A：The company management representatives attending this press conference include Dr. Xu Libo, Chairman and CEO of both groups, Mr. Yang Fan, President of the Large Equipment Business Group, Dr. Wang Xiaogang, Chief Technology Officer and President of the Intelligent Automotive Business Group, Dr. Lin Dahua, Chief Scientist, and Mr. Wang Zheng, Chief Financial Officer.

A：Yes, this meeting may contain forward-looking statements. However, SenseTime Group reminds that forward-looking statements involve inherent risks and uncertainties which may cause actual performance to deviate from current expectations. For detailed information on these risks and uncertainties, please refer to the latest announcements submitted by SenseTime Group to the Hong Kong Stock Exchange. SenseTime Group does not undertake any obligation to update such statements unless required by applicable law. In addition, this discussion also provides non-International Financial Reporting Standards financial measures for comparison.

A：The management will prepare speeches in Chinese and provide simultaneous English interpretation by an interpreter on the English conference line. For specific details, please refer to the performance announcement. If there are discrepancies between the original words and the translation, please follow the original statement of the management. Dr. Xu Li pointed out that in the first half of 2025, SenseTime Group achieved steady growth, with total revenue reaching RMB 23.6 billion, a year-on-year increase of 36%. The revenue generated by the generative AI as the core growth engine reached RMB 18.2 billion, a year-on-year increase of 73%, further consolidating its leading position in the industry. In terms of operating quality, thanks to strategic focus and efficiency optimization, the adjusted net loss narrowed by 50% year-on-year. At the same time, cash inflow performance was outstanding, with a 96% year-on-year increase in receivables collection, and significant improvement in sales and delivery turnover efficiency.

A：Shangtang Group's two core businesses are generative AI and visual AI, forming a dual-drive pattern. The multimodal large model capability of generative AI empowers various scenarios, while the accumulation of visual AI technology helps to train more complex thinking chain data. The two core businesses promote each other and jointly drive overall development.

A：In order to display the growth curve more clearly, SenseTime adjusted its revenue PU caliber in the first half of this year, dividing it into three major segments: generative AI, visual AI, and X innovation business. This adjustment reflects further clarity in the business center, better reflecting the essence and boundaries of each segment, improving the synergy efficiency between business lines, and more accurately conveying the core growth momentum to the capital market.

A：Shangtang has been working in the field of computer vision for a long time. As early as before the wave of large models, they observed the trend of scale laws in visual models for important tasks. Therefore, at the end of 2022 when the trend of large models and large language models emerged, we prospectively considered the possibility of the fusion of visual models and language data scales on a larger scale, which may bring about a leap in technological capabilities. We became the earliest domestic company to invest in the development of multimodal large models. In 2023, we successfully launched our independently developed multimodal large model, and gradually established cognition that integrates visual models with language models in the research and development process.

A：Intelligent upgrade is essentially a gradual process of breaking data boundaries. In the direction of multimodal evolution, it is expected that there will be three important boundary breakthroughs: the first is the combination of language and image, which is currently being done; the second is the interleaving fusion between logical reasoning and intuition, we have made key technological breakthroughs in this field; the third is to break the boundary between digital space and the physical world, allowing multimodal intelligence to be applied around us, this is a very exciting development to look forward to.

A：We have achieved fast iteration of the model by building a multimodal data production line that can produce 5 tons of tokens per day, allowing us to complete a major upgrade in three months, increasing the daily wage from 6.0 to 6.5. The daily wage of 6.5 has become the first commercial-grade large model in China to implement interleaved thinking with images and text, with significantly improved multimodal reasoning capabilities, benchmarking against the international standard G2.5 pro. The three main features of this upgrade are: fast iteration of the model, optimization of the multimodal basic model architecture to triple cost-effectiveness, and integration of multimodal reasoning with real-life intelligent agents.

A：In the field of multimodal data analysis, our product "Little Raccoon Series" has served over 3 million users. It is expected that the installed base will exceed tens of millions in the next 12 months. Our product leads the industry in accuracy and performance in complex data analysis. For example, in financial data analysis, it can improve efficiency by 300% and has formed deep partnerships with companies such as Ant Group. In the education sector, Little Raccoon Education Edition helps teachers significantly improve work efficiency. Currently, over 500 schools and 250,000 teachers and students are using it, with a test correction accuracy rate of over 99%.

A：In the field of audio-visual fusion interaction, SenseTime is at the forefront of the industry, providing users with a deep communication experience through its technology system that combines fast interaction and slow thinking. Specific examples include cooperation with Xiaomi's smart glasses to promote the innovation of interactive paradigms based on the combination of glasses and multimodal capabilities, as well as landings in various fields such as smart hardware, robots, cultural tourism, and consumer applications.

A：With the continuous improvement of the day by day multi-modal large model capabilities, the application scenarios continue to expand, not only supporting the B2B model, but also directly providing services to end users, forming a cloud-integrated product system. In the first half of this year, the number of large model users increased by 280% year-on-year, and the usage time of bull market interactions increased by more than 500%.

