Cheetah Mobile Inc.

Good day, and welcome to the Cheetah Mobile first quarter 2025 earnings conference call. All participants will be in a listen-only mode. Should you need assistance, please signal a conference specialist by pressing the star key followed by zero. After today's presentation, there will be an opportunity to ask questions. To ask a question, you may press star, then one on a touch-tone phone. To withdraw your question, please press star, then two. Please note this event is being recorded. I would now like to turn the conference over to Ms. Helen Jingshu, Investor Relations of Cheetah Mobile. Please go ahead.

Thank you, operator. Welcome to Cheetah Mobile's fourth quarter 2025 earnings conference call. With us today are our company's chairman and CEO, Mr. Fushun, and our company's director and CFO, Mr. Thomas Jin. Following management's prepared remarks, we will conduct the Q&A section. Please note that the management's prepared remarks will be presented by an AI agent. Before we begin, I refer you to the safe harbor statement in our earnings release, which also applies to our conference call today, as we will make forward-looking statements. At this time, I would now like to turn the conference call over to our Chairman and CEO, Mr. Fushun, please go ahead.

Good day, everyone. Thank you for joining us. Q1 2025 earnings call. I am Fushun, the CEO of Chica. We started 2025 with a clear plan to strengthen our position in both our longstanding and new business areas. Q1 2025 marked a strong start to the year, and I'm happy to share some great news about how we are doing. First, our revenue grew significantly and we made solid progress in cutting losses. In Q1, our total revenue went up 36% compared to last year and 9% compared to last quarter. Our internet business did especially well with a 46% increase in revenue year over year. Our AI and recovery segments grew 23% year over year and accelerated to 30% quarter over quarter. Just as important, are lost so sharply while still investing in AI and robotics. And we believe this positive momentum will continue. Second, AI agents are becoming a real game changer. As smarter AI models keep improving, they can now go beyond chatting. They can handle real tasks and solve real problems with our strong background in building and launching products. We believe Chita is well-positioned to take advantage of this big shift. We are actively applying aging technology to upgrade our consumer products and power our innovation pipeline. These smart enhancements are making our products more efficient, user-friendly, and aligned with the expectations of the new AI era. For example, we launched DeepRow MAI tool app that turns videos, audio, PDFs, and other documents into concise summaries and mind maps, making knowledge easier to digest and act on. Deep role, a strong example of how we are planning AI agents to create practical data use tools that improve productivity. Third, AI has always been at the center of our AI strategy. We are investing even more in R&D and using AI agents to upgrade our consumer products and robotics. One of our biggest steps forward is AgentOS, our next generation bullet system for service robots. AgentOS is designed to be a fixable digital purpose AI brain that can handle everyday tasks and further strengthen our leadership in voice-enabled robots. Since its inception, we have been working with our distributors to test AgentOS. Customers say AgentOS makes the interaction with robots feel much smarter. It understands the conversation, notices what people are wearing, and can use tools like They also like that it doesn't get confused. If you pause, say something wrong, or switch between languages. We are already working with schools, therapy centers, libraries, and museums to bring Asian OS into their daily routines. Our goal is to create industry-specific apps on top of Asian OS that's smart, helpful, and personalized. He's asking great people, give presentations, help care of the elderly, and offer companionship. They will use tools and keep learning over time. This will help us grow our market share and move closer to general AI that can handle many tasks. We plan to offer agent OS for free first to enhance our robot performance. We see strong potential for our future subscription-based business model. In the coming months, we will add agents to our existing apps, including our flagship device, and introduce new AI tools to help users work more efficiently in the LLM era. At the same time, our legacy internet business remains strong. It continues to deliver steady revenue and profit, and it gives us a natural entry point for our new AI experiences. Overall, the strength of our legacy business gives us the resources we need to push forward with our AI plans. While staying financially responsible, to wrap up Q1 2025 was a strong quarter. We grew our revenue, reduced our losses, and took steps in our AI journey. We believe agentic AI is driving the Chinese LLM industry into a new phase. Shifting from infrastructure development to application-driven innovation, this change benefits companies like us. Those with a proven track record of turning cutting-edge technologies into real-world products across the PC, mobile, and now ARS. Our ability to productize innovation is what truly sets Cheetah apart in this new phase of AI application. We remain focused on building AI especially utility-focused AI tools and robotics that not only understand people, but also help them get things done. Thank you.

