Nov 17
He Xiaopeng stated during the earnings call that large-scale physical world models are accelerating the advent of the autonomous driving era. Only pre-installed, mass-produced Robotaxis with strong generalization capabilities can truly achieve scaled application and validate the business model.
He revealed that the company will launch several new models
that do not rely on LiDAR, significantly reducing deployment and maintenance costs. Additionally, he emphasized that physical AI is a core strategy for XPeng, with automobiles and humanoid robots sharing the same technological origins, showing emergent capabilities. The company will comprehensively advance the mass production and implementation of large models, Robotaxis, and humanoid robots.
$XPeng (XPEV.US)$XPeng Motors released its Q3 2025 earnings report on Monday, with all key metrics hitting record highs. Revenue doubled year-over-year, and the gross margin exceeded 20%, reaching a historical peak. The net loss narrowed significantly by nearly 80% compared to the same period last year. It is projected that both deliveries and revenue in Q4 will grow over 30% year-over-year. During the subsequent earnings call, He Xiaopeng stated that with the advent of large physical-world models, the era of fully autonomous driving is accelerating.
He noted that only deeply integrated large physical world models can gradually enable machines to interact, communicate, adapt, and even create within the physical environment, thereby reshaping modes of transportation and lifestyles.
Only mass-produced front-end solutions with highly generalized Robotaxi capabilities can achieve widespread adoption while simultaneously proving viable as a business model.
He Xiaopeng revealed that in 2026, XPeng Motors will launch three Robotaxi models. This
technology route avoids using LiDAR, effectively addressing industry-wide challenges such as high costs, particularly deployment expenses, along with mobility restrictions and weak generalization. He believes this will accelerate the large-scale global deployment of Robotaxis.
Additionally, XPeng plans to initiate trial operations of its Robotaxi service in China in 2026, further refining the software, hardware, and operational ecosystem for Robotaxis.
Q3 Performance and Overall Physical AI Strategy
He Xiaopeng stated that in Q3 2025, XPeng achieved record highs across multiple operational metrics: quarterly deliveries reached 116,007 units, up 149% year-over-year; revenue, gross margin, and cash reserves all hit new peaks, with gross margin surpassing 20% for the first time. Net losses continued to narrow substantially. The company aims to break even in Q4. These ongoing improvements reinforce the company’s commitment to advancing research in Physical AI, expediting the mass production and scaled implementation of its second-generation VLA model, Robotaxi, and humanoid robots by 2026.
He pointed out that over the past 11 years, XPeng has adhered to full-stack self-research, evolving from “software-defined vehicles” to “Physical AI.” The company has clearly observed shared origins between automotive and humanoid robot technologies, including common models, chips, and infrastructure, enabling rapid iteration and evolution.
Our Physical AI system is beginning to demonstrate emergent capabilities. In the coming decade, our vision is to position XPeng Motors as a globally-oriented 'giant intelligent life company.'
I believe the value created by physical AI for humanity can only be maximized through an open ecosystem. To this end, we will open-source physical world models, collaborate with mobility platforms to launch Robotaxi services, and provide open hardware SDKs. Through cooperation with commercial and technical partners, we will further expand the physical AI ecosystem and accelerate value creation.
AI-Powered Vehicles: The Era of 'Dual-Energy' Product Cycles Begins
He Xiaopeng stated that XPeng Motors is entering a new product cycle featuring 'dual-energy' (pure electric + super range extender) capabilities, rapidly expanding sales and market share in the new energy sector through a more comprehensive powertrain portfolio.
On November 6, XPeng launched its industry-leading 'large-battery lithium iron phosphate 5C' super range extender model—the XPeng G9 Super Journey—with a combined range of 1,602 kilometers, making it the world’s longest-range, highest-AI-compute-capability, smallest-turning-radius, and most space-efficient large seven-seater SUV. The super range extender will significantly accelerate the migration of fuel vehicle users to new energy vehicles.
He revealed that after the pre-sale launch of the super range extender, there has been a notable increase in attention from northern and inland cities, attracting user groups that pure electric vehicles previously struggled to reach. The pre-sale orders for the X9 Super Range Extender have reached nearly four times that of the previous generation during the same period. The XPeng G9 Super Intelligence will be launched and delivered simultaneously on November 20, with the company expecting December sales to hit a new high.
2026 Product Roadmap: Multiple Super Journey Models + Multiple Dual-Energy New Products
He Xiaopeng mentioned that the company plans to launch three all-new Super Journey models in the first quarter of 2026, addressing core pain points of range extender users with longer pure-electric ranges and 5C ultra-fast charging. Additionally, four new 'dual-energy' models will be introduced in 2026, including a breakthrough product for key segments—similar to the blockbuster M03 model.
