Aug 12, 2026 earnings call — Overseas robotaxi expansion and L2+/L3 commercialization
Revenue RMB 232 million, up 82% year-over-year and 103% quarter-over-quarter, driven by overseas robotaxi growth and L2+/L3 mass production.
Revenue $0.0B (est $0.0B) · EPS $-0.18 (est $-0.13)
Key quotes
"Our overseas operations are built upon an asset-light model. We do not own operating vehicle assets."
"During the quarter, we delivered approximately 30,000 units of our L2+/L3 solutions, making the beginning of a scaled high-growth commercialization for our L2+/L3 business."
"In this quarter, we officially launched the WIT model, which is a physical fact foundation large model, completing a fully -- a full capability loop spanning data comprehension, data generation and end-to-end technical stack."
Analyst Q&A 7
Resource allocation across business lines/regions
Q (Ming Chung · Citi): How do you prioritize resource allocation across L4, L2++, L3 and AI infrastructure? What are your goals for the second half of the year?
A: Management said their Genesis and WIT models let one investment support L4, L2+/L3 ADAS and AI infra together, creating a "double data flywheel." They aim toward an accumulated 0.5 million installation goal, to push more car OEMs to adopt ADAS, expand the robotaxi fleet, and get more robotics companies to use their AI infrastructure.
"So we build up our infra very, very strongly and generate data from L4 and also generate data from ADAS and make these 2 things complement each other, build up what we call it double data flywheel." — Xu Han (CEO·CHAIR)
Overseas expansion plan and repeatability
Q (Tim Hsiao · Morgan Stanley): Can you share more detail on your overseas strategy and whether it can be replicated across different regions?
A: They said overseas is an important growth driver and already contributed approximately almost 40% of group revenue. They operate an asset-light model: license their digital AI driver to local partners, handle localization and permits, and scale quickly without heavy capital; ARR per vehicle is said to exceed USD 40,000 to USD 60,000 and they are focused on growing fleet size in cities where they already operate.
"For the Q2 in the past quarter, we launched 3 new robotaxi deployments and in Spain and in Switzerland and in Denmark, and in the Middle East." — Xuan Li (CFO)
Driver-assist system progress and robotaxi threat
Q (Paul Gong · UBS): What is the latest progress and future plan for your L2+/L3 ADAS business and your competitive advantage with a small team? Also, can OEMs or ADAS companies use ADAS data to build robotaxi and does that threaten your robotaxi business?
A: Management said their ADAS is, in urban China, on par with FSD and the founder demonstrated a live 2-hour drive without intervention. They reported rapid ADAS deployment: 30,000 units installed in a quarter, revenue growth of 2,600%, 26x year-over-year and 220% quarter-over-quarter, a goal of 100,000 units this year and 0.5 million cumulatively next year, and they said their Genesis and WITT models let a small team build strong ADAS while preserving a large safety moat between ADAS and L4 robotaxi.
"And in terms of the installation units, right, our revenue grows by 2,600%, 26x year-over-year and 220% quarter-over-quarter." — Xu Han (CEO·CHAIR)
Future R&D and operating costs
Q (Tianyu Lu · Citic): How do you view the future trajectory of operating expenses, particularly R&D spend?
A: They said operating leverage is emerging: "revenue grew 87% this quarter and R&D only increased like 36%." R&D is shared across L4, L2++ and L3 so it does not scale linearly; they said peak AI infrastructure investment has passed, R&D cost per vehicle should decline, and they target positive cash flow in a single quarter by 2028 and breakeven for full year 2029.
"So revenue grew 87% this quarter and R&D only increased like 36%." — Xuan Li (CFO)
WIP model and video data use
Q (HSBC): You just launched the WIP model. Please share more about it.
A: They explained WIP plus Genesis lets them analyze video into component functions, learn functionality, and reassemble to generate long-tail training data. This boosts training capability, lowers training cost, and lets them distill on-board models that win urban challenges despite using a lower-TOPS controller.
"Genesis gives you another capability of learning the functionality of components and put them together to build another machine." — Xu Han (CEO·CHAIR)
Competitive edge and advantages
Q (Ming-Hsun Lee · BofA): Can you articulate your competitive moat given intensified competition and split market views?
A: They pointed to their foundation models (WIP and Genesis), their large-scale L4 operations globally, and the data feedback loop between L4 robotaxi and L2/L3 ADAS. They said this dual data flow and their results show they are a leader in both L2++ and L4.
"We can gather all kinds of different data and improve our model." — Xu Han (CEO·CHAIR)
Asset-light model and platform relationships
Q (LightShed): Under your asset-light model, what exclusivity or minimum volume commitments do you have with platform partners like Uber, and what leverage do you have if platforms route demand to other partners? Also, how do OEMs like Mercedes evaluate you versus a full NVIDIA stack?
A: They said the asset-light model helps fast global expansion and they work with multiple platform and local partners; they sell assets to local fleet operators, keep a BOM advantage from China, and highlight safety and regulatory permits as key barriers—they have autonomous driving license in 8 countries and operate in 12. On the OEM/NVIDIA follow-up, management said contract details cannot be revealed and spoke broadly about creativity and engineering rather than providing specifics.
"Right now, we have autonomous driving license, official license on 8 countries." — Xuan Li (CFO)