Thank you, Fusheng. Hello, everyone on the call. On that data I stated, all financial figures are presented in R&D. Q1 2025 marked another quarter of meaningful loss reduction and improved efficiency. Building on the momentum from 2024, our Q1 results reflect our team as focused on disciplined execution, operational efficiency, and strategic investment in AI. Let me walk you through the key numbers. In Q1, total revenue reached $259 million, up 36% year-over-year, and 9% quarter-over-quarter. Gross profit increased by 67% year-over-year and 10% quarter-over-quarter to 190 million. Gross margin was 73.2% up from 59.2% a year ago and 72.9% in the previous quarter. Long-up gross profit was 190 million, an increase of 67% year-over-year and 10% sequentially. Lone Gap Girls margin improved 73.2%, up from 59.6% a year ago, and 72.7% in the prior quarter. We also made meaningful progress in reducing losses. Operating loss was $27 million, reduced from $81 million in the year-ago quarter and $207 million in the previous quarter, operating loss narrowed to $14 million, down from $66 million in the year-over-quarter and $42 million in the previous quarter. Net loss achievable to Chika Mobile's shareholders was $33 million, reduced from $80 million in the year-over-quarter and $367 million in the previous quarter, Non-GAAP net loss attributable to Chita Mobile's shareholders decreased to 21 million, a significant improvement from 66 million in Q1 2024 and 202 million in Q4 last year. By segment, our Internet business continues to provide solid cash flow and probability, operating margin nearly double year-over-year to 15.5%. driven by improved mitigation and a leaner cost structure. Losses from our AI and other segments narrowed to 46 million, compared to 82 million a year ago and 228 million in the previous quarter. This reflects ongoing efforts to strike the right balance between investment and efficiency. We remain focused on scalable, modernizable use cases. We also see real improvements in operational efficiency. AI-assisted coding is now part of our daily workflow, improving efficiency and helping our teams scale faster. On the robotic side, we have prioritized use cases that can be deployed at scale and address real customer needs. We also leverage open source models like VAA models. to enhance hardware performance, including the body parts. Following the arrived start position in late 2023, we have continued to consolidate teams and optimize operations. As of March 31st, 2025, our total headcount was approximately 815, down from 860 to a year ago. despite continued cost and expense control. We also launched new products and made our service robots agentic. Looking ahead, we expect further margin expansion and continued loss reduction. At the same time, we will continue to invest in AI, but in a disciplined and focused way. Our balance sheet remains strong. as of March 31st, 2025. We have cash and cash equivalents of approximately $234 million U.S. dollars. Long-term investments of about $112 million U.S. dollars. Looking ahead, our goal is clear. Rich people while maintaining a healthy cash position. We will continue to invest in AI, but in a disciplined and focused way. Ensuring every dollar spent supports sustainable, long-term value creation. Thank you. We are now happy to take your questions.

We will now begin the question and answer session. To ask a question, you may press star then 1 on your touchtone phone. If you are using a speakerphone, please pick up your handset before pressing the keys. If at any time your question has been addressed and you would like to withdraw your question, please press star then two. At this time, we will pause momentarily to assemble our roster.

Everyone, for today's call, thank you. Yes, operator. Hello, everyone. For today's call, management will answer your questions in Chinese, and an AI agent will translate management's comments into English in another line. Please note the translation is for convenience purpose only. In case of any discrepancy, our management statement in Chinese works well. If you are unable to hear the Chinese translation, A transcript will be available on our IRF site within seven working days. Thank you so much. Operator, can we now take questions from analysts? Thank you so much.

The first question today comes from Thomas Chong with Jefferies. Please go ahead.

Good evening. Thank you for introducing my questions. 我们注意到你在这一次财报中提到AI的两个方向, 一方面是工具型AI产品, 另外一方面是服务机器人的方向, 从战略资源投入还有未来三年营收工程的角度来看, 预报未来的发展中心会将偏向于打造AI工具矩阵, 制造西方的AI的效率的提升, 还是会将更多的资源投入到机器人中, 你如何承认这两个方向在技术挑战商业化节奏还有长远护身盒上的差异?

谢谢。 好,我来回答这个问题啊。 就是我觉得您问的这个问题挺好的, 就是我们现在其实经过这么多年, 整个猎豹移动就是C端的AI工具软件和机器人这两个大业务吧。 as well as some commercialization efficiency. In fact, I think these two are not contradictory. Because in essence, all products are ultimately its software capabilities. In fact, Apple, for example, is a company like Apple. In fact, its software is also very strong. Of course, hardware manufacturing is also very strong. But in the end, users use software experience. So I think today's AI tools and robots are related to short-term and long-term. That is to say, today we can use this Internet business to develop at a high speed, especially AI. The programming technology of this wave of agents has matured, so we think that AI tools will develop quickly. Including some traditional software in the past, such as our Jinshan Duba and other AI modifications, it can generate new firepower. So I think the short-term task is that this year we can see the rapid development. It must be AI tools. But the robot itself is a device equipped with AI. You can understand it as a hardware entity equipped with AI tools. So in the long run, I think this robot is after all a long-term development in this direction. So you were talking about the technical challenge. I think these two are actually similar in terms of technology. It is the final productization of AI technology. Of course, the robot is more inclined to the long chain of hardware. Then the AI tool index is more, uh, this is biased against the short frequency. Yes, and then this business, the business rhythm, the commercialization rhythm, I think the AI tool index will be faster. This can be clearly reflected in the industry, and then the machine will be a long-term, not only to improve such a job, but the machine will definitely be a little bigger because he, he is involved in the hardware, involved in the entire business model, but that, uh, In terms of AI tools, we will have to see if we can really change the minds of users in some vertical areas. I think this is Hu Chenghe. But in general, the last simple sentence is that this year, in terms of AI tools, we are in an area where we can produce efficiency very quickly. So you can also see that our Internet business is doing well. Thank you.

