4.A. History and Development of the Company Our Corporate History Pony AI Inc. (the “Company”) was incorporated in November 2016 as an exempted company with limited liability in the Cayman Islands. In the same month, we incorporated Pony.AI, Inc., a Delaware corporation. We then commenced our U.S. operations in Silicon Valley, California through Pony.AI, Inc. In December 2016, Hongkong Pony AI Limited (“Hongkong Pony AI”), a wholly-owned subsidiary of the Company, was incorporated under the laws of Hong Kong. In April 2017, Beijing (HX) Pony AI Technology Co., Ltd. (“Beijing (HX) Pony”), was incorporated in the PRC. Beijing (HX) Pony is currently a wholly-owned subsidiary of Hongkong Pony AI. In January 2018, Guangzhou (HX) Pony AI Technology Co., Ltd. (“Guangzhou (HX) Pony”), was incorporated in the PRC. Guangzhou (HX) Pony is currently a wholly-owned subsidiary of Hongkong Pony AI. In June 2019, Beijing (YX) Pony AI Technology Co., Ltd. (“Beijing (YX) Pony”) was incorporated in the PRC. Beijing (YX) Pony is currently a wholly-owned subsidiary of Hongkong Pony AI. In April 2021, Shenzhen (YX) Pony AI Technology Co., Ltd. (“Shenzhen (YX) Pony”) was incorporated in the PRC. Shenzhen (YX) Pony is currently a wholly-owned subsidiary of Hongkong Pony AI. 76 Table of Contents In March 2022, Shanghai (ZX) Pony AI Technology Development Co., Ltd. (“Shanghai (ZX) Pony”) was incorporated in the PRC, which is a wholly-owned subsidiary of Hongkong Pony AI. Beijing (HX) Pony and Hongkong Pony AI entered into a series of contractual arrangements, as amended and restated, with Beijing (ZX) Pony and its shareholders, through which we obtained control over Beijing (ZX) Pony and its subsidiaries. In addition, Guangzhou (HX) Pony and Hongkong Pony AI entered into a series of contractual arrangements, as amended and restated, with Guangzhou (ZX) Pony and its shareholders, through which we obtained control over Guangzhou (ZX) Pony and its subsidiaries. Pony AI Inc. operated its businesses this way primarily in order to preserve the flexibility to engage in businesses that are subject to foreign investment restrictions under applicable PRC laws and regulations. As a result, we were regarded as the primary beneficiary of Beijing (ZX) Pony, Guangzhou (ZX) Pony and their subsidiaries. For financial reporting purposes, we consolidated the operation results and financial position of the former VIEs in accordance with U.S. GAAP. We refer to each of Beijing (HX) Pony and Guangzhou (HX) Pony as our wholly foreign owned entity (“former WFOE”), and to each of Beijing (ZX) Pony and Guangzhou (ZX) Pony and their subsidiaries as the consolidated variable interest entity (“former VIE” or “former VIE Entity”) in this annual report. We terminated the contractual arrangements among our former WFOEs, the former VIEs and their respective nominee shareholders, and acquired the shares of the former VIEs from their respective nominee shareholders, after which the former VIEs have become wholly-owned subsidiaries of our company since February 2024. In September 2024, PONY.AI EUROPE S.à r.l. was incorporated in Luxembourg. PONY.AI EUROPE S.à r.l. is currently a wholly-owned subsidiary of Pony AI Inc. In April 2025, Pony AI – FZCO was incorporated in the United Arab Emirates. Pony AI – FZCO is currently a wholly-owned subsidiary of Hongkong Pony AI. In May 2025, Company Pony AI was incorporated in Saudi Arabia. Company Pony AI is currently a wholly-owned subsidiary of Hongkong Pony AI. In November, 2025, PONY.AI 株式会社 was incorporated in Japan. PONY.AI 株式会社 is currently a wholly-owned subsidiary of Pony AI Inc. On November 6, 2025, Hong Kong time, our Class A ordinary shares commenced trading on the Main Board of the Hong Kong Stock Exchange under the stock code “2026.” We raised from our global offering in connection with the listing in Hong Kong approximately HK$6,454.4 million in net proceeds after deducting underwriting commissions, fees and the offering expenses. In February 2026, Shanghai (HX) Pony AI Technology Co., Ltd. was incorporated in the PRC, which is a wholly-owned subsidiary of Hongkong Pony AI. In March 2026, Zhejiang (YX) Pony AI Technology Co., Ltd. and Hunan (YX) Pony AI Technology Co., Ltd. were incorporated in the PRC, each of which is a wholly-owned subsidiary of Hongkong Pony AI. ​ 77 Table of Contents 4.B. Business Overview Our Company Pony is a leader in achieving large-scale commercialization of autonomous mobility. Starting from scratch and bringing our technology to people’s lives is by itself a testament to our commitment to autonomous mobility. Yet the progress we have made to date is what sets Pony apart from our peers: · We were among the first in China to secure licenses for fully driverless robotaxi operations across all four Tier-1 cities in China (namely Beijing, Shanghai, Guangzhou and Shenzhen). In addition, we are also the first company in China to receive regulatory approval for driverless robotruck platooning tests on cross-provincial highways. · As of March 31, 2026, we operated a fleet of over 1,400 robotaxis, which has accumulated over 65.0 million kilometers of autonomous driving mileage cumulatively. Following our recent expansion into Croatia, Hangzhou and Changsha in March 2026, we are on track to deploy over 3,000 Robotaxis in more than 20 cities globally by the end of 2026. · We have formed a joint venture with Toyota and GTMC to advance the mass production and large- scale deployment of fully driverless robotaxis in China under the joint deployment model. In addition, we have partnered with other leading OEMs, such as BAIC, GAC and SAIC, to leverage their mature supply chain and after-sales network and jointly deploy vehicles in the overseas markets. · Our Robotaxi business reached city-wide unit economics breakeven in Shenzhen in February 2026, following the same milestone achieved in Guangzhou in November 2025. In terms of scale, the number of total paid orders in Shenzhen from January to mid-February 2026 exceeded the aggregate number for the entire year of 2025. In addition, the daily net revenue per Gen-7 vehicle on the record peak day reached an all-time high of RMB394 in Shenzhen on March 22, 2026, with 25 orders per vehicle for the day. Building upon our initial market success in China, Pony is steadfastly committed to providing this safe, sustainable, and accessible autonomous mobility on a global scale. To date, our presence has extended beyond China to encompass Europe, East Asia, the Middle East and other regions, paving the way for widespread accessibility to our advanced technology. We have made substantial progress in advancing the commercialization of our robotaxi business, marked by the scaling of production and deployment of our Gen-7 robotaxi fleet, the expansion of fully driverless commercial operations and citywide coverage across major cities in China, and the advancement of our international expansion through strategic partnerships. Specifically: · In July 2025, we began mass production and road testing of multiple Gen-7 robotaxi models, representing a significant milestone in transitioning from pilot programs to scalable fleet deployment. In August 2025, we accelerated the production of our Gen-7 robotaxi vehicles, with initial batch production completed and large-scale manufacturing initiated with OEM partners. In October 2025, we reached a key milestone with the production of the 300th Gen-7 robotaxi (BAIC model), further scaling our manufacturing capabilities toward mass deployment. By March 2026, our Gen-7 robotaxi fleet continued to scale, supported by expanding deployment and improved operational efficiency, contributing to increased ride volumes and revenue generation. 78 Table of Contents · In July 2025, we received a permit to provide fully driverless commercial robotaxi services in Shanghai’s Pudong New Area, making us among the first companies authorized to operate such services and further expanding our presence in Tier-one cities. In October 2025, we secured Shenzhen’s first citywide permit for fully driverless commercial robotaxi operations, enabling expansion from district-level operations to citywide deployment. In November 2025, we commenced fully driverless commercial operations across multiple Tier-one cities, including Guangzhou, Shenzhen and Beijing, further expanding service coverage and accelerating commercialization. · In May 2025, we announced a strategic partnership with Uber to deploy our robotaxis onto the Uber platform, marking a key step toward integrating our autonomous mobility solutions into global ride-hailing ecosystems and expanding into international markets. In July 2025, we announced our robotaxi deployment plan in collaboration with the Roads and Transport Authority in Dubai, representing a key step toward entering the Middle East market. In September 2025, we expanded into additional international markets through partnerships with local operators, including Singapore and the Middle East, subject to applicable regulatory approvals. With these milestones, Pony is on track to achieve large-scale commercialization of our Virtual Driver technology. Specifically, we aim to develop a commercially viable and sustainable business model that enables the mass production and deployment of vehicles equipped with our Virtual Driver technology across transportation use cases, providing autonomous mobility to people and businesses around the world. Our Vehicle-Agnostic Virtual Driver We have built the proprietary vehicle-agnostic Virtual Driver, our full-stack autonomous driving technology that seamlessly integrates our proprietary software, hardware and services, to deliver safe and reliable autonomous mobility in diverse use cases. Our Virtual Driver can be deployed across multiple vehicle platforms and applications to bring a compelling, customized autonomous driving experience to a wide user base in all road conditions. · Proprietary AV Software Stack We pioneered in introducing the world model methodology, PonyWorld, to train our Virtual Driver, enabling our autonomous driving system to “learn by practicing” in AI-generated scenarios. We have advanced our autonomous driving solutions by leveraging end-to-end (E2E) technology, while still integrating the key strengths of individual modules such as perception, prediction, planning and control, and simulation. This approach enhances the efficiency and reliability of our intelligent systems, delivering seamless integration and superior performance in real-world scenarios. · PonyWorld – the Path to Fully Driverless Level 4 Solutions Driven by the vision to create the next-generation autonomous driving solutions that could outperform human drivers, we positioned ourselves at the forefront of the industry as a pioneer in implementing a world model methodology. Currently, most autonomous driving solutions rely on “learning by watching” human-driving data, which inherently limits their performance to optimize for safe, comfortable and efficient driving outcomes in fully driverless Level 4 operations. For Level 4 autonomous driving, we believe the object is not merely to mimic human driving behavior, but to drive well — safely, comfortably, and efficiently. World model is a “coaching” methodology for training Level 4 autonomous driving systems by enabling the Level 4 autonomous driving systems to “learn by practicing” in AI-generated scenarios. World model is not merely an algorithm or a standalone system; instead, it “teaches” Level 4 autonomous driving systems how to handle diverse and intricate driving conditions. Driven by the innovativeness of world model methodology, we have developed PonyWorld, which utilizes the latest technology and advanced theories (including reinforcement learning (RL), imitation learning (IL), stable diffusion and LLMs) to create a high-fidelity environment that precisely replicates real-world conditions in both visual detail and dynamic response. PonyWorld consists of three key components: the ability to generate realistic scenarios and sensor data, a high-fidelity simulation system, and a comprehensive set of evaluation metrics. Together, these elements allow PonyWorld to effectively “coach” our Level 4 autonomous driving system to handle real-world challenges. 79 Table of Contents In April 2026, we launched PonyWorld 2.0, the latest upgrade to our proprietary world model and a major advancement in the core training system behind our autonomous driving stack. PonyWorld 2.0’s most important advance is its ability to diagnose its own weaknesses and guide targeted improvement. The upgrade brings three core capabilities: self-diagnosis, targeted data collection in scenarios where the model still falls short, and more efficient training focused on the hardest cases. The launch comes as the autonomous driving industry enters a new commercial phase. The challenge is no longer just proving that driverless technology works. It is now about improving performance quickly and consistently enough to support broader deployment, stronger unit economics, and sustained technical leadership. For a world model to function effectively in training a Level 4 autonomous driving system, we believe it needs to perform well in three respects. First, it should support a learning-based framework for defining desirable driving behavior, so that the system can optimize for safety, comfort and efficiency rather than merely imitate human driving patterns. Second, it should provide a high-fidelity representation of the physical world, accurately reflecting the kinematics of our vehicle and surrounding road users. Third, it should model how other traffic participants may react to the behavior of an AI driver in both routine and complex scenarios. We believe that only when these capabilities work together can a world model generate training outcomes that meaningfully improve the performance of our Level 4 autonomous driving system. PonyWorld consists of a high-fidelity simulation system which leverages the latest technology to create a high-fidelity environment that precisely replicates a wide range of real-world traffic scenarios in both visual detail and dynamic response. Unlike conventional Level 4 autonomous driving systems that depend on human-driving data, our simulation system creates a high-fidelity simulation environment that automatically generates driving scenarios and corner cases for the Level 4 autonomous driving system to comprehend, adapt and make driving decisions. The simulated scenarios include interactions between pedestrians and other traffic participants, as well as factors such as weather conditions and road surface slipperiness. In a virtual environment created by PonyWorld, traffic participants are designed to mirror human driver behavior and interact with the Level 4 autonomous vehicles in a human-like manner, thereby simulating a driving scenario that is more representative of real-world traffic conditions. The interaction modelling is critical because surrounding traffic participants may respond differently to an AI driver than to a human driver. In addition, It empowers us to push the boundaries of system capabilities with exceptional accuracy, simulating critical scenarios such as a child suddenly appearing, an uncovered manhole, or falling debris from vehicles ahead. The engine provides a highly efficient and flexible virtual development and testing environment, significantly enhancing safety metrics while reducing both testing time and associated costs. By using PonyWorld to answer various what-if questions, the model is given the ability to repeatedly test and make mistakes in order to find the optimal strategy in reality, thereby ensuring the safety, efficiency, and comfort of the system. This is in contrast to validating various possibilities through road tests, since even extensive road testing may not cover all scenarios. 80 Table of Contents The reinforcement learning component then activates the Level 4 autonomous driving system to learn from these AI-generated driving environments and decision-making experiences to continuously refine its performance. We have also developed a comprehensive set of evaluation metrics to assess the performance of our Level 4 autonomous driving system in various simulated driving scenarios. By measuring the performance of the Level 4 autonomous driving system against these metrics, we could implement real-time improvements to enhance its Level 4 autonomous driving systems. World model also enhances reinforcement learning, interaction prompts and comprehensive metrics: ​ ​ World model also enhances reinforcement learning, interaction prompts and comprehensive metrics: Reinforcement Learning. Reinforcement learning trains AI agents through environmental interaction, rewarding optimal behaviors and penalizing poor ones. This learning by practicing approach allows AI agents to progressively improve their decision-making until reaching optimal performance. This learning process can be further enhanced through the involvement of human guidance, where domain experts directly identify preferred behaviors, supplementing or even replacing the standard reward-penalty system. This approach achieves dual benefits — faster convergence during training (i.e., an acceleration of the learning process) and better alignment of the AI’s behavior with human values and expectations. Interaction Prompts. Interaction prompts serve as an additional layer of information encoding that enriches the model’s understanding of its environment. These prompts encapsulate human knowledge and rules that are explicitly injected into the model. By encoding this information, interaction prompts help to improve the model’s interpretability. For instance, traffic rules or social norms can be encoded as prompts, ensuring the model adheres to these guidelines during interaction scenarios. This aspect enhances the transparency and trustworthiness of the model’s decisions. Comprehensive Metrics. A thorough set of metrics and evaluation criteria is essential for assessing the model’s performance from multiple perspectives. Comprehensive metrics enable a holistic understanding of where the system excels and where it needs improvement. These metrics include not only the accuracy or effectiveness of the model’s actions but also aspects such as safety, fairness, and user satisfaction. By evaluating the model across multiple dimensions, the developer can identify the strengths and weaknesses with higher precision, guiding further refinement and training. The world model methodology enhances reinforcement learning by leveraging human insights, structured interaction prompts, and detailed evaluation frameworks consist of comprehensive metrics to create models that are not only skilled at performing tasks but are also aligned with human norms, highly interpretable, and rigorously assessed. This holistic approach ensures the development of systems are both effective and trustworthy. 81 Table of Contents Our world model methodology has allowed our Level 4 autonomous driving system to become “smarter” with each training cycle, strengthening our established presence in the industry in terms of both technology development and commercialization. Since the implementation of world model methodology in 2020, we could train the Level 4 autonomous driving system using virtual data, thus breaking through the limitation of data quantity. Using PonyWorld to train our Level 4 autonomous driving system, it is estimated that over ten billion miles of simulated test data could be generated per week, which is at least 100,000 times more than the road test data generated prior to its implementation. Before the implementation of the world model, we could rely solely on real world road testing data to train our Level 4 autonomous driving system, amounting to several million kilometers per year max. The introduction of the world model enables the generation of virtual data, free from the physical constraints of traditional road testing. Consequently, the volume of data available for training our Level 4 autonomous driving system was estimated to grow exponentially. In addition, PonyWorld could generate scenarios for extreme weather conditions, accounting for the influence on sensor data and subsequent alterations in vehicle control. With the simulated scenarios generated by PonyWorld, it is expected that our Level 4 autonomous driving system could be trained in 95% of extreme weather conditions. PonyWorld could therefore facilitate training of Level 4 autonomous driving systems to navigate inclement weather safely and efficiently. We believe, the PonyWorld-trained Level 4 autonomous driving system is poised to surpass human drivers to make safer and more efficient driving decisions that significantly enhance passenger experience. · “E2E” — Closed-loop Evolution Driven by Advanced Technology Our models are designed for interpretability and are powered by technology with outstanding generalization capabilities. This end-to-end approach has demonstrated safety performance that significantly exceeds that of human drivers, while also reducing the costs of scaling operations across new regions and cities. The traditional architecture is hierarchical and it involves step-by-step perception, prediction, and planning and control, which is popular among autonomous driving companies due to its ease of implementation. By embracing E2E technology, we have streamlined the system architecture of our Level 4 autonomous driving system while enhancing overall performance, delivering exceptional functionality in diverse and complex road conditions and extreme weathers such as storms. We believe that streamlining different modules enables the system to achieve more reliable, efficient and safer performance while minimizing lag and data loss. Learnable Metric Space: The E2E system leverages a learnable metric space, integrated with a generative model, to simulate realistic behaviors of vehicles and other road agents that align with real-world scenarios. A key component of this system is its discriminator, which plays a vital role in closed-loop training by measuring loss and in closed-loop testing by evaluating performance metrics. By accurately modeling and analyzing these behaviors, we could ensure our Level 4 autonomous driving system is both realistic and highly effective across a wide range of driving conditions. As a result, the E2E system could reduce the reliance on maps, enabling robotaxi services to rapidly expand to more cities at a relatively low cost. Knowledge Distillation from LLMs: By employing smart labeling and feature distillation techniques, we transfer the vast knowledge embedded in large language models (LLMs) into our resource-constrained, on-board E2E models. This process effectively incorporates the nuanced understanding of complex, long-tail scenarios into compact models, enabling the system to address diverse and intricate driving conditions while maintaining high efficiency and performance under limited computational resources. Self-Supervised Interpretation Model: Our self-supervised interpretation model utilizes data which lacks meaningful labels or tags and consists of easily accessible human-created samples (“unlabeled data”) to train the E2E model through advanced representation learning. This representation model significantly reduces the need for labeled data by enabling the system to interpret E2E inference results, such as perception outputs, prediction outcomes, specific decisions, and detailed scene descriptions. With the assistance of the model, the labeling efficiency was improved by three times. Human labeler could rely on the model to label data first and then only need to cross-check the model-labeled data. This approach ensures comprehensive interpretability, offering deeper insights into the Level 4 autonomous system’s functionality and decision-making processes. 