Pony AI Q3 Earnings Call Transcript: Seventh-Generation Robotaxi Achieves City-Level Unit Profitability, Accelerating Fleet Expansion to 3,000 Vehicles by Next Year

Following the earnings release, Peng Jun, Founder and CEO of Pony AI, Tiancheng Lou, Founder and CTO, and Haojun Wang, Co-founder and CFO, attended the earnings call to discuss key highlights of the financial results and address analysts' inquiries.

– Successfully completed a dual primary listing on both the Hong Kong Stock Exchange and the New York Stock Exchange, marking the largest IPO in the global autonomous driving industry in 2025. Following its Hong Kong listing, the company secured over USD 800 million (approximately RMB 6 billion) in new cash, which will be used to accelerate mass production, large-scale commercialization, technology research and development, and expansion into new markets;

– The seventh-generation Robotaxi has commenced fully driverless commercial operations in Guangzhou, Shenzhen, and Beijing. The seventh-generation Robotaxi has achieved positive unit-level profitability on a city-by-city basis, officially ushering in a new phase of commercial profitability and rapid expansion;

– Revenue from the Robotaxi business has experienced rapid growth, with income reaching RMB 47.7 million, representing an 89.5% year-over-year increase, including a more than 200% year-over-year rise in passenger fare revenue;

– Pony AI’s total fleet of Robotaxis has reached 961 vehicles, of which 667 are seventh-generation Robotaxis. The company is expected to surpass its target of over 1,000 vehicles within the year ahead of schedule and plans to triple its fleet size to over 3,000 vehicles by the end of 2026;

– The production cost of the seventh-generation Robotaxi will further decrease in 2026, with the BOM cost of its autonomous driving suite dropping by an additional 20% compared to this year’s level, supporting the company's 2026 mass production plan;

– The company continues to deepen its global strategic layout and consolidate its first-mover advantage, with its Robotaxi business now covering eight countries. Through partnerships with multinational automotive groups such as Stellantis, as well as mobility platform companies like Uber, ComfortDelGro, and Qatar National Transport Company, Pony AI is further expanding its Robotaxi operations in Europe and the Middle East, officially commencing Robotaxi testing in Dubai (UAE), Punggol (Singapore), and Doha (Qatar);

As early as 2020, we recognized the importance of surpassing human driving capabilities through reinforcement learning. In the same year, Pony AI began developing its World Model (PonyWorld), which creates highly realistic training environments and behavioral evaluation criteria by integrating high-fidelity simulation, the ability to recreate long-tail scenarios, and a reward evaluation mechanism based on AI self-learning. This ultimately achieves unsupervised, self-evolving closed-loop training, becoming a key technology for our scalable deployment and core competitive barrier. The World Model can generate 10 billion kilometers of test data per week, with over 99% of it being interactive data between autonomous vehicles and other road users, helping virtual drivers self-improve and train models to achieve performance surpassing human drivers. In recent years, the entire autonomous driving and robotics industry has gradually converged toward the World Model approach, validating the forward-thinking nature of the technical path we adopted back then.

In the third quarter, we achieved rapid revenue growth and further strengthened our financial foundation. Total revenue for the third quarter increased by 72.0% year-over-year, primarily driven by strong performance in autonomous mobility services and technology licensing and application services. The net proceeds from our recent IPO in Hong Kong significantly bolstered our balance sheet, combined with the key milestone of achieving profitability for individual seventh-generation autonomous ride-hailing vehicles operating citywide in Guangzhou. These net proceeds will enable us to quickly expand our commercial footprint and inject strong momentum into long-term growth.

Could the management provide an update on the fleet size this year and the outlook for additional vehicles by 2026? What is the deployment plan for fleets across different cities?

As you can see, since the launch of the seventh-generation Robotaxi, our production and deployment pace has far exceeded expectations. Therefore, we are on track to surpass our year-end target of 1,000 Robotaxis this year, and this strong momentum will continue into 2026.Our conservative goal is to expand the fleet to over 3,000 vehicles.This is primarily due to the positive feedback loop created by the rollout of the seventh-generation vehicles. Increased fleet density reduces passenger wait times, thereby enhancing user experience; in turn, a better user experience drives higher vehicle utilization, allowing us to implement premium pricing. This virtuous cycle lays a solid foundation for our rapid expansion.