Operator, please move to the next question.

The next question comes from Ziping Zhao with ICBC. Please go ahead.

Hello, Mr. Guan. Thank you for giving me this opportunity. I'm Zhao Zeping, a professor at T&G International. We see that the robot is now building a data factory as a key strategic investment. The goal is to accumulate data on high-quality physics in the world using model training. We also see that the robot has already accumulated a lot of scene data in the actual process. I would like to ask you to share some of your thoughts on the company's data market construction and self-improvement. In the future, will you consider providing external data or creating a 2B service? Thank you.

Okay, thank you for your question. It's a very humble question. Because today we see the robot industry, especially the service robots we talked about, or today's popular humanoid robots, are facing a lot of difficulties. A very important point is that Today, it is difficult for us to turn human labor data into robot data. This is very different from autonomous driving. Autonomous driving, the data of human driving is the data of the robot. So indeed, we see that there are many companies in the industry that go to make data factories, including some初创公司. But what I want to say is that up to today, the robot industry has turned this data factory into a real product. and commercialization, I think it's still a very early stage. In the three years that can be seen, even within five years, I don't dare to say three years, five years, because AI changes very quickly. It can't turn into a real commercialized product. So today, you said we accumulated some data, indeed. But today's data, so far, 我们还不能够认为它能够在公司的这个 产品化方面起到巨大的帮助 但是做一些前沿的探索我们的确在做 但是我觉得就是一方面 我对整个机器人行业的大的前景我很乐观 但另一方面我觉得在当下我是非常谨慎的 因为我们已经投入这个行业 投入了七年八年吧 然后研发也投入了超过 We have been working on this development for a long time. I think that from a technical paper to some technical direction, to the final productization, there is still a long way to go. And there will be changes in various industry models, including the opening of a company, and so on. In short, for Alibaba, today's data factory issue Or the robot data factory is not our focus. We will keep this up-to-date, but we will not invest blindly. And then in the future, will we provide data or 2B services to the outside world? I don't have this idea at the moment. Because at the moment, I think it's really practical. I'm also in Guigui recently. I also talked to many people in Guigui, including startup companies in this area. In fact, everyone is... I don't know. Operator, please move to the next question.

Thank you.

The next question comes from Yanlong Chen with City Securities. Please go ahead.

Thank you very much to the management of the company for giving me the opportunity. I am Chen Yanlong, an engineer at Central Securities. I just heard that the company mentioned that we are using some open-source VLA models to push these smart glasses. We know that artificial intelligence and large-scale model-related technologies have developed very quickly and have changed a lot. Now there are many open-source models that can be used. The first question I would like to ask is, how does the company balance the use of open source models and our own technical path in actual deployment, especially in terms of efficiency, safety, operability, and cost? How do we consider and do resource configuration? The second question is, I would like to ask, Thank you for your professional question. This is a very professional question.