82 Table of Contents Learnable Optimization Model: The learnable optimization model integrates model- based and optimization-based methods, combining the adaptability of data-driven approaches with the precision and controllability of optimization techniques. The model-based method utilizes both deep learning layers such as Convolutional Neural Networks (CNN) and Vision Transformers (ViTs), whereas the optimization-based methods adopt mathematical formula such as learnable Extended Kalman Filter (EKF) and Neural Ordinary Differential Equations (NODE). As a result, the system can learn the behavior of human drivers, while ensures that the output maneuver of Level 4 autonomous vehicle obeys physical laws for realistic results. For instance, the learnable EKF is specifically designed for end-to-end tracking, while the end-to-end planning models incorporate model-based cost functions alongside optimization-based solvers. This hybrid methodology ensures precision, adaptability, and efficiency, reinforcing the reliability of our Level 4 autonomous driving system. In addition, compared to module-by-module systems, an E2E system possesses stronger modeling capabilities. With E2E system in place, the same level of manpower can yield better results. After introducing E2E, engineers could focus more on tasks such as data collection, filtering, and analysis, leveraging data to enhance the model’s performance for iteration. In contrast, before introducing E2E, engineers had to manually add rules, which required significant manpower for complex systems. As a result, the E2E system enables engineers to spend less time on manual tasks and focus more effectively on system upgrades, thereby accelerating iteration efficiency and reducing iteration costs. · “Perception and Prediction” — “Zero Critical Missing” Achieved Through Large Multimodal Models with Prompt Learning Perception. The perception module enables our Virtual Driver to see and understand the world around our autonomous vehicle, from puddles on the road to a plastic bag flying in the air. The following diagram demonstrates how our perception module works to produce the data output required to enable autonomous driving: ​ ​ By fusing and processing relevant data collected by our comprehensive sensor suite, our perception module enables object segmentation, detection, classification, tracking and scene understanding automatically. In inclement weather conditions such as sandstorms and heavy rains and snow, our perception module demonstrates superior perception capabilities compared to human drivers. 83 Table of Contents To ensure performance, we leverage a hybrid solution that combines our state-of-the-art deep learning technology and the heuristics approach to process, refine and use the relevant data collected by our sensors. In order to bridge the simulation-to-reality gap for our deep learning technology, we apply heuristics, which is an expression of human knowledge and common sense, by adding deterministic math formulas and rules to the decision-making layer — for instance, a car usually does not cross road barriers, and a pedestrian’s speed usually does not exceed 10 meters per second. This hybrid solution enables accurate detection, classifications and tracking in dense and complex environments. Prediction. The prediction module forecasts how other vehicles, pedestrians and other objects may move and behave based on a number of data, including the output of our perception module, raw sensor data, and data regarding historical decisions made by similar road agents. The following diagram demonstrates how our prediction module works to anticipate the trajectories of other road agents: ​ By using a mix of deep learning and heuristics to enable rapid learning and adaptation, our prediction module delivers a series of predicted trajectories for each observed road agent, with each trajectory having an assigned probability of occurrence. These predictions are subsequently used by our other modules, such as the “planning and control” models, to inform the decision-making process for route selection and maneuver execution. Large multimodal models with prompt learning. We design and train our perception and prediction modules to be capable of achieving “zero critical missing,” meaning they are able to accurately detect and classify all objects on the road and anticipate all potential object trajectories. In addition, our perception and prediction modules both utilize a large transformer framework that is multi-modal, multi-tasking and prompt-tuned, ensuring a highly reliable and accurate system with low latency. Powered by prompt learning technologies, our perception module integrates inputs from various modalities, including point cloud, images, and electromagnetic responses, to accurately detect a variety of distinct object types based on a single model. This approach allows us to significantly reduce latency while improving the precision of perception as compared to the traditional multi-task learning technology. Our prediction module employs a multimodal deep learning model that fuses information from both perception observations and human common senses. These common senses are represented by knowledge graphs extracted from traffic rules and human-designed prompts. Transformer structures capture the correlations between different modalities. To address corner cases such as aggressive or erratic driving behaviors, we add extra learnable and specially-designed prompts for each case, in addition to large-scale dataset of daily driving records. This approach ensures that our prediction module can efficiently respond to unexpected behaviors. · “Planning and Control” — Ensuring Safety Through Game Theory and Learning-Based Planning Our planning and control module is designed to plan and execute safe, comfortable and efficient road maneuvers based on input from our perception and prediction modules. We leverage our strong AI capabilities to create a robust planning and control module that is capable of smoothly navigating complex road layouts — from streets and alley ways to bustling eight-lane intersections, while being prepared for outlier behaviors or unexpected events caused by other road agents. 84 Table of Contents Importantly, our data-driven planning and control module does more than directing the vehicle’s movement based on its surrounding environment and the behavior of nearby road agents — it chooses the best route, accelerates and decelerates smoothly and changes lanes appropriately, which together contribute to a safe, comfortable and efficient autonomous driving experience. This is achieved by using game theory and conditional prediction to analyze the probabilistic prediction results, and make the best driving decision under each prediction while always being prepared for the worst-case scenario. Game theory is utilized to model and analyze the interactions between our vehicle and other road agents such as pedestrians and cyclists, with the aim of creating safer and more efficient transportation systems. For instance, if our autonomous vehicle and a human-driven car approach an intersection simultaneously, game theory can help determine the optimal decision for our vehicle. This decision can affect the overall outcome of the system, and game theory can identify the best combinations of actions to minimize conflicts, improve safety, and enhance efficiency. Using game theory results in a one-magnitude improvement in safety during rush hours and congested roads, as well as potential erratic driving behaviors. The following photos include some examples where game theory is utilized: ​ ​ ​ ​ ​ ​ Example 1: Unprotected Left Turn in a Chaotic Intersection Example 2: Moving Through Hectic Traffic Flow with Numerous Pedestrians and Cyclists ​ ​ ​ ​ ​ ​ Example 3: Snowy Day. Crossing Pedestrian Interaction Example 4: Large Construction Area. Merging with Vehicles. ​ To ensure that our autonomous vehicle drives like humans, we have tuned our decision maker using reinforcement learning from human feedback (RLHF). We have utilized human labelers to provide feedback on the safety, comfort and efficiency of the autonomous driving system in various scenarios. This feedback is then used to train a reward function, which guides the tuning of our deep learning decision-making on a much larger dataset. As a result, our decision-making system has sufficient generalization ability to handle both common cases and extreme scenarios. 85 Table of Contents · “Data & Infrastructure” — Foundation to Rapid Iteration, Scalable Deployment and Efficient Testing Successful autonomous driving technology deployment and scale rely on a complete set of supporting software infrastructure. From the real-time onboard operating and monitoring systems to offline simulation and machine learning training, and from data collecting and recording system to offline data analysis and mining, we have built a full suite of capabilities that drives rapid iteration, scalable development of a high-quality system, and efficient testing in all aspects of software and hardware development. To ensure safe and reliable autonomous driving, all of our perception, prediction and planning and control modules are designed to deal with complex corner cases, which require a special data mining mechanism to dig from tons of road testing data that our autonomous vehicle collects on a daily basis. Therefore, we purposefully design our data mining system to automatically identify such corner cases in which our algorithms and modules need to be refined. Those corner cases will be recorded and added to our training data and simulation system. Our onboard monitoring system schedules, runs and monitors all of our software modules underlying our Virtual Driver. It implements a unified application programming interface for seamless module communications, which maintains a stable data flow from the upstream sensors all the way to the downstream planning and control module. This has helped to ensure safety and performance of our Virtual Driver. · “Tool Chains & Metrics System” — Advancing Autonomous Driving Through Rapid Iteration and Trackable Metrics We have developed a comprehensive, scalable, and user-friendly tool chains and metrics system to support every major stage of our technology development, ensuring both safety and reliability while accelerating the iterative cycles. Such powerful proprietary tool chains and metrics system allows us to develop and train autonomous driving systems that are capable of adapting to new cities or regions and operating effectively typically in less than two weeks. The development of autonomous driving technology heavily relies on data and involves multiple stages, such as data analytics, data mining, code development, data labeling, model training, simulation-based evaluation, continuous integration/continuous delivery (CI/CD), and feature release. We have developed powerful, automated tool chains that provide one-stop solutions to engineers with low latency covering the entire AV software development process. Adding server resources can further reduce latency without any technical changes required. All of our tool chains are based on distributed data and computation platforms that can be easily deployed to cloud or private environments compliant with regulations. Scaling autonomous driving technology is critical for its success as it must adapt to varying road conditions, traffic rules, and driving patterns across geographies. This poses another challenge, as autonomous driving systems must recognize and respond to various situations they may encounter on the road. With our advanced tool chains, we have significantly reduced the time required for us to penetrate a new city from six months to half a month. Developing a consistent and accurate way to measure the performance of autonomous driving systems is critical to ensuring the safe and trackable deployment of this technology. We have created a comprehensive metrics system that combines real-world road testing and simulations to effectively measure the safety of the autonomous driving system. By leveraging the strengths of both real-world road testing and simulations, we can evaluate the system’s performance in a wide range of scenarios, which may be difficult to recreate on the road. Our metrics system is highly sensitive and has high credibility in detecting any subtle regression in the system’s performance. This is critical to ensuring the safety of passengers and other road agents. Additionally, we have established a process for ongoing monitoring and improvement of the system’s safety performance. This includes regular updates and improvements to the system’s software and hardware, as well as ongoing testing and evaluation to ensure that the system remains safe and effective over time. 86 Table of Contents Sensors and hardware According to Frost & Sullivan, we are the world’s first autonomous driving technology company to develop Level 4 autonomous vehicles with automotive-grade, factory-installed sensors and hardware (including LiDAR and computation platform) integrated, while our peers currently have not integrated automotive-grade components into their Level 4 autonomous vehicles. To commercialize and scale our Level 4 autonomous vehicle fleet, transitioning to more cost-efficient automotive-grade hardware was a necessary step. By adopting automotive-grade hardware, we have achieved significant cost reductions that support mass production, enhance the viability of our business model, and open up new commercial opportunities in the Level 4 autonomous vehicle market. We have established in-depth collaboration and mass production arrangements with multiple global and Chinese OEMs, while our peers have not had any mass production arrangements with global and Chinese OEMs, according to Frost & Sullivan. We employ a multi-sensor approach that incorporates LiDARs, high-resolution cameras as well as radars to accurately and precisely perceive and understand the environment surrounding our Level 4 autonomous vehicles. Our current Level 4 autonomous vehicle model is equipped with a robust sensor suite comprising nine LiDAR units, 14 high-resolution cameras, four millimeter-wave radars, four microphones, two wade sensors, and one collision sensor suite. This comprehensive sensor array enables 360-degree, blind-spot-free perception around the vehicle, with environmental detection capabilities extending up to 650 meters. Our upgraded sensor suite could enable navigation in urban and highway environments, as well as reliable operation in inclement weather conditions such as rain, snow, and fog, and corner cases such as sudden pedestrian appearances and yielding to emergency vehicles. The following diagram illustrates our sensor designs on our 7th generation Level 4 autonomous vehicle model: ​ ● LiDARs. LiDAR uses laser beams to accurately detect objects around our Level 4 autonomous vehicle, which allows high resolution range sensing in all lighting conditions. We deploy nine LiDARs on both the top and sides of our current Level 4 autonomous vehicle model, which can generate precise and real-time three-dimensional images of the surrounding, from cars to traffic lights to pedestrians, in a wide range of environments and under diverse lighting conditions day and night. 87 Table of Contents · Cameras. By equipping our Level 4 autonomous vehicle with 14 high-resolution cameras at every angle, the vehicle is capable of maintaining a 360-degree view of its surrounding environment without major blind spots, thereby providing a broader picture of the traffic conditions around it. In our latest Virtual Driver, we install in-house designed cameras tailored specifically for our Level 4 autonomous vehicle fleet needs. For example, the customized automotive-grade cameras improved image quality significantly, enhancing sensor data input to ensure safety, reliability and optimal performance of our Level 4 autonomous driving solutions. · Radars. Radar emits radio waves that detect objects and gauge their distance and speed in relation to our vehicle in real time. As compared to LiDAR and camera, radar works best in inclement weather such as rain, snow and fog. Our Level 4 autonomous vehicle is equipped with four long range radars at the front, back and two corners of rear side, respectively. Our 4D radars featured increased sensing capabilities with detection range up to 300 meters, higher accuracy, improved object recognition, real-time adaptability and robustness. · GNSS / IMU. In addition to perception sensors, we also use two other types of sensors in our system, a high-accuracy global navigation satellite system (GNSS) and inertial measurement units (IMU). These sensors work together with our high-definition maps and localization module to ensure accurate positioning of our Level 4 autonomous vehicles. By synchronizing inputs from the sensor suite, we effectively balance the inherent strengths and weaknesses of the different sensors, leading to improved precision in outlining the environment around the vehicle. In addition, we also integrate information from multiple sensors of the same object type to yield a more accurate and reliable representation of the surrounding environment by taking advantage of partially overlapping fields of view. Through the integration of our redundant sensor coverage and intelligent software modules, our Virtual Driver is able to sense the surrounding environment and objects at all times and in complex weather conditions, resulting in safer and more reliable Level 4 autonomous driving solutions. Through years of dedicated research and practical application of sensor technology, combined with our in-depth industry insights, we have developed a highly comprehensive sensor evaluation and selection system. This system enables us to select the best sensors available in the market while offering valuable suggestions to our suppliers on how to improve their product design and quality for Level 4 autonomous driving applications. Furthermore, drawing on our unique insights into the specific demands of sensor design and functionality for Level 4 autonomous vehicles, we work closely with our suppliers to develop customized sensor products that are better suited for Level 4 autonomous driving scenarios. Via rigorous testing and design improvements, the hardware of our system has been customized to tackle challenging driving conditions. For instance, to enhance detection performance and accuracy, we have developed an advanced sensor cleaning solution, ensuring that vehicles can maintain real-time awareness of complex road conditions, meeting the demanding operational and safety requirements for robotaxi services to operate in both urban areas and highways during inclement weather conditions. This collaboration supports the continued optimization of our highly integrated AV software and hardware, ensuring that we provide our customers with the safest, most reliable and efficient Level 4 autonomous driving solutions. Built upon our extensive industry expertise and robust partnerships, we distinguish ourselves as the sole autonomous driving technology company that designs our Level 4 autonomous vehicles, integrating with auto-grade, factory-installed sensors and hardware, according to Frost & Sullivan. Computation System Our computation system is responsible for processing the data collected from the sensors and running our proprietary algorithms in real time to enable our vehicles to drive autonomously. In designing and configuring our computing system, we focus on performance, reliability and resource efficiency. Each piece of the computation system is validated by well-defined compliance tests. For example, our latest computing unit is built on NVIDIA DRIVE Orin, an automotive-grade processor purpose-built for Level 4 autonomous vehicles, for high performance and scalable compute. Our in-house development of the autonomous driving computation unit (ADCU), a fully automotive-grade computing platform, has enabled us to define a computation architecture that is tailored specifically for Level 4 autonomous driving applications. We have customized the memory system, data pipeline, and time synchronization topology to ensure that all processor capabilities are utilized to their maximum potential. Furthermore, our in-house ADCU can be fine-tuned to balance performance and resource consumption, making it a more sustainable and cost-effective solution. Additionally, the ADCU can be more easily adapted and upgraded as new technologies become available, enabling greater flexibility and scalability. This transition from industrial-grade to automotive-grade computing platform allows us to deploy and scale safer, more efficient and cost-effective Level 4 autonomous vehicle fleets. 88 Table of Contents Vehicle Integration Vigorous vehicle engineering brings together every piece of our Virtual Driver, from our robust AV software to the high-quality hardware sourced from our business partners. By integrating them on diverse vehicle platforms, we design our Level 4 autonomous vehicle, as a carrier of all, to offer the safest and smoothest passenger experience. We utilize an automated standard integration process to enhance vehicle consistency. Rigorous validation and testing at both component- and system-level, including mechanical shock, vibration, thermal chamber and waterproof tests, are conducted to ensure optimal performance, reliability, and stability. Together with our partners, we have established an integrated and streamlined assembly line, fully prepared for the mass production of vehicles incorporating our advanced Virtual Driver technology. This endeavor is bolstered by our utilization of our OEM partners’ industrialized designs, quality control and supply chain management process, yielding improvements in reliability, efficiency, and scalability. Years of testing and design improvement with our OEM partners over the course of our six generations of Level 4 autonomous vehicle models underpin our confidence in our purpose-built Level 4 automation. With each new generation of our Level 4 autonomous vehicle model, we strive to deliver improved and more sophisticated hardware designs that better integrate with the vehicle platform, while also enhancing cost efficiency and adaptability. Our dedication to ongoing improvement means that each iteration of our vehicle represents the latest advancements in Level 4 autonomous driving technology, ensuring that we provide passengers with the safe and most advanced Level 4 autonomous driving experience possible. Our highly integrated Level 4 autonomous vehicles were designed to closely resemble mass-produced cars in terms of weight, power consumption, size, and other key aspects. Our 6th generation Level 4 autonomous vehicle model, developed in partnership with Toyota, has been deployed for public-facing robotaxi services since July 2023. In April 2025, we launched the 7th generation robotaxi, including three vehicle models co-developed through our strategic partnerships with Toyota (Toyota bZ-4X), BAIC (Alpha-T5) and GAC (AION V). ​ Our 7th generation Level 4 autonomous vehicle model features a redundant vehicle platform. With redundant sensors, computation systems, power, and actuators in our vehicle platforms, we can avoid single points of failure. For example, in our computation system, different processors cross-check and function as backup systems for each other, and certain algorithms running on the GPU can fall back to the CPU if an error occurs. Another example is that if the main power system fails, the backup power system will engage and ensure continuous power to the computation system, and thus the continued operation of our Virtual Driver as a whole. We are currently co-developing 7th generation Level 4 autonomous vehicle models with BAIC, GAC, and SAIC Motor, in addition to our strategic collaboration with Toyota. Our 7th generation vehicle marks a critical step to advance the large-scale deployment of our fully driverless Level 4 robotaxis. Our 7th generation Level 4 autonomous vehicle models will feature multiple layers of redundancy in system design coupled with industry-leading remote assistance capabilities. We designed our 7th generation Level 4 autonomous vehicle to offer smoother passenger experience with enhanced HMI system and efficient trip plan. Our approach to ensuring the reliable and safe operation of our Level 4 autonomous vehicle includes a three-layer system comprised of (i) normal operation mode, (ii) degraded safe mode, and (iii) minimal risk condition mode. The degraded mode and minimal risk condition mode operate on physically independent redundant platforms, which include redundant sensors and computations. In the event of faults occurring during normal operation, we detect these faults and transition the system to a degraded safe mode, allowing the vehicle to drive to a safe location. If critical faults occur that cannot be addressed by the degraded safe mode, the minimal risk condition mode will be triggered, allowing the vehicle to at least stop in its lane without collision. 