I believe that the seven dual-energy models equipped with super range extenders, set to launch next year, will greatly expand our total addressable market (TAM) and drive significant sales growth.
Global Market Expansion
He Xiaopeng stated that the company's overseas sales continue to grow rapidly. In September 2025, overseas monthly deliveries exceeded 5,000 units for the first time, marking a year-on-year increase of 79%. During the third quarter, 56 new overseas stores were added, bringing the sales and service network coverage to 52 countries and regions worldwide.
The localization production project in Europe has officially launched at the Magna plant in Graz, Austria, with the first batch of G6 and G9 models already in mass production. The R&D center in Munich has been put into operation simultaneously.
He revealed that the company plans to launch at least three new models overseas by 2026, including the highly anticipated compact SUV, to meet diverse global demands.
Second-generation VLA model: A tenfold increase in parameters, with full deployment scheduled for 2026.
XPeng Motors' AI computing power and data are being transformed into the capabilities of large physical-world models. He Xiaopeng stated that the parameter count of the second-generation VLA model is ten times that of the previous generation, achieving a significant generational leap in both the safety and user experience of autonomous driving.
He Xiaopeng mentioned that he personally test-drove VLA 2.0 recently, describing its performance under complex road conditions as 'extremely impressive, with no comparable end-to-end model at the same level.' He anticipates that the pioneer user co-creation experience will be available by the end of December, with VLA 2.0 set to roll out to all O7 and above models in the first quarter of 2026.
The mass production of this generation of models will represent a tremendous leap forward, delivering a cross-generational upgrade experience and encouraging more people to choose XPeng due to its leading autonomous driving capabilities.
He disclosed that XPeng will open-source VLA 2.0 to global business partners, with Volkswagen becoming the first customer. Additionally, XPeng's self-developed Turing AI chip has secured a designated project from Volkswagen, and the jointly developed models are expected to enter mass production in early 2026.
In our external technology collaborations, we will generate licensing revenue, which will be reinvested into R&D for iterative upgrades of the Turing chip and VLA models, creating a positive feedback loop. We welcome more OEMs and Tier 1 suppliers to collaborate with us on the Turing chip and second-generation VLA model to jointly promote the widespread adoption of intelligent technologies in China and globally.
Robotaxi: Trial operations to commence in China in 2026
He Xiaopeng pointed out that traditional end-to-end large models can only achieve the best L2-level assisted driving at most, whereas physical-world large models truly accelerate the advent of the autonomous driving era. He believes that only pre-installed mass-produced Robotaxis with extremely strong generalization capabilities can achieve scalability and commercial success.
He revealed that XPeng Motors will launch three Robotaxi models in 2026. These vehicles will not rely on high-definition maps or LiDAR, significantly reducing the industry cost and deployment complexity of Robotaxis while enhancing urban adaptability and reducing mobility restrictions. In 2026, XPeng Motors will initiate Robotaxi trial operations in China, while continuously refining its software, hardware, and operational ecosystem.
He stated that the Robotaxi SDK will be fully open, with AutoNavi becoming the first ecosystem partner. XPeng Motors also welcomes more mobility companies to join in collaboration.
Humanoid Robots: Mass Production by the End of 2026, Sales Target Exceeding 1 Million Units by 2030
He Xiaopeng noted that the company's humanoid robot (Iron) adopts the same large model approach for the physical world. With support from the vehicle and powertrain team, XPeng showcased a new generation of Iron robots at its Tech Day, whose agile gait and human-like performance sparked cross-disciplinary discussions.
He mentioned that the capabilities currently demonstrated by Iron represent only a small fraction of its potential. The company plans to achieve full integration of cross-domain capabilities by the second quarter of 2026, with advanced humanoid robots entering mass production by the end of 2026. Initially, these robots will be deployed in commercial scenarios such as guided tours, retail assistance, customer service, and security. By the end of 2026, Iron will be introduced into XPeng’s stores, campuses, and factories, becoming a 'new colleague.'
XPeng will open up the robot SDK to global developers, enabling Iron to acquire stronger capabilities across more niche application scenarios.
I believe the market potential of humanoid robots is greater than that of automobiles. Once they surpass the tipping points of intelligence and electrification, there will be explosive growth. I look forward to XPeng’s robot models achieving annual sales exceeding 1 million units by 2030.