我觉得今天就是应该说绝大部分公司就在你的第一个问题比如如何权衡模型的开源还自研路径上已经非常清楚了 就是作为绝大部分公司是不会区分到底开源还自研了 就开源如果更好用肯定用开源 对吧 因为开源你做私有化部署和自研是没有区别的 而且能够大量的节约你的自研经费 没有必要重复造轮子 Right? Today, even Tencent's Yuanbao uses DeepSeq. Baidu uses DeepSeq, right? So I think today, no company is so persistent that it must be self-reliant. Maybe companies like Google and OpenAI will. Maybe companies like DeepSeq will. But I think for a company like us, it must be using this. And the big principle is to use open source. If there is open source, we will not self-reliant. There's no need to build wheels again. The power of open source communities. In the past two or three years, I've been saying this repeatedly on my own short video shows. Open source in the AI industry is very powerful. It's hard for a company to fight against so many geeks in the world. We've thought about this very clearly. We started thinking about it a long time ago. Including what we're doing today, including Aging OS, we've used a lot of open source. So this is... And then in terms of efficiency, security, reliability, cost structure, three aspects. I think today we talk a lot about this so-called VLA model, or talk about some of this model. A lot of it is just from the advanced model. But very little is everyone talking about this efficiency or scene. 我举个例子来说,比如说你今天用DeepThink,你问他个问题,他在想半天然后给你,你在电脑前面你是能够去承受的,尤其让他给你写个文章这样的事。 但你比如我们做机器人的交互,你说句话他等五秒钟再给你一个回复,那人都走了,对吧? 所以我们今天这三个维度里面,你说推理,效率,安全,口控性和成本结构,我们肯定是推理,推理,效率是第一位的。 That is, we are doing this kind of service for commercial occasions. That is, its response is low. So there are a lot of details to be done in this. Of course, safety is also very important. We must be a foundation. But I think this push efficiency is the first place. The cost structure is a little better. The token is still constantly decreasing. Today, if you go to any company that does AI, He considered the real one. If he is a company that really faces the market, he will definitely put the efficiency itself in a very, of course, combined with the scene. In our situation, the efficiency must be in the first place. So we put a lot of money on technology and resource configuration to really improve the reaction speed of our machines. You may have seen some models today. If you really run to a robot or let him use it, Yes, that delay may not be acceptable to ordinary users. And then you asked, from a medium-term point of view, do you think that Leopard should be built on model capabilities or scene data? I think this is undoubtedly built on scene data assets. Today, we can have the so-called model capability. Actually, I have a very big question. What is the model capability today? It's your ability to train a model. Actually, I said it a long time ago. Training a model capability, as long as it's in the background of China and the United States, it's not a problem. Many companies can train it. Because there are a lot of open source papers and open source models. You go buy a GPU and get it up. It's not a top problem. But companies with the ability to change the underlying logic structure of the model, right? I think this is a bit of a mystery on the earth. There are very few companies that can really optimize the underlying ability of the model. So today's industry is very obvious. This wave, everyone has become more and more clear. Most companies are on the data asset of the scene. Its advantage must be on the data asset of the scene, not on the model ability. 我认为这个我没有交流过,这也是为什么腾讯会大力推向元宝,其实内置的是DeepSeq,因为只有场景数据反过来去反补你的模型,等到时机成熟了去适应一个符合你自己场景数据模型,我觉得这是绝大部分公司的自身之道吧。 Or we look at, for example, the field of self-driving, right? Google's modeling ability must be very strong. But in fact, today's Tesla's FSD experience, right? It must have exceeded Google's self-driving. Because I think this is the Tesla of that year. It was built on its large amount of land, this end and scene. So you say, is it right? Yeah, that's probably it. So I think we have this today. The strategy is very clear, that is, we will not invest in the so-called self-sacrifice model, but really pay attention to the landing, pay attention to our more interaction with our users, and then build our own scene data. And then this data is reversed, right? And then it goes back, even if it is an open model ability. Okay, operator, please move to the next question. Thank you.

The next question is Yi Xu with Founder Securities. Please go ahead.

Thank you for the opportunity to ask questions. I am Zhu Yi from Founder Correct. I would like to ask how the company considers the commercialization path of AI tools. One is whether users will consider the subscription of SaaS products or the authorization of 2C. The second is how the company plans the specific development path from the background of gradually changing from concept verification to actual commercial value in the current AI application layer. Thank you.