89 Table of Contents Remote Assistance We have a cost-effective and scalable remote assistance system (“RA system”) to ensure that our Level 4 autonomous vehicles can handle unexpected situations with ease. Unlike remote safety drivers who would assume control over the vehicle from time to time, our remote assistance operators only monitor and provide driving guidance to vehicles without any direct control over the vehicle. Our Virtual Driver handles all driving decisions autonomously, from perceiving surroundings to executing maneuvers such as acceleration, braking and lane changing. It operates safely and efficiently with minimal human intervention, primarily due to rigorous training on various scenarios and corner cases generated by PonyWorld. This reduced reliance on remote control over vehicles also reduces exposure to potential cyber intrusions, which greatly improves the overall safety of our Level 4 autonomous driving system. Our remote assistance operators intervene only in extreme scenarios, such as temporary road closures or severe congestion, providing driving guidance to vehicles including rerouting or pulling over based on real-time data. Using multi-screen workstations, each remote assistance operator can oversee a fleet of Level 4 autonomous vehicles simultaneously, which significantly reduces labor and operational costs for our fleet management. Automotive-Grade Mass Production with Cost Efficiency Our Level 4 autonomous driving solution is fully prepared for mass production. While we do not have an in-house production function, we have established deep partnerships with several OEMs and manufacturing partners, who support our mass production efforts. We have developed our technology to ensure that all hardware complies with comprehensive automotive-grade standards, enabling scale up of production. One of the key advantages of our mass production capability is the significant reduction in costs, making our business model viable and unlocking new commercial opportunities in the Level 4 autonomous vehicle market. Beyond Autonomous Driving: The Physical AI Engine As PonyWorld continues to be refined by tens of millions of kilometers of real-world data, we are exploring its potential applicability beyond structured road environments. We believe the efficiency and targeted learning mechanisms underlying PonyWorld 2.0 are highly relevant not only for autonomous driving but also for broader, higher-complexity physical AI training. Ultimately, this foundational technology has the potential to support highly advanced automation in a variety of complex tasks outside of our core driving operations. Commercialization Models and Services Offerings We have been commercializing our autonomous driving technology by integrating it with vehicles of various models, classes and levels of autonomy to enable multiple commercial applications. We mainly focus on vehicles and use cases that maximize our commercial opportunity, including electric vehicle passenger “robotaxis” and long-distance, heavy-duty “robotrucks.” We also capitalize on our robust technology capabilities by offering POV intelligent driving solutions and value-added technological services. · Robotaxi Services We provide robotaxi services to drive passengers autonomously on a ride-hailing basis in vehicles integrated with our Virtual Driver. As of March 31, 2026, we operated a fleet of over 1,400 robotaxis, with over 65.0 million kilometers of autonomous driving mileage cumulatively. 90 Table of Contents A Fun and Safe Ride with Our Robotaxi We endeavor to offer a compelling autonomous driving experience to our passengers. Passengers can enjoy a safe, comfortable and convenient ride with us via a few clicks on our PonyPilot mobile app, which is currently available for download on both Apple and Android app stores. Once a passenger hails a ride, our mobile app will direct the passenger to a nearby travel station for pick up and drop off. We have built our proprietary human machine interface application PonyHI to improve passenger experience. PonyHI provides passengers with significant information about the journey, including the vehicle position, trip route, vehicle trajectory and road conditions. For our fare-charging autonomous robotaxi services, passengers can view the fare for the ride on both our PonyPilot mobile app and the in-car interactive interface, and they can complete payment on their mobile devices using major mobile payment apps including Weixin and Alipay. The following are screenshots of our proprietary PonyPilot mobile app: ​ ​ We are currently working with a growing number of leading TNCs in China to roll out our robotaxi services across their apps, increasing the visibility and accessibility of our services. Today, passengers can easily hail our robotaxis on our proprietary PonyPilot mobile app, WeChat mini program, Amap, WeChat mobility services platform, and Alipay mini program. Commercialization Roadmaps We launched our autonomous vehicle fleet on open roads with a safety driver in Guangzhou in February 2018, and have since then rapidly scaled our public-facing robotaxi operations in China. As of March 31, 2026, we operated a fleet of over 1,400 robotaxis. Our Robotaxi business reached city-wide unit economics breakeven in Shenzhen in February 2026, following the same milestone achieved in Guangzhou in November 2025. In terms of scale, the number of total paid orders in Shenzhen from January to mid-February 2026 exceeded the aggregate number for the entire year of 2025. In addition, the daily net revenue per Gen-7 vehicle on the record peak day reached an all-time high of RMB394 in Shenzhen on March 22, 2026, with 25 orders per vehicle for the day. Total users approached one million in China by late March 2026, nearly tripling year-on-year. 91 Table of Contents Obtaining a regulatory permit represents a critical milestone of an autonomous driving company’s technological and operational readiness towards commercialization. In China, local regulators have established rigorous, comprehensive criteria to ensure the safety and commercial viability of autonomous vehicles before granting permits for road testing and commercial operations. These criteria take into account a wide range of highly specialized and technical metrics and indicators, including the proportion of autonomous driving mileage, critical intervention and accident rate performances, simulation and other road testing results, the quality of safety drivers and remote control / assistance capabilities, contingency plans, and the number of passenger orders. By carefully evaluating these factors, regulators assess the technological and operational readiness of autonomous driving companies to safely and effectively operate vehicles on public roads. Therefore, the regulatory permit review and approval process serves as a critical safeguard to ensure that only the most advanced and reliable autonomous driving technologies are allowed to be tested and deployed on public roads. With all available regulatory permits essential for providing public-facing robotaxi services received in all four Tier-1 cities in China, namely Beijing, Shanghai, Guangzhou and Shenzhen, we are the frontrunner in advancing commercialization of public-facing robotaxi services in China, according to Frost & Sullivan. The following diagram further illustrates the key progress we have made in obtaining major regulatory permits for providing public-facing robotaxi services in China as of December 31, 2025. Note: Information about issuance status of applicable regulatory permits in the table above is based on the publicly available information and our company’s best knowledge as of the date of this annual report. Under the current regulatory framework, each of the four Tier-1 cities in China issues two categories of robotaxi permits: one for robotaxis operating autonomously with a safety driver present, and another for fully driverless robotaxis. Within each category, there are three specific types of permits: testing permits, public-facing permits and fare-charging permits. Tier-1 cities usually grant robotaxi permits in stages, with each successive stage imposing stricter technical and operational requirements such as test mileage and disengagement rate. The initial permit is a testing permit which allows an autonomous driving technology company to test its autonomous vehicle within testing areas. Then the company could apply for public-facing permits that allow testing vehicles to carry passengers without charges in all open roads. When reaching the most advanced stage, an autonomous driving company could obtain a fare-charging permit which allows it to operate autonomous vehicles for commercial services (such as ride-hailing service). We were among the first in China to obtain licenses to operate fully driverless vehicles in all four Tier-1 cities in China. In addition, we are the only autonomous driving technology company to secure all available regulatory permits essential for providing public-facing robotaxi services in all four Tier-1 cities in China, according to Frost & Sullivan. We believe this first-mover advantage for staying ahead in regulatory approval, combined with our robust technology and partnership with OEMs, has positioned us to commercialize public-facing robotaxi services at scale in China in the future. As of December 31, 2025, we had obtained all 24 categories of robotaxi permits available in Tier-1 cities in China. We also have operations on a limited scale in the U.S., with employees mainly engaged in R&D. Furthermore, we are strategically pursuing expansion opportunities in other growing markets worldwide. Collaborative partnerships have been forged with local governments, industry leaders, and technology innovators across Europe, East Asia, the Middle East and other regions, focusing on local development, deployment, and commercialization of our technology. To date, our business footprint extends to Luxembourg, South Korea, Saudi Arabia, Singapore, Croatia and the United Arab Emirates. 92 Table of Contents As our robotaxi services continue to scale, we are well positioned to connect and empower different stakeholders along the value chain with diversified monetization models with OEMs, TNCs and fleet companies, as illustrated by our go-to-market strategies depicted in the diagram below: For our robotaxi services, we operate our own Level 4 robotaxi fleet. We collaborate with OEM partners to co-develop vehicle designs that are engineered to accommodate and support Level 4 autonomous driving functionality. This partnership enables us to leverage OEMs’ expertise in vehicle engineering and mass production. We either procure robotaxis with our Virtual Driver system installed from OEMs or purchase vehicles from OEMs and install our Virtual Driver system afterwards, for deployment in our robotaxi fleets. We reach passengers through two channels. First, we integrate our services with TNCs, utilizing their platforms and user interfaces. In this model, passenger fares are processed directly through the respective TNC platforms. Passengers make payments through the TNC platforms. TNCs collect and then transfer the fare to Pony when the ride is completed, and will deduct the customer acquisition fees afterwards. Additionally, passengers can access our robotaxi services directly through our PonyPilot app. In this channel, passenger fares are paid directly to us. Revenues are recognized over time as we provide the fare-charging services. In addition, we enter into technology licensing arrangements with strategic partners for our proprietary virtual driver systems. Revenues are recognized over time as we provide the licensing services. The pricing of ride-hailing services is usage-based and calculated primarily based on trip mileage, with rates varying depending on route, traffic, and other dynamic factors. This go-to-market strategy is demonstrated by the following graphics. ​ We are developing a second business model for our Level 4 robotaxi ride-hailing service. While the core operations remain identical to the model discussed above, the key distinction lies in how the robotaxi fleet is managed. Instead of procuring robotaxis ourselves and operating and maintaining our own fleet, we will lease robotaxis from third-party fleet operators. 93 Table of Contents Under this upcoming model, third-party fleet operators will purchase robotaxis co- developed by Pony and OEMs and take responsibility for fleet maintenance. We then will lease the robotaxis from these operators, paying both leasing and maintenance fees. This approach shifts the majority of the capital expenditure associated with acquiring robotaxis, as well as ongoing fleet maintenance costs, to the third-party fleet operators. This business model is demonstrated by the following graphics. In addition to Level 4 robotaxi ride-hailing services, we provide a comprehensive suite of AV engineering solutions, including AV software deployment and maintenance, vehicle integration and engineering and road testing, to OEMs and TNCs, enabling them to integrate our Level 4 autonomous driving technology with their vehicle models or establish their own robotaxi fleets with vehicles equipped with our Virtual Driver system. These engineering solution contracts typically involve a combination of software, system integration, hardware components, and related services. These contracts and pricings are customized based on each customer’s technical and operational requirements. We typically source hardware components directly from suppliers, either for product development or to fulfill service requirements requested by our customers. Revenue is generally recognized based on the progression schedule of collaboration projects. This process is illustrated in the following graphics. ● OEMs. We believe OEMs will help us commercialize our robotaxi services at scale. We work with OEMs to co-develop and produce Level 4 autonomous vehicles across various vehicle platforms. We will deepen our collaboration with an increasing number of OEM partners to constantly upgrade and optimize our Level 4 autonomous vehicle models, delivering improved Level 4 autonomous driving experience to passengers. The Level 4 autonomous vehicles manufactured by our OEM partners will be then sold to comprise the fleets owned by ourselves or third-party fleet companies. 94 Table of Contents · Self-owned fleets & joint deployment model. While we currently serve passengers entirely with our self-owned robotaxi fleets to directly engage with them, we expect the future robotaxi fleets to be largely owned by a growing network of third-party fleet companies funded by third-party fleet owners. Under this long-term operating model, we expect to generate revenues by operating robotaxi fleets for these fleet companies. In addition, we may also generate revenues by selling Level 4 autonomous vehicles co-developed with different OEM partners to them where our Virtual Driver system has been incorporated to OEMs’ vehicle models. Under this model, we could charge fleet companies licensing fees for the use of our Virtual Driver system. We anticipate this model to enable a potentially asset-light and high-margin revenue stream, while allowing us to continuously focus on technology innovations and scale more rapidly across our geographic markets. In April 2024, we reached a significant milestone in our commercialization roadmap by unveiling a joint venture with Toyota and GTMC. The joint venture is held by Guangzhou (HX) Pony as to 50.0% and Toyota and GTMC together as to 50.0%. Under the terms of the joint venture agreement, Toyota will supply the joint venture, acting as a fleet company, with Toyota-branded battery electric vehicles. These vehicles, furnished with Toyota’s Level 4 autonomous driving-compatible redundant systems our advanced Virtual Driver technology and, can be accessed through our PonyPilot mobile app. The ownership of the robotaxi fleet shall belong to the joint venture, and we shall pay fees to the JV under a leasing arrangement, which is determined based on, among others, the number of robotaxis to be leased by us and technology to be deployed on the leased robotaxis. Any intellectual property newly created or developed based on the pre-existing intellectual property owned by any party to the joint venture shall be deemed “Joint Intellectual Property” and shall be owned by the joint venture. The parties to the joint venture, along with their respective affiliates, shall have an equal and royalty-free right to use such Joint Intellectual Property. The profits shall be distributed in proportion to the actual capital contributions paid. In addition, backed by Toyota’s platform and manufacturing expertise, we are executing our joint deployment model. Gen-7 mass production has begun, and we have secured 1,000 bZ4X vehicles this year to form the backbone of our 2026 expansion. We upgraded our strategic partnership with BAIC and GAC, to leverage their mature supply chain and after-sales network and jointly deploy vehicles in the overseas market. · TNCs. Under our go-to-market strategies, TNCs will serve as an effective conduit connecting our robotaxis with their expansive user bases. Both us and third-party fleet companies may offer robotaxi services to passengers through various TNCs, and receive a portion of fare paid by passengers as revenues under certain revenue sharing arrangements. In 2025, we formed a partnership with Sunlight Mobility, further broadening our downstream deployment network and supporting fleet rollout in a wider range of operating scenarios. We elevated our collaboration with OnTime Mobility in Guangzhou to scale our Robotaxi fleet and secure recurring licensing revenues. Concurrently in Beijing, our January partnership with Beijing ATBB Travel & Express Service Co., Ltd. (“ATBB”) expands our footprint into premium business travel, successfully penetrating higher-value mobility segments. We also continued to deepen integration with Tencent’s WeChat Mobility Services platform in Shenzhen and Guangzhou, enhancing service connectivity and user access through established mobility platforms. Taken together, these collaborations reinforce the integrated and scalable foundation supporting our continued commercialization efforts. ● Passengers. As our robotaxi fleets continue to scale, passengers may access our robotaxi services either directly on our PonyPilot mobile app or through the mobile apps operated by different TNCs. Passenger fare will be charged by us and/or the applicable TNCs, as the case may be, for each ride on a robotaxi. ● Robotruck Services Building upon a common set of underlying technology, we rolled out our hub-to-hub autonomous freight solutions in March 2021 in China to capture tremendous opportunities in the truck freight market. We have obtained autonomous driving public road testing permits in Beijing and Guangzhou, and we operated a fleet of 210 robotrucks as of March 31, 2026, consisting of both Level 2++ intelligent trucks and Level 4 driverless trucks, covering all major commercially active areas and transportation arteries throughout China. To validate our technology and business model in anticipation of large-scale commercialization in the future, we are also running our robotrucks in a variety of business scenarios. 95 Table of Contents Adapting Our Virtual Driver to Robotruck Use Cases While the key autonomous driving technology used in our robotruck services largely overlaps with our robotaxi services, we meticulously customize certain modules, such as sensor suite and control, to cater to the specific robotruck use cases. For example, we expand our vehicle’s detection range to approximately 500 to 1,000 meters, allowing our robotruck to drive safely at a high speed. We equip our robotrucks with back-facing cameras and radars, which are considered optimal for trucks to change or merge lanes. Additionally, short-range and wide-angle LiDARs are added to our robotrucks, eliminating any potential blind spots to improve safety. The following diagram illustrates our sensor designs optimized for our robotrucks: In addition, trucks are significantly less nimble than passenger cars, as characterized by longer gearshift timeframes and higher actuation accuracy requirements. Our control module is designed to dynamically adapt with high precision to varying truck trailer cargo weights as well as crosswind speeds which can both drastically alter the movement of and create unique challenges for robotrucks. 96 Table of Contents Commercialization Roadmap The blueprint of our robotruck services is built upon our strategic relationships with truck OEMs on the one hand and logistics platforms on the other. We believe our collaboration with truck OEMs will allow us to rapidly scale the production of robotrucks integrated with our technology, whereas our cooperation with leading logistics platforms will help us apply our robotruck services to commercial use cases including intelligent hub-to-hub autonomous freight solutions. The diagram below illustrates our go-to-market strategies and our current and planned monetization models for our robotruck services. ​ Sinotrans partnership. In December 2021, we announced our partnership with Sinotrans, China’s largest freight logistics company according to CIFA. We formed Cyantron as its controlling shareholder to build a mixed capacity freight service provider with our autonomous driving technology. As of December 31, 2025, Cyantron has commenced operations with a fleet of over 140 robotrucks, consisting of Level 2++ intelligent trucks and Level 4 autonomous trucks. In the short term, Cyantron offers hybrid logistics capacity, including its robotruck services, to Sinotrans for logistics fees. As the robotruck fleet size continues to grow, Cyantron is expected to serve a growing number of Sinotrans’ freight orders across China, and offers paid robotruck services to customers at a large scale in the long term. As a result, Sinotrans and other logistics platforms in China’s truck freight market will gain access to safe, reliable and environmental-friendly freight capacity at reduced labor and other costs. Through Cyantron, we will also use data analytics to improve loading and dispatching efficiency and reduce accident rates. In January 2025, we entered into a new agreement with Sinotrans to enhance Cyantron’s investments in R&D, while continuously expanding its logistics capacity. Cyantron aims to provide intelligent, efficient, safe, and green logistics services. In alignment with the transition to renewable energy, we are also assessing the opportunities to increase the proportion of new energy vehicles in our fleet to embrace evolving business demands. SANY partnership. As another firm step towards commercial applications of our robotruck services, we entered into a strategic partnership with SANY, a leading truck manufacturer in China, in May 2022, pursuant to which we will co-develop automotive-grade Level 4 trucks powered by our technology. As part of our strategic cooperation agreement, we are responsible for licensing our autonomous driving technology and providing technical support for the development of robotrucks, and SANY has agreed to, among other things, manufacture robotrucks at arm’s length prices, and help to market our robotrucks through its sales channels. Both parties may terminate the agreement by mutual consent or in the event of force majeure. 