How can the Robotaxi industry achieve rapid growth?
The large-scale commercialization of Robotaxi is not a single-line advancement of a technical issue but a multi-dimensional, multi-stakeholder co-evolution process. To move from the current ODD "greenhouse pilot" to real large-scale implementation, it is necessary to simultaneously break through four key thresholds: technology and engineering, economic advantages, regulations and ethics, and social responsibility.
Robotaxi (autonomous taxi) is generally considered one of the most imaginative tracks in the era of intelligent driving. Once the L4/L5 autonomous driving technology is truly mature, it will have a potentially subversive impact on the traditional taxi and ride-hailing industries, and even the entire automotive industry structure.
The market has different predictions about the market capacity and development speed of Robotaxi. The most radical prediction comes from Cathie Wood, the star fund manager of ARK Investment Management. She once estimated that by around 2030, the global Robotaxi fleet size may increase to 50 million vehicles, corresponding to an annual travel revenue of 8 - 10 trillion US dollars, and the total ecosystem is expected to create an enterprise value of about 34 trillion US dollars.
Other mainstream institutions, such as Citigroup, Goldman Sachs, and Frost & Sullivan, give much more conservative predictions. For the Chinese Robotaxi market, the predictions for 2030 are roughly between 10 billion and 70 billion US dollars, and for 2035, they fall between 40 billion and over 100 billion US dollars. For the global market, the predictions for 2030 mostly concentrate in the range of 20 billion to 100 billion US dollars, and for 2035, they increase to the scale of hundreds of billions of US dollars. Even based on these relatively conservative predictions, Robotaxi still points to a burgeoning market with a considerable scale and amazing growth rate.
To understand these predictions, we must understand the current development situation of the industry, find the underlying logic of the industry's development, and then integrate the two and project into the future.
Two Models, Two Scenarios
In all application fields of artificial intelligence, the United States and China are at the forefront of the world, and Robotaxi is no exception.
In 2025, the market size of the US Robotaxi was about 300 million US dollars, accounting for about 0.5% of the total US ride-sharing market of 58 billion US dollars. Waymo is one of the earliest pioneers in systematically promoting Robotaxi. By the end of 2025, it provided paid Robotaxi services in four cities: San Francisco, Los Angeles, Phoenix, and Austin, with a fleet size of about 2,500 vehicles. It plans to expand the fleet to at least 3,500 vehicles this year, extend the service coverage to about 12 US cities, and explore overseas cities. The latest weekly order volume reached about 500,000 (in March 2026), about 50 times that of three years ago (in May 2023). By early 2026, Waymo's fully autonomous driving fleet had accumulated more than 170 million miles (about 274 million kilometers, ranking first in the industry) on real roads, with about 70 - 80 million miles added in 2025 alone.
Tesla is an important variable in the US Robotaxi landscape. It is advancing along two strategic paths: one is to enable existing mass-produced vehicles equipped with FSD (Tesla's assisted driving) to access a distributed Robotaxi network in areas where regulations permit through software upgrades; the other is to develop a dedicated Robotaxi model, Cybercab, without a steering wheel and pedals, and plans to start mass production this year. However, Elon Musk's plans often have significant uncertainties in the implementation schedule, so there are many variables.
In addition, the transition from assisted driving to autonomous driving is not a smooth extrapolation in terms of technology but requires a gradient leap in aspects such as long-tail handling ability, system redundancy, and engineering reliability. Tesla hopes that the FSD, which currently serves supervised driving, can directly undertake autonomous driving tasks in the future without major hardware modifications, which means great pressure in terms of engineering and safety standards.
The Chinese market presents a different picture. According to Goldman Sachs' calculations, the market size of the Chinese Robotaxi in 2025 was about 54 million US dollars, also in the initial stage, but with more participants and more intense competition. The leading players include Baidu's Apollo Go, Pony.ai, and WeRide.
