Pony AI Inc.

our Gen 7 robotaxis. With the aforementioned three things that we are working on, we will focus on reinforcing these critical foundations to realize robust growth momentum of our fleet size, ensuring a scalable and sustainable expansion. As for year 2026, I think our scale-up will be even more accelerated. We will produce more autonomous driving vehicles and then deploy them in China and also the international markets. With this, back to the operator.

Thank you. And our next question today comes from Ting Song at Goldman Sachs. Please go ahead.

Thank you. Congratulations on the results. I have two questions. And the first one is, while you emphasized China's first strategy last quarter, we have now seen some progress on global market this time. So could you elaborate more on your involving global strategy? And to what extent does the China market remain a core focus at this stage? Thank you.

Sure. This is James again. I think I'll take this one. As I mentioned in my prepared remark, PolyAI's mission has always been autonomous mobility everywhere. While we currently prioritize the China market, giving its relatively mature regulatory environment, we believe our established ecosystem, technological advancement, and the scaled operation in China have empowered us to enter new markets with proven capabilities, experiences, and proven business models. At this stage, we are aiming for markets with strong mobility demand, advanced infrastructure, and welcoming regulations. While the commercialization of these international markets is still at the early stage. We relentlessly work hand in hand with our global partners to showcase the technology readiness, move forward local commercial driverless regulations, build momentum for public acceptance, and generate revenues along the course. This approach actually mirrors our achievements that we have already established in the tier one cities in China over the recent years, which we believe our successful track record in China will also help foster greater confidence for these new markets. Recently, we have formed strategic partnerships with key global players. One of them is Uber, with plans to launch our robo-taxi service on their platform, starting in a key market in the Middle East this year and expanding to other international markets. Another partner is ComfoDelgro, one of Singapore's largest transport companies operating across 13 countries, including Europe, China, and Australia. We also continue to make regulatory and testing progress globally. having secured a L4 robo-taxi testing permit from Luxembourg and also initiated road testing in Seoul's Gangnam district in South Korea. These successes have provided us with valuable experience as we explore future opportunities beyond the China market. With this, back to the operator.

Thank you. My second question is you deliver very impressive revenue growth in RoboTaxi. What factors are driving behind this quarter and do you believe this is sustainable in the upcoming quarters? Thank you.

This is Leo and I'll take this question. So the revenue growth in Robotaxi segment was driven by both fair charging and project-based engineering solution services, with fair charging revenues achieving at a much faster growth rate, increasing by roughly 800% year over year. The strong growth rate was attributed to the expansion of our public-facing fair charging Robotaxi operations in tier one cities in China. We also optimized our operations to cater to diverse user group, such as interactive rewards features. I would also like to take this opportunity to explain our Robotaxi revenue structure. Our revenue are currently generated from two main streams. The first stream consists of engineering solution services. which are recognized upon the achievement of project milestones. Hence, this is project-based and could fluctuate among quarters. The second stream is recurring revenue, primarily from our virtual driver operations, such as our robot taxi fare charging services. While the project-based revenues currently make a larger portion in our total robot taxi revenue, We believe the non-recurring revenues we are generating from partners such as ride-hailing platforms, OEMs, and other parties are very critical to enhance and advance our recurring revenue stream. These collaborations also further pave the way for a robust long-term monetization model. As a result, we anticipate some natural volatility in revenues from quarter to quarter in this segment. That being said, we will gradually reduce financial fluctuations and are very confident to deliver a strong growth trajectory in the long term. I'll get back to the operator.

Thank you. And our next question comes from Ben Long with Deutsche Bank. Please go ahead.

Thank you for taking my question. Much more from a technology perspective, you mentioned that in the ADK pricing was dramatic. Did you need to upgrade your software to fulfill this ADK cost reduction? In particular, what's the improvements you're doing for the computing power? You also mentioned that you actually would decline 48% of the cost for computing power. Thank you.