97 Table of Contents As Chinese local governments bolster efforts to promote and regulate the safety and commercial viability of robotruck services, we have made substantial progress in obtaining relevant regulatory permits for road testing and commercial operations. In December 2020, we were the first to obtain the robotruck road testing permit in Guangzhou, according to Frost & Sullivan. In July 2021, we expanded our road testing footprints to include Beijing and were allowed to test our robotrucks on national open highways. In January 2024, we received the very first cross-provincial robotruck road testing permit in China, according to Frost & Sullivan, and began testing on the highway freight network across the Beijing-Tianjin-Hebei region. In early 2024, we made significant progress in the commercialization of robotrucks, obtaining permissions to offer fare-charging robotruck services in cities such as Beijing and Tianjin. Through our partnership with Sinotrans, we have formed a robotruck fleet that provides routine transportation services to clients. In December 2024, we became the first company in China approved to test “driver out” in the follow truck in a “1+N” platoon, which allowed us to begin testing our robotruck convoy on the cross-provincial expressway connecting Beijing, Tianjin and Hebei. While the lead robotruck will continue to have a safety operator in the driver’s seat, the robotruck behind it will be driver out, i.e., no one behind the wheel. The platooning permit marks another major milestone for our robotruck services and could enable us to further reduce the costs of autonomous trucking. · Licensing and Applications Leveraging our extensive vehicle engineering and integration experience, we provide intelligent driving solutions for peripheral of vehicles to empower such vehicles to achieve higher levels of driving automation. We offer a complete suite of intelligent driving solutions to leading vehicle companies, spanning software licensing, hardware and data analytics tools. In addition, we also provide certain value-added technological services, such as vehicle integration services, and software development and licensing services, primarily to sensor and hardware component suppliers, helping them better adapt their products and solutions to autonomous driving use cases. Furthermore, we also offer vehicle-to-everything products and services to enhance road safety, and improve transportation efficiency and experience. Our licensing and applications business also recorded solid progress in 2025. Deliveries of autonomous domain controllers increased significantly year-over-year, reaching approximately six times the level of 2024. In 2025, we further expanded the application scenarios into a wider range of customers, including those in low-speed robot delivery, robot sweepers, logistics and humanoid robotics. We are also actively exploring opportunities in the robotics field. The continued expansion of customer demand and broadening application scenarios is expected to provide sustained support for the growth of the business. Ecosystem of Partners Around our core technology, we have built a thriving ecosystem of industry and technology partners, including OEMs, TNCs and logistics platforms, semiconductor chip suppliers, sensor suppliers, and other types of industry stakeholders. These strategic partnerships allow us to continue to hone our expertise in developing cutting-edge autonomous driving software and technology, but at the same time effectively leverage the manufacturing, product development, customer networks, and service expertise of our partners to scale and monetize our technology globally. 98 Table of Contents The following diagram illustrates our key partners and our cooperation with them. ​ The collaborative ecosystem around our technology and partners connects us with the vehicle and component suppliers, and the mass service demand from TNCs and logistics platforms. This connection has enabled us to scale our autonomous vehicle fleets integrated with our Virtual Driver, while simultaneously applying such autonomous vehicles across diversified commercial use cases in a cost-effective way. Robotaxi services: we have strategically built in-depth collaborations with trusted OEMs such as Toyota, BAIC, GAC and SAIC Motor to co-develop and mass-produce Level 4 autonomous vehicles. Through our collaboration with Toyota, we successfully launched our 6th generation Level 4 autonomous vehicle model in January 2022 to support fully driverless Level 4 robotaxi operations. In April 2025, we launched the 7th generation robotaxi, including three new Level 4 vehicle models co-developed through our strategic partnerships with Toyota, BAIC and GAC. The partnerships with these trusted OEMs have significantly increased our ability to scale our technology globally with reliable, integrated vehicle platforms. In addition, we have formed partnerships with established TNCs, such as Alipay, Amap and OnTime Mobility to scale and expand our robotaxi services and enhance passenger coverage. For example, with strategic investments in OnTime Mobility in April 2022, OnTime Mobility owned and operated a fleet of 50 vehicles integrated with our Virtual Driver on its mobile app to offer paid public-facing robotaxi services in Guangzhou and Shenzhen, China. In addition, we also integrate our services with OnTime Mobility’s platform and utilize their interface for the operation of own robotaxi fleet. Robotruck services: we have formed strategic partnerships with China’s leading truck manufacturer SANY to co-develop robotrucks powered by our technology, and with Sinotrans, China’s largest freight logistics company according to CIFA, to operate both Level 2+ trucks and Level 4 autonomous trucks throughout Sinotrans’ certain existing logistics network. Our Virtual Driver technology, combined with the manufacturing and aftersales capability of truck OEMs and the demand and infrastructure of logistics platforms, has positioned us to capitalize on opportunities in China’s large trucking market. Licensing and applications: we have co-development collaborations with logistics customers, such as Meituan, Neolix and Cainiao, on hardware components, and with NVIDIA. Such deep collaborations with established hardware component companies enable us to customize designs to deliver high performance and cost effectiveness, as well as to secure supply during uptime. 99 Table of Contents As we progressively broaden our overseas presence, we engage in collaborative ventures with a diverse array of business partners in selected global markets. These partnerships, spanning local governments, industry leaders, and technology innovators, are strategically aligned to advance the commercialization of our technology within these markets. For example, we formalized a memorandum of understanding with the Luxembourg government in March 2024 to propel the evolution of Level 4 autonomous mobility within Luxembourg, as our regional hub. Our collaborative efforts with the Luxembourg government and local partners aim to drive technological innovation and customize solutions tailored to the European market. We have collaborated with a local partner in South Korea to initiate road tests of robotaxi in urban areas in Seoul in 2025. Our discussions with various entities, including potential partners in Saudi Arabia, are advancing towards potential deployment in the region, and we have signed a MoU with the Abu Dhabi Investment Office to move forward with deployment in Abu Dhabi in October 2023. Under the contractual arrangements with our major strategic partners, we typically maintain ownership of the intellectual property rights that were developed by us. In cases where joint collaboration results in new intellectual property, the ownership of these rights is typically shared between ourselves and our strategic partners. Our agreements with these partners ensure that all parties benefit from the co-development and deployment of new technologies, while also providing clear guidelines for the protection and management of intellectual property. By working closely with our partners and taking a collaborative approach to innovation, we are able to leverage our collective strengths and drive the continued growth of our business. Customers and Suppliers At the current stage of commercialization, our customers consist primarily of (i) OEMs and TNCs with respect to our robotaxi services, (ii) OEMs and logistics platforms with respect to our robotruck services, and (iii) sensor and hardware component suppliers and other industry participants with respect to our licensing and applications business. To a lesser extent, our customers also include passengers who access our robotaxi services via our PonyPilot mobile app. In 2023, 2024 and 2025, we had 52, 111 and 213 corporate customers, respectively, in addition to individual customers who were passengers of our robotaxi services. These customers include domestic companies and multinational companies operating at various scales along the autonomous driving value chain, including vehicle manufacturing, logistics, and AV software and hardware design and manufacturing. We have historically generated revenues from a small group of customers during the early stage of commercialization. Our top three customers accounted for an aggregate of 65.8%, 43.9% and 59.4% of our revenues in 2023, 2024 and 2025, respectively. These were primarily customers of our (i) engineering solution services, (ii) transportation services provided by our robotruck fleet, and (iii) licensing and applications business. There was one customer who was one of our top three customers in each period of 2023, 2024 and 2025. There is no preexisting relationship between any member of our management team with these customers. As we continue to commercialize our autonomous driving technology through executing our go-to-market strategies, our customer base and profile are expected to constantly change, and we expect to further reduce our customer concentration. Our suppliers include primarily various component and service suppliers, such as semiconductor chip suppliers and sensor suppliers. We collaborate with these suppliers, which co-design with and/or supply to us certain key components used in our sensor suite and hardware, allowing us to focus our endeavors on research and development while improving our ability to mass produce and commercialize our technology. Research and Development We have invested a significant amount of time and effort into research and development of proprietary artificial intelligence, algorithms and software and hardware components to constantly enhance the capability of our Virtual Driver and solidify our technology leadership in the market. As the commercial deployment of our autonomous driving technology progresses, we are also devoted to adapting and optimizing our technology to different commercial use cases. As of December 31, 2025, we had 811 engineers, researchers and scientists whose expertise spans a broad range of disciplines such as vehicle engineering, industry design, AI and machine learning and data analytics. Our research and development teams are responsible for the design, development and testing of our autonomous driving technology. Our research and development presence across multiple locations has enabled us to develop, test and refine our autonomous driving technology based on diverse road, and weather, resulting in more reliable, resilient and scalable autonomous driving solutions. Our multi-center approach, combined with our leadership in the industry, also allows us to attract and retain top talents across the world, which contributes to our long-term business growth. 100 Table of Contents Intellectual Property We rely on proprietary technology and we are dependent on our ability to protect such technology. We rely on a combination of patent, copyright, trade secret and trademark laws as well as contractual restrictions such as confidentiality agreements, licenses and intellectual property assignment agreements to protect our intellectual property. We also maintain a policy requiring our employees, consultants and other third parties to enter into confidentiality and proprietary information agreements for the protection and confidentiality of our proprietary information. As of December 31, 2025, we had registered 386 patents, 213 copyrights, 642 trademarks in China, and 214 patents and 403 trademarks overseas. We have also registered 53 domain names globally. Despite our efforts to protect our proprietary rights, unauthorized parties may attempt to copy or otherwise obtain and use our technology. Monitoring unauthorized use of our intellectual property and proprietary rights is difficult and costly, and we cannot be certain that the steps we have taken will prevent misappropriation. From time to time, we may have to resort to litigation to enforce our intellectual property and proprietary rights, which could result in substantial costs and diversion of our resources. In addition, third parties may initiate lawsuits against us alleging infringement of their intellectual property or proprietary rights or declaring their non-infringement of our intellectual property or proprietary rights. In the event of a successful claim of infringement and our failure or inability to develop non-infringing technology or license the infringed or similar technology on a timely basis, our business could be harmed. Even if we are able to license the infringed or similar technology, license fees could be available only on commercially unreasonable and unfavorable terms, which may adversely affect our business, results of operations and financial condition. For additional information on the risks relating to intellectual property, see the section titled “Item 3. Key Information—3.D. Risk Factors—Risks Related to Our Business and Industry—We rely on patents, unpatented proprietary know-how, trade secrets and contractual restrictions to protect our intellectual property and other proprietary rights. Failure to adequately obtain, maintain, enforce and protect our intellectual property and other proprietary rights may undermine our competitive position and could materially and adversely affect our business, prospects, results of operations or financial condition,” and “Item 3. Key Information—3.D. Risk Factors—Risks Related to Our Business and Industry—We may be sued by third parties for alleged infringement, misappropriation or other violation of their proprietary technology or other intellectual property rights, which could be time-consuming and costly and result in significant legal liability or require us to cease using certain technology or other intellectual property rights, which could harm our business, financial condition, operating results, and reputation.” Data Security and Privacy To enable our autonomous driving solutions, we collect, store, transmit and otherwise process data from vehicles, users, employees, drivers and other third parties, some of which may involve personal data or confidential or proprietary information, such as a user’s name, phone number, place of departure and destination. We have implemented and maintained data protection policies, including our data classification policy and data life cycle specification, which have been designed to ensure that the collection, use, storage, transmission and dissemination of such data are in compliance with applicable laws across jurisdictions in which we operate, including China, the United States and other applicable jurisdictions, and with prevalent industry practice. In particular, data collected in different markets across the globe are stored and maintained locally and separately from each other, in compliance with applicable local laws and regulations. We endeavor to keep our users informed of how their personal information is handled by us throughout its life cycle. Users may access our privacy policy on our official site, which describes the type of personal information we collect, and how we use, share and protect users’ personal information, among other information. We have established an all-round information system designed in compliance with data security requirements and best practices and intend to continually invest heavily in data security and privacy protection. Our information system applies multiple layers of safeguards, including internal and external firewalls, enterprise-standard web application firewalls, and risk management platform. We adopt various technical means including encryption, desensitization, verification and backup to ensure the confidentiality, integrity and availability of the data we collect throughout its life cycle. We implement a robust internal authentication and authorization system designed to ensure confidential and critical data can only be accessed by authorized staff. We have also completed certain information security, privacy and compliance certifications/validations. For instance, our system is on file with the relevant public security authorities in China with a Level 3 information system security level. In addition, we have a designated data security team and an incident response team comprising members across various disciplines to provide daily cybersecurity and data security protection support, including a quick, effective and orderly response to servers and personal information related potential or actual incidents such as virus infections, hacker attempts and break-ins, improper disclosure of confidential information, system service interruptions, breach of personal information, and other events with serious information security implications. Our data security team reports to our board of directors. 101 Table of Contents As of the date of this annual report, we have not received any claim from any third party against us alleging any violation of such party’s data privacy rights, and we have not experienced any material data loss or breach incidents. See “Item 3. Key Information—3.D. Risk Factors—Risks Related to Our Business and Industry—Any unauthorized access, collection, control, manipulation, interruption, compromise or improper disclosure of personal information, cyber-attacks or other security incidents or data breaches that affect our networks or systems, or those of our service providers or our customers and/or passengers, whether inadvertent or purposeful, could degrade our ability to conduct our business, compromise the integrity of our products and services or our platform and data, result in significant data losses or the theft of our intellectual property, damage our reputation, expose us to liability to third parties and require us to incur significant additional costs to maintain the security of our networks and data, in any case of the foregoing, which could adversely affect our business, financial condition and results of operations.” Environmental, Social and Governance We are committed to promoting corporate social responsibility and sustainable development and integrating it into all major aspects of our business operations. Corporate social responsibility is viewed as part of our core growth philosophy that will be pivotal to our ability to create sustainable value for our shareholders by embracing diversity and public interests. Our board of directors will assume the critical role of evaluating and managing corporate social responsibility strategies and policies, including Ms. Asmau Ahmed, who is expected to bring her extensive expertise and experience in technological innovations to develop and implement our social responsibility initiatives as a leading technology company. We fully recognize our important role in society’s sustainable development and are committed to integrating environmental protection, social responsibility, and governance principles into business operations. Accordingly, we have implemented comprehensive internal Environmental, Social and Governance (“ESG”) policies that outline environmental protection and social responsibility objectives while providing practical guidance for daily operations. We have established a top-down ESG management approach. The Board of Directors serves as the highest decision-making body for the Company’s ESG matters, responsible for evaluating and determining the ESG- related risks, ensuring that the Company establishes an appropriate and effective ESG risk management and internal monitoring system, identifying and evaluating the ESG strategies and objectives, and regularly supervising and reviewing the ESG performance and progress in accomplishing the ESG targets. At the management level, the Safety, Compliance and Sustainability Committee is responsible for organizing and carrying out the Company’s ESG management work in accordance with the Company’s overall ESG targets and strategies, formulating relevant systems and processes, and maintaining the ESG indicator framework. At the implementation level, each department and subsidiary of the Company is responsible for completing the tasks associated with ESG implementation, including processes establishments and improvements, and ESG practice activities organization. We are committed to fostering sustainable practices, promoting social responsibility, and maintaining strong governance standards, reflecting our dedication to ESG principles. We have established a set of ESG policies framework, which outlined, among others, (i) the roles and performance requirements of ESG management structure; (ii) ESG strategy formation procedures; (iii) ESG risk management and monitoring; and (iv) ESG reporting whole-process management. Under the supervision of the Board, we are adopting various strategies and measures to identify, assess, manage and mitigate ESG and climate-related risks, including but not limited to: (i) continuously tracking ESG regulatory related issues and updating our internal ESG policy to ensure the policy compliance; (ii) regularly hosting management discussion and meeting to ensure that all material ESG risks are identified and reported; (iii) establishing communication channels and ongoing discussions with key stakeholders to identify material ESG-related issues and risks associated with our business operations; (iv) engaging professional advisers to advise on ESG compliance matters. We are aware of the importance of environmental protection, strictly abide by the Environmental Protection Law of the People’s Republic of China and other relevant laws and regulations, carry out environmental protection publicity and education activities, and improve the environmental awareness of all employees. We are dedicated to providing green energy to society, fostering the sustainable growth of our enterprise, and realizing a mutually beneficial outcome between economic benefits and environmental stewardship. 102 Table of Contents We are committed to contributing to our communities and understand the needs of the communities in which we operate. We have taken the following steps to support our community investment and charitable activities: · Science Popularization and Education: In 2025, we organized a total of 54 science popularization and public welfare activities, reaching over 120,000 participants. Key initiatives included hosting autonomous driving exhibitions at the 2025 WAIC, organizing tech exchanges for international students and university delegations (such as Tsinghua University and the University of Hong Kong), and co-building the Greater Bay Area Autonomous Driving Science Popularization Center. · Educational Support: We raised funds through annual charity auctions and online flea markets to sponsor 11 students from remote areas for a full year of study through the “UU Charity Association.” Overseas, our Fremont office partnered with the local NGO Abode to donate 100 sets of school supplies and backpacks to support homeless school-age children in returning to school. · Community Care and Volunteer Services: Our volunteer teams visited the Beijing Tianyun Rehabilitation Center to donate essential living supplies, clothing, and toys to disabled children, providing warmth and companionship to vulnerable groups. · Rural Revitalization: Through the “Rural Chinese Dream” public welfare project, we hosted 50 youths, teachers, and grassroots workers from rural areas in Hunan and Gansu provinces. We provided them with hands-on AI and autonomous driving learning experiences to broaden their horizons and empower rural development with technology. · Green Public Welfare: We actively participated in the “Zixia Jinfeng” tree-planting activity in Guangzhou to support rural ecological construction. Furthermore, we donated 5,400 recycled plastic bottles to sustainable brands for environmental regeneration, promoting resource recycling and a circular economy. In April 2026, we released our 2025 ESG report, which detailed our ESG performance in 2025 in key areas, driven by our mission: providing safe, advanced, and reliable full-stack autonomous driving technology to revolutionize the future of transportation. Guided by “Safety First” and a rigorous safety culture, we have embedded safety across technology development, operations, and management while advancing autonomous mobility and logistics through industry collaboration. We also monitor our long-term impact on road safety, transportation efficiency, energy structure, industrial development, and social inclusion, ensuring technology serves both people and the planet. Beyond innovation, we invest in employee well-being, community engagement, and public education to build trust and understanding. We will continue to strengthen our institutional frameworks and operational practices, advancing safety and responsibility in parallel, and enabling autonomous driving to contribute more meaningfully to social development. The information contained in our 2025 ESG report is not incorporated by reference into and should not be considered a part of this annual report. Competition We face competition from primarily technology-focused companies building end-to-end technical capabilities for autonomous driving applications, and autonomous players building internal autonomous development programs. The principal competitive success factors in our market include but not limited to: · Technology quality, safety and reliability; · Vehicle engineering and integration capabilities; · Business model and go-to-market approach; · Strategic partnerships; · Cost efficiency; and · Patents and intellectual property portfolio. 