Okay, thank you for your question. I think today's AI tool has a very obvious feature, which is the subscription value. Is the user willing to pay for this AI tool? Because the essence of AI, this AI tool is essentially a productivity tool, right? So we see basically this wave of commercial models, the whole world is basically subscription value, right? Whether it's OpenAI's chat GPT model or QSO, Right? This programming software mode is all subscription. And the subscription is now found to be continuously divided. That is, if you use more, you will pay more according to this higher level. It is essentially to help users improve their efficiency. Users are willing to pay for it. I think this is also the AI different from the last wave of the Internet. This wave is that its business model is very simple, clear and clear. And the user access is very high. Yes, so we also gave an example, we made a small product called DVO, which is this kind of product, right? In fact, users are now starting to ask how to pay, right? And there are already users paying. So we must consider user subscription. It's not about considering, it's about user subscription. This matter is very clear. Maybe you are familiar with us in the past, because of the globalization of tools in the past. So in fact, in the past few years, we have changed a lot of tools to subscription. including our Jinshan Duba. In fact, today, users pay for the mainstream, not the ads. In the past few years, even in Chinese software, many people still don't understand. Today's Chinese tool software, in fact, the subscription is only... Anyway, our own practice can be mainstream. In this way, it pays for the effect, but it will make us more focused on the experience of the user, rather than focusing on how to create an ad. This is why our Internet business has been growing in recent weeks. I think it's a very important basis. We turn user subscription into our core business model today. And then whether to consider promoting corporate SaaS products. We have actually been trying to promote corporate SaaS products. There are some like this. This is what we are talking about. We are talking about the whole AI today. Just to answer your question, Today, how do we plan our business path from the concept of actual commercial value transformation? I think our business path is now more clear. On the Internet business, we have like Jinshan Duba, like some of our business in Japan, etc. I won't talk about this in detail. There are also some software business. 我们就是用AI的agent的技术 让我们的整个的像金山独霸这样的软件 像我们这个一些在海外的软件能够去agent的话 从交付功能向交付结果跨越 这件事是我们今天的非常重要的事 第二件事就是说 由于我们在过去那么多年做工具 因为我认为agent和工具的整个思路是一样的 But the tool was not mature in technology back then, so we could only give the user a function list. But now that we have Agent, we can directly give the user a result. This is able to use our entire R&D system efficiency. So the second part is that we have some innovative applications on the Internet. We should have a lot of presentations recently. And these innovative applications, because we used AI tools to improve efficiency, 我们的研发成本会是过去的低很多吧 所以这个就是一个就是AI的agent 去改造传统的互联网业务 第二个就是在agent 在新的一些我们叫大爆发的时代 就是agent的大爆发的时代 我们也会做很多创新的尝试 这些创新尝试不会以重投入 什么这样的去做 而是以快速的 like a startup company. And then in terms of robots, we focus on our agent OS, which is all the differentiation between us and our competitors. We don't do humanoid, and we don't do robots with too complex mechanical structures, such as grass. These are not what we are best at. What we are best at is robot interaction and companionship. Thank you. Operator, please move to the next question.

The next question comes from Zhang Huang with Everbright Securities. Please go ahead. Jane Huang, your line is now open.

You may ask your question. Okay, okay. I don't know what happened just now. Thank you, Vice President Han, for accepting my question. I'm Huang Zheng, an analyst at Kandahar TMT. We can see that the market, including our company, is very concerned about robot-related business, including what Vice President Han just mentioned, the importance of the actual scene. Then maybe we can ask the management of the company to take this step forward. I don't know if we can share what progress we have made recently. And then I don't know if we can share some specific cases. The second question is from an industry point of view. I don't know if Mr. Fu is concerned about the recent changes in the robot industry, including humanoid and non-compatible robots that we focus on. And how does our company investigate and deal with these changes?

Thank you. Okay, let me talk about the industry first. I think because Yes, I am not alone in China, and I have also run a lot of entrepreneurs in the United States recently. I have seen some industries. Let me talk about some of my views on the industry. We have always believed that man-made robots have a long cycle of commercialization. What I mean by commercialization is the commercialization that can really form a repeat purchase and become a productivity. It's not an exhibition or a rental business. This, including teaching, I think this market has already appeared. And the scale is OK so far. But you said you hope to go to the production line with a human heart. I think there is still a long way to go. My point of view is more than five years. Can you really achieve this kind of commercialization? Actually, I'm at the industry level. And then, on the other hand, I see that in addition to this wave of people who are very hot, there are all kinds of robots in different specialized scenarios, including start-ups. It's very obvious that this industry is changing. So what is our own progress? I think our own progress can be said to be There are several aspects. The first is that we should have a clear idea of the development of our own robot. As I said just now, when we are in the field of robots today, what we are best at is not the mechanical structure of hardware, like the fish-fighting machine like Yushu. Indeed, they are very strong. I admit that we are not the strongest in this regard, right? This is not how strong we are. What we are really good at is the interaction experience of the hardware. It is to be able to really make this, whether it is a robot from point A to point B, the movement of the wheel can be done sufficiently stable and reliable. In the past two years, customers in Japan, Korea and Europe have already been verified. It is still the real interaction ability to be able to get this good performance. For example, we have to explain, receive, introduce, and promote such a scene, right? Add some delivery scenes. I think this path, due to the value of the large-scale model, this scene can completely explode. So you let us share some examples of actual landing. That is, we are in many of these corporate exhibitions today. In many of them, like Beijing City Center, I just shot a short video yesterday, which is their commentator. Right? And we have started to land. And because of the support of the large language model, it reflects the ability of this robot to interact. And you talk to him, his ability to understand. Right? And his ability to plan his tasks. It has been greatly improved than before. Right? Today, I think he can become a relatively excellent interpreter. The time has come. And because of the large language model, he is naturally a translator. In the past, we did this very headache, like multi-language. English, Chinese, Japanese, these multi-language abilities, right? This is natural. In the past, we did not dare to open the language ability of the machine we exported, because otherwise it would go with Google's network to do a series of complex operations, so that it can really have a foreign language ability, and the effect is not good. But today we have also started to launch this multi-language interaction machine overseas. So this is, you say, 我觉得今天我们做机器人 你要真的问我们的优势是什么 我觉得我最大的优势都不是来自于 我们对技术所谓懂一些agent做过工具 或者是说做了多少年的技术积累 而是今天我们真的已经有了 上百个这样的代理商 无论是国内还是海外的 我们有这样的商业渠道 我们可以通过这样的渠道 快速的去获得用户的 . . . China China China China China China China Thank you for your answer. We have no more questions.