A：Visual AI has undergone ten years of development and is entering a second growth curve with the help of multimodal visual intelligent agents. In the domestic market, SenseTime has ranked first in the computer vision field for nine consecutive years. In the first half of this year, it acquired more than 250 new customers with a high repurchase rate of 57%, further increasing market share and influence. The Ark product has upgraded from visual perception capability output to a visual intelligent agent platform, covering nearly 200 cities and regions, with daily intelligent agent calls exceeding 100 million times.

A：Shangtang continues to expand its advantage in intelligent terminals, seeking differentiated advantages in generative AI and visual integration in the AI 2.0 era. In the first half of the year, it empowered over 250 million Android smartphones and 1.02 million smart cars, and expanded into new areas such as sports cameras and AI glasses. At the same time, the X innovation business sector has been successfully established, with significant progress made in various fields such as autonomous driving, medical care, and intelligent retail by companies like Jueying, Shancui, and Shanhui, respectively, achieving innovative breakthroughs and market recognition.

A：In the first half of 2025, the revenue of SenseTime Technology increased by 35.6%, with the revenue from generative AI reaching 1.82 billion RMB, with an annual growth rate of as high as 73%, and the proportion of total revenue from the group rising from 60% in the previous year to 77%. This is due to the development model and ability accumulation of large model research and development in the computing power infrastructure, model application integration. In addition, despite a decrease in revenue in the visual AI sector due to changes in the single customer cooperation model, this sector has resumed growth after excluding the impact of that customer, achieving a good performance with a gross profit margin of 38.5% through optimized high-quality customer repurchase and cash flow. SenseTime has narrowed down the adjusted EBIT loss and net profit loss in the first half of the year by 72% and 50% respectively through strict control of operating costs and deepening the implementation of the "One Plus X" strategy.

A：Sensetime has maintained strong control over operating expenses, with total operating expenses increasing by only 1.1% year-on-year. Management expenses decreased by 18.8%, and sales expenses continued to decrease by 17.3% on the basis of a 20% decrease for the full year last year. Although R&D expenses have increased, they are lower than the growth rate of gross profit. The company strategically allocates sufficient computing resources to the R&D team to maintain Sensetime's leading position in the industry with its daily new large models. In addition, among the three expenses, overall other R&D costs are decreasing, operating capital efficiency has significantly improved, the cash cycle has shortened by about two-thirds, and trade receivables have reached a historical high of 3.16 billion RMB, an increase of 95.5% year-on-year.

A：SenseTime has increased its capital expenditures, mainly for optimizing its large-scale computing power construction. At the same time, its cash reserves have reached 13.2 billion RMB, successfully exiting some strategic investment targets, with the exit amount hitting a historic high. The financial asset impairment loss in the first half of this year decreased by 76% compared to the same period last year, and the company is seeking to achieve a balance between heavy and light asset operation methods. SenseTime will continue to focus on three main pathways, actively embracing the historical opportunities brought by artificial intelligence, consolidating its leading position in multimodal big model technology, and creating industry-leading intelligent body technology to drive the landing of key vertical industries. At the same time, visual AI will rely on long-term technological accumulation and scene insight to achieve commercial breakthroughs, while focusing on innovative businesses with industrialization and capitalization potential to enhance operational vitality and attractiveness to the capital market.

A：SenseTime relies on self-developed multimodal large models and super large-scale computing power clusters, collaborates with partners such as CATL to build an AIDC algorithm-electricity synergy platform, realizes on-demand computing supply, function matching, and efficient utilization, and provides green upgrading solutions. In terms of compliance and controllability, it deeply participates in the formulation of international and domestic standards, is selected as an important case, and expands synergistic effects with international partners such as the Indonesian government. In the future, SenseTime will be guided by green computing power trustworthy standards to meet applications, and achieve long-term value in higher quality development.

A：SenseTime believes that the "Artificial Intelligence Action Plan" issued by the State Council signals an important policy shift from industrial pilot projects to systematic and scaled implementation. In the next 3 to 5 years, the industry will see many positive changes. SenseTime's strategy will deeply align with this direction, focusing on six major industries and eight key foundations. On the supply side, computing centers, data elements markets, and standardized models will form reusable infrastructure, enhancing computing power efficiency and application delivery efficiency, while the role of software platforms will further expand. On the demand side, AI development is entering uncharted territory, with industries shifting from being available to being user-friendly, and general artificial intelligence will achieve end-to-end efficiency. SenseTime will continue to deepen its presence in existing industries such as finance, education, internet, and e-commerce, and combining with the spread of smart hardware, will promote the large-scale deployment of generative AI in productivity tools and interactive tools, accelerating the joint commercialization of industry models and scene intelligence.

A：In existing markets such as urban governance, transportation, retail, and security, visual intelligent entities can upgrade their performance through multi-modal driving models. This transformation enables the originally single-model calls to become intelligent entities with multi-modal capabilities, thereby driving profit and cash flow growth.

