Pony AI Inc.

Hello, ladies and gentlemen. Thank you for standing by, and welcome to Pony AI Inc.' 's third quarter 2025 earnings conference call. At this time, all participants are in a listen-only mode. After the management's prepared remarks, there will be a question and answer session. As a reminder, today's conference call is being recorded, and a webcast replay will be available on the company's investor relations website at irpony.ai under the news and events section. I will now turn the call over to your host, George Hsiao, head of capital markets and investor relations at Pony AI. Please go ahead, George.

Thank you, operator. And hello, everyone. We appreciate you joining us today for Pony AI's third quarter 2025 earnings call. Earlier today, we issued a press release with our financial and operating results. which is available on our Investor Relations website. An earnings presentation, which we'll refer to during this conference call, can also be accessed and downloaded on our Investor Relations website. Joining with me on the call today are Dr. James Peng, Chairman of the Board and Chief Executive Officer, Dr. Tiancheng Luo, Chief Technology Officer, and Dr. Liu Wang, Chief Financial Officer of the company. They will provide prepared remarks followed by a Q&A session. Before we begin, please refer to the safe harbor statement in our earnings release, which applies to this call as we'll be making forward-looking statements. Please also note that we'll discuss non-GAAP measures today, which are more thoroughly explained and reconciled to the most comparable measures reported under GAAP in our earnings release. available on our Investor Relations website and following with the SEC and Hong Kong Stock Exchange. I will now hand it over to our Chairman and CEO, Dr. James Peng. Please go ahead.

Thank you, George. Hello, everyone. Thank you for joining our earnings call. I'm excited to share that we have successfully completed the dual primary listing on the Hong Kong Stock Exchange. under stock code 2026 on November 6th, just one year after our NASDAQ listing. With strong support from both international and domestic investors, we secured the largest IPO in the global autonomous driving sector this year, raising more than 800 million U.S. dollars. This significantly strengthens our balance sheet and provides the dry powder to accelerate mass production and large-scale commercialization. We now expect stronger growth, surpassing 1,000 Robotaxi fleet plan by year-end and expanding to more than 3,000 vehicles for 2026. We have already seeing the flywheel in action. Expanded fleet is driving higher user adoption, shorter wait time, more orders, and a strong revenue growth. After launching Gen7 Robotaxi, we have already seen a citywide unit economics breakeven. This, in turn, gives us more room to increase fleet size. The capital we raised also fuels our business development, research and development, making strategic investments in new markets, new applications, and attracting world-class AI talents. All these are set to further propel our technology leadership and the long-term growth. Our Hong Kong IPO also powers our core mission. bringing autonomous mobility to everyone around the world. We're firmly delivering on this commitment. Earlier this month, we officially launched fully driverless commercial service for Gen 7 robotaxis across Guangzhou, Shenzhen, and Beijing. Today, our management team, including myself, actually arrives at our Shenzhen office in a fully driverless Gen 7 robot taxi to host this conference earnings call. This is more than just a normal ride for us. It actually marks a giant leap in autonomous driving's advancement. We are making level 4 autonomy more accessible than ever to a much broader user base. I'm excited to share a critical milestone. Our Gen 7 robo taxis have reached city level break even in Guangzhou shortly after their official commercial launch. This is pivotal to validate our viable business model. It not only gives us strong confidence to further scale our fleet, but also attract more and more third party partners in labeling them to fund our fleet and to support our asset-light model. The scaling up of a fleet is key to our growth as large-scale operational footprint drives efficiency through the economy of scale. Our robot taxi vehicles are essentially moving billboards, in fact, many new users discover and download our Pony Pilot app after spotting our vehicles on the road for daily operation. Fleet expansion serves as a highly efficient self-reinforcing marketing engine, facilitating user adoption and strengthening brand recognition. This creates a powerful upward spiral, more vehicles generate greater visibility, which attracts more users and establish network effects. The results are already evident. Building on that momentum, new registered users nearly doubled within just one week of launching Gen7 from late October. reflecting robust user demand and effective go-to-market strategy now let me highlight some key advance advance were made in recent months in executing our scale up strategy first we have ramped up production at an accelerating pace since the start of production in the middle of this year by november More than 600 Gen 7 robo taxis had rolled off our assembly lines, bringing the total fleet size to be over 900 vehicles. Thanks to the streamlined production process, we now expect to outperform our full-year target of 1,000 vehicles delivering ahead of schedule. This gives us increasing confidence to sustain robust momentum, driving fleet size to surpass 3,000 vehicles in 2026. Second, in Q3, our robo-tax revenue surged by 90% year over year with fair charging revenues delivering over 200% year over year growth. This was fueled by driving user adoption across all four tier one cities, improved fleet operational efficiency, and tailored pricing strategy for diverse user segments. We have seen that the higher order density leads to lower users' average waiting time and, in turn, higher vehicle utilization rate. This allows us to continuously optimize our pricing strategy. Third, we have continued to expand our operational footprint. For example, in Shanghai, we became the city's first company to launch fully driverless commercial robot taxi operations earlier this July, covering the Jingqiao and the Huamu areas of Pudong. In Shenzhen, we extended commercial fully driverless operations to more and bigger city areas, including Shekou and the overseas Chinese town. Fourth, we're taking major steps toward scalable mobility.