Moreover, we have started collaborating with fleet operators such as West Lake Group and Sunshine Travel to pilot a light-asset model. Moving forward, we plan to bring more partners on board. This model allows usto achieve larger-scale fleet deployment with reduced capital expenditure,which forms part of our growth strategy.

Regarding fleet deployment strategy: we will deepen our presence in existing markets while actively exploring new market opportunities. The milestone of achieving city-wide profitability per vehicle with the seventh-generation model in Guangzhou validates our business model and strengthens our confidence in expanding operations and partnerships in existing markets, such as China’s first-tier cities. As previously mentioned, scaling up the fleet creates a positive feedback loop. Additionally, we aim to expand into more domestic cities and international markets, which will serve as key growth drivers. Our market entry strategy involves close collaboration with local partners and government agencies to establish market presence and prepare for future growth. Stay tuned for more updates as we continue to deliver promising developments.

Will fare revenue maintain growth in the third quarter of 2025? What is the outlook for fare revenue as vehicle deployment scales up?

Fare revenue growth was particularly strong in the third quarter, with a year-over-year increase of approximately 233%. The vehicles deployed this quarter are mainly fifth-generation and sixth-generation Robotaxi models. We believe this growth has been driven by both demand-side factors and operational optimizations.

On the demand side, we have continuously worked to improve the driving experience and overall user experience, resulting in robust organic user demand growth in China's first-tier cities. This reflects high consumer recognition of Pony AI’s Robotaxi services. For example, the total number of registered users in the third quarter more than doubled compared to the same period last year.

On the operations side, we have continuously optimized fleet management, improving vehicle utilization rates and order completion rates. As mentioned in previous remarks, we have refined our fleet dispatching and deployment strategies,reducing average wait times by approximately 50% compared to the same period in 2024.Meanwhile, we have continued to expand pick-up and drop-off points to create a smoother user experience. In Shenzhen, for instance,we now have over 10,000 pick-up and drop-off points., representing a 300% increase from the end of June.

Driven by dual optimization on both the demand and operational sides, coupled with the ongoing deployment of the seventh-generation vehicles, we believe that as our fleet size continues to expand, fare revenue will maintain strong growth momentum. First, our fleet size is growing exponentially: last year it was 270 vehicles, this year it will exceed 1,000 vehicles, and next year we aim to surpass 3,000 vehicles. This scaled expansion will further enhance network effects, leading to shorter wait times, higher vehicle utilization, and greater user acceptance.

We will also gradually expand service coverage in cities such as Shanghai and Shenzhen. Relevant initiatives are already underway and will further improve coverage and extend drivable mileage. Through these measures, we expect to increase the average ticket size per order.

Congratulations on achieving the city-wide single-vehicle profitability milestone. Could you elaborate on the core assumptions behind this breakeven target, including daily order volume, pricing strategy, daily operating hours, and remote assistance ratios? Thank you.

As you mentioned, we all recognize that achieving city-level single-vehicle profitability is a key milestone for the company and the industry as a whole. It is important to note that since the commercial launch of our seventh-generation Robotaxi, we have already achieved this critical milestone in Guangzhou. We have always believed that China represents the largest ride-hailing market globally, with first-tier cities accounting for a significant share of the national mobility market.

The economics of a single vehicle primarily consist of revenue and cost components. On the revenue side, the key metric is daily revenue per vehicle. Since the launch of the seventh-generation Robotaxi in Guangzhou, as of the most recent two weeks ending November 23, daily revenue per vehicle has reached RMB 299. This revenue figure reflects total ride-hailing service income after deducting discounts and refunds.This average daily revenue of RMB 299 corresponds to an average of 23 daily orders per vehicle, driven by broad and robust user demand.

On the cost side, the economic cost per vehicle is primarily composed of two parts: first, the hardware depreciation cost of the seventh-generation Robotaxi (calculated as annual depreciation over a six-year useful life); second, operating costs, including charging fees, remote assistance personnel expenses, ground support personnel expenses, vehicle maintenance costs, insurance premiums, parking fees, and network charges, among others. Regarding the ratio of remote assistance, we are proceeding as planned toward the target of 1:30 (i.e., one remote operator supporting 30 vehicles).