Among them, Apollo Go, relying on Baidu's Apollo platform, has obtained various permits for driverless operations in Beijing, Wuhan, Chongqing, Shenzhen, Changsha, and other places. It has achieved 24/7 fully driverless operation in some areas of Wuhan and publicly announced that some single vehicles have achieved break-even on certain routes. It is one of the Robotaxi brands with the largest order volume and driverless operation scale in China at present.
Pony.ai and WeRide are conducting Robotaxi pilots in cities such as Guangzhou, Beijing, Shanghai, and Shenzhen. At the same time, they are actively expanding overseas, cooperating with local governments and platforms in the Middle East, Singapore, and other places to build a dual-line layout of "domestic pilots + overseas commercialization".
The second-tier players include DeepRoute.ai, DiDi Autonomous Driving, Momenta, etc. They have obtained demonstration operation qualifications in some cities and are actively exploring business models. Some automobile manufacturing enterprises, such as XPeng, have also started the technical and operational preparations for Robotaxi-related businesses.
A prominent feature of the Chinese Robotaxi market is top-level design and policy-driven development. Under the policy concept of "central guidance + local pilots", the central government will issue top-level strategic documents to guide the development of the autonomous driving industry and select key cities as test demonstration areas, providing local policy space for first-mover trials.
Cities such as Beijing, Shanghai, Guangzhou, Shenzhen, Wuhan, and Chongqing have introduced phased policies for testing, demonstration operations, and commercialization, allowing enterprises to first operate with human safety drivers in specific areas and then gradually move towards driverless operation inside the vehicle, remote monitoring outside the vehicle, and finally full driverless operation. The management measures in Wuhan, Beijing Yizhuang, and other places also provide relatively clear charging rules and insurance guidelines, which provide an institutional basis for the early commercial closed-loop.
The Question of Safety
Safety is a core consideration in the development of Robotaxi. Robotaxi must be safer than human driving to be widely applied.
According to the latest data disclosed by Waymo, compared with the average human drivers in the same operating cities, the number of accidents causing injuries by Waymo's autonomous driving system has been reduced by about 82%, and the number of accidents causing serious injuries or more severe consequences has been reduced by about 92%. These figures are quite encouraging.
Moreover, Waymo's methodology is relatively rigorous. The human benchmark is not the national average but is constructed based on the police accident and driving mileage data of its operating cities. When making comparisons, the actual driving roads and regional distributions of Waymo are aligned as much as possible through the reweighting method, and corresponding restrictions are imposed on the time period and road conditions to compare the accident rate per million miles under the conditions of "the same city, similar roads, and operating time periods". This is much more mature and reliable than similar comparisons made by other companies.
Does this mean that Robotaxi is already safer than human driving? Not really. It only shows that within the currently carefully selected and strictly restricted ODD (Operational Design Domain), the safety performance of Robotaxi is better than the average level of human drivers in the local area. The fundamental weakness of the current mainstream intelligent driving route is the lack of robust generalization ability for unseen long-tail scenarios.
The excellent safety data shown by Waymo in cities such as San Francisco and Phoenix were obtained within the carefully selected ODD. The ODD is not an arbitrarily drawn circle but a "relatively safe area" defined after manual screening and engineering trade-offs. When selecting blocks, areas with regular roads, clear markings, and relatively low complexity of pedestrian and vehicle flows are preferred. When selecting cities, places with relatively mild climates and fewer extreme weather conditions are preferred. In case of bad weather, it can choose to actively shrink operations or suspend services. Once the operating scope is expanded to the whole of the United States, throughout the year, and under all road conditions, the number of long-tail scenarios will increase exponentially, and "unseen" combinations will become the norm rather than the exception.
The "world model" may be a long-term direction to fundamentally alleviate this problem, but the more realistic issue is whether relying on real data accumulation and simulation amplification to approach the long-tail coverage as much as possible can become an acceptable solution for large-scale operation at both the technical and social levels before the world model can be broken through in the short term.
