Exclusive interview with Han Xu, Founder and CEO of Waymo: L3 will not compete with L4 for business; ultimately, L3 requires collaboration between tech companies and automakers that already possess the full autonomous commercial operation capabilities of L4.

Everyday Economic News Reporter | Zhang Rui   Editor | Liao Dan

Following last year’s government work report which emphasized the development of intelligent connected new energy vehicles and next-generation smart terminals and manufacturing equipment, this year’s government work report again clearly states “promote the rapid adoption of next-generation intelligent terminals and intelligent entities.”

As the “world’s first stock for general autonomous driving” and the “world’s first stock for Robotaxi,” in November last year, WeRide officially listed on the Hong Kong Stock Exchange, becoming the first autonomous driving company to be dual-listed in the U.S. and Hong Kong markets, and the only tech company worldwide holding autonomous driving licenses in 8 countries.

Founder and CEO Han Xu is an internationally renowned expert in computer vision and machine learning, and a pioneer in autonomous driving research and applications. Last year, Han Xu was invited to join Singapore’s Autonomous Vehicle Advisory Committee, participating in the formulation of Singapore’s future national policies, operational standards, and technical routes for autonomous vehicles.

Industry experts generally believe 2026 will be the “decisive year” for autonomous driving. What are the key quantitative indicators for breaking through the commercialization of L4? Will the policy opening for L3 impact the pace of L4 commercialization? After dual listing, has WeRide set more specific milestones for breakeven? Is there consideration to release humanoid robot products? Where will the company’s most critical resource investments be in 2026?

With these questions in mind, recently, Daily Economic News (NBD) conducted an exclusive interview with Han Xu.

Scene data + cost control are core technological barriers for China’s L4 companies

NBD: You have repeatedly emphasized that China’s complex mixed traffic scenarios are an “excellent testing ground” for autonomous driving technology. Now that WeRide’s technology has been validated in 8 countries worldwide, compared to international giants like Waymo, what do you believe are the core technological barriers for China’s L4 companies?

Han Xu: The deep accumulation of scene data and the ability to control the costs of core hardware are the main barriers for China’s L4 companies.

On one hand, China’s complex mixed traffic (pedestrians, vehicles, non-motorized vehicles coexisting; complex urban traffic and interchanges; vast rural towns and suburban roads) has strengthened and refined autonomous vehicles’ ability to handle highly unpredictable and challenging scenarios. This is the fundamental guarantee that allows us to quickly adapt in overseas markets such as the Middle East.

On the other hand, China has a mature vehicle manufacturing and autonomous driving industry chain. Our autonomous driving kits built on automotive-grade chips and high-performance sensors can ensure safety redundancies while achieving better cost control.

NBD: Currently, industry consensus sees 2026 as the “decisive year” for autonomous driving. What development trends do you foresee in the global autonomous driving industry over the next 3–5 years?

Han Xu: Over the next 3–5 years, L4 autonomous driving will accelerate into large-scale commercial applications, with some cities entering full-city open deployment. We will see more “thousand-vehicle” and even “ten-thousand-vehicle” Robotaxi fleets in more cities.

NBD: By 2026, L3 passenger vehicles will see large-scale deployment. Some believe this will divert industry resources and impact the commercialization pace of L4 autonomous driving. How do you view the competition and integration among the three technical routes: L2++, L3, and L4?

Han Xu: In fact, I disagree with the view that “L3 policy opening will impact L4 commercialization,” and I don’t think “L3 will compete with L4 for business.” Technologically, L3 is a simplified version of L4, not just an upgrade from L2++. Transitioning from L2++ to L3 is very difficult.

Therefore, I believe that the realization of L3 ultimately requires collaboration between tech companies like WeRide, which already possess the capability for pure L4 autonomous commercial operations, and traditional automakers.

NBD: You previously stated that the “Fifteenth Five-Year Plan” will be the “breakthrough year” for L4 autonomous driving commercialization. Now that WeRide’s Robotaxi (autonomous taxi) has entered the “thousand-vehicle era,” with a gross profit margin soaring 1123.9% year-over-year in Q3 2025, what do you see as the core quantitative indicator for breaking through L4 commercialization?

Han Xu: I believe that breaking through L4 commercialization involves several aspects. First is fleet size. As of January 2026, WeRide has deployed over 1,023 Robotaxi vehicles globally, gradually forming a scale effect.

Second is fully unmanned operation. Currently, WeRide has achieved fully unmanned operation in Guangzhou, Beijing, and Abu Dhabi, and will soon launch fully unmanned services in Dubai. In November 2025, WeRide also obtained Switzerland’s first license for fully unmanned operation.

Additionally, whether the unit economics of operating vehicles become positive is equally important. By 2025, WeRide’s Middle East subsidiary’s Robotaxi business had achieved operational profitability, demonstrating that under appropriate operational efficiency (average daily trips per vehicle) and cost structure, the business model is viable. Our goal for 2026 is to realize positive unit economics in more global markets.

NBD: To achieve this, where will the company’s most critical resource investments be in 2026?

Han Xu: In terms of resource investment, first is technological R&D, including infrastructure, computing resources, talent recruitment, and model training, to ensure continuous technological leadership and safety.

Second is accelerating overseas market expansion, especially in the Middle East (Saudi Arabia, UAE), Europe, and Southeast Asia, leveraging higher service prices and favorable exchange rates to achieve better economic models.

Third is expanding the fleet size, aiming to build a fleet of 2,000–3,000 Robotaxis.

“Orderly expansion and continuous self-sustaining growth,” with sufficient capital reserves to support about ten years of future business development

NBD: Having achieved dual listing in the U.S. and Hong Kong markets, investor expectations have shifted from “storytelling” to “financial reports.” Has WeRide set more specific milestones for breakeven?

