Author | An Yun
As the afterglow of the National Day holiday lingers, the global autonomous driving industry has hit the "fast-forward" button.
On October 8, Waymo announced the completion of a $5 billion term loan, marking its first debt financing. The loan is collateralized by intellectual property, with lenders including Blackstone, Pacific Investment Management Company (PIMCO), and Sixth Street. On the same day, records from the Texas Department of Motor Vehicles showed that Tesla registered 150 new Cybercabs in a single day, bringing the total to 319, setting the largest single-day increase since the model entered the registration system. Meanwhile, Xiaopeng Group's Robotaxi business officially adopted the Chinese name "Xiaopeng Youyou" ("XPENG YOYO"), and the XPENG YOYO autonomous ride-hailing mini-program was launched simultaneously, inviting users for beta testing.
While some players are raising capital in the financial market to expand their forces, others are accelerating production volume in manufacturing, and new players are also making a strong entry. Behind these moves, a notable change is taking place: the competition in Robotaxi is shifting from verifying whether the technology can achieve autonomous driving to verifying whether a commercial system can operate sustainably. Technology still determines whether vehicles can safely hit the road, but it no longer solely determines the winner. The popularization of End-to-End and VLA (Vision-Language-Action) large models is turning technical capability into a "ticket" for entering large-scale operations, rather than the only advantage to widen the gap.
Going forward, the large-scale supply of vehicles, the continuous growth of orders, and the constant reduction in operating costs will jointly determine whether technical advantages can be cashed in as commercial advantages. The Robotaxi competition has evolved from a contest of single-point technical capabilities to a battle of entire commercial systems. At the data level, the safety baseline for large-scale operations has already been verified.
An independent study by the Insurance Institute for Highway Safety (IIHS) on approximately 50 million miles of autonomous driving operational mileage shows that Waymo experiences 1.28 reportable crashes per million miles, compared to 4.06 for human drivers. As of October 2026, Waymo has entered 15 cities in the United States, with a fleet size of over 4,000 vehicles, completing more than 500,000 paid trips per week. To understand this Robotaxi competition, it can be deconstructed through four dimensions:
The differences among players lie not in "who operates," but in who controls which segments and who they rely on.
01 Who Is Accelerating, and Who Is Blocked by Regulation?
The global autonomous driving industry is forming a "three-front race" pattern among China, the US, and Europe, with major markets almost simultaneously entering the critical window for operational implementation. The Middle East and Europe are accelerating implementation. Wenyuan Zhixing has achieved fully driverless commercial operations in Abu Dhabi and Dubai, covering over 70% of the core urban areas, with a local fleet of approximately 400 vehicles.
In Europe, in partnership with Uber, services have landed in Madrid, Spain, and Zurich, Switzerland. Waymo, taking 15 US cities as its base, is simultaneously launching international expansion, initiating its first test outside the US in Tokyo in 2026, and announcing plans to enter over 20 cities in the future, including London and New York.
Chinese companies are also accelerating their overseas steps. In July 2026, Baidu's Luobo Kuaipao, together with Uber and Freenow under Lyft, launched public road testing in London, planning to provide driverless mobility services to citizens starting in 2027. This marks the first on-road test of Chinese autonomous vehicles in the complex road conditions of London, UK. Meanwhile, Xiaoma Zhixing is expanding its global strategic partnership with Uber, planning to deploy over 2,000 Robotaxis in five European cities based on the existing fleet in Zagreb, Croatia.
Tesla's Cybercab simultaneously launched static tours in Beijing and Shanghai, as well as a tour in Hong Kong, but explicitly stated that it does not involve commercial operation arrangements in the Chinese market. However, policy relaxation does not mean regulatory loosening. In September, just as the Cybercab went online, the National Highway Traffic Safety Administration (NHTSA) announced an investigation into approximately 1,000 Cybercabs to evaluate how Tesla determined that this vehicle, which lacks a steering wheel and pedals, meets federal safety standards through self-certification. This investigation illustrates that autonomous driving safety remains under high-pressure regulation.
Waymo's experience in New York is even more typical. Although it already holds a testing permit, New York State requires a safety driver to be in the vehicle, effectively banning fully driverless commercial services. Among the over 20 cities Waymo plans to enter, regulatory barriers in key markets such as New York and London remain the most uncertain variables.
02 Who Controls Key Segments, and Who Relies on Partners
Since 2025, the competition has extended from algorithmic contests to the combination of technology, vehicles, traffic, and urban resources. The advantage of mobility operators lies in their demand pool and ground-level relationships, while the advantage of technology companies lies in the deep coupling of hardware and software. As to which is more suitable for integrating the industry chain, there is currently no definitive conclusion.