103 Table of Contents Because of the depth and breadth of our talents, full-stack autonomous driving technology, differentiated go-to-market approach, and extensive strategic partnerships that drive commercialization at scale, we believe that we are able to compete favorably across these factors. See the section titled “Item 3. Key Information—3.D. Risk Factors” for a more comprehensive description of risks related to competition. Regulation Regulations on Foreign Investment The Company Law of the PRC, promulgated by the Standing Committee of the National People’s Congress of the PRC (the “SCNPC”) on December 29, 1993, last amended on December 29, 2023 and came into effect on July 1, 2024, governs the establishment, operation and management of companies in the PRC, including foreign-invested companies. Unless foreign investment laws provide otherwise, foreign- invested companies shall abide by the Company Law of the PRC. On January 1, 2020, the Foreign Investment Law of the PRC (the “FIL”) became effective and simultaneously replaced the prior laws regulating foreign investment in China, namely, the Sino-foreign Equity Joint Venture Enterprise Law of the PRC, the Sino-foreign Cooperative Joint Venture Enterprise Law of the PRC and the Wholly Foreign-invested Enterprise Law of the PRC, together with their implementation rules and ancillary regulations. Pursuant to the FIL, “foreign investments” refer to investment activities conducted by foreign investors directly or indirectly in the PRC, which include any of the following circumstances: (i) foreign investors setting up foreign-invested enterprises in the PRC solely or jointly with other investors, (ii) foreign investors obtaining shares, equity interests, property portions or other similar rights and interests of enterprises within the PRC, (iii) foreign investors investing in new projects in the PRC solely or jointly with other investors, and (iv) investment of other methods as specified in laws, administrative regulations, or as stipulated by the State Council. Pursuant to the FIL, China has adopted a system of pre-establishment national treatment plus a negative list with respect to foreign investment administration. The negative list shall be issued by, amended or released upon approval by the State Council, from time to time. The negative list sets forth industries in which foreign investments are prohibited and industries in which foreign investments are restricted. Foreign investment in prohibited industries is not allowed, while foreign investment in restricted industries must satisfy certain conditions stipulated in the negative list. Foreign investments and domestic investments in industries outside of the negative list will be treated equally. Additionally, the PRC authorities also maintain a catalogue which identifies industries in China where foreign investments are proactively encouraged. This catalogue often serves as a reference for local governments in China to formulate and implement their foreign investment support policies. The Special Administrative Measures (Negative List) for the Access of Foreign Investment (2024 Version), which was promulgated by the National Development and Reform Commission of the PRC (the “NDRC”) and the Ministry of Commerce of the PRC (the “MOFCOM”) on September 6, 2024 and became effective on November 1, 2024, and the Catalogue of Encouraged Industries for Foreign Investment (2022 Version), which was promulgated by the NDRC and the MOFCOM on October 26, 2022 and became effective on January 1, 2023, replace previous negative list and encouraging catalogue and list the categories of encouraged, restricted, and prohibited industries. On December 30, 2019, the MOFCOM and the SAMR jointly promulgated the Measures for Reporting of Information on Foreign Investment, which became effective on January 1, 2020 and pursuant to which, foreign investors or foreign-invested enterprises shall report investment information to the MOFCOM and its local counterparts when foreign investors carry out investment activities directly or indirectly within China, and its subsequent changes are required to submit an initial or change report through the enterprise registration system. Pursuant to the Measures for the Security Review of Foreign Investment promulgated by the NDRC and the MOFCOM on December 19, 2020 and became effective on January 18, 2021, any foreign investment that has or possibly has an impact on state security shall be subject to security review in accordance with the provisions hereof. 104 Table of Contents Our business is permitted under the applicable PRC foreign investment regulations and no part of our business falls under either the prohibited or restricted categories under the PRC Special Administrative Measures (Negative List) for the Access of Foreign Investment (2024 Version). Additionally, the development and application of autonomous driving system is included in the PRC Catalogue of Encouraged Industries for Foreign Investment (2022 Version) as an industry in China where foreign investments are proactively encouraged. However, we have not received any specific benefits or preferential treatment as a result of this inclusion. With respect to the former VIEs that we used to have, such former VIEs, before the termination of their VIE arrangements, did not engage in any business or hold any license that is subject to PRC foreign investment restrictions or prohibitions. We initially established these former VIEs solely for the purposes of preserving the flexibility to potentially engage in future business that may be subject to such PRC restrictions or prohibitions (although we have never engaged in such business to date). The former VIE arrangements were terminated by our WFOEs’ acquisition of the equity interests in the former VIEs. There has not been any change to the nature or extent of the former VIEs’ business activities after the termination of the VIE arrangements and this termination did not involve the transfer of any operation, technology or license of the former VIEs to any third party. Regulations on Autonomous Driving On July 27, 2021, the Ministry of Industry and Information Technology (the “MIIT”), the Ministry of Public Security (the “MPS”) and the Ministry of Transport (the “MOT”) promulgated the Administrative Norms for Road Testing and Demonstrative Application of Intelligent Connected Vehicles (for Trial Implementation) (the “Road Testing Administrative Norms”), which came into effect on July 27, 2021 and replaced the Norms on Administration of Road Testing of Autonomous Driving Vehicles (Trial Implementation) issued in April 2018. The Road Testing Administrative Norms is the main national level regulation on road testing of autonomous driving vehicles in the PRC, under which, road testing refers to the testing of autonomous driving function of intelligent connected vehicles (a PRC regulatory concept that encompasses autonomous driving vehicles) carried out on the designated sections of highways, urban roads and other roads used for the passage of public motor vehicles, and “demonstrative application” of such vehicles refers to pilot and experimental activities of driving such vehicles with passengers and goods, which are carried out on designated sections of certain roads that are used for passage of public motor vehicles. Pursuant to the Road Testing Administrative Norms, any entity intending to conduct a road testing of autonomous driving vehicles must obtain a road-testing certificate and a temporary license plate for each tested vehicle. To qualify for these required licenses, an autonomous driving applicant entity must satisfy, among others, the following requirements: (i) it must be an independent legal person registered in the PRC with the capacity to conduct businesses in relation to intelligent connected vehicles, such as manufacturing, R&D and testing of vehicles and vehicle parts, which has established protocol to test and assess the performance of autonomous driving system and is capable of conducting real-time remote monitoring of the tested vehicles, and with the ability of event recording, analysis and reproduction of the vehicles under road testing and ensuring the network security of the vehicles and the remote monitoring platforms; (ii) the vehicles must be equipped with a driving system that can switch between autonomous pilot mode and human driving mode in a safe, quick and simple manner and allows human driver to take control of the vehicle instantaneously when necessary; (iii) the vehicles must be equipped with the functions of recording, storing and real-time monitoring the condition of the vehicle and be able to transmit real-time data of the vehicle; (iv) the applicant entity must sign an employment or labor service contract with the driver of the tested vehicle, who must be a licensed driver with more than three years’ driving experience and a track record of safe driving and is familiar with the testing protocol for autonomous driving system and proficient in operating the system; (v) the applicant entity must insure each tested vehicle for at least RMB5 million against car accidents or provide a letter of guarantee covering the same. The testing duration for a road testing should not exceed 18 months in principle, and should not exceed the validity period of the certificate of safety technical inspection and the insurance voucher of the tested vehicle. Pursuant to the Road Testing Administrative Norms, a road-testing entity and a demonstrative application entity must submit a periodic report every 6 months to the competent governmental authority and provide a summary report within 1 month upon conclusion of the road testing or demonstrative application. The entity responsible for the road testing or the demonstrative application must report information on the traffic accidents during the road testing or demonstrative application to the competent authorities on a monthly basis. In case of any traffic violation, the traffic administrative department of the public security department must impose the penalties on the responsible parties in accordance with the laws and regulations on road traffic safety. In the case of serious injuries or deaths of any person or serious damage of a vehicle, the entity responsible for the road testing or the demonstrative application must report such accident to the competent governmental authority within 24 hours, and if such entity fails to report as required, its road testing or demonstrative application activities may be suspended for 24 months. 105 Table of Contents On July 30, 2021, the MIIT issued the Opinion on Strengthening the Access Administration of Intelligent Connected Vehicles Manufacturing Enterprises and Their Products, which strengthens the safety management of products with autopilot function and provides that such products shall at least meet the requirements as follows: (i) being able to automatically identify the failure of the autopilot system and whether the designed operating conditions are continuously satisfied, and to take measures to minimize risks; (ii) having the function of human-computer interaction to display the operating status of the autopilot system; (iii) having the event data recording system and the autopilot data recording system; (iv) satisfying the process assurance requirements, such as functional safety and network security, as well as testing requirements in relation to simulation, roads, network security, software upgrading and data recording. On August 25, 2022, the Ministry of Natural Resources issued the Notice on Promoting the Development of Intelligent Connected Vehicles and Maintaining the Security of Surveying, Mapping and Geo-information, which among others, provides that for any vehicle manufacturer, service provider or autonomous driving software provider that engages in the collection, storage, transmission and processing of certain geo-information that is surveying and mapping data in nature, if it is a domestic enterprise, it shall obtain the surveying and mapping qualification in accordance with the law or engage an agency with such qualification to carry out the surveying and mapping activities; if it is a foreign-invested enterprise, it shall engage an agency with such qualification to carry out the surveying and mapping activities. Pursuant to the Surveying and Mapping Law, which was promulgated by the Standing Committee of the National People’s Congress (the “SCNPC”) on December 28, 1992, and last amended on April 27, 2017 and became effective on July 1, 2017, conducting surveying and mapping activities without obtaining the necessary qualification may be ordered to cease such activities, and the unlawful gains from the surveying and mapping activities shall be confiscated. In addition, a fine of not less than one time but not more than two times of the unlawful gains from the activities may be imposed on. None of our company, our subsidiaries or the former VIEs (before or after the termination of the former VIE arrangements) has engaged in any mapping or surveying activities, or holding the requisite license (which is currently subject to foreign investment prohibition and available exclusively to PRC domestic companies). Rather, our subsidiaries and the former VIEs (which subsequently became our wholly-owned subsidiaries) have been procuring standard and ancillary mapping and surveying services that support their autonomous driving algorithms and functions from independent third-party mapping data and surveying suppliers holding the relevant mapping and surveying license in China. This decision was made not due to a lack of technological capability, but rather both from an economic perspective and for PRC regulatory compliance reasons. In China, such services are generally readily attainable on reasonable terms from a number of renowned domestic technology firms specialized in digital maps and related data services (including, but not limited to, those engaged by us). Those firms provide standard mapping and surveying services not only to autonomous driving companies like us but also to automakers and other ride-hailing and logistics companies. The following table sets forth certain details of our collaboration with the three independent qualified mapping and surveying service providers in China that we transact with in 2025. These providers deliver a range of services pursuant to arm’s-length terms typically ranging from one to three years, some of which include renewal options. We do not substantially rely on any of these service providers, and we consider alternative service providers to be readily available in the PRC market, should we decide to replace any of these service providers. Name of Service Provider Background Information Principal Services Procured by Pony Fee Terms Amap ​ A leading domestic digital map, navigation, and location service provide ​ Real-time maps and transit information on the PonyPilot mobile app and web page ​ One-time fixed service fees for the entire contract term NavInfo ​ A top-tier domestic supplier of integrated solutions for smart mobility applications ​ Provision of high-definition digital maps ​ One-time fixed service fees for the entire contract term Xianli ​ A leading domestic supplier of data annotation services for autonomous driving and other commercial applications ​ The collection and handling of mapping and surveying related data ​ Service fees determined based on factors including the data quantities and complexity of deliverables and outputs ​ 106 Table of Contents On August 20, 2021, the MIIT promulgated the Taxonomy of Driving Automation for Vehicles, which became effective on March 1, 2022. It provides for six levels from Level 0 to 5 for the taxonomy of driving automation, among which Level 2 refers to combined driving assistance, Level 3 refers to conditionally- automated driving, Level 4 refers to highly-automated driving and Level 5 refers to fully-automated driving. On November 17, 2023, the MIIT, the MPS, the Ministry of Housing and Urban-Rural Development and the MOT promulgated the Notice on Carrying out the Pilot Program of Market Access and Road Passage for Intelligent Connected Vehicles, which provides that the aforementioned authorities will select intelligent connected vehicle products with Level 3 and Level 4 qualified for mass production as the pilot program of market access, and launch the pilot program of road passage for intelligent connected vehicle products that have been granted access within specified areas. On November 21, 2023, the MOT issued the Guideline on Transport Safety and Service for Autonomous Vehicles (Trial Implementation), which specifically provides the requirements for commercial operation of autonomous vehicles in respect of the scope of application, basic principles, application scenarios, operators of autopilot transport, transport vehicles, staffing, safety assurance, supervision and administration. On July 26, 2024, the Ministry of Natural Resources promulgated The Notice of the Ministry of Natural Resources on Strengthening the Administration of Surveying, Mapping and Geoinformation Security Relating to Intelligent Connected Vehicles, and emphasized various related matters, including the requirement of conducting surveying and mapping activities related to intelligent connected vehicles in accordance with the law, strengthening the management of surveying and mapping activities involving intelligent connected vehicles, strictly managing confidential and sensitive geographic information data, strictly reviewing electronic navigation maps, implementing the requirements for the storage of geoinformation data and cross-border transfer of such data, strengthening the regulation of geoinformation security, encouraging the exploration of geographic information security application, etc. A number of local governments in China, such as Beijing, Guangzhou, Shanghai and Shenzhen, have also released rules that regulate road testing and application of autonomous driving vehicles. For example, Beijing Municipal Commission of Transport, Beijing Municipal Bureau of Public Security and Beijing Municipal Bureau of Economy and Information Technology promulgated the Implementing Rules for Road Testing Management of Autonomous Vehicles (for Trial Implementation), effective on November 12, 2020, which stipulates the detailed procedures and requirements for road testing and trial operations in Beijing, including those for general technical test, special weather test, highway test, driverless test, etc. On July 8, 2021, Guangzhou Municipal Industry and Information Technology Bureau promulgated the Opinions on Gradually Launching Regional Piloting Policies for the Application, Demonstration and Operation of Intelligent Connected Vehicles (Automatic Driving) under Different Mixed Environments, and the Work Plan for the Application, Demonstration and Operation of Intelligent Connected Vehicles (Automatic Driving) under Different Mixed Environments, which among others, provide that intelligent connected vehicles (autonomous driving) may be used to carry out passenger transport activities, such as taxis and buses, and carry out ordinary road freight transport (except for dangerous goods) and other demonstrative operations, provided that the relevant license, permit and other regulatory requirements are met. On November 11, 2024, Guangzhou Municipal Industry and Information Technology Bureau promulgated the Work Plan for the Application, Demonstration and Operation of Intelligent Connected Vehicles (Automatic Driving) under Different Mixed Environments (Second Version). Regulations on Road Transport Pursuant to the Regulations on Road Transport, which was promulgated by the State Council on April 30, 2004, last amended on January 30, 2026, and became effective on March 20, 2026, operators engaging in the road passenger transport business operations, the road freight transport business operations and road transport related business shall abide by this regulation. An operator may engage in freight transport business only after obtaining a road transport business operation license, except for those operators that use any general freight transport vehicle with a total mass of 4.5 tons or below to engage in the general freight transport business operations. In addition, any vehicle used by freight transport business operators for transportation shall obtain a vehicle operation certificate, except for those vehicles with a total mass of 4.5 tons or below. The road transport business operation license as well as the vehicle operation certificate are also required for operating the road passenger transport business. Pursuant to the Administrative Provisions on Road Freight Transport and Stations promulgated by the MOT on June 16, 2005, last amended on November 10, 2023 and became effective on the same date, an operator of road freight transport shall engage in business operations of road freight transport within the business scope as specified in the operation license for road transport and shall hire drivers with practice qualification certificates as required by the relevant provisions. If an operator intends to establish a branch engaging in road freight transport business, it shall file for record with the competent road transport department of the place where the branch is to be established. 