Upon achieving this milestone, we are confident in capturing a significant share of China’s vast mobility market while laying a strategic foundation for further scaled expansion in both domestic and international markets. This not only strengthens our confidence in expanding fleet size but also attracts an increasing number of third-party enterprises willing to invest in purchasing fleets, aiding our transition to a light-asset model. We believe these factors will collectively drive revenue growth and cost optimization for the company.

We have noticed an increasing number of participants from various sectors entering the Robotaxi industry, particularly original equipment manufacturers (OEMs). Could you share your perspective on these new entrants at the L4 autonomous driving stage? Additionally, could you elaborate on the key technical and operational challenges, such as handling edge cases and managing fleets within the digital economy?

First, I believe it is a positive development to see more companies announcing their entry into the Robotaxi industry, which reflects growing recognition and confidence in the potential for large-scale commercialization of Robotaxis. As awareness increases, more resources and companies will be committed to this space, thereby accelerating the development of the entire industry—a very encouraging sign.

On the other hand,The Robotaxi industry is not one that any new player can easily enter.Currently, none of the new entrants have achieved large-scale deployment of fully driverless Robotaxi services, indicating a high barrier to entry. I believe new players face three main categories of obstacles: business challenges, regulatory challenges, and technological challenges.

In terms of commercial challenges, Robotaxi is not only about the autonomous driving technology itself,but also involves multiple aspects such as user acquisition, vehicle production, fleet dispatching, and fleet maintenance (including cleaning, charging, etc.).As a pioneer and leader in the industry, Pony AI has an early-mover advantage: more vehicles on the road, higher brand recognition, optimized costs across various business operations, and established partnerships that collectively form significant barriers for new entrants.

From a regulatory perspective, Robotaxi has extremely high safety requirements. Regulatory bodies around the world generally impose stricter safety standards on Robotaxis compared to traditional taxis. New players need to progressively demonstrate their safety before expanding to full-scale operations in any city,typically starting with a few dozen or even fewer pilot vehicles, accumulating safety records, gradually increasing the number of vehicles, and expanding operational areas,while obtaining necessary licenses and permits. This process is inherently time-consuming and difficult to accelerate.

The technological challenge, the third and highly critical one, will be further explained by our CTO from a technical standpoint.

From a technical perspective, the industry has widely begun adopting the 'World Model' approach, with both Robotaxi manufacturers and vehicle OEMs leveraging reinforcement learning methods within simulation training environments. We started developing reinforcement learning for autonomous driving about five years ago, which gives us an early technical edge. As the World Model matures, the training loop will increasingly rely on the interaction between the 'virtual driver' system and the World Model:The 'virtual driver' generates feedback in simulations, and the World Model improves accordingly, thereby reducing reliance on real-world data.

Within this closed loop, the world model can generate more training samples for complex scenarios through feedback from 'virtual drivers,' progressively enhancing the model's ability to handle complex or edge cases. However, the real technical challenge lies in how to validate the safety of new technologies in the real world and achieve scaled deployment. With our proven track record in Robotaxi operations, we believe we can quickly seize the next wave of technological innovation opportunities. Additionally, listing in Hong Kong will further accelerate our R&D and iteration cycles, solidify our leadership in technology and operations, and broaden our competitive edge.

What are the main factors contributing to your faster expansion in operational areas? Beyond technology, what are the key drivers? From a technical perspective, do you utilize large language models (LLMs)? If so, how do they help enhance autonomous driving capabilities?

I will divide your question into two parts: generalization capability and large language models.
First, on generalization: Technically, our tech stack is inherently designed for generalization. For instance, in the third quarter, we expanded our operational areas to more districts in Pudong, Shanghai, and Nanshan, Shenzhen. In both cases, it only took a few weeks from completing safety validation to launching fully driverless operations for the public, without requiring additional model training. The key reason is thatour technical architecture is already adept at handling various extreme scenarios, which are highly consistent across different regions,such as small obstacles on the road, cardboard boxes, temporary construction, crowd gatherings, sudden lane changes by vehicles ahead without observing following traffic, etc. The differences mainly lie in the probability distribution of these occurrences.

Therefore, I hope this helps everyone understand why our tech stack naturally possesses generalization capabilities. At this stage,I believe one of the key constraints to new regional expansion is the number of Robotaxi vehicles.If the expansion happens too quickly without a corresponding increase in the number of vehicles, fleet density will be diluted, impacting passenger wait times and overall experience. This is why the pace of operational area expansion cannot significantly outpace the growth of the fleet size.

