--- title: "XPEV (Trans): Robotics unit valued at $6.2bn, mass production by year-end" type: "Topics" locale: "en" url: "https://longbridge.com/en/dolphin/post/43568758.md" description: "The robotics biz raised over $900mn in its first round at a post-money valuation above $6.2bn, and is slated to enter volume production by year-end. However, the core auto unit's per-vehicle GPM remains just 12.1%, and quarterly net loss widened to RMB 1.34bn. For Q4, the company targets 60k+ monthly deliveries, with overseas quarterly deliveries aiming for 40k." datetime: "2026-08-24T13:55:15.000Z" locales: - [en](https://longbridge.com/en/dolphin/post/43568758.md) - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/43568758.md) - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/43568758.md) author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)" generator: "portal-rs" --- # XPEV (Trans): Robotics unit valued at $6.2bn, mass production by year-end **Dolphin Research's Trans of XPeng FY26 Q2 earnings call:** **I. Key takeaways** 1\. **Q3 guide: deliveries of 115k–121k; revenue of RMB 21.7–23.4 bn.** This implies QoQ growth of 11.3%–17.1% for deliveries. It implies 9.9%–18.5% for revenue. Further out, the company targets monthly deliveries of over 60k in Q4. Overseas quarterly deliveries are targeted at over 40k. 2\. **First-round funding for the robotics biz exceeded $0.9 bn with a post-money valuation above $6.2 bn, fully consolidated.** IDG Capital led the round, with Gov. funds participating and Tencent and Alibaba as strategic investors. Management said both scale and valuation set a new record in China’s humanoid robotics private funding. Per the announcement, the biz can be carved out over an 18-month window. Given the equity structure, it will be consolidated 100%, and any planned separation will not affect financial statements. 3\. **Overall GPM was 20.7%, while vehicle GPM fell YoY to 12.1%.** Overall GPM was 20.7% vs. 17.3% a year ago and 20.6% in Q1. The company aims to keep it above 20%. Vehicle GPM was 12.1% vs. 14.3% a year ago and 12.1% in Q1. Management attributed the YoY decline to model refreshes. Service and other revenue reached RMB 2.7 bn, up 93.9% YoY. This was driven by milestone revenue recognition from tech R&D services delivered to Volkswagen Group and by parts sales. 4\. Key financials this quarter a. Totals: revenue of RMB 19.74 bn (+8% YoY, +51.5% QoQ). Auto sales revenue was RMB 17.05 bn (+1% YoY, +55% QoQ). b. Opex: R&D was RMB 2.91 bn (+32.1% YoY, +0.3% QoQ). SG&A was RMB 2.5 bn (+15.2% YoY, +32.5% QoQ), with QoQ growth driven by franchise commissions and marketing. c. Losses: OP loss was RMB 1.14 bn vs. RMB 0.93 bn a year ago and RMB 1.87 bn in Q1. Net loss was RMB 1.34 bn vs. RMB 0.48 bn a year ago and RMB 1.7 bn in Q1. d. Cash: cash was RMB 40.48 bn. This was as of Jun 30. **II. Earnings call details** **2.1 Management remarks** 1\. Deliveries and products a. Q2 deliveries were 103,295 units, up 65% QoQ, outpacing industry YoY growth. New non-cancellable orders in Q3 rose 50% QoQ to a record high. b. The flagship 6-seat SUV GX exceeded 7,000 domestic deliveries in Jul. It entered the top three among NEV SUVs priced above RMB 300k in China. c. MONA L03 became a hit at launch with record orders for XPeng models. Extreme weather and supply chain disruptions slowed the ramp, and two-shift production has now started. d. The flagship 5-seat SUV G9L will launch and start deliveries in Sep, and MONA L05 will launch domestically in Q4. With four new SUVs, the lineup will cover all key SUV subsegments. 2\. Overseas a. Overseas quarterly deliveries exceeded 20k for the first time in Q2, up 81% YoY. Intl revenue accounted for over 25% of total revenue in 1H. b. The Avg. export ASP exceeded EUR 40k. Management said unit revenue and margin rank among the top of Chinese automakers going abroad. c. MONA L03 made its global debut in Munich in Jul, with overseas deliveries expected to start in Q4. Several hero models, including range-extended, will be introduced overseas in 2027. 3\. Smart driving VLA 2.0 a. Starting late Aug, version 6.3.0 is rolling out. On-device model parameters are up 3.5x, perception sensitivity is up 300%, and ultra-long temporal reasoning and prediction are introduced with industry-leading frame rates. b. VLA and VLM are integrated for smart driving and cockpit functions, and portions of L4 capabilities developed for Robotaxi are being down-tiered to production cars. This includes voice-assisted parking. c. On-road tests were completed in Germany. Models trained mainly on China data performed close to domestic levels on European city streets with minimal local data. The goal is to secure approval and roll out in Europe in 1H 2027. 