Han Xu: Since its founding, WeRide has adhered to the development strategy of “technology productization and commercialization,” emphasizing creating sustainable long-term value for institutional and individual shareholders through high-quality technology, products, and services. I firmly believe profitability is inevitable. The industry consensus is that autonomous driving tech companies will turn profitable before 2030, and WeRide is steadily progressing toward this goal.

NBD: In the current capital environment, does the company prefer to continue expanding globally through financing, or focus more on healthy cash flow, even considering divesting non-core businesses to accelerate profitability?

Han Xu: In the current capital environment, our strategy is “orderly expansion and continuous self-sustaining growth.” Our target markets for expansion are those with clear profit expectations; our major investments are in projects with definite returns. Meanwhile, we maintain healthy and prudent capital reserves. As of September 30, 2025, WeRide had approximately 5.4 billion RMB in cash reserves, enough to support about ten years of future operations.

NBD: Currently, WeRide has a product matrix covering mobility, freight, sanitation, with Robotaxi as the flagship. What is the core logic behind multi-scenario coordinated development? Will the company adjust resource allocation to focus more on autonomous freight or sanitation, which are believed to be “easier to monetize”?

Han Xu: The core technology base for high-speed passenger (Robotaxi), low-speed passenger (public transit micro-circulation), low-speed freight (Robovan), and low-speed sanitation (Robosweeper) is interconnected. From the start, we confirmed the development strategy centered on high-speed passenger commercial scenarios, i.e., Robotaxi. Building on this, we expanded into low-speed sanitation, freight, and micro-circulation, and developed L2-level assisted driving solutions.

The core technology behind this is our self-developed WeRide One autonomous platform, which enables high reusability of software, hardware, and cloud components. This allows WeRide’s technology to be scaled across a broader range of products and scenarios, reducing R&D, operation, and supply chain costs, and accelerating commercialization across various fields and customer segments.

Of course, we will not shift our strategic focus to freight or sanitation; Robotaxi remains our primary business focus.

Focus on embodied intelligence, but autonomous driving remains the strategic core

NBD: In January this year, WeRide released its self-developed general simulation model GENESIS, capable of constructing a realistic simulated city in minutes to replicate extreme tail scenarios. Some industry voices suggest that by 2026, the technological focus will shift from “end-to-end” to “world models.” Does the release of GENESIS indicate that WeRide has already bet on the world model route? What are the fundamental differences between GENESIS and the current mainstream “end-to-end” approach?

Han Xu: “End-to-end” refers to building an AI model that takes input data and produces output directly, without going through modular stages (perception—planning—decision—control), unlike traditional segmented, multi-module approaches.

GENESIS, as a “world model,” is a high-fidelity simulation platform and an AI simulator of the physical world. Its core function is to efficiently simulate various extreme scenarios in a short time, making model training and operational optimization safer and more efficient.

These are different technical concepts rather than distinct routes, and data generated by GENESIS can be used for “end-to-end” model training.

NBD: How do you view the weighting relationship between “simulation-generated data” and “real-world road test data” in future model training?

Han Xu: Both simulation data and real-world road test data are very important. Real-world data is a crucial reference for simulation data, helping to improve its fidelity. Conversely, simulation data addresses the difficulty, long cycle, and high cost of collecting real-world data for extreme tail scenarios.

NBD: Transitioning from a road-test-centric R&D model to “simulation-driven development,” how much can GENESIS accelerate WeRide’s algorithm iteration speed? What quantifiable changes will this shift bring to cost reduction, commercialization efficiency, and full autonomous operation?

Han Xu: The application of GENESIS significantly reduces the mileage and manpower needed for physical road testing, lowering data collection and annotation costs by 75%.

It also creates a “digital universe” that can be generated, expanded, and evolved at any time. With GENESIS, our “AI drivers” can familiarize themselves with driving environments in any city worldwide within minutes, speeding up global commercialization deployment.

Additionally, GENESIS’s built-in “AI diagnostics” module can automatically detect and analyze undesirable driving behaviors, continuously evolving, greatly improving the efficiency of autonomous driving algorithm iteration.

NBD: Cross-industry humanoid robots have become a recent trend. Does WeRide’s accumulation in general AI and simulation models have the potential for transfer to embodied intelligence? How do you view the idea that “autonomous driving is the first step toward embodied intelligence”?

Han Xu: More precisely, autonomous driving is the first large-scale application scenario of Physical AI.

Autonomous vehicles can be understood as wheeled robots, both following the “perception—decision—execution” architecture. Many underlying algorithms for perception, decision-making, trajectory planning, and simulation platforms are reusable.

However, autonomous vehicles operate on open roads where safety is the top priority, whereas embodied robots mainly face indoor scenarios with some tolerance for errors. From this perspective, there are fundamental differences.

NBD: Will WeRide consider releasing humanoid robot products this year or in the coming years?

Han Xu: We will continue to pay attention to this track, but our current core strategy remains focused on autonomous driving.

NBD: What are WeRide’s mid- to long-term strategic goals, and how will it continue to consolidate its leading position in the global general autonomous driving field?

Han Xu: Our mid- to long-term goal is to deploy tens of thousands of Robotaxis worldwide by 2030. We will adhere to the principles of technology productization and commercialization; leverage first-mover advantages in overseas markets like the Middle East, Singapore, and Europe; deepen efforts, expand fleet size, operational scope, and market share abroad; and continue investing in R&D to build a strong technological moat, increasing the value of our WeRide One platform and GENESIS simulation platform.

We will continuously strengthen and expand our upstream and downstream partner network, collaborating with Uber, Grab, and others, to realize from small-scale commercial deployment to large-scale global commercialization.