Excuse me. I believe there has been an interruption. Just one moment please. Excuse me. I've rejoined management. Please continue. Thank you.

Sure. I was talking about the scale-up strategy. So, following our collaboration with in June, we recently forged another partnership with Sunlight Mobility. This aligns reflect growing market recognition of our business model with increasing number third parties wanting to fund fleet deployment this actually enables us to speed up further fleet expansion now let me turn to our global expansion we are deeply dedicated to advance robo taxi services while strategically expanding our international fleet Now, we have Robotaxi presence established in eight countries across China, the Middle East, East Asia, Europe, and the U.S. We entered a new market in the Middle East, Qatar, through a partnership with Novoselic in third quarter. Novoselic is the country's largest transportation service provider. As part of this collaboration, Our robo taxis have recently begun testing on public roads in Doha, the capital of Qatar. We have also advanced our presence in South Korea by securing nationwide robo taxi permits, enabling operation across the country's autonomous testing and operational zones. Our collaboration with local partners continue to deepen. We're collaborating closely with , the country's largest transportation service provider, to begin road testing. In Luxembourg, we plan to deploy testing vehicles based on the e-traveler through our alliance with Stellantis. It's a European leader in light commercial vehicles. effort will initially focus on vehicles designed for European diverse mobility need to enable a range of use cases. In addition, we have partnered with global ride-hailing platforms that also participated in our Hong Kong IPO. Those platforms include Uber and Bolt. Bolt is a Estonia-based mobility company operating in over 50 countries and 600 cities. Built upon our collaboration with Uber, we aim to leverage Uber's robust ecosystem to enter the Middle East and then scale into additional international markets. Last but not least, we recently released our fourth-generation robot truck. with production and initial fleet deployment expected in 2026. Featuring fully automotive grade components, optimized software hardware integration, and a transition from internal combustion engine vehicles to electric vehicles. The Gen 4 Robotruck delivers a significant more efficient cost structure and a greater energy saving. The new platform fully leverages the technological foundation and operational expertise developed through our Gen 7 robo-taxi vehicles. In addition, we deepened our collaboration with Sanyi Group and added Liuzhou Motor as a new partner to have multiple vehicles to support our further operations. To sum up, 2025 is a critical year of mass production and commercialization for Pony AI. We take pride in the progress we have made and are steadily delivering on the promise we have made to our shareholders at the time of our US IPO last year. Our recent Hong Kong listing not only marks a major milestone for our company, but also underscores the promising future of the industry. Moving forward, we will drive technological innovation and create lasting values by scaling fast, efficient and comfortable autonomous mobility services toward our mission. Autonomous mobility everywhere. With that, now I'll hand it over to our CTO, Dr. Tiancheng Luo, to share more about our technology strategies. Tiancheng, please go ahead.

