XPeng's second-quarter earnings report for 2026 revealed total revenue of 32.78 billion yuan in the first half, a year-over-year decline of 3.8%, with automotive sales revenue at 28.05 billion yuan, down 10.3%, and a net loss attributable to ordinary shareholders of 3.12 billion yuan. Following the release, the management team led by Chairman and CEO He Xiaopeng, Vice Chairman and Co-President Gu Hongdi, and other key executives participated in the earnings call to address analyst questions. The primary topics centered around the company's humanoid robot initiatives and upcoming product launches.
Morgan Stanley analyst Tim Hsiao congratulated the company on the impressive valuation achieved in the recent financing round for its advanced robotics project and raised two robot-related questions. The first focused on the production ramp-up plan for the Iron humanoid robot, given the stated goal of commencing mass production by the end of this year, with output targets of several thousand units next year and specific delivery goals for 2027. He Xiaopeng responded that robot production capacity differs significantly from automotive manufacturing, noting that while car supply chains are broad and deep, XPeng's Iron robot benefits from extensive full-stack self-development and deep in-house manufacturing capabilities. He emphasized that the initial challenge in mass production will be quality control, followed by sales, though overall capacity requirements will be considerably smaller than in the automotive sector. The plan remains to begin mass production by the end of 2026, followed by initial deployments in XPeng's own retail stores to establish commercial use cases, with external customer deliveries accelerating from the first half of 2027 as product quality and supply chain maturity improve, eventually scaling to thousands of units per month based on market demand.
He Xiaopeng further clarified that Iron's deliveries target retail and service industry scenarios across both China and overseas, ensuring that reported volumes will be genuine and comparable to automotive sales figures. He emphasized that the robot's intelligent solutions for commercial applications like customer guidance will be fundamentally different from competitors in the industry, enabling rapid expansion across various job functions and leading to comprehensive commercialization. Tim Hsiao's second question addressed the expected unit cost of the mass-produced Iron version, pricing strategy, and projected gross margins. He Xiaopeng explained that due to the self-designed native hardware platform and software systems developed for innovation, quality, and production efficiency, more than 85% of the supply chain partners overlap with XPeng's existing automotive suppliers, positioning the company strongly on cost competitiveness.
Regarding market pricing, He Xiaopeng noted that robot industry pricing in China typically differs from automotive, generally set at 2.5 to 3 times the bill of materials cost. Given Iron's positioning as a general-purpose robot with scarce supply and broad pricing flexibility, he expressed confidence that hardware gross margins would substantially exceed the current automotive business margins. Additionally, the company plans to generate incremental revenue from AI model capabilities and software services beyond hardware sales, further enhancing overall profitability.
BofA Securities analyst Ming Hsun Lee posed questions about the synergies between robot models and autonomous driving systems, asking which components are shared versus independently developed. He Xiaopeng addressed the industry debate about unified large models, explaining that while a single all-encompassing model might eventually emerge, current technology requires three distinct model categories: ultra-fast models operating at 100 to several hundred frames per second, medium-speed models like those used in autonomous driving at 10 to 20 frames per second, and slow thinking models at approximately one frame per second. He noted that XPeng's existing automotive models, including VLA and multimodal VLM systems, share substantial commonality with robotics applications, citing the example of VLA technology being developed for navigation in non-road environments such as underground parking garages, which operates on the same logic as indoor navigation for robots.
Conversely, He Xiaopeng highlighted that certain robot-specific models, including thinking models like VLT and open-platform series, will eventually benefit automotive systems in future years. He also identified unique robot models such as whole-body motion control and various safety systems, including data privacy protection and fall detection or low-battery safety protocols. The company maintains a unified development approach across its broader ecosystem, sharing AI architecture, data management, and organizational frameworks between the automotive and robotics divisions, including collaborative development of foundational technologies like world model simulation and digital model generation. Ming Hsun Lee's second question explored XPeng's differentiation in data collection, training, and closed-loop iteration compared to other robotics startups. He Xiaopeng emphasized that data represents the core competitive advantage in physical AI, though he cautioned that sufficient data is necessary but not sufficient for success, noting that while many companies pursue autonomous driving, few achieve excellence. He highlighted XPeng's decade of autonomous driving research and data expertise as foundational, with robotics data management and training systems operating within the same unified development framework as the automotive division, differing only in areas like data collection hardware.
