Guosheng Securities has issued a research report expressing optimism about XPENG-W's (09868) strong product cycle, overseas expansion, advancements in autonomous driving, and emerging business opportunities in robotics/Robotaxi. The firm projects the company's 2025-27 sales at approximately 450,000/680,000/1.01 million vehicles, with total revenues reaching RMB80.6/133.3/188.5 billion and non-GAAP net profit margins of -1.5%/3.4%/5.2%. The target price is set at HK$126 (US$32.4). Additionally, the bank believes XPeng's other growth drivers may gradually be incorporated into its valuation as product development progresses, warranting a maintained "Buy" rating.

XPeng is set to announce a "major breakthrough" in physical AI at this year's AI Tech Day, with its foundational world model achieving "critical progress" in predictive capabilities. Key highlights from Guosheng Securities include:

1. **AI Model Development**: XPeng's AI team has been developing a 72-billion-parameter autonomous driving model—dubbed the "XPeng World Foundation Model"—since April. This model's parameter count is about 35 times larger than mainstream VLA models. - **Training Efficiency**: XPeng's self-developed data infrastructure has increased data upload capacity by 22x and training bandwidth by 15x, while optimized GPU/CPU and network I/O boosted training speed by 5x. - **Scale Advantage**: The model will train on 200 million clips, with Dr. Liu Xianming, head of the project, noting that larger parameters and data volumes consistently enhance performance.

2. **Competitive Edge**: XPeng's "physical AI + Turing chip" strategy underpins its leadership in smart vehicles, Robotaxi, and robotics: - **Software**: The unified AI model for mobility and robotics will be deployed via cloud-distilled smaller models, enabling rapid global rollout of Turing AI autonomous driving and powering robots/flying cars. - **Hardware**: XPeng's self-designed Turing chips deliver higher cost-effective computing power (one chip ≈ 3 Orin-X units). Current models like the P7 and G7 Ultra feature three chips, while 2026 L4 vehicles will boast 3,000 TOPS.

Guosheng highlights XPeng's potential to outpace domestic peers in assisted driving and robotics, citing full-stack R&D, superior brain-hardware integration, and cost advantages. Similarly, Tesla is advancing high-performance chips and a "World Simulator" neural network, which trains AI using virtual scenarios equivalent to 500 human-years of driving daily. Both companies share aligned technological and commercial strategies, with Guosheng anticipating XPeng's breakthroughs in physical AI.