A：One potential direction in Shenzhen City is to integrate green computational power software platforms (such as Cisco software platform) with green computational power standards, including differentiated competitive advantages such as collaborative optimization and standardized services, which can help bring new value-added space. In addition, the replication and scaling of the Adrey A Real World Assets Acceleration Tri-Union Collaboration Project is also a development point worth noting.

A：We believe that the robotics sector has long-term development potential and competitive advantages in the race track. The generated visual and visual AI technologies integrated into the physical platform can fully empower this field and bring new points of explosive growth. In the first half of the year, the robotics sector was active in financing, and the industry developed rapidly. The company has established partnerships with many leading companies, and there is obvious upstream and downstream linkage.

A：The popularity of smart body terminals will bring great possibilities to downstream, so we will integrate the direction of terminal intelligence with the direction of vision to promote the development of smart terminals, and use vision and generative technologies to drive both wheels, seize the windows of policy and market demand, achieve industry replication and deterministic growth.

A：The One Plus X strategy has made important contributions to the group's loss reduction, with the growth of core business coming from the rapid growth of the E sector. The X sector is facing vertical scenarios with huge market potential. Although the profit points are relatively far away, through relatively independent operation, it can ignite the entrepreneurial passion of the team and effectively utilize the overall capabilities of the group. The new revenue split is to allow investors to have a clearer understanding of the direction of One Plus X business. It is expected that the Visual AI sector will resume annual growth in the second half of the year and maintain long-term stability and health; as for X innovative business, due to the possibility that some businesses may gradually phase out, the future financials lack comparability, but initial success has been achieved, with financing reaching the level of 1 billion RMB.

A：The core competitiveness of the large-scale device lies in the integration of hardware and software. By deeply exploring customer needs, especially the needs of top artificial intelligence research institutions, we have accumulated rich experience and lessons. In the market competition, for several consecutive years, we have been able to maintain a top three position in both technical capabilities and market performance, possessing certain differentiated advantages, such as a deeper understanding of customer needs and usage habits, as well as rapid iteration of new technologies, including efficient parallel training efficiency, tracking and adaptation of new hardware, and optimization of model inference efficiency. We maintain close cooperation with model developers and users to ensure that we keep up with changes in customer needs and respond quickly.

A：The 3D body plays a key role in SenseTime, which refers to close cooperation between SenseTime's own model development team and the large equipment team to form a scale advantage, allowing for faster response and iteration. Inter-departmental cooperation is more closely knit and information exchange is smoother.

A：When Deep C was just starting to gain popularity, Sensetime quickly incorporated these technologies and optimization strategies into its products by leveraging the experiments and exploration capabilities of its research institute and large equipment team before the introduction of DFC and models. This enabled rapid iteration and the launch of Deep C's API services, resulting in achieving top scores in authoritative evaluations, being among the best two in the industry.

A：SenseTime has been conducting forward-thinking for a long time, such as planning the training technology of signature card parallel as early as 2018, laying out a computing center in 2019, and proposing the concept of the three-in-one of computing power, data, and model for the infrastructure of artificial intelligence in 2022, as well as providing professional services to help customers complete the closed loop. At the same time, SenseTime has been deeply cooperating with domestic chips since 2019.

A：In the first half of this year, the overall computing power scale of SenseTime has reached 25,000 horses, compared with the double-digit growth maintained last year. At the same time, it is also carrying out updates and iterations of computing power resources, integrating high-quality resources in the market to enhance customer service capabilities and efficiency.

A：Shangtang has made progress in multiple popular tracks such as body intelligence, Wensheng video, and other areas, covering many leading enterprises and research institutions, and realizing new customer coverage and intelligent transformation pilots in the new generation of intelligent terminals with AI capabilities such as glasses, robots, small hardware, etc.

A：SenseTime has laid out its strategy in the field of artificial intelligence, forming a good ecosystem synergy effect with over 20 leading companies through investment and business cooperation. At the WAIC conference, SenseTime unveiled the SensLink intelligent platform, which includes three core engines necessary for embodied intelligence: full-modal perception, precise navigation, and multi-modal interaction. The platform aims to give robots powerful brains and interesting souls. SenseTime will continue to leverage its visual capabilities and autonomous driving technology to empower the SensLink platform, integrating large devices and edge chips to form a hardware and software integrated solution with edge-cloud collaboration.

A：SenseTime believes that the deep integration of multimodalities and intelligent agents will expand the application boundaries and commercial prospects of large models and AI. Following the continuous development of large models, the focus will be on enhancing the multimodal processing capability and deep integration of intelligent agents, exploring new training paradigms to improve spatial understanding and logical reasoning abilities, optimizing multimodal model architectures, improving computing modes in combination with hardware features, and significantly improving model computing efficiency. Additionally, SenseTime will continue to drive technological innovation in productivity and interaction, such as solidifying multimodal reasoning, strengthening reinforcement learning in open environments, promoting the combination of audiovisual and visual imagery, and promoting active interaction modes to achieve self-improvement and evolution of large models in real environments, as well as more intelligent and user-friendly interactive experiences.