Thank you, James. Hello, everyone. This is Tiancheng. Let me first share my thoughts on automotor driving technology stack. From day one, we believe that full stack integration across software hardware operations was the only way to build a truly scalable autonomous mobility. That conviction has been validated again and again, especially for this critical year of scaling up. With the achievement we have made, it is clear that our early technology best help us achieve the leading position and it will further accelerate our future growth. Our deep foresight into tech stack is what is positioning us as a leader in the industry today. as we become one of the few companies to operate large-scale, fully driverless robot tech services. So as early as 2020, we recognized the importance of a training course rule based on reinforced learning unit simulation. In that year, we transitioned over tech stack into a world model, which is what we call a pony world today. Through years of R&D effort and real-world validation, our top driving model has evolved into a closed-loop training. We achieved unsupervised, self-improving iterations. In recent years, we are seeing the broader autonomous and robotic industry converge on the world model. Validating the approach we adopted today This foresighting AI tech stack has given us a meaningful head start, and we're confident that we will stay ahead for multiple years. Then let me dive into the three criteria that put us at the forefront of world model development. First, the high-fidelity interactive simulation. This is far beyond our ability to just generate the scenarios and render sensor data. Driving is by nature interactive. The Robotex's actions directly affect how surrounding agents behave, such as other vehicles and pedestrians need to react to overdriving behavior. It must understand and adapt to new situations and complex physical interactions in real time, mirroring true on-road interactions. It enables Robotex operations that are safe, smooth, and social aware. Of the 10 billion kilometers of test miles that only were generated each week, More than 99% capture weak agent infections, while less than 1% are synthetic environments such as sensor rendering. Okay, second, the ability to reproduce scale and realistic corner cases. While these long-term scenarios don't occur frequently, they are critical to safety in autonomous driving. More importantly, every scenario must be something that could really happen in the real world. not those useless educatives with no basics in reality. So the third, the AI-based learning evaluator. This is the reward-based evaluation mechanism. Driving is a multiple object optimization problem. What is considered as a good driving also changes in various driving scenarios. Within the closed-loop training environment, Pony World and our virtual driver are continuously evaluated on key driving metrics. This assessment does not rely on real-world data, human-level data, or rules. Instead, it uses AI-empowered models to learn what good driving looks like, directly from outcomes. Turning real and simulated shares into a powerful cycle of self-improvement. A best-in-class world model must meet all three criteria to enable truly unsupervised and self-improving closed-loop training. This is critical to realizing large-scale driverless autonomous driving. And leveraging over full-stack technology as a core strength, I will now turn to how to drive business progress during the third quarter. First, on cost and operational efficiency. We pioneered 100% automotive-grade auto-driving kits for Gen 7 robot taxis. We've optimized the design, reducing volume cost by 70% compared with the previous generation. The Gen 7 vehicle has been officially operating for public in Guangzhou, Shenzhen, and Beijing, fully validating our safety standard and operational efficiency. We've built on our momentum and deliver further progress, driving by scaled production, and enhanced R&D, we've already realized an additional 20% reduction in the Atom Driving Kit BOM cost for the Gen7 platform designed for 2026 production compared with 2025 baseline. This lights foundation for sustained cost savings. Our robust AI algorithm and fleet management exercise has proven effective at driving operational efficiency. to better identify user demand in hotspot areas during rush hour hours, and we enhance our algorithm for order dispatch, matching, and scheduling, thereby ensuring sustained efficiency of taxi utilization. We have also improved our virtual driver to recognize more and more complex scenarios. This allows us to improve our remote assistance to vehicle ratio substantially on the track to reach 1 to 30. by year-end. Over-superior service experience has become the key reason users choose Huni Air Robotaxi. After the launch of Gen7 Robotaxis, we have earned a widespread positive feedback and generated great social media buzz from users. As we deliver high-quality experience, users are increasingly willing to pay a premium for the enhanced effort, reliability, and the safety of the autonomous journey. For ride comfort, OVA advanced interactive planning capability intelligently optimized for the frequency and the magnitude of acceleration, braking, and steering. This delivers smooth, natural motion control tailored to the electronic vehicles and the ride-sharing market, offering consistent comfort experience for every fully AI-approved taxi ride. This enhancement has reflected a measurable improvement for Gen7, such as emergency brakes and steering over the past few months. Additionally, our RopeTax features are superior in cabin experience. We also pioneered the innovative Smart View Positioning feature. With one tap, users can remotely adjust their vehicle position for more convenient pick-up and drop-off. We introduced Voice Activity features, we call the Popo Voice Assist, allow users to do stop trips and control air conditioning, etc. We will continue to upgrade the cabin into an AI-powered mobility terminal. Together, this upgrade creates a more accessible and streamlined user experience. So third, our tech stack is also built for generalization. The L4 native tech architecture allows us to adapt quickly to new markets and platforms. In terms of cross-region generalization, all virtual drivers and show-ins can quickly understand and adapt to diverse traffic conditions around the world. For example, leveraging our high-fidelity training environment and variation mechanism powered by Pony World will extend our fully driverless coverage in Fudong District in just a few weeks. In addition, when expanding to Europe, the system intelligently identified and adapted key differences in local road conditions, such as unique traffic signal configuration, and various driving patterns. Our technology boosts generalization power across platforms as well. The latest generation robot truck will commerce production and operations from next year. This demonstrates our capability to create synergy between Robotaxi and Robotruck tech stack. Looking ahead, we will leverage our success Hong Kong listing to reinforce our technological leadership increasing R&D investment, and attract top AI talent to advance over robo-taxi, robo-truck, and new market initiatives. We will continue pushing the frontier of the autonomous mobility and refining what is possible in the transportation. Okay, this concludes my prepared remarks. I will now pass the call over to our CFO, Dr. Leo Wang, for a closer look at all the financial results. Leo, please go ahead.

