According to an analysis report, XPENG-W (09868) is currently in a phase of quadruple resonance, characterized by a new robust vehicle launch cycle, global expansion, Robotaxi initiatives, and robotics development. The firm maintains its revenue forecasts for the company for 2026, 2027, and 2028 at 91.4 billion yuan, 126.5 billion yuan, and 154.6 billion yuan, representing year-on-year increases of 19%, 38%, and 22%, respectively. Net profit attributable to shareholders is projected to be -2.1 billion yuan, 1.5 billion yuan, and 5.8 billion yuan for those years. Earnings per share (EPS) are estimated at -1.08 yuan, 0.78 yuan, and 3.03 yuan, with corresponding price-to-earnings (P/E) ratios of 65x and 17x for 2027 and 2028. XPENG's strategic positioning in physical AI is considered forward-thinking and leading, and its current valuation is seen as significantly undervalued, warranting the maintained 'Buy' rating.
Key Points from the Analysis
The world model research from XPENG was presented at the prestigious International Conference on Computer Vision and Pattern Recognition (CVPR) in June 2026. The second-generation Visual Language Action (VLA) model and the world model are not seen as replacements or competitors but as two facets of the same foundational model for the physical world, trained under different signals. The VLA model is supervised by human driving actions to address 'how to act,' while the world model uses next-frame prediction for self-supervision, teaching dynamics and causality to address 'how the world changes after an action.' Both apply Scaling Law to train a unified base.
Advancements in the Second-Generation VLA
The second-generation VLA jointly models vision and language at the foundational layer, omitting explicit translation through an intermediate language layer to reduce information loss and improve inference efficiency. It is trained on a cloud cluster of approximately 30,000 cards, with R&D investment exceeding 2 billion yuan, enabling a full-chain iteration cycle of about five days. Since its deployment in March 2026, the proportion of assisted driving mileage in equipped models exceeded 50% for the first time in its first month. The company plans to release a new version in the third quarter of 2026 to further enhance its capabilities.
Academic Foundation for the World Model
The academic underpinnings of the world model consist of three key projects—X-World, X-Foresight, and X-Cache—alongside DrivePTS, which was officially accepted by CVPR. X-World utilizes VAE latent space diffusion for controllable multi-camera scene generation (spatial compression 16x, temporal compression 4x). X-Foresight, built upon the VLA, employs block-wise autoregression to simultaneously predict multi-camera views and output actions. X-Cache uses cross-block caching for the denoising backbone, achieving up to approximately 2.7x inference acceleration and reducing redundant computations by about 70%. DrivePTS enables controllable generation of rare scenes through progressive learning, mutual information constraints, and frequency-guided structural loss.
Commercial Acceleration of Physical AI Applications
Three major physical AI applications are advancing simultaneously: the second-generation VLA, Robotaxi, and the humanoid robot IRON. The Turing chip began mass production and vehicle integration in the third quarter of 2025, with a shipment target of nearly 1 million units for 2026 and a full model lineup transition starting in the second quarter. Volkswagen is the first external partner for the Turing chip and second-generation VLA. The first mass-produced Robotaxi vehicle rolled off the production line in Guangzhou on May 18, 2026, marking China's first fully in-house developed solution by an automaker, with demonstration operations expected to commence in the second half of 2026. Mass production of the IRON robot is planned for the end of 2026, with a target monthly capacity exceeding one thousand units. On June 10, 2026, He Xiaopeng announced via an internal letter that he would personally assume leadership of the robotics business.
Potential Risks Outlined
The report highlights several potential risks: an intensifying price war in the passenger vehicle market beyond expectations; a slower-than-expected recovery in terminal demand; delays in the rollout of policies supporting L3/L4 autonomous driving; and slower-than-anticipated progress in the industrialization of humanoid robots.
Comments