Operator, can we move to the next question?

The next question comes from Xiaodan Zhang with CICC. Please go ahead.

Good evening, Manager. Thank you for accepting my question. I am Zhang Xiaodan, an analyst at Zhongjin. My question is about Agent OS. I would like to ask you to share some feedback on Agent OS. 包括说这个用户的粘性,客户的满意度,以及是否已经有这个定制化的部署,或者是客户的主动询价,那另外就是也想请管理层分享一下,就是公司内部是如何评估H&OS的这个商业化的节奏的,谢谢。 对,你提的问题这个非常的这个,应该说细节,而且很关键,就是我觉得一个所谓的,

I think at least so far, we have done some user satisfaction surveys. Users generally react when they really talk to people today, when they really react to the noise in the case of many people. Most customers think it's much better than the previous generation. This data is not there yet. We may send some articles to do it later. Yes, and then customized deployment. This wave is what you see. You can say that. Just a word from Ford. You ask the customer what he wants. He won't say he wants a car, but a faster car. Before, due to the previous NLP technology, The ASR is a language-language translation, and the text becomes an LP technology. It can't satisfy the user's qualification point. So the user will feel that this thing is useless. If you buy an audio box, you will know. Just click on a song, and you can say a few more words, and he will be stupid. Today, it's like when XGP appeared, he found that he could understand such complicated words, and it triggered the development of the app. So I think that today, AgenOS, through this, multi-sensitivity, including vision, right? And then the microphone, plus even some radar. In fact, it detects your intention to go up the steps than before. I think this is able to really make the user, the whole market is opened. We already have some customized deployment and pricing today. But we won't disclose the specific details. Today we have started, first of all, from the domestic market. We started to train our agents. Then we did the authorization and training of the second-hand platform. They can make their own applications on it. We estimated the commercialized rhythm of AgilentOS by looking at the sales progress of our voice-to-speech type of robot. We think that in Q2, we think that in Q3, whether it can reach our goal is its commercialized rhythm. 我们最关心的一个点 总体来说呢 还是这个阶段就非常关键 就是在这个时候 就是我们能不能把用户的需求 跟我们的整个这个产品 高速的叫做更高效率的 让它一体化 用户真的能够像我们叫讲解员 随时可以上岗 对吧 然后这个推销员做得比人好 这点上如果能做到的话 我觉得它就是真正能够变成 Okay, operator, please move to the next question.

The next question comes from Chen Gruli with Gyoen Securities.

Please go ahead. Thank you, Thomas.

Thank you, Mr. Chen, for your question. Thank you for paying attention to the size of our cash reserves and the direction of our strategic investment. We do have more than $200 million in cash. It does provide us with a lot of strategic flexibility. In fact, in recent years, the R&D investment department in the field of artificial intelligence, including large-scale models, such as vertical AI applications, robot technology, and industrial upstream and downstream. We have been keeping a close eye and a positive evaluation. We also believe that through external cooperation or integration, we can accelerate the construction of our capabilities and build up key chains. This is also an important way to promote AI's long-term competitiveness in the field of AI. But in terms of the strategy of acquisition, our core criteria are mainly in two aspects. One is the compatibility with the strategy of R&D, and the other is the potential to improve the overall value creation of R&D. As for some of the potential target companies, we usually do systematic evaluation from the following aspects. One is the co-ordination of technology and business with our company. On the other hand, it is the strategic value that can be brought to us, as well as the balance of the entire team and the culture and culture. Finally, the fourth is a reasonable assessment of financial value. If the above aspects are completely in line with our standards, and both sides can reach a consensus on the strategic vision, we will purchase it as an important strategic image to accelerate the construction of the ability to integrate documents in the AI or robot industry. Of course, we are also flexible in this evaluation process, and we also have an open-minded attitude. Based on different milestones, different stages of development, and the deep need for cooperation, we will also use small-scale equity investments, strategic investments, or joint investments in various forms to participate in the construction of the entire ecosystem. In summary, how do we use our cash reserves? The core principle is to maximize the long-term value of equity. Thank you, Mr. Guan.