Thank you, Ken Chen. Hello, everyone. This is Leo. I will focus on year-over-year comparisons for the third quarter, unless otherwise noted. Q3 2025 was a landmark quarter. We delivered a robust revenue growth, specifically with solid progress in robo-taxi large-scale commercialization. And now we expect to outperform our full-year fleet target of 1,000 vehicles. Moreover, our newly deployed Gen7 Robotaxi fleet have reached a pivotal city-wide unit economic break-even milestone. This lay out a solid foundation for further scaling up and the implementation of asset-light business model, which will be further accelerated by our successful Hong Kong IPO capital raise. In this quarter, Revenue finished at $25.4 million, growing by 72%. This strong performance was primarily driven by the continuous optimization of our robo-taxi services and the sustained demand in our licensing and application business. Firstly, robo-taxi services revenue reached $6.7 million. representing a remarkable growth of 89.5% year-over-year and 338.7% quarter-over-quarter. Specifically, fair charging revenue continued to deliver a triple-digit growth surging 233.3%. This was achieved even before the commercial rollout of our Gen 7 robo-taxis. Supported by a stable commercial fleet of our Gen 5 and Gen 6 vehicles, the strong growth during Q2 and Q3 stemmed from growing user demand in tier one cities in China. Our continuous effort to optimize fleet operation and pricing strategy altogether leading to increased fleet utilization and efficiency. This is a testament to growing user recognition and the brand royalty to Pony Pilot Service. Going forward, as we follow this strong momentum towards a significant fleet expansion of over 3,000 vehicles by 2026, we expect Robotaxi revenue growth to accelerate even further, driving more orders and a higher operational efficiency. In Q3, another key Robotaxi update is the implementation of our asset light model for fleet expansion. As we have shown promising numbers in vehicle unique economics, we received a strong interest from third parties who are willing to purchase Gen 7 vehicle to run as Robotaxi operators. Such partners include but are not limited to leading ride-hailing or taxi operators. For instance, Shenzhen Xihu Group and Sunlight Mobility. The asset-led model has contributed revenues through technology licensing fee and the vehicle sales, while giving us further leverage and capital efficiency for further fleet expansion. Aside from strong top-line growth domestically, we are also seeing fast growth of robot taxi revenues from overseas market. Moving forward, we expect Robotaxi revenues from overseas market to continue to grow. Currently, our Robotaxi footprint have already expanded into a country globally, serving as a promising foundation in our exploration of the international opportunities. Secondly, moving to Robotruck. Robotruck service revenues were $10.2 million, growing by 8.7%. Moreover, as we launch our Gen 4 fully auto-grade RoboTruck, we expect to reduce the bound cost of its ADK, automatic driving hardware kit, by 70% and reach a thousand unit scale of RoboTruck fleet going forward. This new generation of RoboTruck will powerfully accelerate the progress of RoboTruck commercialization at scale. Thirdly, licensing and application revenues were $8.6 million, growing significantly by 354.6%. We continue to see robust and growing demand of our ATOMS domain controller, primarily from robo-delivery clients. Turning to gross margin, we delivered a significant gross profit margin improvement from 9.2% in Q3 2024 to 18.4 percent in q3 2025 with gross profit of 4.7 million us dollars in the third quarter this remarkable improvement was firstly driven by our strategic initiatives to optimize the revenue mix and secondly by a greater contribution from robo taxi services which carry a relatively higher margin The unique economic break-even achievement validates our due focus on go-to-market execution and optimized operational efficiency. Since the launch of Gen 7 commercial operations in Guangzhou, the daily net revenue per vehicle has reached 299 RMB. The net revenue refers to the total RMB value generated from ride-hailing services after deducting discounts and refunds. Notably, daily average orders per vehicle have reached 23, fueled by a robust widespread user demand and our operational optimization. Meanwhile, we have also optimized hardware depreciation as well as operational costs, including charging, remote assistance, ground support, service and maintenance, insurance, parking, and network costs. This will further improve our margin down the road. The total operating expenses were $74.3 million, up by 76.7%. Excluding share-based compensation expenses, non-GAAP operating expenses were 67.7 million US dollars up 63.7%. The increase primary reflects the one of R&D investment in Gen 7 vehicles and the expansion of our R&D personnel, critical to securing and extending our technological leadership. Specifically, approximately half of the increase in research and Development expenses stemmed from one-time customized development fee of $12.7 million for Gen 7 vehicles. Net loss for the third quarter was $61.6 million, compared to $42.1 million in the same period of last year. Non-GAAP net loss was $55 million. compared to $41.4 million last year. Looking ahead, we expect to sustain disciplined investment to accelerate large-scale commercial deployment. Turning to the balance sheet, our cash and cash equivalents, short-term investments, restricted cash, and long-term debt instrument for wealth management were 587.7 million us dollars as of september 30th 2025 compared to the balance as of june 30th 2025 of 747.7 million us dollars around half of this creek discreet comes from one of cash outflow including capital injection to dray phone our joint venture with toyota to support a Gen 7 mass production and deployment. All of the capital commitment in Drayton has been completed. The remaining cash balance reduction primarily reflects our mass production and the large-scale deployment status, including, firstly, ongoing operational cash outflow, and secondly, capital expenditure for the procurement of Gen 7 vehicle in Q3 to support our goal of 1,000 vehicle fleet by year end. For the nine months ending September 30, 2025, we have an accumulated free cash outflow of $173.6 million. With the completion of our recent Hong Kong IPO, We have over $800 million cash newly added, providing us with substantial fuel for the next phase of growth. The IPO proceeds will help us accelerate fleet expansion into key addressable markets, further optimize our platform for scale and deepen our R&D investments to further solidify our technology mode. Looking ahead, Our mass production momentum continues to strengthen, and we are on track to exceed our full-year vehicle target of 1,000, achieving this milestone ahead of schedule. This acceleration reinforced our confidence in scaling rapidly, and we now anticipate to grow our fleet to be more than 3,000 vehicles by 2026. In addition, we've already transitioned to a asset-light model for a meaningful portion of our new vehicles. This will enhance our capital expenditure efficiency and provide a greater leverage for scalable fleet expansion. With the proven operational model and the financial runway from the recent Hong Kong IPO, we are uniquely positioned to accelerate our business plan, turning momentum into sustained profitable growth I will now turn the call over to the operator to begin our Q&A session. Thank you.