The "vehicle-road-cloud integration" path adopted in some Chinese solutions can be regarded as a way to compress the long-tail problem. Some pilot areas do not completely rely on vehicle-side perception but simultaneously install roadside sensors, signal priority mechanisms, dedicated lanes, and cloud control platforms. Through V2X (Vehicle to Everything), part of the "vision" and "judgment" is moved forward to the infrastructure level, which to some extent makes up for the shortcomings of single-vehicle intelligence. The exploration of vehicle-road-cloud collaboration, by using roadside equipment and V2X to pre-provide key information such as over-the-horizon, occlusion, and blind spots to the vehicle, essentially uses the "external attachment" of the infrastructure to make up for the shortcomings of single-vehicle intelligence.
Even with the support of internal and external intelligent driving, Robotaxi still cannot completely eliminate the long-tail problem. The current ultimate solution is to have a human take over remotely when Robotaxi encounters a long-tail problem that it cannot solve. To some extent, this can be regarded as a "cheating" configuration for Robotaxi as an autonomous vehicle. It may not be the ideal state of autonomous driving, but in practice, it is a smart support for engineering reliability and a necessary means to achieve "controllable failures and orderly exits" at present.
Crossing the Chasm of Technology, Cost, and Institution
The autonomous driving technology on which Robotaxi relies is not a simple enhancement of functions compared to assisted driving but a transformation of the operating paradigm.
What Robotaxi really needs to cross is not just a single technical threshold but a whole set of thresholds from an experimental business to a social basic service. In terms of technology and engineering, can it compress the long-tail risks to a socially acceptable level within a wider ODD and establish sufficiently robust system reliability and degradation mechanisms? Economically, can it get rid of long-term subsidies and capital infusion and run a profitable model? Institutionally, can regulations, liability division, insurance mechanisms, and ethical consensus keep up simultaneously? Socially, is there an acceptable exit and job transfer arrangement for the group of professional drivers who will be partially replaced?
Only when progress is made in these aspects simultaneously can Robotaxi gradually move from the current limited-scenario commercialization to large-scale commercialization in a broader sense.
First, in terms of technology and engineering, there is a huge gap between a Robotaxi being "able to run" and being "able to run on a large scale and safely". Engineering reliability is a challenge that is easily underestimated. A Robotaxi needs to operate for more than ten hours and cover hundreds of kilometers every day, and it needs to maintain an extremely low failure rate and extremely fast fault recovery ability under various working conditions. This involves a whole set of engineering systems such as sensor redundancy and degradation detection, thermal management and long-term stability of the computing power platform, safety verification of OTA upgrades, and the robustness of the fleet scheduling system.
Second, the large-scale commercialization of Robotaxi ultimately depends on whether the economic ledger is feasible. From the perspective of a single vehicle, the additional costs include high-precision sensor suites, high-computing power platforms, high R & D investment, and the cost of safety drivers in the initial operation. The biggest cost reduction comes from the labor cost of drivers. Among the cost items, a very crucial and often underestimated variable is the number of vehicles that a remote safety driver can cover, which directly determines the extent to which the labor cost can be diluted.
According to the data disclosed by some companies, the single-vehicle economic model has begun to approach the break-even point within a local ODD. For example, the management of Pony.ai disclosed at the third-quarter earnings conference in 2025 that its seventh-generation Robotaxi deployed in Guangzhou had achieved positive single-vehicle profitability at the city level in November of that year, with an average of about 23 orders per vehicle per day. In terms of the ratio of safety drivers to vehicles, when Baidu's Apollo Go became well - known in Wuhan in 2024, it once publicly announced a ratio of "1:3", that is, each remote safety driver monitored about three Robotaxis simultaneously. By 2025, by enabling the vehicles to better handle difficult scenarios, such as automatically decelerating and pulling over to enter the minimum risk state, remote safety drivers no longer need to be ready to take over at all times but provide decision - making suggestions more when the vehicles encounter difficult scenarios to help the vehicles leave local complex road sections. As a result, the ratio of safety drivers to vehicles has been significantly improved. Currently, leading Robotaxi enterprises can achieve a ratio of about 1:20 in some cities, greatly reducing the operating labor cost per vehicle.
