaign by Xpeng. At the end of last year, He Xiaopeng made a bet with his chief autonomous driving officer, Liu Xianming, stating that if Xpeng’s second-generation VLA autonomous driving system could match the performance of Tesla’s FSD V14.2 in Silicon Valley by August 2026, he would fund the construction of a Chinese-style restaurant in Silicon Valley. If he lost, Liu Xianming would have to run naked across the Golden Gate Bridge. Three days before the bet was due, Xpeng officially declared that they had “won,” and the restaurant was immediately scheduled to open, sparing Liu Xianming the need to run naked.
Throughout the process, there was no third-party judge; Tesla remained completely silent, and Lei Jun, who had been invited to act as the judge, pretended not to see anything. The outcome was entirely up to Xpeng. However, this publicity stunt greatly increased the popularity of Xpeng’s autonomous driving technology, introducing a bunch of new technical concepts to the public. Essentially, Xpeng used the “matching Tesla” as a pretext to reclaim the title of “best in autonomous driving” from its competitors. Whether the technology is truly superior is something ordinary car owners have no way of verifying.
---
Analysis of the Event from Four Perspectives
1. The Bet Was Designed to Be a Sure-Win Marketing Strategy
If you look closely at the terms of the bet, it’s clear that Xpeng had no way of losing:
- The two scenarios being compared were incomparable: it’s like comparing the scores of your own practice exams with those of an American student’s SAT exams, which use different evaluation criteria. The roads in Silicon Valley are well-organized, with pedestrians and electric vehicles behaving in a predictable manner. FSD has never had to handle the unpredictable conditions of Chinese roads, where vegetables are piled up by the side of the road, unlicensed tricycles make sudden turns, and children appear out of nowhere in narrow alleys. Xpeng’s system, trained on such roads, naturally outperforms FSD in complex scenarios—using its strengths to counter its competitor’s weaknesses.
- The “judges” were all on Xpeng’s side; Lei Jun ignored the bet, and Tesla didn’t say a word about it. It’s like betting that your cooking is better than a restaurant’s, with you as the judge—how could you possibly lose?
In the end, Xpeng’s move was a huge success: the cost of opening a restaurant in Silicon Valley was merely a few million, far less than the billions spent on advertising. The technology’s prominence, previously overshadowed by competitors like Huawei, was restored, and terms like “second-generation VLA” and “physical AI,” which most people had never heard of, became familiar through the media. It was a low-cost way to achieve significant results.
2. The Hype Around “Emergent Technology and 4D Spatial Understanding” Isn’t What It Seems
Many people were impressed by the fancy terminology in the promotions, thinking Xpeng had developed something truly groundbreaking. In simple terms:
- Previous autonomous driving systems were like static images taken from cartoons, recognizing obstacles and issuing braking commands with a delay. Xpeng’s new VLA system, on the other hand, functions like the human eye, continuously analyzing the road and predicting future events. This is a mainstream direction that all leading autonomous driving companies are working on, not an exclusive invention by Xpeng.
- The so-called “emergent technology” is essentially the result of investing billions in high-performance computing power (30,000 advanced GPUs) to process massive amounts of road data. Complex scenarios that previously took months to solve were suddenly solved. However, autonomous driving is about safety, and consumers want 100% reliability, not sudden, unproven capabilities.
3. The “Superior” Autonomous Driving System Is Out of Reach for Most Car Owners
Many people were excited, thinking they could use this technology immediately. However, the reality is different:
- The full version of VLA2.0 requires three high-end AI chips developed by Xpeng, with a total computing power of 2250 TOPS. Few cars on the market currently have such hardware, so most owners won’t be able to use it.
- The “update” expected in September is actually a simplified version with only 3 billion parameters, reducing the complexity of the system significantly. Even this reduced version will only be available in a few top-tier models, and most car owners will have to wait for months or even years to get it.
- The current rapid iteration of autonomous driving concepts outpaces their actual reliability, creating a situation where new terms and features appear before they’re fully tested and proven.
4. The Rapid Pace of Conceptual Development in the Autonomous Driving Industry
The domestic autonomous driving industry is characterized by rapid changes: new terms emerge faster than smartphone updates. What was called a “world-class model” last month might be called “physical AI” this month, and who knows what will come next. The promotional videos look like scenes from science fiction, showing cars driving on unmanned mountains or avoiding traffic cameras. In reality, however, these systems still have significant shortcomings when used in real-world conditions—such as poor visibility in rain or fog, or slow reactions to unexpected obstacles.
The reason companies exaggerate their claims is to capture consumers’ attention. If they don’t shout the loudest, customers will turn to competitors’ products. But autonomous driving is not about winning bets or fancy press releases; it’s about the millions of kilometers driven on real roads by ordinary users. Only when a system can prove its reliability in real-world conditions—whether in busy cities or rural roads—can it truly be considered the best. A marketing bet won’t win customers’ trust.