Thank you. We will now begin the question and answer session. To ask a question, you may press star then 1 on your telephone keypad. If you're using a speakerphone, please pick up your handset before pressing the keys. To withdraw your question, please press star then 2. For the benefit of all participants on today's call, please limit yourself to one question. If you have more questions, please re-enter the question queue. If you ask questions in Chinese, please repeat them in English. And the first question comes from Ming Shun Li with Bank of America. Please go ahead.

Thank you, Sync Management, to give the opportunity for me to ask a question. So I just have one question. So could the management team give us more updates on the fleet size for this year and also the outlook in 2026? For the new vehicles added, what is the fleet deployment plan across different cities? Thank you.

This is James. I'll take this one. So, as you can see that since the launch of our Gen 7 robotaxi, we actually have seen a much faster than expected production and deployment. So, for this year, we certainly expect to outperform our previous target of 1,000 robotaxis by the year end. We certainly expect this strong momentum to continue into 2026. now with conservative target of over 3,000 vehicles. This is mainly because we have already seen upward spiral with the launch of our Gen 7 vehicles. Essentially, the fleet density creates a much shorter wait time for the passengers, and then that creates a better user experience, and the user experience leads to much higher utilization for our vehicles. and then we can actually then charge a better pricing. So this spiral really created a strong momentum for us to expand much faster. In addition, we also started experimenting with the asset-light model by collaborating with fleet managers such as Shihu, Sunlight, and certainly we'll add more This asset-like model allows us to deploy at a much larger fleet with less . So this is our growth plan. Then in terms of the fleet deployment plan, we'll go deeper on our existing markets, and at the same time, we'll go much wider to explore some new opportunities. The citywide UE breakeven for the Gen 7 in Guangzhou, in my view, is a pivotal milestone to validate our business model. This gives us a huge confidence and allow us to deepen our collaboration and our operation in the existing markets, which are the tier one cities in China. This is because, as I already mentioned, expanded fleet size creates an upward spiral. But at the same time, we also expand into many more domestic cities and also the overseas markets. We see those for our future growth. Our go-to-market strategy on those markets is that we'll collaborate deeply with the local partners and the local government agencies to establish presence and prepare for our future growth. So stay tuned. I think we'll have great news ahead of us. With that, back to the operator.

Thank you. The next question comes from Ben Wong with Deutsche Bank. Please go ahead.

Hi, measurement. Thank you for taking my question. I just have one question, which is about charging. I'd like to know fair charging revenues in another growth in 3Q25. So what is the outlook for fair charging revenues as we deploy more vehicles? Thank you.