4\. Robotaxi and tech monetization a. A pre-installed mass-production Robotaxi with VLA 2.0 completed over 2,000 internal test rides in Guangzhou. The cloud-based teleoperation platform is ready, and the goal is to begin rides without a safety operator next year. b. In 2027, XPeng plans to expand in key cities with leading mobility platforms at home and abroad. Monetization will come from vehicle sales, tech services, and revenue sharing. c. A group-level BD team has been formed to export Turing AI chips, VLA 2.0, Robotaxi, and humanoid robotics tech to partners worldwide. 5\. Humanoid robot IRON a. Since Jun, He Xiaopeng has also served as CEO of the robotics biz. Robotics R&D has exceeded eight years, and he believes the tech challenge of general-purpose humanoids is at least 20x that of smart EVs. b. Hardware: the first fully enclosed flexible skeletal structure in the industry. It offers 76 total DoF and 21 DoF per hand, and chips, controllers, motion modules, and dexterous hands are all self-developed. c. Intelligence: powered by three Turing AI chips with effective compute up to 2,250 TOPS. The physical AI base model runs directly on-device. d. Roadmap: capabilities will be released and demoed starting Sep. Mass production will start by year-end, with initial deployments in XPeng stores and campuses, and large-scale deliveries to retail and services customers in 2027, ramping to several thousand units per month. **2.2 Q&A** **Q: What is the planned capacity once IRON reaches commercial mass production? What is the FY27 delivery target?** A: **No FY27 delivery target was provided, but mass production will start by year-end with monthly capacity reaching several thousand next year.** Management noted a very different capacity logic vs. autos, with a broader and deeper supply chain, and capacity challenges 'early in quality, later in sales'. The ramp will start with mass production by end-2026 and initial commercialization in XPeng’s own stores. In 2027, deployments will expand across in-house scenarios, then accelerate commercialization and use cases externally, with R&D stepping up from mid to late next year and monthly output rising beyond several thousand units. Deliveries will focus on retail and services in China and overseas. Management emphasized that delivery and sales data will be credible and that IRON leads peers in sales-assist capability, intelligence, and environmental adaptability. **Q: What are the unit cost, pricing strategy, and margin trajectory at scale for IRON?** A: **No unit cost or pricing was disclosed, but hardware GPM will be better than the current auto biz.** Both hardware and software are full-stack in-house. Over 85% of the parts supply overlaps with the auto supply chain, supporting a cost advantage for IRON. As a reference, current market pricing for robots is roughly 2.5–3.0x bill of materials, and IRON is a general-purpose robot with very limited market supply. Management therefore expects higher hardware margins than in autos. Beyond hardware sales, models, software services, and subscriptions will also contribute profit. **Q: Where do robotics and autonomous driving share the most synergy, and what is developed independently?** A: **VLA/VLM and the overall architecture are shared, while safety and low-level generative, imitation, and simulation models are independent.** Management refuted the notion that a single brain and single model can power robots, saying that might be viable years from now but not today. Models come in three tiers: ultra-fast at 100 fps or even several hundred fps, mid-speed large models at 10–20 fps, and a slow 'thinking large model' at 1 fps. This multi-speed design addresses different tasks. Shared modules include VLA and VLM across autos and robots, where car VLA roaming off planned paths is similar to robot mobility. Some open-platform reasoning capabilities in robots next year will feed back to cars, and AI apps and the overall architecture are shared. Independent modules include robot-specific safety models such as data privacy, anti-fall, and anti-battery depletion. Low-level generative, imitation, and simulation models differ significantly. **Q: What differentiated advantages does XPeng have in data collection, training, and closed-loop iteration for robots?** A: **The edge lies in training methods and data quality rather than sheer volume, with a flywheel driven by real-world and human demonstration data post mass production.