Operator, please move to the next question. The next question comes from Guangten Zhang with Sealand Securities. Please go ahead.

Okay. Thank you, Manager, for accepting my question. My name is Zhang Wanpeng from Wuhai Securities. Will the company be able to achieve the overall loss balance in the second half of 2025? 我想进一步了解一下 就是在收入的话 未来实现盈利 更多是依赖互联网业务的 恢复性增长 还是AI业务的新增动能驱动 同时我们注意到 未来过去几个季度 互联网业务增长 其实还挺好的 请问这背后的 主要驱动因素是哪些 这些因素是否具备可持续性 对于未来几个季度的话 管理层是否可以给出 一些方向性的一个判断 Thank you. Thomas, please answer this question.

Okay, thank you, Mr. Guangfeng. What are your goals, including the balance of profit and loss, growth momentum, and business prospects? Let me give you a separate answer. Regarding the company's overall balance of profit and loss in the second half of the year, in fact, achieving the energy in the second half of the year is actually a major challenge for us internally, but it will indeed face some challenges. Whether we can achieve this goal depends on our core business, especially the speed of the progress of AI business and the overall market environment. Of course, our management team will do their best. We will also update our expectations to the market based on the actual progress. Regarding the driving force of future profits and some analysis of the internet business, I think in the future, The main driving force is certainly from our AI and other businesses. The Internet business is an important basis for us. It is expected to maintain a stable growth. One of the key factors for the growth of the Internet business in recent years is what Mr. Fu just mentioned. We have actually passed through a traditional advertising model in the past few years. We have completely transformed into a user-appropriate model. and return the value of a product. After many years of working together, we have managed to increase our product value by insisting on the user and bringing about a stable growth of users, as well as stable long-term cooperation partners and customer channels. I think these driving factors are sustainable. The subsequent growth of the Internet is mainly based on whether it is possible . . . . . . . . . . . . the development of new partners or channels, or the development of new functions related to AI agents. We will also actively promote the development and landing of these new functions. Finally, regarding the investment and planning of AI, Mr. Kang has just talked about the IT business of AI. Our main task is to invest in how to make our products Thank you. Thank you. Operator, please move to the next question.

Next question comes from Yanpeng Gao with Goitai Haitong. Please go ahead.

管理层好,非常感谢给我这次提问机会。 我的问题是关于AI业务减亏和AI投资策略相关的。 . . . . . . . .

Let me tell you a little bit about this. It's like this. I don't think AI's loss and other models are just about the collection. Of course, on the one hand, we think that today there are some explorations that have been completed. For example, we think that the training of large models is not very meaningful in our company. 所以我们节约了不少的算力成本去这个不再去就是从预训练打模型 我讲预训练那个那个后面的做什么这个增强调优这些我们还在做就预训练这件事 我们训练两个模型分别是140级参数的和一个7层8的moe 我的团队也具备了整个技术链条的这种能力和理解的时候 我们停止了这方面训练因为我们认为 就是未来这个模型的提供商不会那么多 对吧 这个要以模型制生的公司会极少极少 也许DeepSync是一个 但是这个未来应该没有太多的所谓模型公司 都是应用公司 就谁能把应用做好 对吧 谁就有可能这个变成这个巨头 然后呢 反向的你有这应用以后 你可能未来再去改模型 那再说吧 就现在这个阶段就做应用 So this is one aspect. On the other hand, we have some research that is more efficient. So this is an important reason for us to complete this exhibition. Including some of the exploration projects that you mentioned have been eliminated. For example, we can also say that we did some projects on the ground. to do some large system projects. But later we felt that this was not suitable, so we decided to adjust it quickly. Yes, so we are indeed, now basically, the entire company, almost all of them, are like ROI transformation. No matter it is the Internet business, including our robot business, including the robot new products we are developing, they are all very strong ROI transformation, ROI orientation. I think today's robot... I've already mentioned a few questions. It's meaningless to talk about technical advancedness. Who can find a scenario and make the first version? Who will have a chance to win the constitution in this round of competition? It's not that you did a few technical points today. What is the advantage of the constitution? I think there is no advantage of the constitution in technical points. The advantage is your scenario. It's your users. It's your real growth spiral. Technology itself has no advantage. So we did turn to this ROI direction. But there is one thing I might be a little bit like that. We won't say that all exploratory projects have been eliminated. We still need a lot of exploratory projects. Today, the entire AI agent wave, we think it's a wave of applications that have been rewritten. There is no company that can be very clear about what will happen. So at this time, we have to put the original business, whether it is the Internet business, which has just been repeated, the Internet business, including the robot business, to be modified with AI technology. On the other hand, we still have to do exploratory business. Exploratory business, but ROI orientation is very clear. It will emphasize ROI extremely. And as I said, today, due to the agent business, the user payment subscription mode is a very clear business mode, so it is easy to measure. So this is our idea of running the whole crack today. In short, it is AI agent to modify old products to increase efficiency. And then the robot also uses this idea of AI to do this, this is called, whether it is interaction or this task planning, this ability to improve its AI ability, instead of going to the complexity of the mechanical structure. That is, the whole sense of experience of AI, this productization is added on all lines. And then we will still do 我们内部孵化的项目 Operator can we move to the next question?