107 Table of Contents On August 11, 2021, the MOT promulgated the Provisions on the Administration of Cruising Taxi Operating Services (2021 Amended Version), which provides that cruising taxi operating services refer to the business activities of cruising on the street for attracting customers or waiting for passengers at taxi ranks, spraying and installing the taxi logos, providing travelling services for passengers through the passenger cars with seven seats or less and driving services, driving according to the wishes of passengers, and charging fares by mileage and time. Operators shall apply to the local government for providing cruising taxi services, and the local government shall issue a written decision on approving administrative licensing for cruising taxi operation, specify the business scope, the operating areas, the number of vehicles and the requirements therefor, the valid period of the right to operate cruising taxis, and other matters, and issue the road transport business license to the applicant, if the applicant is satisfied with the requirements. After verifying that the vehicles comply with the relevant requirements, the licensing authority shall issue the road transport certificates to the vehicles. Regulations on Road Traffic Safety The Road Traffic Safety Law, which promulgated by the SCNPC on October 28, 2003, and last amended and became effective on April 29, 2021, sets out the basic framework for road traffic safety and provides the rules for the drivers of vehicles, pedestrians, passengers and the entities and individuals involved in road traffic activities. Pursuant to the Road Traffic Safety Law, the relevant traffic control department of the public security authorities shall be in charge of determination of responsibilities in traffic accidents, which is also reiterated and brought into details by the last amended Regulation for the Implementation of the Road Traffic Safety Law of the PRC promulgated by the State Council on October 7, 2017. On March 24, 2021, the MPS issued the Draft Proposed Amendments of the Road Traffic Safety Law (the “MPS Proposed Amendments”). The MPS Proposed Amendments clarify, among others, the requirements related to road testing of, and access by, vehicles equipped with autonomous driving functions, as well as regulating how liability for traffic violations and accidents will be allocated. The MPS Proposed Amendments stipulate that vehicles equipped with autonomous driving functions should first pass tests in closed roads and venues and obtain temporary license plates before embarking on road testing. The MPS Proposed Amendments provide that when vehicles equipped with autonomous driving functions and human driving modes are involved in road traffic violations or accidents, the responsibility of the driver or the autonomous driving system developer shall be determined in accordance with laws, as well as the liability for damage. For vehicles on the road that are equipped with autonomous driving functions without human driving modes, this liability issue should be separately dealt with by relevant departments of the State Council. However, the last amended Road Traffic Safety Law did not adopt the aforementioned proposed amendments. Regulations on Cybersecurity, Information Security, Privacy and Data Protection Cybersecurity Pursuant to the National Security Law of the PRC promulgated by the SCNPC on February 22, 1993 and latest amended and became effective on July 1, 2015, the state shall establish systems and mechanisms for national security review and supervision, conduct national security review on key technology, network information technology products and services related to state security, so as to prevent and neutralize state security risks in an effective way. On November 7, 2016, the SCNPC promulgated the Cybersecurity Law of the PRC (the “Cybersecurity Law”), which became effective on June 1, 2017 and was last amended on October 28, 2025 with such amendments effective on January 1, 2026. The Cybersecurity Law requires network operators to perform certain functions related to cyber security protection and strengthen the network information management. For instance, under the Cybersecurity Law, network operators of critical information infrastructure generally shall, during their operations in the PRC, store the personal information and important data collected and produced within the territory of the PRC, fulfill additional obligations of security protection, and are subject to cybersecurity review when purchasing of network products and services that may threaten the national security. When collecting and using personal information, in accordance with the Cybersecurity Law, network operators shall abide by the “lawful, justifiable and necessary” principles. Network operators shall collect and use personal information by announcing rules for collection and use, expressly notify the purpose, methods and scope of such collection and use, and obtain the consent of the person whose personal information is to be collected. Network operators shall not disclose, tamper with or destroy personal information that it has collected, or disclose such information to others without prior consent of the person whose personal information has been collected, unless such information has been processed to prevent specific person from being identified and such information from being restored. 108 Table of Contents On July 30, 2021, the State Council promulgated the Regulations on Security Protection of Critical Information Infrastructure (the “CII Regulations”), effective on September 1, 2021. Pursuant to the CII Regulations, a “critical information infrastructure” has the meaning of an important network facility and information system in important industries such as, among others, public communications and information services, energy, transport, water conservation, finance, public services, e-government affairs and national defense science, as well as other important network facilities and information systems that may seriously endanger national security, national economy, people’s livelihood, or public interests in the event of damage, loss of function, or data leakage. The competent regulatory authorities as well as the supervision and administrative authorities of the aforementioned important industries and sectors will be responsible for (i) organizing the identification of critical information infrastructures in their respective industries in accordance with certain identification rules, and (ii) promptly notifying the identified operators and the public security department of the State Council of the identification results. On December 28, 2021, thirteen regulatory authorities, including the Cyberspace Administration of China (the “CAC”), the China Securities Regulatory Commission (the “CSRC”), jointly released the Cybersecurity Review Measures (the “Cybersecurity Review Measures”) which became effective on February 15, 2022. Pursuant to the Cybersecurity Review Measures, network platform operators holding personal information of over one million users shall apply for cybersecurity review before listing abroad. The cybersecurity review will evaluate, among others, the risk of critical information infrastructure, core data, important data, or the risk of a large amount of personal information being influenced, controlled or maliciously used by foreign governments after going public, and cyber information security risk. On July 22, 2020, the MPS issued the Guiding Opinions on Implementing the Multi-Level Protection Scheme for Cybersecurity and the Security Protection System for Critical Information Infrastructure (the “Guiding Opinions on MLPS and CII”). The Guiding Opinions on MLPS and CII restates the basic principles and work objectives of implementing the requirements on multi-level protection scheme and security protection of critical information infrastructure, and requires network operators to undertake the assessment and filing of their own network systems in time under the multi-level protection scheme. In addition, according to the Guiding Opinions on MLPS and CII, the industrial regulatory authorities shall develop the rules for the identification of critical information infrastructure in such industries, promptly notify the relevant operators of the identification results and report the same to the MPS for record. Data Protection On June 10, 2021, the SCNPC promulgated the Data Security Law of the PRC (the “Data Security Law”), which became effective on September 1, 2021. The Data Security Law provides for data security obligations on entities and individuals carrying out data activities. The Data Security Law also introduces a data classification and hierarchical protection system based on the importance of data in economic and social development, as well as the degree of harm it will cause to national security, public interests, or legitimate rights and interests of individuals or organizations when such data is tampered with, destroyed, leaked, or illegally acquired or used. The appropriate level of protection measures is required to be taken for each respective category of data. For example, a processor of important data shall designate the personnel and the management body responsible for data security, carry out risk assessments for its data processing activities and file the risk assessment reports with the competent authorities. In addition, the Data Security Law provides a national security review procedure for those data activities which may affect national security and imposes export restrictions on certain data and information. On August 16, 2021, five regulatory authorities, including the CAC, promulgated the Several Provisions on the Administration of Automotive Data Security Management (for Trial Implementation) (the “Provisions on Automotive Data Security”), which became effective on October 1, 2021. The Provisions on Automotive Data Security clearly defines the definition of “automotive data”, “processing of automotive data”, “automotive data processor”, “personal information”, “sensitive personal information” and “important data”, and further elaborate the principles of and requirements for the automotive data operating activities within the PRC. Furthermore, the Provisions on Automotive Data Security also prescribes the implementation of classified protection of cybersecurity, the obligations of automotive data operators to inform, anonymize and obtain individuals’ consents, and the specific requirements for processing sensitive personal information, as well as the risk assessment when operating important data and the security assessment when providing data abroad. 109 Table of Contents On December 8, 2022, the MIIT issued the Measures for Data Security Administration in the Industry and Information Technology Field (for Trial Implementation) (the “MIIT Measures for Data Security”), which became effective on January 1, 2023. In accordance with the MIIT Measures for Data Security, data processors in the field of industry and information technology shall classify data firstly based on the data’s category and then match the corresponding organizational and technical measures. It also imposes certain obligations on them in relation to, among others, implementation of data full-life security protection system which includes data collection, data storage, data usage, data transmission, data disclosure, safety audit and emergency plans. On July 7, 2022, the CAC promulgated the Measures for the Security Assessment of Data Cross-border Transfer (the “Measures for Data Cross-border Transfer”), which became effective on September 1, 2022. The Measures for Data Cross-border Transfer provides four circumstances, under any of which data processors shall, through the local cyberspace administration at the provincial level, apply to the national cyberspace administration for security assessment of data cross-border transfer. These circumstances include: (i) where the data to be transferred to an overseas recipient contains important data; (ii) where a personal information processor that has processed personal information of more than one million individuals or a critical information infrastructure operator provides personal information overseas; (iii) where a data processor has provided personal information of 100,000 people or sensitive personal information of 10,000 people in total abroad since January 1 of the previous year; or (iv) other circumstances prescribed by the CAC for which declaration for security assessment for cross-border data transfers is required. On February 22, 2023, the CAC promulgated the Measures for the Standard Contract for Cross-border Transfer of Personal Information (the “Measures for Standard Contract”), which became effective on June 1, 2023. The Measures for Standard Contract requires that any personal information processor transferring personal information abroad by entering into the standard contract shall meet all of the following conditions: (i) it is not a critical information infrastructure operator; (ii) it processes the personal information of less than 1 million individuals; (iii) it has cumulatively transferred abroad the personal information of less than 100,000 individuals since January 1 of the previous year; and (iv) it has cumulatively transferred abroad the sensitive personal information of less than 10,000 individuals since January 1 of the previous year. Where there are other relevant provisions in any laws, administrative regulations or rules of the CAC, such provisions shall apply. It also emphasizes that any personal information processor shall not use methods such as quantity splitting of the personal information that is required by law to undergo the security assessment for data cross-border transfer under the Measures for Data Cross-border Transfer. The standard contract shall be concluded in strict accordance with the annex of the Measures for Standard Contract, and the personal information processors shall, within 10 working days after the standard contract enters into effect, apply for filing with the local cyberspace administration at the provincial level. On March 22, 2024, the Provisions on Promoting and Regulating Cross-border Data Flows (the “New Provisions on Cross-border Data Flows”), which’s promulgated by the CAC, became effective. Under the New Provisions on Cross-border Data Flows, to provide the data collected and generated in such activities as international trade, cross-border transport, academic cooperation, transnational manufacturing and marketing, which do not contain personal information or important data, to overseas parties, it is exempted from declaring security assessment for data to be provided abroad, concluding a standard contract for personal information to be provided abroad or passing authentication for protection of personal information. It also emphasizes that, where a data processor other than a critical information infrastructure operator provides abroad the personal information (excluding sensitive personal information) of not more than 100,000 persons accumulatively as of January 1 of the current year, it may be exempted from declaring security assessment for data to be provided abroad, concluding a standard contract for personal information to be provided abroad or passing authentication for protection of personal information. On September 24, 2024, the Regulation on Network Data Security Management (the “Network Data Regulation”) was promulgated by the State Council, and became effective on January 1, 2025. The Network Data Regulation restates and further specifies the legal requirements for personal information, important data, cross-border data transfer, network platform services, and data security. Among others, if the network data processing activities have or may have impacts on national security, such activities shall be subject to national security review in accordance with relevant laws and regulations. Personal Information Protection On May 28, 2020, the SCNPC adopted the Civil Code of the PRC (the “Civil Code”), effective on January 1, 2021. Pursuant to the Civil Code, individuals have the right of privacy. No organization or individual shall process any individual’s private information or infringe an individual’s right of privacy, unless otherwise prescribed by law or with the consent of such individual or such individual’s guardian. In addition, any processing of personal information shall be subject to the principles of legitimacy, legality and necessity. An information processor shall not divulge or falsify the personal information collected and stored by it, or illegally provide the personal information of an individual to others without the consent of such individual, except for information that has been processed so that specific person cannot be identified and that cannot be restored. 110 Table of Contents On August 20, 2021, the SCNPC promulgated the Personal Information Protection Law (the “Personal Information Protection Law”), which integrates the scattered rules with respect to personal information rights and privacy protection and became effective on November 1, 2021. The Personal Information Protection Law applies to personal information processing activities within China, as well as certain personal information processing activities outside China, including those for provision of products and services to natural persons within China or for analyzing and assessing acts of natural persons within China. The Personal Information Protection Law provides the circumstances under which a personal information processor could process personal information, which include but not limited to, where the consent of the individual concerned is obtained and where it is necessary for the conclusion or performance of a contract to which the individual is a contractual party. It also stipulates certain specific rules with respect to the obligations of a personal information processor, such as to inform the purpose, the method of processing, the type of personal information processed and retention period to the individuals, and the obligation of the third party who has access to the personal information by way of co-processing or delegation etc. Processors processing personal information exceeding the threshold to be set by the CAC and operators of critical information infrastructure are required to store, within the territory of the PRC, the personal information collected and produced within the territory of the PRC. Furthermore, the Personal Information Protection Law also provides for the rights of natural persons whose personal information is processed, and takes special care of the personal information of children under 14 and other sensitive personal information. On July 16, 2013, the MIIT promulgated the Regulations on Protection of Personal Information of Telecommunication and Internet Users, which took effect on September 1, 2013, to regulate the collection and use of users’ personal information in the provision of telecommunication services and Internet information services in China. Telecommunication business operators and Internet service providers are required to constitute their own rules for the collection and use of users’ personal information and they cannot collect or use their information without users’ consent. Telecommunication business operators and Internet service providers must specify the purposes, manners and scopes of personal information collection and usage, and keep the collected personal information confidential. Telecommunication business operators and Internet service providers are prohibited from disclosing, tampering with, damaging, selling or illegally providing others with, collected personal information. Telecommunication business operators and Internet service providers are required to take technical and other measures to prevent the collected personal information from any unauthorized disclosure, damage or loss. On January 23, 2019, the CAC, the MIIT, the MPS and the SAMR jointly issued the Notice on Special Governance of Illegal Collection and Use of Personal Information via Apps, which restate the requirement of legal collection and use of personal information, and announces that the above departments jointly organize the special governance against the illegal collection and use of personal information in China from January to December 2019. In March 2019, the Personal Information Protection Tasks Force on Apps issued the Guide to the Self-Assessment of Illegal Collection and Use of Personal Information by Apps, which’s recommended to be used by App operators to carry out self-check concerning their collection and use of personal information, in the aspects of texts of privacy policies, activities of collection and use of personal information by Apps, and protection of users’ rights by App operators. On November 28, 2019, the CAC, the MIIT, the MPS and the SAMR jointly promulgated the Notice on the Measures for Determining the Illegal Collection and Use of Personal Information through Mobile Applications, which aims to provide reference for supervision and administration departments and provide guidance for mobile applications operators’ self-examination and self-correction and social supervision by netizens, and further elaborates the forms of behavior constituting illegal collection and use of the personal information through mobile applications including: (i) failing to publish the rules on the collection and use of personal information; (ii) failing to explicitly explain the purposes, methods and scope of the collection and use of personal information; (iii) collecting and using personal information without the users’ consent; (iv) collecting personal information unrelated to the services they provide and beyond the necessary principle; (v) providing personal information to others without the users’ consent; (vi) failing to provide the function of deleting or correcting the personal information according to the laws or failing to publish information such as ways of filing complaints and reports. On March 12, 2021, the CAC, the MIIT, the MPS and the SAMR jointly promulgated the Provisions on the Scope of Essential Personal Information for Common Types of Mobile Internet Applications with effective date from May 1, 2021. In relation to ride hailing applications, the basic functional services are “online taxi booking service, cruise taxi call service”, for which the necessary personal information includes mobile phone numbers, place of departure, place of destination, location information, whereabouts and tracks of passengers and payment information. In addition, Internet application operators shall not refuse users from using the basic functions of the Internet application on the ground that users do not agree to the collection of unnecessary personal information. 111 Table of Contents On February 12, 2025, the CAC promulgated the Administrative Measures for the Compliance Audit of Personal Information Protection (the “Measures for Compliance Audit”), which became effective on May 1, 2025. The Measures for Compliance Audit provide detailed provisions on the scope of compliance audit activities, implementation procedures, selection of audit institutions, audit frequency, and obligations of personal information processors and professional institutions during compliance audits. Among others, personal information processors handling the personal information of more than 1 million individuals are required to designate a person in charge of personal information protection to oversee the compliance audit activities. Regulations on Foreign Exchange Control and Dividend Distribution Regulations on Foreign Currency Exchange Pursuant to the Foreign Exchange Administrative Regulations of the PRC promulgated by the State Council on January 29, 1996, and last amended and became effective on August 5, 2008, Renminbi is freely convertible for payments of current account items such as trade and service-related foreign exchange transactions and dividend payments after the relevant financial institutions have reasonably examined the authenticity of the transactions and their consistency with foreign exchange receipts and payments, but are not freely convertible for capital expenditure items such as direct investment, loans or investments in securities outside the PRC unless the approval of the State Administration of Foreign Exchange of the PRC (the “SAFE”) or its local counterparts is obtained in advance. On March 30, 2015, the SAFE promulgated the Circular on Reforming the Administration of Foreign Exchange Settlement of Capital of Foreign-Invested Enterprises (the “Circular 19”), which became effective on June 1, 2015 and amended in 2019 and 2023. The SAFE further promulgated the Circular of the State Administration of Foreign Exchange on Reforming and Regulating Policies on the Control over Foreign Exchange Settlement under the Capital Account (the “Circular 16”) on June 9, 2016, which, among other things, amended certain provisions of the Circular 19. According to the Circular 19 and the Circular 16, the flow and use of the Renminbi capital converted from foreign currency denominated registered capital of a foreign-invested company is regulated such that Renminbi capital may not be used for business beyond its business scope or to provide loans to persons other than affiliates unless otherwise permitted under its business scope. Violations of the Circular 19 or the Circular 16 could result in administrative penalties. On January 26, 2017, the SAFE promulgated the Notice on Improving the Check of Authenticity and Compliance to Further Promote Foreign Exchange Control, which stipulates several capital control measures with respect to the outbound remittance of profit from domestic entities to offshore entities, including (i) under the principle of genuine transaction, banks shall check board resolutions regarding profit distribution, the original version of tax filing records and audited financial statements; and (ii) domestic entities shall hold income to account for previous years’ losses before remitting the profits. On October 23, 2019, the SAFE issued the Notice on Further Promoting Cross-border Trade and Investment Facilitation and last amended on December 4, 2023 by the Notice on Further Deepening the Reform to Facilitate Cross-border Trade and Investment (the “Circular 28”), which expressly allows foreign- invested enterprises that do not have equity investments in their approved business scope to use their capital obtained from foreign exchange settlement to make domestic equity investments as long as the investments are real and in compliance with the foreign investment-related laws and regulations. In addition, Circular 28 stipulates that qualified enterprises in certain pilot areas may use their capital income from registered capital, foreign debt and overseas listing, for the purpose of domestic payments without providing authenticity certifications to the relevant banks in advance for those domestic payments. On April 10, 2020, SAFE issued the Notice of the SAFE on Optimizing Foreign Exchange Administration to Support the Development of Foreign-related Business (the “Circular 8”). The Circular 8 provides that under the condition that the use of funds is genuine and compliant with current administrative provisions on use of income relating to capital account, enterprises are allowed to use income under capital account such as capital funds, foreign debts and overseas listings for domestic payment, without submission to the bank prior to each transaction of materials evidencing the veracity of such payment. 