Turning to large language models (LLMs): L4 in-vehicle models must meet two strict requirements—safety and low latency. However, LLMs and chat models neither meet these requirements nor are they designed for this purpose. In terms of safety, LLMs may exhibit 'hallucination' risks, which could compromise the safety of L4 systems. Regarding latency, LLMs are more optimized for 'throughput' metrics (e.g., tokens/second), whereas L4 systems require ultra-low latency and must run fully autonomous driving systems on low-power, cost-effective in-vehicle chips. Additionally, LLMs heavily rely on human data, naturally constrained by existing human knowledge boundaries, and inevitably learn from human driving errors and bad habits.

Of course, we do extensively use LLMs in our R&D work, such as enhancing human-computer interaction, boosting R&D productivity (coding, documentation tools), and analyzing passenger feedback for continuous improvement. However, for the reasons mentioned above, LLMs are not suitable as in-vehicle autonomous driving models.

Currently, Pony AI is collaborating with several original equipment manufacturers (OEMs) across different regions, including BAIC, GAC, and Toyota. From a financial perspective, does partnering with OEMs help improve operational efficiency and accelerate deployment?

The reality is that, in the global Robotaxi market,Governments and local residents generally prefer taxis from local brands.This is a very realistic point. For smaller Robotaxi fleets, brand influence may not be as significant; however, when large-scale deployment is required, local brand preferences become crucial.

Therefore, it is necessary for us to collaborate with multiple local vehicle manufacturers in different regions, which can help us enter various markets more quickly. This is also why we are currently partnering with three vehicle manufacturers to produce our seventh-generation Robotaxi vehicles.

Adapting the autonomous driving suite to different models does present technical challenges. However, from another perspective,our ability to standardize the technology and successfully deploy it across multiple models demonstrates the versatility of our technology.This versatility will become a long-term competitive advantage, enabling us to add new models more quickly, thereby accelerating expansion into new regions. For instance, in Europe, we have established a partnership with Stellantis to support our expansion in the region.

When Robotaxis encounter difficulties, why does Pony AI opt for 'Remote Assistance' instead of 'Remote Control' or human intervention? What are the technical differences between the two?

A previous question mentioned remote assistance, so let me explain in more detail. First,Remote assistance never involves controlling the vehicle via the steering wheel or pedals; instead, it provides support and advice from remote personnel when the vehicle initiates a service request.Most of the time, our vehicles can make driving decisions independently without requiring remote assistance. Remote assistance only intervenes when the vehicle actively requests it, rather than remotely 'driving' the vehicle. Even if the vehicle receives suggestions from remote personnel, the onboard autonomous driving system will still make timely decisions based on real-time conditions. Since the vehicle does not wait for remote instructions to act, it maintains safe operation regardless of network latency. A typical example is temporary traffic control: the system may request remote assistance, and remote personnel provide high-level suggestions to help confirm the strategy for the vehicle to navigate the scenario. Meanwhile, as I mentioned, we will continue to optimize AI algorithms, leveraging stronger general AI capabilities to enhance understanding of complex scenarios, thereby further improving the efficiency of remote assistance and increasing the ratio of remote assistance personnel to vehicles to approximately 1:30 by the end of the year. I hope this answers your question.

Thank you, management, for taking my question. According to our understanding, some countries in the Middle East have recently issued fully driverless autonomous taxi operation licenses. What is the company's view on this? What is Pony AI's overseas strategy?

Pony AI's mission has always been 'to make autonomous driving accessible.' We have always harbored a global vision, hoping that our technology can create value worldwide. Currently, our global focus is on markets with high growth potential, which typically exhibit strong mobility demand, well-established infrastructure, and a supportive regulatory environment.

When evaluating potential markets for entry, we primarily consider three core factors: the first is the target market size; the second is the local government's openness and policy implementation regarding the commercial operation of fully driverless vehicles; and the third is the resource availability and operational capability of local partners.

Our current progress in global expansion is as follows:Pony AI’s Robotaxi business has entered eight national markets.For instance, in the third quarter, we entered Qatar as a new market through our collaboration with Mowasalat. In the third quarter, our overseas Robotaxi business revenue experienced rapid growth, and we anticipate this momentum will continue. Going forward, upon identifying promising growth opportunities, we will further expand into other global markets, which is part of our international strategy.