Yeah, this is Leo. I'll take this question. Yes, in Q3, our fair charging revenue actually surged even faster. It was growing about 233%. Though at that time, our fleet were still with the Gen 5 and Gen 6 vehicles. So we believe such growth was driven by both the demand side as well as the operational side. On the demand side, we have been continuously to do our effort to improve the whole riding experience and also the user experience. So with this effort, we've seen robust and organic user demand in T1 cities. This is also a signal of a strong consumer adoption of our robot taxi service, giving you an example that the total registered user was more than doubled year over year in Q3. And on the operational side, we have also been optimizing the fleet operation to improve our vehicle utilization and the order fulfillment. as Tianzhao already mentioned in his remarks. So, for example, we enhanced our fleet dispatching and the deployment. This has consistently reduced our wait time. It's approximately 50% shorter compared to the same period in 2024. And we also continue to expand our pickup and drop-off points to create a much more smooth user experience. For example, in Shenzhen, now we have more than 10,000 such points. More than 300% increase since the end of June this year. With all this, you know, demand side and operational side improvement, I believe we could see sustained strong growth momentum through the continuous fleet expansion with more and more Gen 7 vehicles are into our service. First of all, we expect that our fleet has been growing exponentially from 270 last year and to be more than 1,000 this year and a target of more than 3,000 next year. This scaling up would also create a better network effect, which means shorter wait time and higher vehicle utilization and higher user adoption. We would also progressively expanding our service area in cities such as Shanghai, Shenzhen. We've already been doing so today. We would increase the population coverage and expanding to more drivable mileages, et cetera, et cetera. With all these being done, I think we can boost the average order value per chip. Okay. Now I'll get back to the operator.

Thank you, sir. The next question comes from Kyle Wu with Citi Research. Please go ahead.

Thanks for taking my questions. This is Kyle from Citi Research. Congratulations on achieving the milestone of Citi-wide UEFA event. Could you elaborate more about the assumption behind the UEFA event, including daily order, pricing, daily operating hours, and the ratio of remote assistance? Thank you.

Yes, I'll take this question. Like you said, we all believe the city-wide unique economic breakeven is a pivotal milestone for the company and also for the industry. First of all, we, you know, achieved this pivotal milestone in Guangzhou City since our Gen 7 vehicle has been put into commercial service. We always believe China is the largest market of global right-hitting market. And for the tier one cities, the total time accounts for a huge percent of right-hitting market in China. So achieving this milestone in this market is far more meaningful from commercial perspective. Then if we talk about the unique economic, there is the revenue side, there is always the cost side. On the revenue side, first of all, on the daily net revenue per vehicle, as I mentioned, our daily net revenue per vehicle has hit 299 RMB. It's based on a two-week daily average figures as of November 23rd, following the launch of our Gen 7 vehicle in Guangzhou. And this net revenue also refers to the total RMB value generated from ride-hailing service after deducting discounts and the refunds. And in terms of daily orders from this 299 RMB number, it was average 23 orders per day. It's fueled by robust widespread of user demand. Now let's look into the cost side. So the cost side of the unique economic basically has two major components. First of all, it's the hardware depreciation. For Gen 7 vehicle, the annual vehicle depreciation is based on a six-year use for life. The other major component on the cost side is the operational cost, which includes the charging. remote assistance and ground supporting staff, vehicle service and maintenance, insurance, parking, internet network costs. So, regarding the remote assistance, we are on track to achieve our 1 over 30 vehicles. And from this milestone that we achieved, we are very confident to capture the China huge time. Meanwhile, it also established a strategic foundation for further scale, scaling up domestically and internationally. This not only give us strong confidence to further scale our fleet, but we also see more and more third-party companies are enabled to fund their fleet and helping us to transition into a asset-light model. So all these together we believe will drive our top-line growth and also the cost optimization. Okay, I'll go get back to the operator.

Thank you. The next question comes from with Huatai Securities. Please go ahead.

Hello, James, Dr. Low, and Leo. Thank you for taking my question, and congratulations on the results. We've observed a surge in diverse employers attempting to ascend into the robot taxi operation, particularly the EV makers. So what's your take on these new entrants in the level four autonomous driving space? And also specifically, could you elaborate on the main technical and operational challenges such as tackling corner cases in food management or regional commerce?