112 Table of Contents Regulations on Foreign Exchange Registration of Overseas Investment by PRC Residents On July 4, 2014, the SAFE promulgated the Circular on Relevant Issues Relating to Domestic Resident’s Investment and Financing and Roundtrip Investment through Special Purpose Vehicles (the “SAFE Circular 37”) for the purpose of simplifying the approval process, and for the promotion of the cross-border investment. Under the SAFE Circular 37, (i) before the PRC residents or entities conducting investment in offshore special purpose vehicles with their legitimate onshore and offshore assets or equities, they must register with local SAFE branches with respect to their investments; and (ii) following the initial registration, they must update their SAFE registrations when the offshore special purpose vehicle undergoes material events relating to any change of basic information (including change of such PRC citizens or residents, name and operation term, increases or decreases in investment amount, transfers or exchanges of shares, or mergers or divisions). The SAFE further promulgated the Notice on Further Simplifying and Improving Foreign Exchange Administration Policy on Direct Investment on February 13, 2015, which came into effect on June 1, 2015 and allows PRC residents or entities to register with qualified banks in connection with their establishment or control of an offshore entity established for the purpose of overseas investment or financing. The qualified banks, under the supervision of SAFE, directly examine the applications and conduct the registration. Failure to comply with the registration procedures set forth in the SAFE Circular 37 may result in restrictions being imposed on the foreign exchange activities of the relevant onshore company, including the payment of dividends and other distributions to its offshore parent or affiliate, and may also subject relevant PRC residents to penalties under PRC foreign exchange administration regulations. PRC residents who control the company from time to time are required to register with the SAFE in connection with their investments in the company. Moreover, failure to comply with the various SAFE registration requirements described above could result in liability under PRC laws for evasion of foreign exchange controls. Regulations on Dividend Distribution Under applicable PRC laws and regulations, foreign investment enterprises in China may pay dividends only out of their accumulated profits, if any, determined in accordance with PRC accounting standards and regulations. In addition, foreign investment enterprises in China are required to allocate at least 10% of their respective accumulated after-tax profits each year, if any, to fund statutory reserve funds unless these reserves have reached 50% of the registered capital of the respective enterprises. A PRC company may, at its discretion, allocate a portion of its after-tax profits based on the PRC accounting standards to other reserve funds. These reserves are not distributable as cash dividends. A PRC company shall not distribute any profits until any losses from prior fiscal years have been offset and the reserve funds have been funded. Profits retained from prior fiscal years may be distributed together with distributable profits from the current fiscal year. Regulations on Intellectual Property Patent The Patent Law of the PRC, which was promulgated by the SCNPC on March 12, 1984 and last amended on October 17, 2020 and became effective on June 1, 2021, provides for three types of patents, namely, “invention”, “utility model” and “design”. Invention patents are valid for twenty years, design patents filed no later than May 31, 2021 are valid for 10 years while design patents filed on or after June 1, 2021 are valid for 15 years and utility model patents are valid for ten years, from the date of application. The Chinese patent system adopts a “first-to-file” principle, which means that where more than one person files a patent application for the same invention, a patent will be granted to the person who files the application first. To be patentable, invention or utility models must meet three criteria: novelty, inventiveness and practicability. A third party must obtain consent or a proper license from the patent owner to use the patent. Otherwise, the use constitutes an infringement of the patent rights. 113 Table of Contents Copyright The Copyright Law of the PRC (the “Copyright Law”), which first became effective on June 1, 1991, and was latest amended in 2020 and became effective on June 1, 2021, provides that PRC citizens, legal persons, or other organizations shall, whether published or not, own copyright in their copyrightable works, which include, among others, works of literature, art, natural science, social science, engineering technology and computer software. Copyright owners enjoy certain legal rights, including right of publication, right of authorship and right of reproduction. The Copyright Law extends copyright protection to Internet activities, products disseminated over the internet and software products. In addition, the Copyright Law provides for a voluntary registration system administered by the China Copyright Protection Center. Pursuant to the Computer Software Copyright Protection Regulations promulgated by the State Council on June 4, 1991, and amended on January 30, 2013 and became effective on March 1, 2013, software copyright owners may go through the registration formalities with a software registration authority recognized by the State Council’s copyright administrative department. Software copyright owners may authorize others to exercise that copyright, and is entitled to receive remuneration. Trademark Pursuant to the Trademark Law of the PRC, promulgated by the SCNPC on August 23, 1982, and last amended on April 23, 2019 and became effective on November 11, 2019, the Trademark Office of China National Intellectual Property Administration is responsible for the registration and administration of trademarks and is also responsible for resolving trademark disputes in China. A registered trademark is valid for ten years from the date the registration is approved. A registrant may apply to renew a registration within twelve months before the expiration date of the registration. If the registrant fails to apply in a timely manner, a grace period of six additional months may be granted. If the registrant fails to apply before the grace period expires, the registered trademark shall be deregistered. Renewed registrations are valid for ten years. Domain Names Internet domain name registration and related matters are primarily regulated by the Measures on Administration of Internet Domain Names promulgated by the MIIT on August 24, 2017 and became effective on November 1, 2017, and the Implementing Rules on Registration of National Top-level Domain Names promulgated by China Internet Network Information Centre and took into effect on June 18, 2019. The domain name services follow a “first come, first file” principle. Applicants for registration of domain names shall provide their true, accurate and complete information of such domain names to and enter into registration agreements with domain name registration service institutions. The applicants will become the holders of such domain names upon the completion of the registration procedure. Regulations on Taxation Enterprise Income Tax According to the Enterprise Income Tax Law of the PRC (the “EIT Law”), which was promulgated by the SCNPC on March 16, 2007, and was last amended and became effective on December 29, 2018, and the Enterprise Income Tax Implementation Regulations of the PRC (the “EITIR”), which was promulgated by the State Council on December 6, 2007, and was last amended and became effective on January 20, 2025, enterprises are classified as “resident enterprises” and “non-resident enterprises”, and both resident enterprises and non-resident enterprises are subject to tax in the PRC. Pursuant to the EIT Law and the EITIR, PRC resident enterprises typically pay an enterprise income tax at the rate of 25% while non-PRC resident enterprises without any branches in the PRC should pay an enterprise income tax in connection with their income from the PRC at the tax rate of 10%. Enterprises established under the laws of foreign countries or regions whose “de facto management bodies” are located in the PRC are considered to be PRC resident enterprises, and will generally be subject to enterprise income tax at the rate of 25% of their global income. The EITIR defines “de facto management bodies” as “establishments that carry out substantial and overall management and control over production and operations, personnel, accounting, and properties” of the enterprise. 114 Table of Contents Pursuant to Notice of the State Taxation Administration on Issues about the Determination of Chinese- Controlled Enterprises Registered Abroad as Resident Enterprises on the Basis of Their Body of Actual Management (the “STA Circular 82”) issued by the State Taxation Administration of the PRC (the “STA”) in April 2009 and amended in December 2017, an overseas registered enterprise controlled by a PRC company or a PRC company group will be classified as a “resident enterprise” with its “de facto management body” located within China if the following requirements are satisfied: (i) the senior management and core management departments in charge of its daily operations are mainly located in the PRC; (ii) its financial and human resources decisions are subject to determination or approval by persons or bodies located in the PRC; (iii) its major assets, accounting books, company seals, and minutes and files of its board and shareholders’ meetings are located or kept in the PRC; and (iv) no less than half of the enterprise’s directors or senior management with voting rights reside in the PRC. The STA issued additional rules to provide more guidance on the implementation of STA Circular 82 in July 2011, and issued an amendment to STA Circular 82 in January 2014 delegating the authority to its provincial branches to determine whether a Chinese-controlled overseas-incorporated enterprise should be considered a PRC resident enterprise. Although the STA Circular 82, the additional guidance and its amendment only apply to overseas registered enterprises controlled by PRC enterprises and not those controlled by PRC individuals or foreigners, the determining criteria set forth in the circular may reflect STA’s general position on how the “de facto management body” test should be applied in determining the tax resident status of offshore enterprises, regardless of whether they are controlled by PRC enterprises, PRC individuals or foreigners. If our offshore entities are deemed PRC resident enterprises, these entities may be subject to the EIT at the rate of 25% on their global income, except that the dividends distributed by our PRC subsidiaries may be exempt from the EIT to the extent such dividends are deemed “dividends among qualified resident enterprises.” In addition, pursuant to the EIT Law, enterprises qualified as “High and New Technology Enterprises” are entitled to a 15% enterprise income tax rate rather than the 25% uniform statutory tax rate. The preferential tax treatment continues as long as an enterprise can retain its “High and New Technology Enterprise” status. Dividends Withholding Tax According to the EIT Law and the EITIR, dividends paid by foreign-invested companies to their foreign investors that are non-resident enterprises as defined under the law are subject to withholding tax at a rate of 10%, unless otherwise provided in the relevant tax agreements entered into with the central government of the PRC. Pursuant to the Arrangement Between the Mainland of China and the Hong Kong Special Administrative Region for the Avoidance of Double Taxation and the Prevention of Fiscal Evasion with respect to Taxes on Income promulgated on August 21, 2006, if a Hong Kong resident enterprise is determined by the competent PRC tax authority to have satisfied the relevant conditions and requirements under such tax arrangement, the withholding tax rate on the dividends the Hong Kong resident enterprise receives from a PRC resident enterprise may be reduced to 5% from 10% applicable under the EIT Law and the EITIR. However, based on the Notice of the State Taxation Administration on Certain Issues with Respect to the Enforcement of Dividend Provisions in Tax Treaties promulgated by the STA and effective on February 20, 2009, if the relevant PRC tax authorities determine, in their discretion, that a company benefits from such reduced income tax rate due to a structure or arrangement that is primarily tax-driven, such PRC tax authorities may adjust the preferential tax treatment. Based on the Notice of the State Taxation Administration on the Recognition of Beneficial Owners in Tax Treaties, which was promulgated by the STA on February 3, 2018 and came into effect on April 1, 2018, a comprehensive analysis will be used to determine beneficial ownership based on the actual situation of a specific case combined with certain principles, and if an applicant was obliged to pay more than 50% of its income to a third country (region) resident within 12 months of the receipt of the income, or the business activities undertaken by an applicant did not constitute substantive business activities including substantive manufacturing, distribution, management and other activities, the applicant was unlikely to be recognized as a beneficial owner to enjoy tax treaty benefits. Furthermore, the Administrative Measures for Convention Treatment for Non-resident Taxpayers, which became effective on January 1, 2020, require that non-resident taxpayers claiming treaty benefits shall be handled in accordance with the principles of “self-assessment, claiming for the enjoyment of treaty benefits, and retention of the relevant materials for future inspection.” Where a non-PRC resident taxpayer self-assesses and concludes that it satisfies the criteria for claiming treaty benefits, it may enjoy treaty benefits at the time of tax declaration or at the time of withholding through a withholding agent, simultaneously gather and retain the relevant materials pursuant to the provisions of these Measures for future inspection, and subject to subsequent administration by relevant competent tax authorities. 115 Table of Contents Value-added Tax and Business Tax Before August 2013 and pursuant to applicable PRC tax regulations, any entity or individual conducting business in the service industry is generally required to pay a business tax. In November 2011, the Ministry of Finance (the “MOF”) and the STA promulgated the Pilot Plan for Imposition of Value-Added Tax to Replace Business Tax. In May and December 2013, April 2014, March 2016 and July 2017, the MOF and the STA promulgated five circulars to further expand the scope of services that are to be subject to value-added tax (the “VAT”) instead of business tax. Pursuant to these tax rules, from August 1, 2013, a VAT was imposed to replace the business tax in certain service industries, including technology services, and from May 1, 2016, VAT replaced business tax in all industries on a nationwide basis. On November 19, 2017, the State Council further amended the Interim Regulation of the People’s Republic of China on Value Added Tax to reflect the normalization of the pilot program. On March 20, 2019, the MOF, the STA and the General Administration of Customs jointly issued the Announcement of Strengthening Reform of VAT Policies (the “Announcement No. 39”), which provides certain VAT reduction arrangements. According to the Announcement No. 39: (i) for general VAT payers’ sales activities or imports that are subject to VAT at an existing applicable rate of 16% or 10%, the applicable VAT rate is respectively adjusted to 13% or 9%; (ii) for the agricultural products purchased by taxpayers to which an existing 10% deduction rate is applicable, the deduction rate is adjusted to 9%; (iii) for the agricultural products purchased by taxpayers for production or commissioned processing, which are subject to VAT at 13%, the input VAT will be calculated at a 10% deduction rate; (iv) for the exportation of goods or labor services that are subject to VAT at 16%, with the applicable export refund at the same rate, the export refund rate is adjusted to 13%; and (v) for the exportation of goods or cross-border taxable activities that are subject to VAT at 10%, with the export refund at the same rate, the export refund rate is adjusted to 9%. On December 25, 2024, the SCNPC adopted the Value-added Tax Law of the PRC (the “VAT Law”). On December 25, 2025, the State Council promulgated the Regulations for the Implementation of the Value-added Tax Law of the PRC (the “Implementation Regulations”). Both the VAT Law and the Implementation Regulations took effect on January 1, 2026, replacing the Provisional Regulations on Value-added Tax of the PRC and the Detailed Rules for the Implementation of the Provisional Regulations on Value-added Tax of the PRC. Pursuant to the VAT Law and the Implementation Regulations, entities and individuals (including individual businesses) engaged in the sale of goods, services, intangible assets and immovables, as well as the importation of goods, within the territory of the PRC are VAT payers and shall pay VAT in accordance with such laws and regulations. Unless otherwise specified in the aforesaid laws and regulations, taxpayers engaged in the sale of goods, the provision of processing, repair and replacement services, tangible movable property leasing services, or the importation of goods are subject to a VAT rate of 13%. Enterprise Income Tax on Indirect Transfer of Non-PRC Resident Enterprises On December 10, 2009, the STA issued the Notice on Strengthening the Administration of Enterprise Income Tax on Equity Transfers of Non-PRC Resident Enterprises (the “Circular 698”). By promulgating and implementing the Circular 698, the PRC tax authorities enhanced their scrutiny over the indirect transfer of equity interests in a PRC resident enterprise by a non-resident enterprise. The STA further issued the Public Announcement on Several Issues Concerning Enterprise Income Tax for Indirect Transfer of Assets by Non-PRC Resident Enterprises (the “STA Circular 7”) on February 3, 2015, which replaces certain provisions in the Circular 698. The STA Circular 7 introduces a new tax regime that is significantly different from that under the Circular 698. The STA Circular 7 extends its tax jurisdiction to capture not only indirect transfer as set forth under the Circular 698 but also transactions involving transfer of immovable property in China and assets held under the establishment and place, in China of a foreign company through the offshore transfer of a foreign intermediate holding company. The STA Circular 7 also provides clearer criteria than the Circular 698 on how to assess reasonable commercial purposes and introduces safe harbor scenarios applicable to internal group restructurings. Where a non-resident enterprise indirectly transfers equity interests or other assets of a PRC resident enterprise by implementing arrangements that are not for reasonable commercial purposes to avoid its obligation to pay enterprise income tax, such an indirect transfer shall, in accordance with the EIT Law, be recognized by the competent PRC tax authorities as a direct transfer of equity interests or other assets of the PRC resident enterprise. On October 17, 2017, the STA promulgated the Announcement on Matters Concerning Withholding and Payment of Income Tax of Non-PRC Resident Enterprises from Source (the “STA Circular 37”), which replaced the Circular 698 and certain provisions in the STA Circular 7 on December 1, 2017 and was partly amended on June 15, 2018. The STA Circular 37, among other things, simplifies the procedures of withholding and payment of income tax levied on non-resident enterprises. Pursuant to STA Circular 37, where the party responsible for withholding such income tax did not, or was unable to, withhold the taxes that should have been withheld to the relevant tax authority, the party may be subject to penalties. Where the non-PRC resident enterprise receiving such income failed to declare and pay taxes that should have been withheld to the relevant tax authority, the party may be ordered to rectify within a specific time limit. 116 Table of Contents Regulations on Employment and Social Welfare The Labor Contract Law The PRC employment laws and regulations mainly include the Labor Law of the PRC promulgated by the SCNPC on July 5, 1994, and last amended and became effective on December 29, 2018, the Labor Contract Law of the PRC promulgated by the SCNPC on June 29, 2007, and amended on December 28, 2012 and became effective on July 1, 2013, and the Regulations on the Implementation of the Labor Contract Law promulgated by the State Council and became effective on September 18, 2008. According to such employment laws and regulations, labor relationships between employers and employees must be executed in written form. Wages may not be lower than the local minimum wage standard and must be paid in a timely manner. Employers must establish a system for labor safety and sanitation, strictly abide by state standards and provide relevant training to its employees. It is required that employers provide safe and sanitary working conditions for employees. Social Insurance and Housing Fund Pursuant to the Social Security Law of the PRC, which was promulgated by the SCNPC on October 28, 2010, and was amended on December 29, 2018, and other relevant PRC laws and regulations such as the Interim Regulations on the Collection and Payment of Social Insurance Premiums effective on January 22, 1999 and amended on March 24, 2019, Regulations on Work Injury Insurance implemented on January 1, 2004 and amended on December 20, 2010, Regulations on Unemployment Insurance promulgated on January 22, 1999 and Trial Measures on Employee Maternity Insurance of Enterprises implemented on January 1, 1995, the employer shall contribute to social insurance plans covering basic pensions insurance, basic medical insurance, maternity insurance, employment injury insurance and unemployment insurance. Basic pension, medical and unemployment insurance contributions shall be paid by both employers and employees, while employment injury insurance and maternity insurance contributions shall be paid only by employers, and employers who failed to promptly contribute social security premiums in full amount shall be ordered by the social security premium collection agency to make or supplement contributions within a stipulated period, and shall be subject to a late payment fee, and where late payment fee is not made within the stipulated period, the relevant administrative authorities shall impose a fine. Pursuant to the Regulations on the Administration of Housing Fund, which was promulgated by the State Council on April 3, 1999, and last amended and became effective on March 24, 2019, enterprises in the PRC must register with the competent managing center for housing provident funds and upon the examination by such center, these enterprises shall complete procedures for opening an account at the relevant bank for the deposit of employees’ housing provident funds. Enterprises are also required to pay and deposit housing provident funds on behalf of their employees in full and in a timely manner. Employers that violate these regulations and fail to process housing provident fund payments or deposit registrations with the housing provident fund administration center within a designated period are subject to a fine. Labor Dispatch Pursuant to the Interim Provisions on Labor Dispatch issued on January 24, 2014, and implemented on March 1, 2014, by the Ministry of Human Resources and Social Security, employers may only use dispatched workers for temporary, ancillary, or substitute positions. The aforementioned temporary positions shall mean positions lasting for no more than six months; ancillary positions shall mean positions of non-major business that serve positions of major business; and substitute positions shall mean positions that can be substituted by other workers for a certain period during which the workers who originally hold such positions are unable to work as a result of full-time study, being on leave or other reasons. Pursuant to the Interim Provisions on Labor Dispatch, employers should strictly control the number of dispatched workers, and the number of the dispatched workers shall not exceed 10% of the total amount of their employees. Pursuant to the Labor Contract Law, where rectification is not made within the stipulated period, the employers may be subject to a penalty ranging from RMB5,000 to RMB10,000 per dispatched worker exceeding the 10% threshold. Employee Stock Incentive Plan Pursuant to the Notice of Issues Related to the Foreign Exchange Administration for Domestic Individuals Participating in Stock Incentive Plan of Overseas Listed Company, which was issued by the SAFE on February 15, 2012, employees, directors, supervisors, and other senior management who participate in any stock incentive plan of a publicly-listed overseas company and who are PRC citizens or non-PRC citizens residing in China for a continuous period of no less than one year, subject to a few exceptions, are required to register with the SAFE through a qualified domestic agent, which may be a PRC subsidiary of such overseas listed company, and complete certain other procedures. 