The next question today comes from Xi Zhizhong with Wells Fargo. Please go ahead.

Hello. Thank you.

Yes, your question is also something that I've been thinking about a lot lately. I think it's because of the ability of big models to integrate. Today, on the other hand, if you can make an agent-based product, you won't be marginalized by the platform. In other words, how do I put it? First of all, I think today's so-called platform is built on past experiences. So at this time, the experience you really made with Agent is completely new to the user. Of course, because of the answer to your question, we won't go into the details of what AI Agent is. What are the specific differences it brings to the experience? But today we can see very clearly. I will give a few examples to make it clear. For example, in the field of search, Google has been doing it for so many years. Today, PlugCity, right? Including what you can think of as OpenAI's ChatGPT, it is essentially a new search. 对吧,都已经导致了像谷歌这样的巨头,它的有些垂液搜索流量都在下降。 对吧,然后那个在国内也非常明显,对吧,这个我还发了一期专门的视频叫做搜索引擎的黄昏。 除了这个搜索之外,对吧,还有像比如编程领域,像QSO,CURSO这样的东西,对吧,它在以前那些大的像威斯扣的这些编程工具很强大的时候,它就是个套客产品,但是它的发展极其迅速。 And the user experience is completely different. Today, for example, similar to what I know, many companies are all using Kosovo products. So what I want to say is that today, first of all, if we start from the perspective of the user, instead of the perspective of the machine, you will find that today you do aging, you bring the experience to the user, the traditional software technology is very difficult to bring. So at this time, you can create a new awareness on the user's end. This awareness has nothing to do with the past. 所以今天反而是平台应该很担心的。 就今天各种各样的agent是有可能, 我认为啊,就有些地方真的是有可能颠覆平台的。 对吧,这是第一个。 第二个呢,正是因为大模型的能力趋同也好, 或者说开源的模型能力在起来也好, 今天大模型厂商并没有一个超出的杀手锏。 也就是说今天你去做agent,你的体验, 你可以用GM9,也可以用DeepSeq, 对吧,也可以用API, You find that your experience is above all the comprehensive capabilities of the model. So the user experience is very good. The user experience is good. It's not because, for example, today you see the entire financial system, all the big model companies have to open various APIs. So you see, we gave an example called DeepVO. You can use it. It has the ability to do this meeting summary. Today, I can be very responsible to say that it is far higher than flying books. This is our... . . . This, this, this, this, this, this, this, this, this, this, this, this, this, So I just talked about it. Today, you said to do an agent. Why is it so imaginative? The first is that you can let users re-establish recognition. The second is that your underlying ability is provided by the big model. Right? Just like you use electricity today. The more stable it is, the better your electricity performance will be. In the past, when we were young, electricity came and went. At that time, you would feel that electricity was not easy to use. Right? The electric fan will be off for a while. But the more stable the electricity, the better the electricity will be sold. So the second thing is, in fact, the new product model of agent today, in fact, it is benefited by the ability to promote large models. You know, companies like Tencent, right? They also promote this large model. But I think recently Tencent is also trying to make a real agent. Right? This platform will do it. Right? Of course, the third one, you said, if the platform does all this, then the small company will not be able to live. But what I want to say is that in the era of a large number of technological changes, the speed of this platform is relatively slow. Then at this time, you can quickly get enough users through this technological change, and slowly form a growth wheel. This growth wheel is not a platform that can be replaced. It can be replaced by the platform of the previous era. Okay, this is probably the logic. You just have to say this, of course you can say, 大公司有很强的掌控力。 但是我觉得今天看到的更多的是 这种中小公司这个创新的机会。 这种创新机会你一旦能够获得 这个用户的认可, 其实你就可以构建出你的护城河, 你不太容易被边缘化。 也可以再说个例子, 当这个和A진目前还没有关系, 但是我觉得我们很快就会有关系。 像金山独霸这样的产品都已经 . .

this concludes our question and answer session i would like to turn the conference back over to management for any closing remarks okay so thank you so much for joining our conference call today so if you have other questions please just let us know. You can send us an email or just give us a call. Thank you so much.

Thank you.

The conference is now concluded. Thank you for attending today's presentation. You may now disconnect.