He Xiaopeng expressed strong confidence in the company's full closed-loop capabilities from data acquisition to application, noting that Iron's mass production will generate valuable real-world data and demonstration data from human usage. He stressed the importance of high-quality data over volume, as poor-quality data degrades overall system precision, and described how the company aims to create a flywheel effect where quality data accelerates Iron's evolution in real-world environments. Citi analyst Jeff Chung asked about the strategic choice of sales and guidance scenarios as initial deployment use cases, target customers, and plans for industrial and household applications. He Xiaopeng explained that XPeng's commercial thinking differs significantly from competitors who typically start from factory or household entry points with B2B approaches. Instead, the company targets large C-customers and small-to-medium B-customers, entering through commercial applications before expanding to industrial and household sectors with minimal SKU variation. The rationale for starting with guidance roles relates to the four comprehensive capabilities the company has developed: physical capabilities, environmental awareness, business functions, and emotional value delivery.
He Xiaopeng stated that integrating these four capability areas best serves diverse industries across China and overseas markets, beginning with select sectors. The company will also open its SDK after Iron deployment to support app development and ecosystem expansion, leveraging XPeng's existing distribution channels while creating new online sales channels. This direct-to-consumer approach targeting large C-customers and small-to-medium B-customers represents a novel commercial sales logic distinct from current robotics industry players. Jeff Chung's second question addressed progress on the company's self-developed dexterous hand, its design approach, and performance-cost positioning relative to industry standards. He Xiaopeng described the robotic hand as an extremely critical component requiring one hardware system, one perception system, and three software capabilities, all of which XPeng has chosen to self-develop. He expressed satisfaction with development progress, confirming that the dexterous hand will enter mass production alongside the Iron robot by year-end, with substantial investment already made in manufacturing processes and equipment. The hand features direct-drive actuation with 21 degrees of freedom per hand and dimensions matching adult human hands, positioning the company's grip and load capabilities as industry-leading.
He Xiaopeng explained the deliberate choice against tendon-driven hand designs that many in the industry are exploring, considering them not optimal at the current stage. Rather than pursuing extreme force or precision, the design philosophy focuses on balanced capabilities including safety considerations, with even the hand's skin under development, along with cost optimization, reliability, maintainability, quick-change functionality, and mass production consistency. The goal is to achieve scale production of the dexterous hand based on balanced performance across these dimensions.
JPMorgan analyst Nick Lai inquired about VLA 2.0 deployment plans, specifically which vehicle models would initially feature the technology and whether all overseas-launched models next year would include it, also asking about subscription and separate fee considerations for the overseas market. He Xiaopeng acknowledged the unclear signal but provided a rapid response, confirming that the Turing-chip-equipped L03 has already launched overseas, and all future vehicle releases will include Ultra versions internationally, meaning VLA hardware capabilities will cover the L03 and all subsequent models across all markets. Progress is proceeding smoothly, including regulatory preparation and localized testing. The company is indeed exploring commercial software subscription models and will announce details at an appropriate time, with a business development team already established and multiple partners in discussions about bringing VLA and other capabilities to global markets.
Goldman Sachs analyst Tina Hou congratulated the company on its first-round financing announcement and raised questions about humanoid robot mass production timelines, profitability outlook at various production scales, and whether XPeng plans to separately report financial results for the robotics and automotive divisions. Gu Hongdi responded that discussing financial projections for the robotics business is premature at this stage, with current focus on executing key milestones: achieving SOP mass production by year-end, deploying in internal scenarios first, gradually delivering to external customers from the first half of next year, and then continuing production ramp-up. He declined to provide specific sales volume guidance required for profitability but expressed expectations that humanoid robot gross margins will significantly exceed automotive levels, given that hardware alone carries higher margins than automotive, complemented by high-margin incremental revenue from AI model iterations and software-related services. Additionally, capital expenditure requirements for the robotics business are considerably smaller than automotive, suggesting that once production capacity ramps successfully, the robotics division should achieve profitability faster than the automotive business.
Regarding potential financial separation, Gu Hongdi explained that both businesses currently operate together without division separation, though the company has an 18-month window to progressively pursue business separation as previously announced. However, the current priority is maximizing business synergies by fully leveraging shared AI, advanced manufacturing, and supply chain capabilities to achieve higher efficiency and stronger product competitiveness across both divisions. Only after mass production and commercialization paths become clearer will the company evaluate separation further, and even if pursued, the robotics business would remain 100% consolidated without changing group financial statements. In the near term, the group will continue consolidated accounting for all robotics financial data while seeking the most efficient and reasonable operational model.
Comments