117 Table of Contents In addition, the STA has issued certain circulars concerning employee stock options and restricted shares. Under these circulars, employees working in the PRC who exercise stock options or are granted restricted shares will be subject to PRC individual income tax. The PRC subsidiaries of an overseas listed company are required to file documents related to employee stock options and restricted shares with relevant tax authorities and to withhold individual income taxes of employees who exercise their stock options or purchase restricted shares. If the employees fail to pay or the PRC subsidiaries fail to withhold income tax in accordance with relevant laws and regulations, the PRC subsidiaries may face sanctions imposed by the tax authorities or other PRC regulatory authorities. Regulations on Anti-Monopoly and Anti-Unfair Competition Pursuant to the Anti-Monopoly Law promulgated by the SCNPC on August 30, 2007, which was amended on June 24, 2022 and became effective on August 1, 2022, where the concentration of business operators reaches the filing thresholds stipulated by the State Council, business operators shall file a declaration with the SAMR, and no concentration shall be implemented until the SAMR clears the anti-monopoly filing. On February 7, 2021, the Anti-Monopoly Committee of the State Council promulgated the Anti-monopoly Guidelines for the Platform Economy Sector (the “Anti-monopoly Guideline”), aiming to improve anti-monopoly administration on online platforms. The Anti-monopoly Guideline, operating as the compliance guidance under the existing PRC anti-monopoly regulatory regime for platform economy operators, specifically prohibits certain acts of the platform economy operators that may have the effect of eliminating or limiting market competition, such as concentration of undertakings. Pursuant to the Anti-Unfair Competition Law promulgated by the SCNPC on September 2, 1993, which was amended on June 27, 2025 and became effective on October 15, 2025, operators are prohibited from engaging in unfair competition activities such as market confusion, commercial bribery, misleading false publicity, infringement on trade secrets, price dumping, and illegitimate premium sales. Any operator in violation of the Anti-Unfair Competition Law may be ordered to cease illegal activities, eliminate the adverse effect thereof or compensate for the damages caused to any other party. The competent authorities may also confiscate any illegal gains or impose fines on these operators. Regulations on M&A Rules and Overseas Listings On August 8, 2006, six PRC regulatory agencies, including the MOFCOM, State-owned Assets Supervision and Administration Commission of the State Council, STA, State Administration for Industry and Commerce of the PRC, CSRC and SAFE, issued the Regulations on Merger with and Acquisition of Domestic Enterprises by Foreign Investors (the “M&A Rules”), which became effective on September 8, 2006 and was amended on June 22, 2009. The M&A Rules, among other things, require that if an overseas company established or controlled by PRC companies or individuals intends to acquire equity interests or assets of any other PRC domestic company affiliated with such PRC companies or individuals, such acquisition must be submitted to MOFCOM for approval. The M&A Rules also require offshore special purpose vehicles that controlled by PRC companies or individuals and formed for overseas listing purposes through acquisitions of PRC domestic companies or subscription of new shares issued by PRC domestic company using the equity of offshore special purpose vehicles or using its new shares as consideration, to obtain the approval of China Securities Regulatory Commission prior to publicly listing their securities on an overseas stock exchange. On February 3, 2011, the General Office of the State Council promulgated a Notice on Establishing the Security Review System for Mergers and Acquisitions of Domestic Enterprises by Foreign Investors (the “Circular 6”), which officially established a security review system for mergers and acquisitions of domestic enterprises by foreign investors. Further, on August 25, 2011, MOFCOM promulgated the Regulations on Implementation of Security Review System for the Merger and Acquisition of Domestic Enterprises by Foreign Investors (the “MOFCOM Security Review Regulations”), which became effective on September 1, 2011, to implement Circular 6. Under Circular 6, a security review is required for mergers and acquisitions by foreign investors having “national defense and security” concerns and mergers and acquisitions by which foreign investors may acquire the “de facto control” of domestic enterprises with “national security” concerns. Under the MOFCOM Security Review Regulations, MOFCOM will focus on the substance and actual impact of the transaction when deciding whether a specific merger or acquisition is subject to security review. If MOFCOM decides that a specific merger or acquisition is subject to security review, it will submit it to the Inter-Ministerial Panel, an authority established under the Circular 6 led by the NDRC, and MOFCOM under the leadership of the State Council, to carry out the security review. The regulations prohibit foreign investors from bypassing the security review by structuring transactions through trusts, indirect investments, leases, loans, control through contractual arrangements or offshore transactions. 118 Table of Contents On February 17, 2023, the CSRC published the Trial Administrative Measures of the Overseas Securities Offering and Listing by Domestic Companies and five supporting guidelines, collectively the Overseas Listing Filing Rules, which came into effect from March 31, 2023 and regulate both direct and indirect overseas offering and listing of PRC-based companies by adopting a filing-based regulatory regime. According to the Overseas Listing Filing Rules, if the issuer meets both of the following criteria, the overseas securities offering and listing conducted by such issuers shall be deemed as indirect overseas offering and listing: (i) more than 50% of the issuer’s operating revenue, total profit, total assets or net assets as documented in its audited consolidated financial statements for the most recent accounting year is accounted for by domestic companies; and (ii) the main parts of the issuer’s business activities are conducted in China, or its main places of business are located in China, or the senior managers in charge of its business operation and management are majority Chinese citizens or domiciled in China. The Overseas Listing Filing Rules provide that (i) the filing applications be submitted to the CSRC within three business days after the issuer submits its application documents relating to the initial public offering and/or listing in overseas; (ii) a timely report be submitted to the CSRC and update its CSRC filing within three business days after the occurrence of any of the following material events, if any of the following events occurs before the completion of the overseas offering and/or listing but after the completion of its CSRC filing: (a) any material change to principal business, licenses or qualifications of the issuer, (b) a change of control of the issuer or any material change to equity structure of the issuer, and (c) any material change to the offering and listing plan; (iii) after the completion of the listing, a report relating to the issuance information of such offering and/or listing be submitted to the CSRC and a report be submitted to the CSRC within three business days upon the occurrence and public announcement of any of the following material events after the overseas offering and/or listing: (a) a change of control of the issuer, (b) the investigation, sanction or other measures undertaken by any foreign securities regulatory agencies or relevant competent authorities in respect of the issuer, (c) change of the listing status or transfer of the listing board, and (d) the voluntary or mandatory delisting of the issuer; and (iv) where there is material change in the main business of the issuer after overseas offering and listing, which does not apply to the Overseas Listing Filing Rules therefore, such issuer shall submit to the CSRC a report and a relevant legal opinion issued by a domestic law firm within three business days after occurrence of such change. The Overseas Listing Filing Rules states that, any post-listing follow-on offering by an issuer in the same overseas market, including issuance of shares, convertible notes and other similar securities, shall be subject to filing requirement within three business days after the completion of the offering. Therefore, any of our future offering and listing of our securities in an overseas market shall be subject to the filing requirements under the Overseas Listing Filing Rules. Based on the Overseas Listing Trial Measures, violation of the Overseas Listing Trial Measures or the completion of an overseas listing in breach of the conditions listed in the Overseas Listing Trial Measures may result in rectification, warning and a fine ranging from RMB1,000,000 to RMB10,000,000. Furthermore, the controlling shareholders and actual controllers of the relevant PRC domestic companies that organize or instruct such violations or enable such violations by concealing relevant matters, may be subject to a fine ranging from RMB1,000,000 to RMB10,000,000; and the directly responsible supervisors and other directly liable persons may be subject to warning and a fine ranging from RMB500,000 to RMB5,000,000. On February 24, 2023, the CSRC, together with other governmental authorities, released the Provisions on Strengthening the Confidentiality and Archives Administration Related to the Overseas Securities Offering and Listing by Domestic Enterprises (the “Confidentiality and Archives Administration Provisions”), which became effective from March 31, 2023 and aims to expand the applicable scope of the regulation to indirect overseas offerings and listings by PRC domestic companies and emphasize the confidentiality and archive management duties of PRC domestic companies during the process of overseas offerings and listings. The Confidentiality and Archives Administration Provisions require, among others, that PRC domestic enterprises seeking to offer and list securities in overseas markets, either directly or indirectly, shall establish the confidentiality and archives system, and shall complete approval and filing procedures with competent authorities, if such PRC domestic enterprises or their overseas listing entities provide or publicly disclose documents or materials involving state secrets and work secrets of PRC government agencies to relevant securities companies, securities service institutions, overseas regulatory agencies and other entities and individuals. U.S. Regulations While autonomous driving laws and regulations are expected to continue to evolve in numerous jurisdictions in the United States, there has been relatively little mandatory government regulation of the autonomous driving industry to date in the United States. At both the federal and state levels, the United States provides a positive regulatory environment to permit safe testing and development of autonomous vehicle functionality. Currently, there are no Federal Motor Vehicle Safety Standards (“FMVSS”) that relate to the performance of autonomous driving technology. Further, there are currently no widely accepted uniform standards to certify autonomous driving technology and its commercial use on public roads. 119 Table of Contents As of now, the safety of commercial motor vehicles is regulated by the NHTSA and the Federal Motor Carrier Safety Administration (the “FMCSA”). NHTSA establishes the FFMVSS for motor vehicles and motor vehicle equipment and oversees the actions that manufacturers of motor vehicles and motor vehicle equipment are required to take regarding the reporting of information related to defects or injuries related to their products and the recall and repair of vehicles and equipment that contain safety defects or fail to comply with the FMVSS. FMCSA regulates the safety of commercial motor carriers operating in interstate commerce, the qualifications and safety of commercial motor vehicle drivers, and the safe operation of commercial trucks. Motor vehicle equipment manufacturers are subject to existing stringent requirements under the Vehicle Safety Act, including a duty to report, subject to strict timing requirements, safety defects. The Vehicle Safety Act imposes potentially significant civil penalties for violations including the failure to comply with such reporting actions. They are also subject to the existing TREAD Act, which requires motor vehicle equipment manufacturers to comply with “Early Warning” requirements by reporting certain information to the NHTSA, such as information related to defects or reports of injury. The TREAD Act imposes criminal liability for violating such requirements if a defect subsequently causes death or bodily injury. At the state level, states, such as Arizona, Florida, Nevada, and Texas, continue to attract autonomous driving companies with a welcoming regulatory climate that provides the predictability necessary to deploy autonomous driving technology in those communities. Some states, particularly California, impose restrictions and enforce some operational or registration requirements for certain autonomous functions, and many other states are still considering them. In addition, the autonomous driving industry is also subject to trade, customs product classification and sourcing regulations as well as various federal, state and local laws and regulations governing the occupational health and safety of our employees and wage regulations. Specifically, it is subject to the laws and regulations of export controls, including the U.S. Department of Commerce’s Export Administration Regulations, and the requirements of the federal Occupational Safety and Health Act, as amended, and comparable state laws that protect and regulate employee health and safety. Moreover, it’s subject to environmental regulations, including water use; air emissions; use of recycled materials; energy sources; the storage, handling, treatment, transportation and disposal of hazardous materials; and the remediation of environmental contamination. Compliance with these rules may include permits, licenses and inspections of company facilities and products. The U.S. federal government and various states and governmental agencies also have adopted or are considering adopting various laws, regulations, and standards regarding the collection, use, retention, security, disclosure, transfer, and other processing of sensitive and personal information. In addition, many states have laws that protect the privacy and security of sensitive and personal information. Certain state laws may be more stringent or broader in scope, or offer greater individual rights, with respect to sensitive and personal information than federal, international, or other state laws, and such laws may differ from each other, which may complicate compliance efforts. For example, in 2018, California enacted the California Consumer Privacy Act, which came into effect on January 1, 2020, and has since been amended by the California Privacy Rights Act which came into effect on January 1, 2023 (collectively, the “CCPA”). The CCPA creates individual privacy rights for California residents, including rights to opt out of certain processing such as the transfer of personal information for the purpose of cross contextual behavioral advertising, the processing of sensitive personal information for certain purposes, as well as “sales” of personal information, and increases the privacy and security obligations of entities handling personal information of California consumers and meeting certain thresholds. The CCPA is currently enforceable by the California Attorney General, and provides for civil penalties for violations as well as a private right of action for certain data breaches that result in the unauthorized access to, or exfiltration, theft or disclosure of certain types of personal information. This private right of action is expected to increase the likelihood of, and risks associated with, class action data breach litigation. Though regulatory fines have been imposed, the CCPA has not been subject to significant litigation and judicial interpretation and it remains unclear how various provisions will be enforced. Additionally, the CCPA’s further expansion under the California Privacy Rights Act may impact our business particularly given its establishment of a new regulatory agency dedicated to enforcing the CCPA’s requirements in addition to the California Attorney General, potentially resulting in further uncertainty and requiring us to incur additional costs and expenses, and potentially change our business practices, in an effort to comply. In addition, many similar laws have been proposed at the federal level and in other states. For instance, the state of Nevada recently enacted a law that went into force on October 1, 2019 and requires companies to honor consumers’ requests to no longer sell their data. Violators may be subject to injunctions and civil penalties of up to $5,000 per violation. New legislation proposed or enacted in Illinois, Massachusetts, New Jersey, New York, Rhode Island, Washington, and other states, and a proposed right to privacy amendment to the Vermont Constitution, imposes, or has the potential to impose, additional obligations on companies that collect, store, use, retain, disclose, transfer, and otherwise process confidential, sensitive, and personal information, and will continue to shape the data privacy environment throughout the United States. State laws are changing rapidly and there is discussion in the U.S. Congress of a new federal data protection and privacy law to which the autonomous driving industry would become subject if it is enacted. 120 Table of Contents For detailed discussion of other applicable U.S. Regulations, see “Item 3. Key Information—3.D. Risk Factors—Risks Related to Doing Business in China—We could be negatively impacted by possible changes to the U.S. government’s new China-focused Outbound Investment Program; these developments may adversely affect our business, financial condition, results of operations, and the value of the ADS,” “—Tensions in international trade and investment and rising political tensions, particularly between the United States and China, may adversely impact our business, financial condition, and results of operations,” and “—We are subject to U.S. export controls that could restrict our ability to transfer certain of our products and technologies, both within our company or to external parties, including potential customers; increasingly restrictive U.S. export controls directed toward China, in particular its artificial intelligence industry, could also limit our ability to obtain advanced semiconductors and other technology that could be needed to develop our products.” 4.C. Organizational Structure The following chart shows our corporate structure as of the date of this annual report. ​ ​ Notes: (1) In February 2022, Cyantron Logistics Technology Co., Ltd. was incorporated under the laws of the PRC. Shareholders of Cyantron Logistics Technology Co., Ltd. are Beijing (HX) Pony and Sinotrans, each holding 51.0% and 49.0% of its equity interests, respectively. (2) Yancheng Poplar LLP is a limited partnership incorporated under the laws of the PRC. Beijing (ZX) Pony AI Technology Co., Ltd. is the general partner of Yancheng Poplar LLP, holding approximately 62% of its interest. The remaining 38% interest in Yancheng Poplar LLP is held by another limited partnership as the limited partner. (3) Hongkong Pony AI Limited has another two wholly-owned subsidiaries, including Company Pony AI (registered in Saudi Arabia) and PONY AI-FZCO (registered in United Arab Emirates). Prior Contractual Arrangements with the Former VIEs and Their Shareholders Historically, we established a series of contractual arrangements with the former VIEs and their shareholders although our business was not subject to any foreign ownership restrictions under the applicable PRC laws and regulations. 121 Table of Contents Pony AI Inc. is an exempted company registered in the Cayman Islands. Beijing (HX) Pony and Guangzhou (HX) Pony, our PRC subsidiaries, are considered foreign-invested enterprises. Beijing (HX) Pony and Hongkong Pony AI entered into a series of contractual arrangements with Beijing (ZX) Pony as well as its shareholders. Guangzhou (HX) Pony and Hongkong Pony AI entered into a series of contractual arrangements with Guangzhou (ZX) Pony as well as its shareholders. As a result of these prior contractual arrangements, we exerted effective control over, and were considered the primary beneficiary of, the former VIEs and consolidated their operating results in its financial statements under the U.S. GAAP, for accounting purposes, for the years ended December 31, 2023 and 2024. We terminated the contractual arrangements among our former WFOEs, the former VIEs and their respective nominee shareholders, and acquired the shares of the former VIEs from their respective nominee shareholders, after which the former VIEs have become wholly-owned subsidiaries of our company since February 2024. 4.D. Property, Plant and Equipment Our principal executive office is located in Guangzhou, China, with an aggregate of 25,406.85 square meters, primarily for corporate administration as well as research and development. We currently do not own any properties. As of December 31, 2025, we had leased properties in Beijing, Shanghai, Shenzhen and some other cities in China, with a total of 58,575.61 square meters, primarily for office, research and development and fleet operation uses. In addition, we operate internationally with leased offices and facilities in the United States, including Fremont, California, with an aggregate of 36,403 square feet. We believe that our current facilities are adequate to meet our current needs. ​