Most recently, Uber has integrated with over six technology providers simultaneously, including Wenyuan Zhixing, Xiaoma Zhixing, Luobo Kuaipao, Waymo, Zoox, and Wayve. Its autonomous driving mobility services have grown by more than 10 times year-on-year. Currently, these services are live in 8 cities, with a target to expand to 15 cities by the end of the year, covering Hong Kong, Madrid, Zurich, Houston, London, Munich, Los Angeles, and the San Francisco Bay Area.
In Europe, Bolt and Lucid have launched a deployment plan for 25,000 Robotaxis, which is also an important step for Bolt to achieve its goal of deploying 100,000 autonomous vehicles on its platform by 2035. In Southeast Asia, Grab has entered the field through a tens-of-millions-of-dollars equity investment in Wenyuan Zhixing.
Currently, we can observe three typical paths: Vertical extension by technology companies (Waymo): developing its own driving system and partnering with Hyundai to produce IONIQ 5 Robotaxis at the Savannah, Georgia factory, while building its own operational system. This features high hardware-software coupling, but requires large capital investment and a long cycle.
Platform aggregation (Uber, Grab, Bolt): relying on existing demand pools and dispatch capabilities, integrating multiple technology providers in an asset-light manner. This allows for rapid expansion but requires coordinating standardized integration with different suppliers, and relies on external partners for technology and vehicle costs.
Full-chain by automakers (Tesla, Xiaopeng): integrating vehicle manufacturing, technology, and operations. Tesla's Cybercab was put into operation in Texas in early September, and plans to support 24/7 operations through FSD iterations; domestically, Xiaopeng Motors launched its Robotaxi business "XPENG YOYO". Manufacturing capability is an advantage, but it faces uncertainties in regulatory certification.
The differences among the three paths lie in the varying degrees of control over four segments: vehicles, algorithms, demand, and urban operations, which in turn determine per-vehicle costs and expansion speed. The judgment of Han Xu, founder of Wenyuan Zhixing, provides another perspective: "The Robotaxi track is very crowded, and the ones that will ultimately be profitable are definitely the companies that solidly solve users' real needs."
03 How to Translate Scale into Efficiency?
When technical capability becomes a prerequisite for entering large-scale operations, the competition will further extend to fleet utilization rate, order density, vehicle depreciation, dispatch efficiency, and cross-city replication capability.
This barrier is composed of three levels.
The first level is the complexity of ground operations. Urban implementation is far more than just technology deployment; it involves a series of "hard and tiring jobs" such as regulatory communication, insurance system construction, parking dispatch, and the formation of ground operation and maintenance teams. Entering every new city means building a ground ecosystem from scratch. This is precisely the core capability that mobility operators have accumulated over the years.
The second level is the flywheel effect of safety mileage. Wang Haojun from Xiaoma Zhixing has an incisive elaboration on this: "Fully driverless is not about deploying first and supplementing data later, but continuously accumulating safety mileage through large-scale operations to gain public and regulatory trust." He draws an analogy to commercial aviation, noting that the accumulation cycle for safety records is relatively long, and ultimately, far fewer players will succeed in running a reliable and cost-effective operation. This "flywheel effect" means that the advantage of first movers will continuously amplify over time. Latecomers, even if their technology is not inferior, need to re-accumulate mileage data to prove L4 safety. The data from Luobo Kuaipao provides another observation sample. As of June 2026, it has cumulatively provided over 23 million autonomous driving mobility service orders to the public, with the fleet's cumulative autonomous driving mileage exceeding 350 million kilometers, of which fully driverless mileage exceeds 240 million kilometers. This shows that large-scale operations not only mean more vehicles put into use, but also mean that enterprises are accumulating longer actual road operation mileage, larger-scale mobility service experience, and opportunities to continuously verify system safety.
The third level is the validation of the profitability model. The prerequisite for large-scale operations is the viability of the per-vehicle economic model, which tests operational efficiency—that is, whether algorithmic capabilities can be translated into stable orders, a lower idle rate, and controllable unit costs. Wenyuan Zhixing provides an industry reference in this regard: over the past six months, the average daily orders per vehicle reached 15, rising to 26 during peak periods, the average passenger waiting time dropped to under 10 minutes, and registered users in the fourth quarter of 2025 increased by over 900% year-on-year. The total cost of ownership (TCO) per vehicle for its Robotaxi in the Chinese market has decreased by approximately 38%, and the Robotaxi business of its Middle East subsidiary has achieved operational profitability, making it one of the few autonomous driving companies globally to achieve profitability in Robotaxi operations in a regional market. However, it is necessary to see clearly what this set of data proves and what it does not prove: order volume, TCO reduction, and regional operational profitability belong to different levels of indicators and cannot replace each other.