The economic viability of Robotaxi largely depends on the scale effect. Only when the fleet size is large enough can the high R & D investment, vehicle procurement cost, and operation and maintenance cost be diluted. At the same time, a sufficiently high market penetration rate can bring high - frequency orders, ensuring the maximum utilization rate of vehicles and thus reducing the cost per service order. This requires the technical maturity to support stable operation in a wide area and also requires consumers' wide acceptance and trust in Robotaxi services. Large - scale deployment and operation will bring a positive cycle of data accumulation and algorithm iteration for operating enterprises, further optimizing the operation efficiency and reducing the cost.
Third, for Robotaxi to be widely used on the road, technology and economic viability alone are not enough. A whole set of matching governance structures is also needed, including legal liability, regulatory mechanisms, ethical boundaries, and social acceptance, all of which need to be redesigned and continuously adjusted. Robotaxi is a new type of business that involves both public safety and the right to use the road, and the depth of its commercialization highly depends on the completeness of the regulatory and governance systems.
When an accident occurs to a Robotaxi, liability determination will be a core legal issue: is it the responsibility of the manufacturer, the software developer, or the operator? In some special cases, is there any responsibility of the roadside infrastructure or the traffic management system? What kind of fault liability or no - fault liability each link should bear is directly related to the incentive structure of all parties in safety investment, and all these require clear legal definitions.
Due to the continuous operation characteristics of Robotaxi and its complex technology with a black - box nature, the complexity and difficulty of regulation are higher than those of the ride - hailing and automobile manufacturing industries, covering aspects such as access standards, operation permits, data management, and network security. For example, what safety indicators, test mileage, and ODD definitions are qualified for commercialization; whether each version upgrade of software and models needs to be re - verified and filed; how long the data should be stored and in what way it should be opened to regulators and accident investigations.
The ethical issues are directly related to the social trust boundary of Robotaxi. The classic "trolley problem" is just an extremely abstract version. More common disputes in reality include: how conservative or radical the system should be under uncertain conditions, and whether this trade - off will actually form algorithmic discrimination against certain groups; whether the system is allowed to increase the risk of individuals outside the vehicle to protect the passengers inside the vehicle in an inevitable collision scenario; whether these strategies should be determined by the enterprise itself or need to be determined within a certain open and transparent social negotiation framework.
Only when these issues are openly discussed and clear "ethical traces" are left in algorithm design, product descriptions, and regulatory rules can Robotaxi obtain real - world social trust.
Fourth, the large - scale commercialization of Robotaxi means that the taxi and ride - hailing industries will undergo structural changes, and millions or even tens of millions of professional drivers may face unemployment. How to design an orderly and bearable exit and transformation mechanism to deal with the social impact brought about by this large - scale employment substitution is a social fairness issue that must be considered in advance and systematically planned.
Based on comprehensive data estimates, the current scale of active ride - hailing/taxi drivers in China is about 10 million. If the development scenario of Robotaxi is to quickly replace traditional travel services, the spontaneous adjustment of the employment market is likely to lag behind, and in the absence of proper coping strategies, a large - scale employment shock may lead to social problems.
With the rapid and comprehensive development of AI, the potential impact on employment will actually affect all industries. The social security system needs to make early preparations to deal with this structural unemployment. This may include improving the unemployment insurance system and exploring new social welfare models such as the Universal Basic Income (UBI) to build a social mechanism that can buffer the technological shock and promote the orderly transformation of the labor force.
From a macro perspective, the popularization of Robotaxi will also have a profound impact on urban planning, transportation infrastructure, and the business ecosystem. For example, the reduction of parking lots and the evolution of public transportation models. While promoting technological progress, we must also consider the possible social problems it may cause and ensure that technological development can benefit a wider social group through top - level design.
The Mobility Revolution under Co - evolution
The large - scale commercialization of Robotaxi is not a single - line advancement of a technical issue but a multi - dimensional, multi - stakeholder co - evolution process. To move from the current ODD "greenhouse pilot" to real large - scale