Xpeng CEO He Xiaopeng made a declaration at a media briefing during Auto China 2026 on April 25, 2026, that would have seemed improbable two years ago: his company’s autonomous driving system will fully outperform Tesla’s Full Self-Driving in the Chinese market by August. The target is specific, the timeline is tight, and the competitive context, Tesla’s FSD, is still awaiting regulatory approval in China, making it both bold and strategically calculated.
“We’ve set ourselves a goal: to fully outperform Tesla’s FSD in the Chinese market by August,” He said, according to the South China Morning Post. “China’s driving environment is more complex. If we can surpass it here, our overall capability should be stronger.” The statement was made at Auto China 2026, the 10-day Beijing International Automotive Exhibition running through May 3, which has become the primary showcase for China’s AI-driven automotive ambitions.
VLA 2.0 and the Leap Toward Level 4
The system He is referring to is Xpeng’s Vision Language Action 2.0, which the company previewed in April ahead of the Beijing show. VLA 2.0 is designed to advance Xpeng’s vehicles from Level 2 driver assistance toward Level 4 autonomy, the point at which a car can handle the full driving task without human intervention in most conditions. He has repeatedly argued that Xpeng will skip Level 3 altogether, a position that reflects both a technical philosophy and a regulatory strategy: Level 3 autonomy, which requires the driver to be ready to take over at any moment, creates complex liability questions that He believes are better avoided.
The VLA architecture is significant because it integrates vision, language, and action in a single model, enabling the system to reason about driving scenarios in ways that more narrowly designed systems cannot. Rather than relying on a fixed set of rules for specific situations, a VLA model can generalize from its training to handle novel scenarios, the kinds of unpredictable situations common on Chinese urban roads and historically the hardest challenge for autonomous driving systems.
By next month, Xpeng expects to enable autonomous driving without reliance on navigation maps — a capability that would allow the system to navigate in areas where high-definition map coverage is incomplete or outdated. This is a meaningful advance: map-dependent autonomous driving systems are limited to routes with up-to-date map data, which excludes a significant portion of China’s road network.
Tesla’s China Timeline Slips Again
Xpeng’s August target is made more achievable by the fact that Tesla’s FSD has still not received regulatory approval for broader use in China. During Tesla’s earnings call on April 23, executives said FSD approval in China had been pushed into the third quarter of 2026 — a setback from CEO Elon Musk’s earlier projection that approval could come by February or March. The delay means that Tesla’s most advanced assisted-driving features remain unavailable to Chinese consumers, giving domestic competitors like Xpeng a window to establish their systems as the standard against which FSD will eventually be judged.
He acknowledged that direct comparisons are difficult while FSD remains unapproved in China, but said Xpeng’s VLA system already outperformed Tesla in “certain complex driving scenarios.” The claim is difficult to verify independently, but the framing is a deliberate competitive move to position Xpeng as the benchmark-setter rather than the challenger.
A $2 Billion R&D Bet
Xpeng plans to invest nearly $2 billion in research and development in 2026, a figure that includes work on autonomous driving, flying cars, and the full-stack technology approach that He has made central to the company’s identity. The full-stack strategy encompasses self-developed chips, world models (AI systems that build internal representations of the physical environment), and electronic architectures. He has argued that owning the full technology stack is the only way to achieve the rapid iteration required by the autonomous driving race.
(Related: Chinese EV Makers Bet on In-House Chips at Beijing Auto Show — Xpeng, NIO and Hesai Lead the Charge)
The company is also pursuing an aggressive international expansion strategy. Between the fourth quarter of 2026 and the second quarter of 2027, Xpeng plans to roll out VLA 2.0 test rides in several European markets, which accounted for more than half of its overseas sales last year. He’s argument is that training on China’s dense and complex urban roads gives Xpeng’s system an advantage in overseas markets with similar driving conditions, a claim that will be tested when the European rollout begins.
Xpeng was the third-bestselling Chinese EV startup last year. The company’s aggressive autonomous driving timeline and its $2 billion R&D commitment signal that it intends to compete on technology rather than price, a differentiation strategy that is increasingly necessary as the Chinese EV market becomes more crowded and margin-eroding price competition intensifies.
Geely’s Eva Cab and the Broader Competitive Picture
Xpeng is not the only Chinese automaker making bold autonomous driving announcements at Auto China 2026. Geely, China’s second-largest private carmaker after BYD, unveiled its self-developed robotaxi, the Eva Cab, at the Beijing show, with mass production expected in 2027. The Eva Cab is designed for fully driverless commercial operation and represents Geely’s bet that the robotaxi market will become a significant revenue stream as regulatory frameworks for Level 4 autonomy mature.
The broader picture from the show is of a Chinese automotive industry that has moved decisively from competing on hardware, battery range, charging speed, and interior design to competing on software and AI. Xpeng’s August target, Geely’s Eva Cab, Huawei’s ADS 5 platform, and the deep localization strategies of foreign automakers like Hyundai all reflect the same underlying reality: in China’s car market, the software stack is now the primary battleground. Xpeng’s $2 billion R&D commitment for 2026 is a statement that it intends to fight that battle on its own terms, with its own technology, rather than licensing capabilities from a platform provider.