Thank you. This is James. I'll take this one. So first and foremost, I think it's definitely as we see more and more companies announcing that they're going to enter into robo taxi industry. I think itself is actually a great thing because it indicates increasing recognition and the confidence in Robotaxi imminent potential for the large scale of commercialization. As the awareness increase, more resources, more companies come in, more resources will pour into this Robotaxi industry to actually accelerate its development. So, overall, I view this as a good thing, but on the flip side, The robo taxi industry is actually not a one that any new player can easily enter because as you can see, the fact is that currently, none of the new entrants being OEM maker or being ride-hailing platforms, none of them have fully driverless vehicles deployed on the open road. So, it's clear evidence this is not an easy industry to be entered. I think there are certainly three huge hurdles for any new players, and those hurdles are business side, regulatory side, and also technical challenges. Let's probably look at the business challenges first. Because World Taxi, as you see, it's not just about L4 driving itself. It also has many more aspects, such as user acquisition, vehicle production, fleet dispatching, fleet maintenance, such as the cleaning, charging, and everything else. So, as a leader and a first mover in this industry, we certainly enjoyed the early mover advantages as we have a much bigger L4 fleet on the road. We generated better brand awareness. We have optimized the cost on every aspect of the business, as Leo already mentioned in his answer to the last question. And we, because of early mover, we also have secured more partners. I think all those are important and it creates big hurdle for any new entrants. The second hurdle that I want to mention is on the regulatory front. Because L4, a robo-taxi needs very high safety requirements. All the policymakers worldwide have fundamentally will require a much, much higher safety requirements for the robo-taxis compared with the traditional taxi. That means in any city, A new player needs to prove its safety step by step before they can expand even into a fully driverless fleet. Typically, a new player will start with a testing with just a few dozen or maybe even less vehicles. And then once those vehicles prove to be safe, they add more vehicles and then expand operational areas. after they can accumulate the safety records. And along the way, they also need to acquire all the required licenses and permits. And this, in itself, is actually a lengthy process. So, overall, the whole process takes time. And this code starting process cannot be easily accelerated. So that's the second challenge. The third challenge is certainly, in my view, is on the technical side. And probably for this one, I'll turn to Tiancheng to elaborate.

Yeah, sure. So I'm Tiancheng. So let me continue from a technology perspective. So as I said in my preliminary remarks, we are now seeing the broader industry starting to using word model, such as Robotech players and automakers. Essentially, they are all using about using reinforcement learning based on simulation training environments. First and foremost, I will say we started developing reinforcement learning for account driving five years ago. This gives us an early-mover advantage, making us one of the most experienced companies in the world model. We believe that it will continue to stay ahead as more peers follow the same path. So once the world model matured, the human feedback and the real-world data are no longer used for further iterations. So as the stage of new training comes through, the world model and the virtual driver co-evolve into a dual spiral cycle. This means the world model is training the virtual driver, and at the same time, the world model improves itself through feedback of the virtual driver. This sharply reduces the lines on the real-world data. The question will touch on the technical challenges for meeting the corner cases. Maybe an example here that once the virtual driver meets some corner cases, so they can give feedback to the world model, and the world model will improve its distribution of the corner cases, then the next version of our model will be able to create generated testing and also improving the capability of the version where we handle the corner cases. Okay, so looking ahead, our real advantage lies in ability to validate new technologies quickly and deploy them at scale. So based on our proven track record of scaling Robotech's operations, So we believe we can quickly capture the next wave of innovation. Also, last but not least, our current Hong Kong IPO will further accelerate R&D and integration cycles, reinforcing our technical leadership and widening our competitive mode. Yeah, with that, I'll back to the operator.

Thank you.

The next question comes from Jia Yi Li with Jefferies. Please go ahead.

Thanks for taking my question. I have one as well. My question is about what do you see as the main factors behind the faster expansion of your operational areas? And beyond technology, what else do you think really matters? And from the technical perspective, are you using large language models? And if so, how are they helping push L4 autonomy forward? Thank you.

Thank you. This is Qin Chen. I will continue to answer this question. I think your question consists of two parts. Let me answer your question on generalization first. Then I will address the other one on large-language model later. For generalization, I would say technically, over tech stack, it's by nature built for generalization. So a good example is that over operational area expansion into new areas in Shanghai, Fudong, and Shenzhen, Nanshan District, in the third quarter, in both cases, It only took us only a few weeks from verifying the safety to truly realizing fully serviced operations to the public. There was no need for additional model training. The key reason that an overall native architecture is built for hand-in-corner cases and to gym cases, while these cases are actually very consistent across different regions, they are really nothing more than things like small obstacles, boxes on the road, attachments that they are closing, and suddenly lane change from other cars without looking at the vehicle behind, et cetera. So it's just about the likelihood and the probabilities of each one happening. So I hope that can help understand why the L4 tech stack is by nature built for generalization. So at this moment, I will say the key to over new area extension is number of robot taxi vehicles. If we extend to too many areas without adding more cars, it will instead dilute the density. So that is the reason why the speed of operational area extension cannot significantly faster than that of fleet size. So then let me share my thought on the second part, that is large-language model. First, I will say, first and foremost, there are two non-negotiable requirements for L4 onboard driving model. Uncompromising safety, and also low latency. There are the large language model and chatbot don't meet and are not designed to meet as well. So for safety, large language model generally have issues like mode health emissions, which is unacceptable for L4 in terms of safety. And for latency, large language models are optimized for throughput like tokens per second, in contrast, L4 is optimized for low latency and the ability to run fully driverless over-text on chips that have both low power consumption and cost-efficient. Moreover, large-language models overly run human data, which fundamentally limits them to the boundary of existing human knowledge as it inevitably makes them pick up human errors and bad habits from human drivers. So, We also extensively used a large-language model in the IMD effort, such as AI-enhanced human-machine interaction, engineering productivity tools for coding and documentation, and analysis for the writer feedback for external improvement. But however, due to the multiple reasons mentioned above, large-language model is by nature not good for driving model on board. So with that, go back to the operators. Thank you.