To truly judge the profitability of Robotaxis, the following questions must also be answered: how many hours a day does a single vehicle actually carry passengers, rather than driving empty, charging, or waiting; what are the revenue and cost per order respectively; how are vehicle depreciation, insurance, remote assistance, and ground operation and maintenance apportioned; and how many vehicles and orders are needed to amortize the initial investment in a new city.
Most of these data are currently undisclosed, so at this stage, it can only be said that the improvement of operational efficiency is a necessary condition for profitability, not profitability itself.
04 The Dilemma of Large-Scale Operations: Why Does Efficiency Not Equal Profitability?
Even with continuous improvement in efficiency, large-scale implementation still faces multiple bottlenecks, and the "major test" for autonomous driving is far from over.
First, emergency response capability under urban infrastructure failure. The massive power outage in San Francisco in December 2025 remains an alarm bell for the industry. After traffic lights went out, multiple Waymo vehicles broke down at intersections, causing citywide traffic paralysis, and even emergency vehicles could not pass. The Mayor of San Francisco subsequently wrote to regulators, demanding a suspension of Waymo's expansion until it meets stricter safety conditions. During the Independence Day fireworks celebration on July 4, 2026, a similar scene was replayed, with dozens of Waymo vehicles causing massive congestion near the Golden Gate Bridge. These incidents expose more than just technical issues. When urban infrastructure fails, the lack of emergency degradation plans in autonomous driving systems also indicates that a mature collaboration mechanism has not yet been established between autonomous driving companies and urban managers regarding emergency plans and data transparency.
Second, the dual test of per-vehicle economics and operational stability. The validation of the per-vehicle economic model requires simultaneous optimization across multiple dimensions such as fleet size, dispatch efficiency, vehicle depreciation, and energy costs. However, the growth in order scale does not mean that operations have entered a stable state. The operational data of Luobo Kuaipao provides a noteworthy sample. In the first quarter of 2026, its fully driverless operational orders reached 3.2 million; in the second quarter, it was about 1 million. Baidu partially attributed the order fluctuation in this stage to operational adjustments in some domestic cities due to regulatory factors. This shows that the operational scale of Robotaxis depends not only on vehicles and user demand but is also affected by regulatory arrangements and urban operational conditions. Even after accumulating a large number of orders and road operation mileage, enterprises still need to continuously cope with changes in operational rhythm and verify the stability of the per-vehicle economic model in different cities and under different regulatory conditions. Therefore, to measure the commercialization progress of a Robotaxi company, one cannot just look at cumulative orders or single-quarter peaks, but also observe whether order growth can be sustained, whether vehicle utilization rate can be improved, and whether unit costs can decrease as the scale expands.
Third, the fragmentation of regulatory and liability systems. From the NHTSA's investigation into the Cybercab to New York State's de facto ban on fully driverless operations, the fragmentation of regulatory frameworks is the biggest obstacle to global deployment. The establishment of an insurance system is equally lagging. When accident liability is distributed among algorithms, vehicle manufacturers, and operators, the traditional auto insurance model faces fundamental challenges. The 68% safety data from IIHS is a positive signal, but the shift in public perception requires time and continuous safety records for verification. Every occasional accident can cause a disproportionate impact on industry trust.
05 Technical Advantages Must Be Tested by the Operational System
Han Xu once used Churchill's words after the Dunkirk evacuation to position the current Robotaxi process: "This is not the end. It is not even the beginning of the end. But it is, perhaps, the end of the beginning." This autonomous driving marathon is already past the halfway point. The first half compares whose technology can hit the road, and the second half compares whose operational system can survive, transforming technical advantages into advantages tested by the operational system, thereby converting them into commercial advantages. From the operational status of major players globally, fleet size does not equal operational efficiency, and operational efficiency does not equal profitability.
Waymo is a benchmark case of "increasing orders without increasing vehicles," pushing weekly orders from 50,000 to 500,000 with approximately 3,000 vehicles. In contrast, Chinese fleets exceed 4,000 vehicles, yet their utilization rate is relatively lagging. However, since 2026, Xiaoma Zhixing has achieved per-vehicle profitability in Guangzhou and Shenzhen, and Wenyuan Zhixing's Middle East operations are close to break-even, both exploring the possibility of further improving efficiency and profitability.
Why is the fleet size not proportional to the order size? The answer lies in the relationship between fleet utilization rate, order density, and cost structure, which are also the core indicators most worth tracking in the future. It can be said that scale can be bought, efficiency can be ground out, but profitability is the true critical point. The next phase of Robotaxi depends on who can organize vehicles, orders, urban resources, and safety systems into a sustainable and replicable commercial system. Technical advantages will only become true commercial advantages after being tested by the operational system.