** Management stressed that 'enough data' is necessary but not sufficient, and XPeng’s strengths are better training and higher-quality data. The basis is over 10 years in autonomous driving with leading R&D experience and data accumulation. Robot data management, training, and quality are developed within the same ecosystem, and apart from different hardware, the pipelines from collection to application are aligned across the two biz lines. Post-launch, IRON will continuously collect real-world and human demonstration data. Management said both are high quality and distinct from low-value data, enabling a data and R&D flywheel. **Q: Why choose sales assistance and guided tours as the first deployment scenarios? Who are the target customers? Will it expand into industrial and household use?** A: **Commercial scenarios first, then industrial and home, and guided sales/tours were chosen to showcase four integrated capabilities.** Management’s path differs from peers that start in factories and homes, as XPeng targets large enterprises and SMEs. It will start with commercial use cases, then expand into industrial and home, with smaller SKUs later. The rationale is that sales assistance and guided tours best demonstrate the robot’s four integrated strengths: the body and hardware, the environment, the services delivered, and the emotional value created. These scenarios exist both domestically and overseas. IRON will later open SDKs for secondary development and expanded use cases. Commercial collaborations will also open both offline and online sales channels, covering key accounts and SMEs. **Q: What is the latest on the in-house dexterous hand and the design philosophy? How does it compare to peers in performance and cost?** A: **One hardware and one software stack with three sensing systems, with detailed mass production plans to be shared with analysts by year-end.** The dexterous hand has 21 DoF and the same size as an adult human hand. Beyond R&D, the company has invested heavily in hand manufacturing and related equipment. The design favors a balance of safety, reliability, maintainability, and cost over brute force or precision. It also features a bionic skin similar to a human hand. Versus peers, management believes XPeng leads in payload and grasping capability. They also stated that another commonly discussed dexterous hand path in the industry is 'not a wise choice'. **Q: How many models will deploy VLA 2.0 overseas?** A: **No model count was given, but VLA will be deployed on L03 in all markets.** Management noted audio issues and answered based on key words heard. The Turing AI chip is already installed on L03 units exported overseas, all in the Ultra variant. VLA will be deployed on L03 in all markets, with compliance, regulation, localization, and testing progressing in parallel. **Q: What is the business model for smart driving overseas, subscription or one-off?** A: **Only that software subscriptions are under consideration, with no pricing or timeline disclosed.** Management said updates will be announced in due course. A BD team has been set up and is discussing VLA usage and expansion into other domains with partners. **Q: What is the profitability timeline for the humanoid robot biz? How many units are needed to break even?** A: **It is too early to comment on financial outlook and volume, but profitability should come much faster than in autos.** The focus is on milestones: achieving SOP capability by year-end, deploying internally first, and ramping to external customers from 1H next year. Management believes humanoid GP potential far exceeds autos, with hardware margins already well above autos. Future revenue from AI model training, upgrades, and software will carry even higher margins. Capex and investment needs are far lower than autos. Once volume ramps, profitability could materialize much faster than in the auto biz. **Q: Will profitability of the robotics and auto businesses be disclosed separately?** A: **No separation has begun; the announcement allows 18 months, and reporting will remain 100% consolidated either way.** The two businesses currently operate together with a focus on tight synergy, reusing capabilities in AI, advanced manufacturing, powertrain, and supply chain. For now, they will be treated as one. As mass production and commercialization scenarios become clearer, separation may be considered, but based on the equity structure, the biz will remain 100% consolidated, and even a planned separation would not affect financial statements. \ **Risk disclosure and statement:**[**Dolphin Research disclaimer and general disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [VOW.DE](https://longbridge.com/en/quote/VOW.DE.md) - [09868.HK](https://longbridge.com/en/quote/09868.HK.md) - [XPEV.US](https://longbridge.com/en/quote/XPEV.US.md) - [09988.HK](https://longbridge.com/en/quote/09988.HK.md) - [BABA.US](https://longbridge.com/en/quote/BABA.US.md) - [89988.HK](https://longbridge.com/en/quote/89988.HK.md) - [HBBD.SG](https://longbridge.com/en/quote/HBBD.SG.md) --- > **Disclaimer: This article is for reference only and does not constitute any investment advice.**