Thank you. That's very helpful.

The next question comes from Jinyu Fang with UBS. Please go ahead.

Hi. Thank you, management, for taking my questions. I have one question here. It is currently that Pony cooperate with multiple OEMs for robot taxi manufacturing, including BAIC, GAC, and Toyota. Does management see potential for improving operating leverage through working with only one OEM team staff? Thank you.

This is Jeff. I'll take this one. So the reality is that in the whole global taxi industry, local governments and the local residents actually have a strong preference for the local branded taxi vehicles. So, that's a reality. Typically, while Robotech's fleet is relatively small, the brand doesn't really matter much. But if we need to deploy a significant fleet size, the requirements certainly is no longer true, and the local branded OEMs is much more preferred. So, it is necessary for us to cooperate with multiple local OEMs in different regions, it actually can help us to expand into different markets much quickly. And that's why we are now collaborating with three OEMs to produce our Gen 7 robot taxis. It is true that fitting our autonomous driving kit into different vehicles actually poses a huge technical challenge. But if you look at it from the other side, The mere fact that we were able to standardize our technology and being able to fit our setup into different vehicles, that shows our technical generalization. And down the road, it actually can create a huge competitive edge. So, as a result, we can add new models. much faster to accelerate our expansion into new regions. For example, in Europe, we currently added the partnership with Stellantis. So with that, back to the operator.

The next question comes from with . Please go ahead.

Thanks for taking my question. I have one question. Why Pony can use remote assistance on Robotaxi when the car meets difficulty instead of remote control or human takeover? And what is the technology difference behind that?

This is Tiancheng. I will take this one. I think one of the previous questions also touched on the remote assistance for Robotaxi. So let me elaborate on that in a little more detail. First and foremost, as the over-remote assist never control the vehicle through the steering wheel or pedal. Instead, they provide remote support and suggestions by responding to service requests. For all the time, the vehicle can independently make decisions without remote assistance. Assistance only initiates when a vehicle requests it, rather than through the remote driving. So when vehicle receives the assistance response, the onboard driving system will still make timely decision based on the actual situation because the vehicle never wait for remote command to act. So it remains safe operation without any dependence on network latency. So one typical example of remote assistance is the situation of a temporary traffic control. In such cases, the system may request remote assist which can provide high-level suggestions to confirm the cost decision navigating through a scenario. Also, as I mentioned, we have continuously improved AI algorithm and also leveraged a general AI capability to recognize more and more complex contexts. This allows us to improve remote assist to vehicle ratio in the third quarter to reach 1 to 30 by year-end. Hope that can answer your question. So back to the operator.

Thank you. The next question comes from Serena Lee with China Securities. Please go ahead.

Okay. Thank you for taking my question. This is Serena Lee from China Security. As far as we know, some countries in the Middle East have issued fully driverless Uber taxi license recently. What's our view on that? What policy is overseen to stretch it?

Sure. This is James again. Let me take this one. Our company's mission has always been autonomous mobility everywhere. So, we certainly have the global ambition since our funding to actually utilize our technology to benefit the local societies worldwide. Current, our global efforts are focused on the markets with hyper-growth potential. So those are the markets with typically strong mobility demand, well-developed infrastructure, and a supportive regulatory environment. When we evaluate a potential market to enter, on the high level, three factors we all consider. One is the adjustable market size, which is 10. Second is the openness the execution of the local government to support and issue permit for the fully driverless commercial operation. Third is how strong is the local partner for their on-the-ground resources and operational capacities. So, as you can see, our current global expansion status is that we have already entered eight countries for our global taxi, and we also, For example, in Q3, we added Qatar as a new market by collaborating with Mova Saleh. In Q3, we have also saw a rapid revenue growth, especially for the robo taxi from our overseas markets, and we certainly expect this momentum to continue. So going forward, we will enter other global markets if we see there's a good growth opportunities. So this is our overseas strategy. With this, back to the operator.

As there are no further questions, I'd like to turn the call back over to the company for closing remarks.

Thank you, operator. This is George again. If anyone has any more questions, feel free to contact our team. We will conclude our call today. Thank you, everyone.

This concludes today's conference call. We thank you all for attending today's presentation. You may now disconnect your line.