Thank you. I will take this question. So this is Tiancheng. So before I answer your question, I would like to say that we believe in the field of upon-driving technology, Pony AI is a point to represent China's leading company in embracing the deep-seek moment. So by optimizing our Pony world and enhancing engineering capabilities, we have designed a cost-effective hardware and software system. This enables us to significantly improve inference performance while reducing associated cost, even with auto-grade SOC and the lower-precision LiDAR sensors. So for instance, the online phoning work had effectively improved computing power efficiency by three times through AI-influenced optimization, auto-distillation, and other innovations, significantly outperforming the broader L4 industry. As a result, we're able to adopt more cost-efficient computing power with a total capacity of 1,016 tuf, compared with industry pairs typically ranging from 2,000 to 5,000 tuf. In terms of the inference computing, we have implemented numerous optimizations, such as optimizing operators in AI models, increasing computational parallelism, and improving model memory efficiency to enhance inference performance. So all these efforts help us to improve cost efficiency without give-up performance, proving that we're able to realize the cost-effective and scale L4 drive-based autonomous driving. Thank you. I will give back to the operator.

Thank you. And our next question today comes from with Hightower Securities. Please go ahead.

Thank you for taking my question. So, congrats on your expansion in local taxi services. So we noticed that the Ministry of Industry and Information Technology of China has recently issued some regulatory requirements regarding driver assist. So I just wonder that could this potentially have an impact on Pony AI? Thank you.

Yeah, this is Tiancheng. I will take this one. So I think a lot of people mistakenly equate L2 driver assist with L4 autonomous driving. Recently, the Ministry of Industry and Information Technology, MRIT, issued a notice, clearly states that L2 is not equal to L4. The key requirements from MRIT include, first, the manufacturers or solution providers must avoid using misleading terms, such as autonomous driving, intelligent driving, when promoting L2 driver assist system. Second, manufacturer or solution provider are required to clearly define the capabilities and safety measures of driver assist system. Terms like zero takeover or hands off must not be used and the responsibility of the driver for continuous monitoring must be emphasized. So we believe this is a clear beneficial for Pony AI as it helps foster a comprehensive and a clear understanding of distinctions between L2 and L4 for the public. That's also the reason why we consistently emphasize that L2 and L4 are fundamentally different in value add to the customer. Being more specifically, only L4 can truly fulfill users' need in the situation where they are looking for relaxation or even wish to take a nap while autonomous driving system is on. So I would like to go into more detail about the technical difference between L2 and L4 systems. So L2 systems widely use imitation learning. The AI drivers learn by copying human behavior from real-world driving data. The limitation of imitation learning is that AI drivers cannot understand the reasoning behind the driving behavior. So as a result, it is not safe enough to handle ever-changing traffic scenarios. For L4 systems, we use the reinforcement learning and over-generative pony world. So under Pony World, our virtual driver teaches itself through a large amount of generative data. This allows our virtual driver to understand why by analyzing the outcome of every action, teaching them to make smarter decisions in different scenarios, and eventually surpass the safety of human drivers. So over time, our virtual driver trains under Pony World developers advanced skills needed for complex tasks. such as multi-navigating urban areas, handling unpredictable traffic scenarios, or safely operating for 500,000 hours without any human intervention. More importantly, the key competitive edge differs significantly between the two approaches. So imitation learning requires a large amount of data, while reinforcement learning relies heavily on AI model capabilities. This underlying distinction creates a considerable barrier, making it challenging to transit from one to another. It basically requires a company to start over and build the whole team of exercise, which means it cannot be simply accelerated through prior experience. Thank you. I will forward back to the operator.

Thank you. And our next question comes from with Jefferies. Please go ahead.

Hi, thanks for taking my question. My question is regarding the US-China tariff issue, which appears to be easing at the moment. But I'm still wondering, will it still have any potential negative impact on the operations? How many materials are you sourced from the overseas market?

Thank you. And this is Leo. I'll take this question. We believe the potential impact from the tariffs issue will be very minimal to our operation. First, the majority of our supply chain is domestically sourced. Second, over the past few quarters, we have enhanced our supply chain resilience in response to the evolving geopolitical landscape. This includes diversifying suppliers and also increasing inventories when necessary. As a result, we are well prepared to manage this risk. In addition, I would like to highlight that our Gen 7 mass production plan has also been reflected with these assumptions and also uncertainties. Therefore, we are confident that our full-year target of deploying a 1,000-unit fleet size is on track and will not be affected by the changing trade environment. I will now get back to the operator.

Thank you. This concludes our question and answer session. I'd like to turn the conference back over to management for closing remarks.

Thank you once again for joining us today. If you have any further questions, please feel free to contact our IR team. We look forward to speaking with you in the next quarter.

Thank you. This concludes today's conference call. We thank you all for attending today's presentation. You may now disconnect your lines and have a wonderful day.